SOUTH AFRICAN INVENTORY OF INLAND AQUATIC ECOSYSTEMS (SAIIAE)

The National Biodiversity Assessment of 2011 found freshwater ecosystems to be highly threatened and poorly protected. However, a number of studies have shown that the National Wetland Map (NWM) Version 4 represents less than 54% of wetlands mapped at a fine scale. A more comprehensive South African Inventory of Inland Aquatic Ecosystems (SAIIAE) would greatly improve the assessment of wetland ecosystem types and their condition and conservation status, and is crucial for monitoring trends to inform decision making and planning. In preparation for the third National Biodiversity Assessment of 2018, a review was undertaken to identify possible data sources that could contribute to the SAIIAE. The objectives of the study were to (i) assess which type of information is available for developing a SAIIAE; and (ii) list and understand the availability of fine-scale wetland data for updating the NWM. A variety of data related to species occurrence and distribution, extent and type of inland wetlands and rivers, as well as datasets which describe regional settings of inland aquatic ecosystems, were found across a number of institutions. Fine-scale spatial data amounted to more than double the extent of inland wetlands mapped by remote sensing at a country-wide scale. Nearly 5 million ha of fine-scale data were collected from a diverse number of institutions, with the majority (73%) of these data mapped by Government (3 681 503 ha or 3% of South Africa). It is estimated that < 8% of the sub-quaternary catchments of South Africa had complete wetland data sets, primarily in the Gauteng, Mpumalanga and Western Cape Provinces. Accuracy assessment reports and confidence ratings were however not consistently available for the wetland datasets. Inland wetlands in the majority of South Africa (84%) therefore remain poorly represented. We recommend future steps to improve the SAIIAE, including improving the representation of inland wetland ecosystem types and focusing on accuracy assessment. Keywords : wetland inventory, National Wetland Map, river/wetland ecosystem types, inland aquatic ecosystems, National Biodiversity Assessment


LIST OF BOXES
Box 1: Selective definitions of terms related to inland aquatic ecosystems from the National Water Act (RSA 1998:8-9

GLOSSARY OF TERMS
Artificial waterbodies / artificial wetlands: include terrestrial and aquatic ecosystems which have been modified by humans to varying degrees either within or in the vicinity of aquatic ecosystems. Such datasets may include point or polygon data on the extent and distribution of for example artificial wetlands, aquaculture facilities, bridges, canals, excavations, railways, reservoirs, roads, salt works, sand mining, treatment works and weirs.
Aquatic ecosystems: 'an ecosystem that is permanently or periodically inundated by flowing or standing water, or which has soils that are permanently or periodically saturated within 0.5 m of the soil surface' (Ollis et al. 2013:94) Biodiversity: 'The diversity of genes, species and ecosystems on Earth, and the ecological and evolutionary processes that maintains this diversity' (SANBI 2016:5) Classification: refers to the categorisation of watercourses according to their resource quality objectives as per Chapter 3 of the National Water Act (RSA 1998:13).
Commission error: for the purpose of this report, terrestrial areas mapped as wetlands.
Coordinate system used for data: is the South African Albers Equal Area conical projection with the central meridian at 25°E, parallels at 24°S and 33°S and the spheroid and datum the World Geodetic System of 1984 (WGS84).
Floodout: 'Lobate/fan-shaped sediment body that radiates downstream from an intersection point of a discontinuous channel. Typically comprise sandy materials immediately downstream of the intersection point, but may terminate in swamps or marshes as fine-grained sediment accumulates downstream.' (Brierley & Fryirs 2005:113).

Inland aquatic ecosystems:
'an ecosystem that is permanently or periodically inundated by flowing or standing water, or which has soils that are permanently or periodically saturated within 0.5 m of the soil surface' (Ollis et al. 2013:1) and which occurs inland and is not estuarine or marine in nature.
Lacustrine: (non-vegetated wetlands) 'Relating to a system of inland deep-water and wetland habitats associated with lakes and reservoirs and characterized by the absence of trees, shrubs, or emergent vegetation.' (Collins 2014).
Lakes: 'are depressions in the valley bottoms which may be temporarily, seasonally or permanently inundated. Unlike pans, they are not deflationary erosional features, but instead they have, or would have had, an outlet at the downstream end of the valley (a low point); which has been variously blocked or otherwise restricted by dune deposits; terminal moraines (e.g. Lake District; U.K.), landslides or other depositional features across the valley bottom. Their shape is therefore determined by the surrounding slopes/higher ground (in contrast to the deflational processes creating the typical circular or oval depressional pan shapes)' (DWAF 2008:20).
Limnetic: '(cf. littoral) inundated to a maximum depth of 2 m or more at the average annual low-water level of an open waterbody ' (Ollis et al. 2013:98).
Littoral: '(cf. limnetic) inundated to a maximum depth of less than 2 m at the average annual low-water level of an open waterbody ' (Ollis et al. 2013:98).
Omission error: for the purpose of this report, wetland areas not mapped in a National Wetland Map (NWM).
Palustrine: (vegetated wetlands) 'Relating to a system of inland freshwater wetlands, such as marshes, swamps, and lake shores, characterized by the presence of trees, shrubs, and emergent vegetation.' (Collins 2014).
Riparian habitat: 'includes the physical structure and associated vegetation of the areas associated with a watercourse which are commonly characterised by alluvial soils, and which are inundated or flooded to an extent and with a frequency sufficient to support vegetation of species with a composition and physical structure distinct from those of adjacent land areas' (RSA 1998:9).
Rivers: a linear inland aquatic ecosystem with clearly discernible bed and banks, which permanently or periodically carries a concentrated flow of water. A river is taken to include both the active channel and the riparian zone as a unit (Adapted from Ollis et al. 2013:100).
Strategic Water Source Areas: 'Strategic Water Source Areas (SWSAs) are defined … as areas of land that either: (a) supply a disproportionate (i.e. relatively large) volume of mean annual surface water runoff in relation to their size and so are considered nationally important; or (b) have high groundwater recharge and where the groundwater forms a nationally important resource; or (c) areas that meet both criteria (a) and (b).' (Le Maitre et al. 2017:17).

Sub-quaternary catchments (SQ4s):
A fifth level of catchment division which has been derived in South Africa from a 50 m spatial resolution Digital Elevation Model (DEM) for the National Freshwater Ecosystems Priority Areas (NFEPA) project   (Weepener et al. 2012) were used to constrain the calculation . WRC Report No. 2020. See also Sub-quaternary catchments (SQ4s).
Wash: a term used in the arid and semi-arid regions which refers to the broad, gravelly, normally dry bed of an intermittent stream (derived from Bates & Jackson 1987:730, Glossary of Geology). A wash is a type of floodout.
Watercourse: 'Watercourses' includes both the channel and banks of all rivers, springs, and natural channels 'in which water flows regularly or intermittently' (RSA 1998:9).
Wetland: 'means land which is transitional between terrestrial and aquatic systems where the water table is usually at or near the surface, or the land is periodically covered with shallow water, and which land in normal circumstances supports or would support vegetation typically adapted to life in saturated soil' (RSA 1998:9).
Wetland delineation: 'the determination and marking of the boundary of a wetland… [and] … marking the outer edge of the temporary zone of wetness' (adapted from DWAF 2008:11). In this document the term is used to refer to both in field and desktop delineation.

EXECUTIVE SUMMARY
A South African Inventory of Inland Aquatic Ecosystems (SAIIAE) was established during the National Biodiversity Assessment of 2018 (NBA 2018). The SAIIAE offers a collection of data layers pertaining to ecosystem types and pressures for both rivers and inland wetlands.
The SAIIAE builds on previous efforts while also introducing improvements and several new elements. An inventory of inland aquatic ecosystems responds to a multi-stakeholder need for the planning, conservation and management of these systems, as mandated by a number of legislative Acts, including the South African National Water Act No. 36 of 1998 (NWA) and the National Environmental Management: Biodiversity Act No. 10 of 2004 (NEMBA). This report provides a full overview of historical efforts in the inventorying of inland aquatic ecosystems, as well as efforts undertaken between 2015 and 2018 in the update of data layers associated with river and inland wetland ecosystem types for the NBA 2018.

Key highlights
 A South African Inventory of Inland Aquatic Ecosystems (SAIIAE) has been established as a collection of data layers that represent the extent of river and inland wetland ecosystem types as well as pressures on these systems;  The extent of inland wetland ecosystems has been increased by 123% compared to the NFEPA wetlands;  Eight unique freshwater lakes or limnetic wetlands have been identified; and  A confidence map of inland wetlands guides users to make appropriate use of the National Wetland Map (NWM).

Key findings
 Inland aquatic ecosystems are represented by a river lines dataset as well as polygons of river and inland wetland types in NWM5. Both datasets should be used to represent inland aquatic ecosystems;  NWM5 represents nearly 4 million hectares (ha) of aquatic ecosystems which cover 3.3% of the surface area of South Africa. These include: o Inland wetlands which constitutes > 2.6 million ha or 2.2% of the surface area of South Africa; o > 1 million ha of river channels; and o 201 381 ha of estuarine ecosystems (< 0.2% of the surface area of South Africa);  The rivers dataset represent 200 955 km of river length of which 164 018 km (82%) is situated within South Africa. The majority of the rivers in South Africa (90%) has a river ecosystem types assigned of which 10% are ephemeral and episodic systems within the arid Northern Cape Province and less than 0.1% coincide within an estuary. Mainstream rivers constitute 76 830 km (47%) of the total length of the South African rivers, and tributaries 87 188 km (53%);  Eight freshwater lakes (limnetic wetlands) have been identified where the depth of the water-level at low tide is > 2 m. These systems constitute 13 376 ha of which Lake Sibayi is the largest; and  Artificial wetlands have been mapped as a separate layer, totalling almost 600 000 ha. Knowledge gaps  Addressing uncertainties pertaining to the conceptual ecosystem types implemented and species biodiversity observed; and  Inland aquatic ecosystems in the arid region are poorly understood. In the rivers data layer these are typed as ephemeral and episodic rivers. In the NWM a high uncertainty of the hydrogeomorphic (HGM) unit is associated with these systems. Further work should be done to better define and distinguish water courses, floodouts and washes in these regions.

Key Messages
 Baseline data related to inland aquatic ecosystems are crucial for the planning, conservation and management of inland aquatic ecosystems. Currently the baseline datasets provide a poor representation of inland aquatic ecosystem types, as well as their pressures and impacts. The inland wetlands, for example, showed a 69% low confidence for representing the extent of ecosystems, with an estimated 50% omission error. Confidence and accuracies of other data layers, such as rivers and artificial wetlands, are deficient.

Priority actions
Institutional collaboration across all organisations and stakeholders for the improved understanding, mapping, conservation, monitoring and management of inland aquatic ecosystems should be established and sustained. Responsibilities related to ecosystem datasets should be listed by relevant data custodians under the South African Spatial Data Infrastructure (SASDI) Act No. 54 of 2003.
 The research priorities are to: o Improve understanding of the relationship between the ecosystem types and species biodiversity; o Improve understanding and classification of watercourses, particularly in arid systems; o Broaden regional representation of Level 2 of the Classification Systems that should be informed by analysis of relevé data from the National Wetland Vegetation Database; o Improve the extent of river and inland wetland ecosystem type; o Improve national modelling and monitoring of inland wetlands across long-term hydrological regime cycles. This will improve our understanding of ecosystem types and their functions. Currently Geographical Information Systems (GISs) are used for once-off  (2): [184][185][186][187][188][189][190][191][192][193][194][195][196][197][198][199].
This chapter provides an overview of the original mapping of hydrological features and aquatic ecosystems in South Africa since the 1940s, and subsequent development of directories of aquatic ecosystems since the late 1980s. The National Freshwater Ecosystem Priority Areas (NFEPA) project of 2011 marked the first milestone where both rivers and wetlands were typed into ecosystem types at a country-wide scale. This supported the conservation planning outputs of NFEPA, and the National Biodiversity Assessment (NBA 2011) in 2011. The first South African Inventory of Inland Aquatic Ecosystems (SAIIAE) now builds on these previous efforts and expands the directory to a wider representation of watercourses, and additional information for the purpose of inventorying, assessment and planning. An update of the Classification System and terminology is given, as well as unique ecosystem types identified for wetlands.

The history of the development of a directory and map of freshwater ecosystems
The South African National Water Act (NWA), Act No. 36 of 1998(RSA 1998, provides governance to ensure equitable rights, use and protection of water resources in South Africa. The Department of Water and Sanitation (DWS) is the primarily implementing and governing agency of the NWA. The condition of the aquatic ecosystems, in terms of quantity and quality, is addressed by DWS through the classification of water resources in which resource quality objectives are defined. The NWA includes all watercourses, surface water, estuaries and aquifers as water resources governed by the Act. 'Watercourses' includes both the channel and bands of all rivers, springs, natural channels 'in which water flows regularly or intermittently' (RSA 1998:9), wetlands, lakes or dams. Definitions are provided for aquifers, riparian habitats and wetlands (Box 1).
In addition to the NWA and the National Environmental Management: Biodiversity Act (NEMBA), Act 10 of 2004 (RSA 2004) provides for the management, planning, monitoring and conservation of the biodiversity associated with water resources, whether ecosystems or species. The South African National Biodiversity Institute (SANBI) is the implementation agency responsible for monitoring and reporting on the status of biodiversity in South Africa.
In response to both the NWA and NEMBA, an inventory of all water resources, including information related to the quantity, quality, use, biodiversity and protection of water resources, is required. International definitions of the term 'inventory', especially as related to water resources have been primarily defined in the ecological domain, and include multiple components required for supporting monitoring, assessment and planning (Finlayson & Spiers 1999;Finlayson et al. 1999). A 'directory', in contrast, is considered merely a list of, for example, wetland ecosystems and their coordinates. Components of an inventory may therefore be more than just a list or spatial map of the water resources and would include information on (compiled from Finlayson & Spiers 1999):  Supra-habitat / System (e.g. estuarine, lacustrine, marine, fresh)  Habitat (e.g. saltmarsh, peat, mangrove)  Floral/faunal groups (orders or taxa; migration and nesting behaviour or sites)  Climate  Impacts (land use, invasives, pollution)  Function  Hydrology (e.g. hydroperiod)  Biodiversity value(s)  Cultural value(s).
Owing to the multitude of components in an inventory, it may well be that the components are stored in multiple databases in a variety of formats. Aspects of water resources and related biodiversity are governed by more than one Act and government departments in South Africa. It follows that it will require a diversity of institutions to compile and coordinate a thorough inventory of water resources.  Melly et al. 2016;), but the spatial extent of river channels is inadequate for finescale planning, while features such as aquifers and riparian habitats are poorly represented. Further work was required to improve these base layers from how they are served to the public by the above institutions, in order to use rivers in hydrological modelling and to develop river and wetlands data into ecosystem types for response to the NEMBA.
Directories of the diversity of aquatic ecosystems in South Africa date back to the early 1970s and 1980s (Noble & Hemens 1978;O'Keeffe 1986). Subsequently, the Department of Environmental Affairs and Tourism (DEAT) compiled directories in the late 1990s (Cowan & Van Riet 1998). Since 2004, SANBI has been involved in the update of the National Wetland Map (NWM) and the rehabilitation of wetlands through the Working for Wetlands Programme of DEAT. The use of spatial data and Geographical Information Systems (GISs) has become prevalent since the mid 1990s, supporting improved spatial representation and analysis of water resources and their associated biodiversity. This allowed river ecosystem types to be included in South Africa's first National Spatial Biodiversity Assessment of 2004. River types were modelled using a combination of attributes from DRDLR:NGI and DWAF (Nel et al. 2004). The former DWAF and current Department of Water and Sanitation (DWS) has been supplementing the major river systems (1:500 000 scale rivers dataset) with information on the condition of the water resource. This partially fulfils the needs of an inventory for the NWA and enables the assessment of river ecosystems in an NBA for the NEMBA.
More recently, the Water Research Commission (WRC) funded a project to investigate the development of a draft National Wetland Classification System (NWCS) following previous work done internationally and locally (Ewart-Smith et al. 2006). The NWCS was subsequently adopted by SANBI as a policy document (SANBI 2009). The typing of wetland ecosystems for the purpose of conservation planning of freshwater ecosystems in South Africa was applied in the National Freshwater Ecosystems Priority Areas (NFEPA) project . The NFEPA wetland types, as well as the river ecosystem types, were used for the first time in the assessment of the threat status and protection levels of freshwater ecosystems at a national scale in the NBA 2011. During the same period of time, the NFEPA wetlands were also used for classification in Reserve Determination, a requirement of the NWA which, until that time, had not been possible to achieve spatially. Although neither of the datasets constituted a complete inventory of water resources or aquatic ecosystems, the effort has drawn much attention to the challenges associated with completeness and accuracy of representation of these datasets in response to the NWA and NEMBA. The momentum gained after 70 years of mapping hydrological features, and more than 30 years of attempting to classify ecosystems, with ecosystem typing finally being enabled through GIS, has resulted in a growing number of users and professionals interested in the improvement of water resources in South Africa. The NFEPA project, therefore, serves as an enabling bench-mark. From this grew a robust inventory that can be used in planning and assessment.

The South African Classification System of Inland Aquatic Ecosystems
In a recent update of the NWCS, under the new name Classification System of wetlands and other aquatic ecosystems (hereafter referred to as the Classification System), an 'aquatic ecosystem' is defined as 'an ecosystem that is permanently or periodically inundated by flowing or standing water; or which has soils that are permanently or periodically saturated within 0.5 m of the soil surface' (Ollis et al. 2013:1). Aquatic ecosystems include 'rivers; lakes, ponds, dams and other waterbodies; estuaries; and (shallow) marine systems' (Ollis et al. 2013:1). The Classification System divides aquatic ecosystems into three broad systems, being Inland, Estuarine and Marine Systems and adopts a tiered or hierarchical structure, with Inland Systems being divided in several ways from the upper to the lower levels:  the ecosystem context at Level 2 (regional setting) and Level 3 (landscape units);  the functional unit at Level 4 (hydrogeomorphic unit) and Level 5 (hydrological regime); and  descriptive characteristics at Level 6.
Inland Systems are further subdivided into three broad types, including rivers, open waterbodies and inland wetlands. This is in keeping with the South African NWA, which distinguishes rivers from inland wetlands and other watercourses. In the Classification System, these three broad categories are collectively referred to as 'inland aquatic ecosystems' which corresponds to the broader sense of the term 'wetland' as it is applied by the Ramsar Convention. However, the South African NWA differentiates between rivers, wetlands and other watercourses.

Developments during the course of the NBA 2018
The Classification System was adopted for the update of the NWM in preparation for the freshwater component of the NBA 2018. During the course of the NBA 2018, a number of Ecosystem Classification Committees (ECCs) were established, including those for Rivers (RECC) and Wetlands (WECC). A National Ecosystem Classification Committee (NECC) was established to coordinate alignment between the Terrestrial, Marine, Estuarine and Freshwater ecosystems as components of the NBA 2018. Discussions during these meetings, as well as those at the reference committee workshop held between 2 and 4 October 2017 for the freshwater ecosystems (see Appendix A), resulted in a number of key decisions and changes for the freshwater ecosystems. These decisions include: (i) The alignment between ecosystem types for the NBA 2018 necessitates clear boundaries between the four major ecosystems (Figure 1.1). This resulted in a reconsideration of the representation of freshwater ecosystems within the vegetation map. It was concluded that the vegetation map be used to represent terrestrial ecosystems, with the exception of only a select set of estuarine and inland aquatic ecosystems. Changes to the representation of freshwater ecosystems in the vegetation map and amalgamation of the ecosystem map will be documented in a report by the vegetation map team at SANBI and the National Ecosystem Classification System Report (Dayaram et al. 2017). The mapping of coastal ecosystems during the first semester of 2018 resulted in further refinement of the boundaries between inland wetlands, coastal ecosystems and estuarine systems.

(ii)
Using the term 'Inland Aquatic' to align with the Classification System. It is recognised that inland aquatic ecosystems include other systems which are not technically 'freshwater', namely brackish and saline systems. To avoid confusion and ensure alignment with the Classification System, it was agreed to formally adopt the term 'Inland Aquatic' as opposed to 'Freshwater' for the inventory and assessment reports of the NBA 2018. (iii) The consideration and inclusion of coastal ecosystem for the NBA 2018 was requested by the leads of the estuarine and marine ecosystem components. Noble & Hemens (1978) provides criteria and examples of coastal aquatic ecosystems ( Attributes of these systems were updated in the field related to the hydrological regime or Level 5 of the Classification System in NWM5. No specification is given in the Classification System of the minimum extent of this depth or duration of flooding in a cycle of the hydroperiod of the system. Further studies would be required to refine the definition and evaluate the suitability of the listed candidate wetlands. Although Lake Futululu and Teza (KwaZulu-Natal) were considered for the list, Lake Futululu has reduced in size and depth to such a degree that it is considered shallower (Grenfell et al. 2010;Grenfell pers. comm.) and insufficient evidence exist for Lake Teza (Scott & Steenkamp 1996) for it to be classified as limnetic. Both Lake Futululu and Teza may also be coastal ecosystems.

(v)
Rivers and inland wetlands were the only two sub-components of inland aquatic ecosystems. Inland Aquatic Ecosystems with open water features were considered to be either Estuarine Systems (estuaries or lagoons) or Inland Systems (artificial or depressions, whether littoral or limnetic). Therefore, the subcategory 'open waterbodies' of the Classification System was omitted from the inventory and assessment of inland aquatic ecosystems.

(vi)
Updates to the Estuarine Functional Zones (EFZ) as a result of new Light Detection and Radar (LiDAR) data resulted in adjustments to the NWM. In addition, micro-estuaries were also added to the polygon dataset of the estuaries. Alignment between the inland wetlands and estuaries was ensured through erasing all the estuaries from the inland wetlands and then merging it back into the NWM5. Updates to the names of the estuaries were also transferred to the river systems ( Figure 1.

1). (vii)
Level 2 and 4A should be used to describe and assess wetland ecosystem (bio)diversity. Levels 2 and 4A of the classification system effectively describe the biodiversity of wetland ecosystem types. Level 2 describe the broad biodiversity setting of the types, whereas level 4A includes the HGM or functional diversity of wetlands. Although the landscape unit (Level 3) does not describe (bio)diversity, it contributes to the identification of HGM units (at Level 4A).   Noble & Hemens (1978);Hill (1969);Butcher (1984); https://www.lakepedia.com/; Miller (1998); Parsons (2014); Van der Waal (1997);Whitfield et al. (2017). ***Although these are currently inland aquatic ecosystems, the recent geological uprift of the coastal plain has resulted in a cut-off between the ocean and the inland system. Relics of marine species are still found in these systems and therefore these systems are considered not only unique, but also coastal as a result of the combination of inland aquatic and marine features (Whitfield et al. 2017).

The South African Inventory of Inland Aquatic Ecosystems (SAIIAE)
Compilation of the first South African Inventory of Inland Aquatic Ecosystems (SAIIAE) was undertaken as part of the NBA 2018, with the primary intention of including those datasets related to ecosystems type and condition. It, therefore, partially addresses the aspects of these Systems required to respond to the NWA, with a stronger focus on NEMBA and the listing of ecosystems. Four broad categories of spatial data features were collected, including those related to rivers, wetlands, their condition and others (Table 1.2). The latter consisted of data which could be used for assessing the protection levels of the sub-ecosystems, as well as features related to tourism, cultural value and ecological infrastructure. Peatland points (Grundling et al. 2017) Possible locations of oxbow rivers for the identification of floodplain systems (Oxbows) (results of this report)

Condition (pressures) Other information
Land cover  Data collection was undertaken firstly through a data audit and survey of the available data sources for the SAIIAE, with more than 500 Interested and Affected Parties (I&APs) contacted . The results of the audit showed that the records were mostly related to wetland ecosystems, and few records related to species, the regional context within which inland aquatic ecosystems occur, condition or monitoring history. An initial comparison between the hydrological classes from the most recent land cover dataset (GTI 2015) and the hydrological categories mapped through heads-up desktop digitising of hydrological features, showed that the fine-scale mapping of inland wetlands resulted in a much higher percentage of mapped wetlands with an improved interpretation of the HGM units. The use of sensors with a higher spectral and spatial resolution remains to be assessed for their potential to contribute to improved mapping of wetland ecosystems. The audit paper ) revealed that 5 million ha of fine-scale wetland mapping datasets were received from a number of organisations. The DRDLR:NGI contributed the largest percentage (73%) of hydrological feature mapped at a fine scale, although not done by wetland specialists. Less than 8% of the country's sub-quaternary catchments (SQ4s) have been mapped by wetland specialists or typed to the HGM units of the NWCS or Classification System. Recommendations were made with regards to collaboration between DRDLR:NGI, DWS and SANBI for an improved, coordinated effort of fine-scale mapping of wetlands, as well as attending to the collation and curation of other datasets related to inland aquatic ecosystems.
Secondly, and in parallel to the data audit, improvement of the river and wetland sub-ecosystems has been undertaken. More attention was given in this inventory update to inland wetlands than rivers in recognition of the extensive underrepresentation (omission errors) and overrepresentation (commission errors) of inland wetland ecosystems.
The CSIR has also reconsidered the feasibility of mapping the landscape units, or level 3 of the Classification System, at a country-wide scale. Since NWM5 was undertaken at a fine-scale (see Chapter 4) and not modelled, a dataset reflecting landscape units was not considered an immediate priority. It remains a research topic of interest for the inland aquatic ecosystems, though, and effort has therefore been spent on assessing whether morphometrics of a Digital Elevation Model (DEM) could be used to model landscape units. At the time of this publication, the CSIR has drafted their report on the intermediate findings (Van Deventer 2018).
The chapters of this report detail more information on the following themes of the SAIIAE: The first South African Inventory of Inland Aquatic Ecosystems (SAIIAE) built on previous efforts for the establishment of a directory to represent watercourses and additional information for the purpose of inventorying, assessment and planning.    This chapter provides an overview of the origin, development and improvement of the 1:500 000 spatial dataset of the rivers of South Africa. For the purposes of the National Biodiversity Assessment (NBA) 2018, a rivers network Geographic Information Systems (GIS) layer is required in order to map and classify the river ecosystem types across the country. River ecosystem types represent the diversity of river ecosystems. These subtypes are components of rivers with similar physical features and serve as essential coarse-filter biodiversity surrogate . A river ecosystem is viewed as a system operating within its surrounding surface and sub-surface catchment environment, and includes biotic interactions amongst animals, plants and micro-organisms, as well as abiotic chemical and physical interactions (modified from Angelier 2003). This chapter begins with a summarized history of the 1:500 000 river coverage. It then outlines the updates of this dataset relevant specifically to the NBA 2018 and ends by describing possible future updates to the rivers dataset.

History of the 1:500 000 rivers dataset
In this section, key information is drawn from the former national Department of Water Affairs and Forestry (DWAF 2006). In the early 1990s the Department of Land Affairs: Chief Directorate of Surveys and Mapping (DLA:CDSM) created the original 1:500 000 river network coverage by scanning and vectorising the blue plates of their published 1:500 000 map sheets. In 1995, the Institute for Water Quality Studies (IWQS, now Resource Quality Information Services, RQIS) of the DWAF, saw the need for an accurate regional rivers dataset to serve as a background vector coverage for cleaning up point records of national DWAF monitoring sites. IWQS also had plans for developing a water quality monitoring network, for which a hydrologically-correct river network was required. IWQS was appointed the consulting group GisLAB to produce a contiguous and consistent arc dataset from the DLA:CDSM (now the Department of Rural Development and Land Reform: National Geo-spatial Information or DRDLR:NGI) raw data. GisLAB filled gaps in the network, aligned arcs downstream, ensured that all existing surface monitoring points were on a river arc and ran the first Strahler stream ordering process on the coverage. IWQS allocated reach codes to the river arcs using the quaternary drainage region polygons from the 1990 Surface Water Resources of South Africa Study (WR90) database (Midgley et al. 1994). A reach code is a unique identifier for each reach in the river network, based on the approach applied in the United States Geological Survey National Hydrography Dataset (https://nhd.usgs.gov/). Starting in the late 1990s, the river arcs were adjusted to match the 1:50 000 data, where available, to within approximately 50 m.
The river coverage extends to the limits of international basins (e.g. Namibia and Zimbabwe), at the lower accuracy of the available base data, i.e. at a scale of 1:1 000 000. The fundamental assumption behind this simplified GIS river network is that water flows downhill along single channels. Multiple-thread channels, impounded flow and exchange with the atmosphere or soil are ignored. An important component of the river coverage is the set of attributes attached to each arc which provide useful information for basin network analysis. These include a name, where available, and various network descriptors such as flow variability and Strahler order. In time, this dataset has found widespread practical application in, for example, river management and conservation, hydrological modelling and river profile generation. The latest version of the 1:500 000 river coverage for South Africa is available from the Department of Water and Sanitation (DWS) and can be downloaded from the following website: http://www.dwaf.gov.za/iwqs/gis_data/river/rivs500k.aspx. It is now almost stable in terms of representation and attributes following years of editing procedures, consistency checks, network and name verifications and a consolidation process. The Digital Elevation Model (DEM) determination process of Weepener et al. (2011) also provided several corrections.
Occasionally, GIS users do report errors and inconsistencies that are related to some shortcomings in the dataset. For example, the reach codes within a quaternary drainage region changed unpredictably when new rivers were added to the region, depending on their position in the Strahler hierarchy. This no longer occurs because the DWS shut down the system on which the editing took place and therefore development of the rivers dataset on the RQIS website is frozen. This no longer occurs because the DWS shut down the system on which the editing took place and therefore development of the rivers dataset on the RQIS website is suspended. Ideally, the reach codes should remain attached to the same reach except in the case of splits and additions. Recursive errors could be related to the DWAF monitoring point locations. Monitoring points incorrectly positioned would have influenced the retention or deletion of river arcs. It is also essential to check arcs for errors related to river names although most of the unnamed river arcs do not have names on the original 1:50 000 maps. In addition, at certain places the river arcs still do not spatially match the 1:50 000 rivers. However, the greatest discrepancy is in the flow variability, i.e. the classification of rivers as perennial, non-perennial or dry from the 1:50 000 river coverages, which DLA:CDSM derived from aerial photographs. These data are less accurate because they depended to a large extent on the season during which each aerial photograph was taken and whether the river channel had water in it at the time of mapping ).

Update of the 1:500 000 rivers dataset for the NBA 2018
In the NBA 2018 the National Freshwater Ecosystem Priority Area (NFEPA) rivers GIS layer was used to represent the diversity of rivers nationally (Table 2.1 and Figure 2.1). This layer was also used in the freshwater component of the NBA 2011 assessment. Only minor river arc edits (e.g. deleting duplicate vertices) were performed. This updated rivers dataset represent 200 955 km of river length of which 164 018 km (82%) is situated within South Africa. Mainstems make up 76 830 km (47%) of the total length of the South African rivers, and tributaries 87 188 km (53%).
When comparing the NBA 2011 and NBA 2018 assessments, total river length within South Africa differs only slightly at a decrease in 242 km or 0.1% (Table 2.1). The NFEPA rivers GIS layer was compiled from the DWA 1:500 000 river network coverage (  Table 2.1) from the 1:50 000 rivers GIS layer (DLA:CDSM 2006). These smaller streams were added to the DWA rivers dataset so that all rivers associated with estuaries could be included in the NFEPA analyses . They were connected to estuaries mapped for the NBA 2011 (Van Niekerk & Turpie 2012). The NFEPA rivers GIS layer was coded to distinguish quaternary catchment (Midgley et al. 1994) mainstems and tributaries (Table 2.1), river segments and river reaches. Mainstems typically pass through a quaternary catchment into a neighbouring quaternary catchment while tributaries nest within a single quaternary catchment . The quaternary catchment is used as the basic unit for national water resource management (mean size 650 km 2 ), and each contains hydrological data (gauged or simulated) that is updated occasionally.  A river segment represents the portion of river joining two 1:500 000 river confluences. Sub-quaternary catchments or quinary catchments within a mean size of 135 km 2 were delineated around river segments. They are nested within quaternary catchments and were used by NFEPA as the units of selection or planning units for identifying priority areas . Quinary catchments were refined by . River reaches may be several metres or kilometres in length. They can be made up of a number of river segments and signify the whole river sub-system. In the RQIS river dataset, a reach is the piece of river between two river confluences or nodes. Other attributes attached to the NFEPA rivers GIS layer, for example, included the Level 1 ecoregion (Kleynhans et al. 2005), geomorphic zone (Rowntree & Wadeson 1999;Rowntree et al. 2000;Moolman 2008), river ecosystem type and river condition . The NFEPA rivers were also shifted from the original Cape datum to the World Geodetic System 1984 (WGS84) datum.

Recommendations for updating the 1:500 000 rivers dataset
Recommendations to improve the 1:500 000 river coverage include working towards stable reach codes and a comprehensive hydrological dataset for example, by looking at variables such as channel elevation, stream velocity and connectivity. This would require migrating the rivers dataset from an ArcInfo coverage to an ArcGIS geodatabase with topology recorded within the rivers network, as well as between the rivers network and other datasets. High resolution digital elevation models should be considered for improving drainage network generation and the accuracy of the rivers network.
Future developments ought to also include applying the updated 1:500 000 verification and reach allocation procedures to the 1:50 000 rivers (DWAF 2006;. This would necessitate substantially more computing time since the 1:50 000 river coverages are more accurate and comprehensive, being about 100 times larger than the 1:500 000 river coverage. A finer river network GIS layer for example biodiversity planning or land-use planning may be preferable, as this would indicate smaller streams and habitats that need to be managed and conserved. The 1:50 000 rivers have undergone extensive editing (e.g. flow direction) in the early 2000s in a different process managed by the DWAF and are now hydrologically-corrected as upstream and downstream linkages can now be defined for each river reach. They were vectorised from the 1:50 000 topographical map series in the late 1990s by the DLA:CDSM (DWAF 2006), now DRDLR:NGI.
Other specific recommendations from members of the River Ecosystem Classification Committee (RECC) and Reference Committee for inland aquatic ecosystems, pertaining specifically to future NBA assessments of river ecosystems, include:  Improvement of the spatial accuracy of the rivers layer used for assessments. There is a need for using a higher resolution rivers network, a new national hydrologically-corrected rivers dataset derived from the 1 arc-second (30 m) Shuttle Radar Topography Mission (SRTM) data, or better resolution if available. This include addressing the mismatch of various 1:500 000 rivers with that of the 1:50 000 rivers;  Transferring attributes such as river ecosystem type and condition from the 1:500 000 rivers network to the 1:50 000 river coverages. However, it was noted that the 1:50 000 rivers also have errors and that employing them in national scale assessments could be problematic;  Obtaining higher confidence in the flow variability attributes attached to the 1:500 000 rivers network. In addition, the river flow categories should be revised and more categories must be added, for example those recommended by Uys & O'Keeffe (1997); and  Exploring new approaches to the classification of geomorphic zones developed for non-perennial river systems, which cover a large area of southern Africa. Jaeger et al. (2017), for example, recognise a 'floodout' zone in non-perennial river systems, wherein channel breakdown through transmission losses results in environments of net deposition with diverse and sometimes distinctive geomorphic features, sedimentary deposits and ecosystem characteristics. This chapter provides an overview of the origins of the National Wetland Map (NWM) from versions 1 to 3, as well as the wetlands that were compiled for the National Freshwater Ecosystems Priority Areas (NFEPA) project.

Initiatives towards directories of wetlands
During the 1970s the South African government initiated a number of research projects to investigate the diversity of inland aquatic ecosystems of the country. This resulted in two key reports on the diversity of inland aquatic ecosystems (Noble & Hemens 1978;O'Keeffe 1986). The report offers a directory list and map of the location of some of these inland aquatic ecosystems. The interest in inland aquatic ecosystems has sparked a number of conferences, reports and research projects around the country. For example, the CSIR conference held 15 and 16 October in 1987 on 'Ecology and conservation of wetlands in South Africa' (Walmsley & Botten 1987). Subsequently, the former national Department of Environmental Affairs and Tourism (DEAT) has developed a second national directory of wetlands in 1998, using Geographical Information Systems (GIS) at that time (Cowan & Van Riet 1998). Although the authors of these early reports have taken care in mapping wetland extent, type and condition with great care, the lack of readily available spatial information from this time has resulted in a slow percolation of the information into current inventories.
A period followed during which the capabilities of remote sensing and GIS were evaluated to spatially represent hydrological features including waterbodies and wetlands. This period was also characterised by the increasing use of Global Positioning System (GPS) receivers by civilians. The use of GPSs has enabled more accurate mapping of features. This resulted in the widespread use of the World Geodetic System 1984 (WGS84) spheroid and the Hartebeesthoek94 datum in South Africa since 1 January 1999 (DRDLR:NGI 2013). The formalisation of a framework for the responsibility of spatial data amongst organisations followed with the gazetting of the South Africa Spatial Data Infrastructure (SASDI) Act (No. 54 of 2003).

Origin of the National Wetland Maps 1 to 3
The first three maps of South African wetlands were produced under the National Wetland Inventory project, which moved to the South African National Biodiversity Institute (SANBI) in 2004 as part of the Working for Wetlands Programme. In the Working for Wetlands Programme, remote sensing, field and a desktop approach using GIS, had been used to map and classify wetlands into different wetland types based on available data at the time.
The first National Wetland Map (NWM1) Beta Version or Map 1 layer was derived from the National Land Cover 2000 (NLC2000) GIS layer Van den Berg et al. 2008) in which wetland polygons are described as 'Wetland' or 'Waterbody' (Figure 3.1). NLC2000 was derived from multi-season Landsat imagery primarily of the year 2000 (GTI 2008). NWM1 was released in 2006 for public use and its polygons did not distinguish between natural or artificial waterbodies but merely between waterbodies and wetland categories, as derived from NLC2000. Historical wetlands lost through impacts were therefore not represented in NWM1. Wetlands and waterbodies categories from an improved version of the NLC2000 dataset were used to derive NWM2 in March 2008. Upon request of SANBI and the Water Research Commission (WRC), both NWM2 and the National Wetland Probability Map (NWPM)  were provided to GeoTerraImage (GTI) for the assessment of accuracy (GTI 2008). The assessment considered congruency between the dataset and reference data which primarily consisted of features mapped by the former Department of Land Reform: Chief Directorate Surveys and Mapping (DLA:CDSM) for the topographical maps, as well as data from the Mpumalanga and KwaZulu-Natal provinces.
The majority of the reference datasets represented non-wetland categories and few included fine-scale mapped wetlands and were not mapped by wetland specialists. The congruency assessment found that both NWM2 and the NWPM had an overall agreement of between 30% and 90% for both the wetland and water classes (GTI 2008). A high commission error was also observed for both datasets, in relation to the reference datasets. A number of recommendations were made to improve the NWM2, including the use of a new range of multi-seasonal images and higher spatial resolution imagery. A confidence index was also produced, ranking the suitability of the Landsat imagery for each topographical map sheet, as well as the result of confidence in the mapping of wetland extent.
Subsequent to the report, NWM2 was updated using polygons of dams and lakes and the rivers network (line) data supplied by the former Department of Water Affairs and Forestry (DWAF date unknown). The resultant layer constituted of a number of GIS layers: wetlands, natural waterbodies and artificial waterbodies (Table 3.1). In NWM3, data from DLA:CDSM was included and four layers created to include the data from NWM2, more artificial waterbodies from DLA: CDSM and those in NWM2 from DWAF, as well as natural waterbodies from DLA:CDSM and remaining unclassified waterbodies.
An analysis done by the CSIR showed that all three versions had mapped nearly 2 million hectares of wetlands and other waterbody features (Table 3.2) with an increase in extent and more attribute information in the dataset(s) (Figure 3.2).

Wetlands from the NFEPA Atlas, NWM4
A number of improvements to the NWM3 were made during the NBA2011 and NFEPA projects  in which existing sub-national (datasets at provincial, district or smaller catchment scale) wetland delineations from other biodiversity planning initiatives were added to the NWM3. These included the following sub-national data of wetland delineations: The output of the improved wetlands dataset was released with the NFEPA atlas and has subsequently been termed the 'NFEPA wetlands' by many individuals and institutions. The output was however adopted as NWM4 by SANBI and has been used in a number of applications to date, such as provincial biodiversity plans and Environmental Impact Assessments.    (2) This chapter provides an oversight of the rationale and approach for the update of the National Wetland Map version 5 (NWM5) which will be used in the National Biodiversity Assessment of 2018 (NBA 2018). This chapter documents the principles adopted and methods used for the mapping of HGM units and the integration of existing datasets for various parts of the country. The results of reviews done by experts are provided as well as the results of the final NWM5. A confidence map guides users as to the completeness of the extent of inland wetlands mapped in an area, as well as the accuracy confidence of the attribution of hydrogeomorphic (HGM) units. Recommendations for improvements to the NWM are provided in conclusion.

References
Wetland 'means land which is transitional between terrestrial and aquatic systems where the water table is usually at or near the surface, or the land is periodically covered with shallow water, and which land in normal circumstances supports or would support vegetation typically adapted to life in saturated soil.' NWA 1998:9

Recognition of errors and shortcomings of the NFEPA wetlands (NWM4)
Since  Areas mapped around dams were initially thought to be natural palustrine and seep wetlands however upon closer inspection after NWM4, it was agreed to have rather classified these wetlands as artificial for the purpose of a national map; and  Slivers of inland wetland polygons which resulted from a number of overlays and editing processes.
Errors of omission and commission can both affect the assessment of the Ecosystem Threat Status and Ecosystem Protection Levels, the headline indicators of an NBA. Errors of commission, which result in the overestimation of the extent of an inland wetland ecosystem type, are particularly problematic in that they can lead to significant underestimation of threat status and over estimation of protection levels. For the purposes of the NBA 2018, it was therefore crucial to reduce commission errors. In an attempt to address the vast number of issues reported for NWM4, as well as the underrepresentation of the extent of wetland ecosystem types, an approach was followed to first use the fine-scale wetlands (i.e. high confidence) data collected for NWM5 , and then evaluate whether the NFEPA data could still add value. The following Sections (4.2 to 4.9) describe the way in which the fine-scale data was amalgamated and typed whereas Section 4.10 describes the comparison of this amalgamated dataset (NWM5) with the NFEPA wetlands / NWM4.

Approach, process flow and protocols used for NWM5
A key decision made at the onset of the improvement of NWM5 was to first split fine-scale, desktop mapped wetlands data from wetlands modelled either from a Digital Elevation Model (DEM) or satellite imagery (Appendix B). This implied that all the fine-scale wetlands datasets would be identified, amalgamated and typed to HGM units, separate from any modelled data. Only after the compilation of a fine-scale version of NWM5, would the modelled data be considered for integration.
An update of the hydrological features mapped by the DRDLR:NGI was issued as provincial geodatabases at the end of March 2016. This dataset showed a larger amount of hydrological features captured by the DRDLR:NGI, compared to the 2006 dataset which was readily available at the Council for Scientific and Industrial Research (CSIR). Although the DRDLR:NGI issues updates every three years, the time constraints on the inventory resulted in consideration of only the 2006 and 2016 datasets, which were readily available for use at a national scale (merged). As a first step in the compilation of the fine-scale NWM5 (Figure 4.2), the 2016 provincial geodatabases were merged to a national geodatabase using the GIS software ArcGIS 10. 3 (ESRI 19993 (ESRI -2014 and topological errors cleaned and referred to as NWM5.1 (Figure 4.1; Appendix C). All data were projected to the South African coordinate system used by the NBA 2018, the Albers Equal Area (AEA) Conical projection with the spheroid and datum being the World Geodetic System of 1984 (WGS84). This coordinate system least distorts the surface area extent calculated for ecosystems. This coordinate system uses the 25°E as central meridian with two standard parallels including 24°S and 33°S.
A geodatabase was chosen as the preferred format of the spatial data related to the South African Inventory of Inland Aquatic Ecosystems (SAIIAE), since it could economically and effectively contain multiple feature classes under a single feature dataset and allow for the assessment and correction of topological errors. Most particularly, the use of defined subtypes for multiple categories in all the fields, and relationships between these had the advantage that during distributed editing of the subsequent versions, consistency in classifications will persist which limits timeously editing after amalgamation. Concurrent to the period during which the NWM5.1 was compiled, the CSIR and the South African National Biodiversity Institute (SANBI) undertook a data audit of available data which could be used for the SAIIAE.
In particular, fine-scale wetlands data were sought for improving the NWM. A wide range of wetland datasets had been received during the course of 2016 until 2017 for use in NWM5. This has been compiled as an inventory list . The list was updated for this publication and is available in Appendix E.
In the second step of integration, fine-scale mapped wetlands data which are nationally available from the Working for Wetlands team at the Department of Environmental Affairs (DEA), the Water Research Commission (WRC) project (Grundling et al. 2017) and the estuaries from the NBA 2011 were integrated into NWM5.2 (Figure 4.2; Appendix E). Hereafter the data was clipped to the provincial boundaries of the Municipal Demarcation Board's 2011 dataset and distributed to a number of assistants for further processing. The available fine-scale datasets for the provinces were then evaluated on a fit-for-purpose basis, and a selected number of these chosen for integration (Appendix E).
Fine-scale wetlands data received for the purpose of updating NWM5 ; Appendix E) were incorporated with sub-version NWM5.2 for each province into sub-version NWM5.3. The naming convention of the output feature class was given as [province_name_NWM5.3_coordinate system].
For example, Gauteng would be GT_NWM53_AEA. Firstly, the existing HGM units allocated to fine-scale wetlands map data from the fine-scale data received, were transferred to the subtype fields of NWM5.3. In addition, a field Condition was added and where information on the condition was available this was also transferred.
Secondly, topology was calculated for the feature data class to show where duplicate polygons occurred, as a result of the merging. This allowed the assistant to review these multiple polygons and make a choice of extent and HGM unit. The error inspector tool in ArcGIS makes provision for the topological overlap error to be dealt with through either merging, subtracting or deleting selective polygons. A choice can then be made to select the polygon with the preferred HGM unit and name of the wetland (from NWM5.2 or higher confidence fine-scale wetlands data).
Principles and guidelines were compiled and provided to assistants for the data capturing and integration process (Appendix B & D). Three key principles guided the compilation of NWM5 to be used in a biodiversity assessment:  The original extent of the wetland should always be captured for the purpose of an NBA. The original extent of wetlands is required for the assessment of ecosystem types in the NBA 2018 (SANBI 2017). Even though this project had no funding to capture wetlands from historical imagery, the principle implied that all artificial wetlands and modifications to wetlands were removed, and gaps within large valley bottom systems were filled up. This will be elaborated on in the subsections to follow.
o For planning purposes, however, the current extent of wetlands is relevant and should again be combined with the artificial wetlands and other modifications.
 The maximum extent of a wetland should always be captured. Ideally, multi-temporal imagery across large annual cycles and seasons should be used to determine the maximum inundated extent of a wetland across hydrological cycles.
 The HGM units and condition assigned in fine-scale wetlands datasets received from wetland experts should always be retained. Where HGM units and condition of wetlands has been assigned by wetland specialists, even if desktop assessed, retain these and do not remodel these wetlands. The time-period within which the condition was determined should be documented.
In addition to the guidelines listed in Appendix F, assistants also had to export all artificial wetlands to a separate layer, and then remove artificial wetlands from the NWM layers, within the focus areas as well as later in the remainder of the provinces (following subsections). Although it is internationally recognised that artificial wetlands may contribute to biodiversity, for the purpose of assessing the headline indicators of the NBA 2018, these systems were not considered and have therefore been kept in a separate layer. Artificial wetlands were dealt with in three ways ( Isolated dams were deleted from the NWM5 (Figure 4. 3.A.1-3). In focus areas the extent of the smaller HGM unit, which would be present where a series of small dams occur, had to be captured, while in the remainder of the provinces, these smaller systems were not captured. The following subsections will detail more of steps 4-7 (Figure 4.2) taken at a provincial level for the mapping and integration of the fine-scale version of NWM5. Concurrently to the process of mapping and integrating the fine-scale wetlands data, a wetland probability map was developed by Dr Nacelle Collins from the Free State Department of Economic, Small Business Development, Tourism and Environmental Affairs (FS DESTEA). The errors in NWM4, particularly the omission errors, inspired Dr Collins to improve the modelling of wetlands for the Free State Province. The output has been very impressive and therefore we have asked him whether he would be willing to run the same script for the rest of the country, which he did. The methods and outputs of this dataset are provided in Chapter 5 of this report. A comparison between the fine-scale and modelled wetlands datasets will enable more choices in the compilation of a final NWM5 for the NBA 2018, and could contribute to future fine-scale mapping of NWM updates. The comparison between fine-scale wetland maps and the wetland probability map has been done for selected areas and the results hereof are presented in Chapter 6 of this report.

Mapping of inland wetlands for districts of South Africa
Inland wetlands have been mapped at a fine scale (below 1:10 000) for nine district municipalities of South Africa during the update of the NWM5 and the NBA 2018 (Figure 4.4, see Section 4.4). These updates supplemented the inland wetlands data received for four metropolitan districts, including City of Cape Town (CoCT), City of Johannesburg (CoJ), eThekwini and Nelson Mandela Bay District Municipalities (See Chapter 1 and the inventory list for more details). The former district boundary of the City of Tshwane Metropolitan Municipality (CTMM) has also been incorporated, which constitutes the western half of the current City of Tshwane District Municipality. These districts have been mapped to various degrees of scales and accuracies, and few have aligned the inland wetlands with the estuarine ecosystems. In addition to the available datasets as well as the focus areas, efforts to update the inland wetlands for three other districts were also under way. Improvements to the inland wetlands in the Gert Sibande and Nkangala District were done by Dr Mervyn Lötter and Mr Hannes Marais of the Mpumalanga Tourism and Parks Agency (MTPA). The baseline of the NWM5.3 for the Mpumalanga Province, which included all the data received for the province, was used to improve the extent and HGM unit typing of the inland wetlands. In total, 16 of 52 districts of South Africa have been included in NWM5, which constitutes about 31% of the total number of districts.

Mapping wetlands in nine focus districts in the country
Nine District municipalities within the country were selected as focus areas based on availability of resources and the situation assessment results (   Train delegates in the practice of desktop mapping protocols and procedures as well as the principles and rules in capturing wetland data;  Make use of ESRI's ArcGIS mobile, ArcGIS Online and Quantum GIS's Qfield; and  Expose delegates to working on the wetland inventory workspace i.e. the classification system and the geodatabase. The training workshop was attended by 3 trainees from SANBI, the Freshwater Consulting Group (FCG), as well as 12 data capturers from ICLEI, the CSIR and SANBI (Table 4.2). Thereafter, assistants were expected to map examples seen in the field at a desktop level using GIS. Improvements to these attempts were then suggested by the trainers. Following the training workshop, assistants were expected to capture wetlands in the focus areas allocated to them using the guidelines documented in Appendix F. Further supervision was provided by Dr Heidi van Deventer at the CSIR and Ms Namhla Mbona at SANBI or alternatively through Skype meetings and sharing of screens. A number of datasets were provided to the assistants for guiding data capturing as well as the guidelines for mapping wetlands in South Africa, which was in draft format at the time of the update of NWM5 (Job et al. 2018). For the update of the NWM5, the original intent was to use the most recent images which had national coverage. The 50 cm spatial resolution colour orthophotography through the ArcGIS online viewer from DRDLR:NGI was freely available for the update and dated back to between 2012 and 2013. SPOT imagery was also used in some instances, dating to similar years. Unfortunately, both the colour orthophotos and SPOT images were largely taken during the dry season (possibly to avoid cloud cover) and were therefore less suitable for the purpose of inland wetland mapping. When a wetland deemed particularly difficult to discern the extent and/or HGM units, multiple years of Google Earth imagery was used to facilitate the mapping.
The data capturing process took approximately three months per focus area and was followed by a review period. Most districts were captured and reviewed between 1 September 2016 and 31 March 2017, although a few commenced and were completed earlier.

Integration of wetland datasets for the remainder of provinces
No data capturing was done in the remainder of the provinces, except for a selected number of floodplains, the eight freshwater lakes (limnetic wetlands), and wetlands within the majority of the Ramsar sites. A number of assistants who were mapping wetlands in the focus areas continued with the integration of wetlands in the remainder of the provinces, with help from a number of additional assistants (Table 4.3).
The period of integration took place between 1 April and 16 March 2018, with a reduction in the availability of staff's available hours dedicated towards this effort in the last three months of this period. Hence a number of unforeseen delays were experienced. During the mapping and integration of the NWM5.3 versions, the following errors had to be addressed:  Depression and artificial wetland attributes were accidentally mixed through an automated GIS process when data was integrated into NWM5.2. This may have been a result of limited computer processing power in combining data at a national scale and resulted in the extensive checking of these polygons across the provinces in order to ensure that these errors were eliminated. In the future, processing should be considered only at provincial level, unless computer processing power has improved.
 Polygons from the NWM3, including those modelled through remote sensing and split with the landform dataset, appeared to have crept back in to NWM5.3. This resulted from the fine-scale wetland datasets having included modelled or remote sensing data. These were not necessarily eliminated across all provinces, though some assistants did pick up this error visually and have attempted to eliminate these as far as possible.
 Merging duplicated polygons which result from combining multiple datasets. Noticeably, depressions were mapped by different projects using imagery from different dates. Where the differences were minor, the polygons were merged. Where the differences were larger, the outer part of the depression was made the seep and the inner polygon the depression HGM unit.
 Small slivers were merged with the larger polygon, after exploding all features, to ensure there are no multi-part polygons.
 River channels still posed a contentious matter. Originally, the sand banks and flood bands of the DRDLR:NGI data were merged into river channels or identified as potentially riparian. Later on, the section of the channel running through a floodplain or valley-bottom wetland were split and included in the HGM unit adjacent to the channel. Consensus was not reached amongst members of the team and reference committee on how to deal with the river channels, flood banks and sandbanks. A future update should table these for discussion to resolve a sensible way forward.
 Gaps within wetland and river channels were filled in some instances but not all. Sometimes the gaps are true islands, however in other instances these were data capturing errors resulting from mapping errors or different purposes of mapping inland wetlands and hydrological features by DRDLR:NGI. Future updates should attend to these in more detail.

Review protocol for the focus districts and remaining parts of the provinces
In the focus areas, the reviewers commented on the extent and HGM units of the draft NWM5 through generating points where errors were noted and added suggestions in the attribute field of the shapefile (Appendix F). A total amount of 849 points were received and addressed across all focus areas (Figure 4.5) by the assistants (Table 4.4). A document template was provided to all reviewers who were involved in the reviewing of the draft NWM5 for the remainder of the provinces (Appendix F). A random selection of 30 points per HGM unit was done in ArcGIS, and the resulted point shapefile, as well as corresponding polygons, were sent to the respected reviewers (Table 4.5). The Cape Winelands District was amended at desktop level by the Freshwater Consulting Group (FCG) and therefore no points are reflected in Figure 4.5 for this District.

Assessing whether NWM4 should be added to the fine-scale map version of NWM5
As mentioned in the previous sections the intention was to first compile a version of NWM5 which is purely fine-scale though mapped at a desktop level, and then assess whether there are still areas of the NWM4 polygons which would be useful. In the focus areas, NWM4 was considered during the mapping of wetlands. However, in the remainder of the provinces NWM4 was not included. As such, this section reports the methods and the results of the comparison between NWM5.4 and polygons from NWM4 not within NWM5.4. Both the extent and HGM unit typologies of NWM4 were evaluated to assess whether the extent should be included and the HGM unit be remodelled or manually retyped.
The smallest province, the Gauteng Province, was used for the prototype comparison, assuming that the prevalence of commission and omission errors, generated through the modelling of wetland extent from remote sensing in previous NWMs, may prevail in other provinces.
The following GIS steps were taken in a geodatabase with the outputs indicated after '->' as feature classes: 1. Extract NWM4 for the province -> GT_NFEPA_GeoWGS84 (the latter referring to a naming convention of the province, dataset and coordinate system, being unprojected geographic with the WGS84spheroid); 2. Project 'NATURAL' wetlands to the AEA coordinate system for South Africa -> GT_NFEPA_NATwetlands_AEA; 3. ERASE GT_NWM5.4 from the -> GT_NFEPA_NATwetlands_erased_GTNWM54_AEA; 4. Investigate the statistics related to the sizes of the remaining NWM4 polygons (Figure 4.7): The results showed the majority (74%) of NWM4 polygons are < 1.7 ha in size (mean), or a total number of 2 760 of 3 740. To assess whether there would be a certain threshold of selection large wetland polygons, the Ha field was sorted from the smallest size to the largest size and then randomly selected and viewed increments of approximately 0.5 and then 1 ha at a time. Most of the polygons between 0.1 ha and 18 ha were found to be commission errors, slivers or polygons on the edges of artificial wetlands. Since a thorough check was not done, it may be that there are some polygons which may be useful, however, the majority were found to be commission errors. 5. Polygons from NWM4 outside NWM5.4 were then selected where the hectares field [Ha] was >= 18.9 (an arbitrary threshold selected from viewing each polygon below and above this size). The remaining depressions from NWM4 were also omitted, assuming that these were not modelled from the spectra and may still be considered for inclusion. Therefore: 'Ha' >= 18.9 AND 'NWCS_L4' <> 'Depression'. An assumption here was that one would rather like to pick up large wetland systems from NWM4 not present by NWM5.4, rather than the very small polygons which may have resulted from spurious pixels from Landsat in the National Land Cover of 2000. The selection process resulted in 51 polygons which appeared to be useful. This was then entered as a definition query for the layer to minimise the display and these 51 polygons were then inspected one-by-one. 6. Explode polygons to avoid multi-part polygons: Select all polygons from GT_NFEPA_NATwetlands_erased_GTNWM54_AEA and explode; the UPDATE THE HA field.
The depressions remaining from the NWM4 outside NWM5.4 were then investigated from the largest to smallest ones. Most of these depressions were in fact seeps adjacent to existing depressions from NWM5.4 (Figure 4.8a). In some instances, the depressions were mapped on degraded land or appear to have resulted from interrupted drainage of roads, which resulted in the formation of an artificial wetland (Figure 4.8b). Others merely represent differences in the extent to which a depression was mapped with different images (Figure 4.8c). It was therefore concluded that these polygons would have to be carefully evaluated before integration with the fine-scale version of NWM5.4 and it was agreed that this should rather be done during the improvement of NWM6. 7. Lastly, non-depression HGM units remaining in NWM4 outside the fine-scale version of NWM5.4 were inspected for extent and HGM units -> GT_NFEPA_NATwetlands_erased_GTNWM54_dslv_AEA.
To evaluate the usefulness of these remaining polygons above 5 Ha, the remaining polygons were classified following visual inspection, according to the following rules:  if the majority of the NWM4 polygon falls outside existing data (NWM5.4) and were not considered to be wetland, it was classified as a 'Commission error';  if the majority of the NWM4 polygon was artificial, it was classified as 'ART';  if the majority of the polygon was useful but needed to be reshaped, it was classified as 'Consider';  if the polygon was rather a river, it was classified as a 'River'; and  if the wetland was better represented by the wetland probability map (Figure 4.9), then it was classified as 'wetland probability map'. The majority of the remaining NWM4 wetland types (non-depression wetlands) were found to be slivers or small polygons < 5 Ha in size (33%) or commission errors (32%) ( Table 4.6). Less than 1 122 ha (~19%) were considered useful for inclusion into NWM5.4; however, most of these polygons would require manual editing to reshape it to the correct extent and to avoid the inclusion of terrestrial areas (commission errors). The inclusion of the NWM4 polygons outside the NWM5.4 fine-scale wetlands dataset was therefore not considered for NWM5, and is recommended for consideration during the improvement to NWM6. The size of the polygons that could be considered in Gauteng, ±1 000 ha, could also be added through consideration of large floodplain, valley-bottom and seep systems that could visually be detected and mapped in a quicker way than by inspecting multiple small polygons. Expert-ID wetlands from the NFEPA wetlands which were natural, however, extracted and integrated into the provincial dataset by Ms Namhla Mbona, Ms Millicent Dinala, Dr Andrew Skowno and Dr Heidi van Deventer.
The supplementation of the wetland probability map (see Chapter 5) to the NWM5 was also considered and is detailed in full in Chapter 6. Owing to a large amount of commission errors in this dataset, it was not included into NWM5.
Concurrent to the finalisation of NWM5.4 for each province, Dr Heidi van Deventer and Ms Anisha Dayaram captured large floodplain systems across the country, totalling < 15 systems. All provincial NWM5.4 datasets were reviewed by Dr Heidi van Deventer in January 2018 and sent back to SANBI for final revisions and amendments.

Integration of provincial datasets into a country-wide National Wetland Map 5 (NWM5)
The improved provincial NWM5.4 versions were received by Dr Heidi van Deventer at the end of March 2018 for final integration. A number of steps have been taken during the final integration to ensure quality of the output file:  Edge-matching between provinces, giving preference to the extent and HGM unit of polygons where finer-scale data and mapping was available;  Identification and refinement of the eight freshwater lakes (limnetic wetlands) in the country;  Inclusion of the large floodplain systems;  Improvements of the HGM units: o Floodplain depressions, kept in some provinces separate from the floodplain flats, were correctly attributed at CS_L4B and corrected at CS_L4A and the subtype field to be floodplains. Not all floodplain depressions were included across all the provinces, since many have already been merged by interns into the Level 4A floodplain category. Future updates of the NWM could verify the 'lakes' hydrological category of DRDLR:NGI visually to distinguish artificial ponds from floodplain depressions or ox-bow rivers on floodplains. The latter could be included as floodplain depressions into all floodplains of future NWM updates, and attributed to Level 4B of the Classification System. The DRDLR:NGI 'lakes' polygons are not inland lakes at all and should not be used for reporting lakes. Only the eight limnetic wetlands are considered to be freshwater lakes. The DRDLR:NGI 'lakes' have been converted to a points dataset and cleaned up by two interns, where points were added in order to potentially be used in modelling floodplain systems. These required a final check and update by experts before it is used.
o Wetland flats in most provinces were corrected to Level 4A depressions, except where it originated from data captured by wetland experts, such as the C.A.P.E. fine-scale project. Errors in typing wetlands as wetland flats resulted from the incorporation of some datasets which retained the modelled HGM unit from the NFEPA wetlands. A discussion with the inland aquatic reference committee and a number of wetland experts concluded that these systems are difficult to distinguish at desktop-level, particularly to determine if the source of water is purely groundwater driven, or also from seepage.
 Slivers < 0.125 Ha in size were checked and either merged or deleted. Springs buffered to 2 m resulted in an extent of 0.000313 Ha. These were kept in the dataset and can potentially still indicate the origin of inland wetlands. Many other slivers and small polygons remain in the dataset which can be considered for deletion. A thinness index can be calculated and a threshold selected to delete slivers, though users should be careful not to delete any polygons before visual inspection, since some may still be valid inland wetlands. The smallest depression mapped, for example, was measured at 0.0031 Ha or 3 m²;  Polygons with no HGM unit (435)  o Swamp forests from the NVM were attributed to Level 6 (CS_L6) in the attribute fields. o The Wetland Vegetation Groups (WVGs) ) was re-assigned to the related vegetation types of the NVM. Alluvial vegetation, inland saline, forest and azonal categories were reassigned to adjacent vegetation types. Long linear systems were not split, resulting in potential intrusions of one category onto another. These need to be improved during future updates. The WVGs were then dissolved and intersected with the NWM5. Where a polygon was dominated by one WVG, the WVG was assigned to Level 2 (CS_L2). Where more than one WVG was present within a wetland polygon, the dominant WVG was determined in Excel through an INDEX command using the extent (ha) in each WVG, and then attributed to the polygon under Level 2. Owing to time and budget constraints, the names of wetlands present in NWM5.2, were not assigned to the polygons of NWM5.2 again. Although this is not a priority for the NBA 2018, or decision making and planning, names offer key reference points to users. Consideration should also be given to the standardised names and aliases for estuaries and whether these should be adopted by the river lines and polygons.

Alignment of inland aquatic ecosystems with Estuarine and Coastal Ecosystems
During the investigation of the alignment between the NWM5.2 and the estuaries from the NBA 2011 in the Estuarine Ecosystems Classification Committee workshop March 2017, the estuarine and inland aquatic team noted a number of issues. Polygons mapped by the DRDLR:NGI outside the Estuarine Functional Zone (EFZ) as hydrological features were either inland wetlands, estuarine, coastal or terrestrial in nature. Some of these polygons, for example, included tidal pools, dunes or dune vegetation of which the latter had been captured at a low spatial resolution (lower than 1:10 000). Although it is recognised that all aquatic ecosystems would contribute to biodiversity, and that the ecotone of change is a fuzzy boundary, several steps were taken to ensure alignment of inland wetlands with the estuarine and coastal ecosystem types in the National Wetland Map 5:  The extent of the EFZs as well as micro-systems (micro-estuaries, ephemeral coastal streams, coastal seeps and waterfalls into the ocean) were finalised using the first 5 m contour above the shore, as well as vegetation mapped by NMU estuarine experts. These included Dr Lara van Niekerk  Where an inland wetland exists through the shore as a seep, it was mapped as a point in the outlet dataset of the estuarine ecosystems. These outlets will therefore not be reflected as polygons in the NWM5.
Further updates of the NWM should ensure that for each estuarine system, whether EFZ, micro-system or outlet, an inland wetland should be mapped to reflect the connectivity between inland wetlands, estuaries and the coast.

Creation of an artificial wetlands layer to be used for planning
As part of the development of the national wetland map a dataset of artificial water bodies was built using the DRDLR:NGI hydrological features exported from NWM5.2. All non-natural feature types were included in this feature class and were checked for consistency and then integrated by Dr Andrew Skowno (SANBI) with the national 1:50 000 dams map. However, they were also verified or supplemented by data from the dams register of the Department of Water and Sanitation (DWS), which lists approximately 159 dams (DWS 2015). This formed the foundational layer for the development of the artificial water bodies dataset which was exported in the geodatabase as a 'Modifications' feature class. The process of development of the map included basic desktop validation of features over 25 ha in extent, and the identification of dams built after 1990 -using the national dams register (DWS 2015) and a range of remote sensing products. This final step was conducted in order to facilitate the use of the dataset in the land cover change analyses. Appendix H contains a technical report covering the development of the artificial water bodies map for South Africa, version 1 (Skowno & Van Deventer 2018).
In a second phase of update to the artificial wetlands data layer, Dr Heidi van Deventer (CSIR) refined the extent and names of the 92 large dams monitored in the DWS NEMP. In addition, information from available point shapefiles, including the Aquaculture facilities (DAFF & DEA 2018) as well as the Water Treatment Works (WTWs) and Waste Water Treatment Works (WWTWs) (CSIR 2015) were attributed to the polygon dataset. Not all points had a polygon to attribute the information to (Table 4.7). Some slivers were removed up to an extent of < 0.000313 Ha and some of the adjacent polygons with different but related categories were merged. These include, for example, reservoirs, purification plants, sewerage works and WWTWs which may have been derived from different sources and dates of classification, however were the same facility.
The final map contains just over 200 000 features, covering an area of 598 381 ha, with dams making up the bulk of the artificial water bodies in South Africa (Table 4.7; Figure 4.10).
The accuracy of the underlying DRDLR:NGI hydrological features data dictates the accuracy of this layer to a large degree. Although efforts have been made to eliminate 'gross errors', the large number of features makes a comprehensive, feature by feature, validation process impossible, and many classification and mapping errors are likely to be encountered in the dataset. The aim is to iteratively improve this dataset through the release of an updated version annually or biennially.
The polygon representing artificial wetlands should be used in association with the WTWs, WWTWs, invasive species and other datasets for condition modelling and planning processes to obtain a better reflection of the pressures on inland wetlands.
It is furthermore important to note that the current datasets have not been assessed for completeness. Some of the artificial wetland types are listed in the Classification System but not in the inventory which includes canals, excavations, salt works, aquaculture points and storm water ponds. Future updates should attend to the assessment of the completeness and accuracies of the existing datasets, as well as mapping the point features which are not represented in the Artificial wetlands layer. Furthermore, the full extent of a facility should be captured in the map. The DRDLR:NGI only maps a representative point associated with the WTWs or WWTWs, but not the full extent of the property (Figure 4.11). Similarities between categories should be considered and those polygons amalgamated, following finalisation of definitions.

Results of NWM5 for country
The final National Wetland Map version 5 (NWM5) shows the distribution of the estuaries and inland wetland HGM units across South Africa (Figure 4.12). Not only has the number of datasets represented in the South African Inventory of Inland Aquatic Ecosystems (SAIIAE) increased, but also the number of ecosystem types within the NWM5: we now have Estuarine Functional Zones (EFZs), estuarine microsystems, inland wetlands and rivers included, totalling nearly 4 million hectares (ha) of aquatic ecosystems which cover 3.3% of the surface area of South Africa (Table 4.8; Figure 4.13). Various categories of artificial wetlands are also now mapped, totalling almost 600 000 ha in a separate data layer.
NWM5 shows an increase of 123% of inland wetlands mapped compared to the NFEPA natural inland wetlands. A total of 2.6 million ha of inland wetlands have been mapped in NWM5 and typed to HGM units, making up 2.2% of the surface area of South Africa 1 (Table 4.8; Figure 4.13). A total amount of 201 381 ha of estuaries have been added and > 1 million ha of river channels.    In total we estimate that more than 31 data editors distributed across more than ten organisations have been involved in the data mapping, integration and finalisation of the inland wetlands and estuarine ecosystem represented in NWM5. The cost of generating NWM5 is estimated at > R7 million during a period of only two years.

Results of NWM5 for the focus areas
NWM5 showed an increase in the extent of HGM units for seven of the nine focus areas where inland wetlands had been mapped (Figure 4.15). In three of these focus areas, the extent increase by > 150% of the inland wetland extent mapped in the NFEPA wetlands, including the Eden, Ehlanzeni, Vhembe and Umgungundlovu Districts. In five of the focus areas, namely the Buffalo City, Cape Winelands, Frances Baard and Lejweleputshwa Districts, a reduction in the extent of inland wetlands was observed when NWM5 was compared to the NFEPA natural wetlands. The Buffalo City, Frances Baard and Lejweleputshwa Districts are in more arid regions. Some of these had been attributed to commission errors in the NFEPA wetlands that had been removed, as well as expert knowledge from Dr Nacelle Collins on inactive paleo river systems in the north-western part of the Free State Province, which now had been removed in NWM5. It was quite interesting to note that the extent of inland wetlands for the West Rand District Municipality in Gauteng had increased by nearly 700%! The increase followed the supplementation of NWM5.4 of the Gauteng Province with the wetland probability map (Chapter 5), as well as the reshaping and HGM unit typing of the inland wetlands by Dr Joseph Mulders and Dr Heidi van Deventer. The NWM5.3 for the Gauteng Province, prior to the supplementation of the wetland probability map, showed almost a 200% improvement of the extent of the inland wetlands compared to the NFEPA natural wetlands map, whereas the wetland probability map added another 500% improvement compared to the NFEPA data. While the integration of more than 9 datasets for the Gauteng Province into NWM5.3 took quite some time, the supplementation, reshaping of extent and typing of the wetlands of the wetland probability map took less than two weeks by the two experienced personnel.

Comparison between the NWM5 and the National Wetland Vegetation Database
A comparison was done between the NWM5 and 5 137 relevés of the National Wetland Vegetation Database (NWVD) of Sieben (2015). Nearly 60% of the relevés of the NWVD had no polygons represented in NWM5 (Table 4.9), which emphasize that there remains a large underrepresentation of inland wetlands, regardless of the improvements. Twenty-two percent of the relevés had polygons in the NWM5 which were congruent in HGM units. Another 6% of the relevés had polygons in NWM5, but with discrepancies in the HGM units. Further investigation showed that multiple relevés were taken within one large inland wetland unit represented by one polygon, and if the HGM unit was different, it was detected as multiple differences. In addition, HGM units of floodplains, for example, were typed in the NWVD at Level 4B of the classification system as 'depressions' which occurred on floodplains. Comparisons with Level 4A appeared as discrepancies, but were however similar HGM units. Some relevés within the estuarine ecosystems were also typed as inland wetlands, which may reflect a gradual transition from one system to another, and is not necessarily incorrect in either of the two datasets, but merely that different criteria and scales have been used in the typing of the aquatic system.
The HGM unit assigned to approximately 291 relevés (5% of the total number of relevés) could potentially be updated, however since NWM5 received fine-scale typed datasets from various wetland experts, the changes of these polygons require a process of verifying the type against the source and review points before it is changed. Owing to time and budget constraints, the team decided to leave these points for further investigation and improvement in NWM6.

Confidence map
A confidence map was compiled to identify areas where the extent and HGM units attained a higher level of certainty compared to other areas. For the purpose of this first version of the SAIIAE, higher levels of certainty are associated with areas that have been visited in-field by a wetland specialist(s) over multiple seasons and cycles of the hydroperiod of a wetland, and accurately represented in the dataset. Accuracy, on the other hand, reflects the degree to which the spatial boundaries and/or attributes of the wetlands match those in the real world (Pascual 2011). A higher confidence score would imply an increase in accuracy, assuming that the spatial boundaries of a wetland had been informed by water, soil and vegetation indicators, and recorded with differential GPSs. Low confidence scores would imply the inverse.
A number of confidence ratings have been developed for the inland wetlands included in the NWM5 (Table 4.10). These have been primarily derived from the data received for the SAIIAE . Confidence ratings have been assigned to the sub-quaternary catchments (SQ4s) of South Africa (Figure 4.16). The majority of the country (69%) has a Low confidence rating for NWM5, where the data is a cross-walk of DRDLR:NGI hydrological features to the HGM units.
For 24% of the country, interns trained by wetland specialists have mapped the extent and HGM units at a desktop level resulting in a Low to Medium rating of confidence. For 7% of the country, desktop mapping was done by wetland specialists, resulting in a Medium confidence. For only a few of the SQ4s, or parts thereof, in-field verification was used to modify the extent and HGM units of polygons in the NWM5. None of the SQ4s had studies which considered the long-term hydrological cycles of the inland wetlands, and therefore no part of the country had a confidence rating of High.
The confidence rating map should always be consulted to assess whether the NWM5 is fit for purpose. The confidence of estuarine and coastal ecosystems is not reflected in this confidence map (Figure 4.16). In general, areas mapped as a Low confidence, is likely to still have a 50% omission error, though the commission error may have been reduced to < 10% compared to the previous versions of the NWMs. The appropriate scale of use would be 1:50 000. Where the confidence rating is Low to Medium, Medium and Medium to High, we estimated the omission error to be < 30% and the commission error < 10%. The appropriate scale of use would be 1:10 000.

-Low
Desktop mapping of the extent of inland wetlands was done by non-wetland specialists for a part of, or the full extent of the SQ4.

-Low to Medium
Desktop mapping of the extent of inland wetlands was done by interns trained by wetland specialists for the full extent of the SQ4.

-Medium
Desktop mapping of the extent of inland wetlands and HGM unit typing was done by wetland specialists for the full extent of the SQ4.

-Medium to High
Desktop mapping of the extent of inland wetlands and HGM unit typing as well as field verification and revision by experts was completed for the full extent of the SQ4.

-High
Inland wetlands have been mapped and verified for a period of > 10 years over multiple hydrological cycles for the full extent of the SQ4. Verification may include field observations as well as soil and/or vegetation surveys.  Table 4.9 provides definitions of the various ratings of confidence.

Recommendations of future updates
During the course of the finalisation of the NWM5, Ms Nancy Job has been appointed as the Programme Manager of the Freshwater Biodiversity Unit at SANBI as of 1 February 2018. The Freshwater Biodiversity Unit at SANBI will further address the strategy of updating NWM6 and the identification of priority areas and approaches in doing so.
A summary of recommendations is provided for future updates of the NWMs for the consideration not only by the Freshwater Biodiversity Unit at SANBI but also for many other stakeholders who have interest in the improvement of the NWM.
Adopt the following principles in data capturing:  The original extent of the wetland should always be captured;  The maximum extent of a wetland should always be captured;  Multi-temporal imagery across hydrological cycles should be used to determine the maximum inundated extent of a wetland across hydrological cycles; and  The HGM units and condition assigned in fine-scale wetlands datasets received from wetland experts should always be retained.
Improvements to the extent of the NWM5:  Aim for immediate improvement of 75% of the country to at least confidence level 2 (Low to Medium) within the next 5 years. Immediate updates of the FEPAs are illogical while the majority (69%) of the country's inlands wetlands are mapped at a low confidence with an estimated 50% omission error. Collaboration in reaching this target is crucial;  Desktop reshaping and typing of polygons from the wetland probability map appears to be a very quick and effective way of addressing omission errors in the NWM5;  Desktop mapping of wetlands for the remaining peatlands points and relevé points which were not captured in NWM5. These should be a quick exercise since there are < 350 polygons to map;  Reintegrate the river channels of the DRDLR:NGI into the NWM to remove 'commission errors' of inland wetlands which should be in fact rivers;  Desktop verification of small polygons mapped in NWM5 appear as inland wetlands, but also appear in the Artificial wetlands layer as farm dams;  The improvement of definitions, categories and desktop mapping of artificial wetlands. These systems are important for assessing hydrological regime impacts on wetlands and rivers, fragmentation and planning. Similar to the inland wetlands, a confidence map should be developed for the artificial wetlands, as improvements are progressively done across the country;  Close polygons where dams were or have been within the inland wetland. This improves the original extent of the inland wetland ecosystem types for NBAs. The artificial wetlands can always be used with the NWM for planning purposes; and  Large floodplain and valley-bottom wetlands not mapped in NWM5 can be easily detected and supplemented from the alluvial categories of the 1:250 000 geology and NVM, if not already integrated from the wetland probability map.
Improvements to the NWM that are of secondary importance:  Removing slivers using a thinness threshold. This may be a time-consuming process with minor impact on the extent of inland wetland ecosystem types represented and assessed. This should not be done as a top priority;  Names of wetlands were lost from NWM5.2 to NWM5.4. Spend time assigning names to inland wetlands again. It makes the map user friendly and serves as search facility and common reference; and  This chapter provides an overview of the method used to procedure to create a map of predicted wetland extent.

Introduction
An objective of the National Biodiversity Assessment 2018 (NBA 2018) process was to improve on some of the existing national spatial datasets available at the time. The poor spatial accuracy of the national wetlands layer has been a concern and was identified as a priority layer requiring improvement.
Methods previously employed for mapping wetlands have not produced satisfactory results. Some casual observations have estimated that in certain regions the wetland features of the National Freshwater Ecosystem Priority Areas (NFEPA) wetlands layer, which is informed by National Wetlands Map 3 (NWM3), account for only ±20% of the wetlands in certain regions. Not all wetlands are mapped equally poorly, with wetlands of the valley floor landscape unit being of particular concern. In-field mapping of wetlands by  showed that 68% of wetlands mapped in the Mpumalanga Highveld were located on valley floors, which when considering the above, explains findings that the more recent and improved National Wetland Map 4 (NWM4) is estimated to account for less than 54 % of the wetlands found at the fine-scale level ). The results of these studies also showed hydrogeomorphic (HGM) unit typing and the wetland condition assigned to the wetlands to be inaccurate in NWM4 .
Low mapping accuracies of NWM4 are also expected for the rest of the country . According to  these widespread weaknesses are considered to be of sufficient severity to warrant investment in the improvement of the National Wetland Map as a matter of urgency, hence the efforts of the NBA 2018 process to improve on this layer. In addition to presence and absence, inaccuracies the NWM4 also suffers from poor contiguity of inland aquatic systems within and between individual wetlands.
The NBA 2018 process therefore set out to improve on NWM4 by creating National Wetlands Map 5 (NWM5). Two approaches to improve the NWM5 were pursued, these being (Van Deventer 2016):  A fine-scale wetlands map to be created by heads-up digitising from imagery with inclusion of data from the Department of Rural Development and Land Reform: National Geo-spatial Information (DRDLR:NGI) as well as newly acquired fine-scale data; and  A dataset of probable wetland extent.
This chapter reports on the technical procedure to create the map of predicted wetland extent.

Background
The South African national wetland inventory has seen several improvements over the past decade. Numerous improvements have been enacted on the original wetlands map and the latest release is the NWM4. NWM4 has recently received some criticism, specifically the spatial accuracy, with the latter referring to both wetland representation and boundary delineation. Wetlands associated with the valley floor landscape unit in particular are considered to be poorly mapped considering the fact that previous image-based approaches have not produced satisfactory results. Alternative methods for mapping wetlands were investigated.
A Digital Elevation Model (DEM)-based approached was considered as a possible alternative. It is a desktop approach which, considering the extent of the study area (RSA), is considered to be a practical and appropriate solution. The DEM-based approached stems from the landscape position criterion of the indicators and criteria for identifying and delineating wetlands (DWAF 2005) and is based on the logical assumption that water will accumulate within the lowest position of the landscape which therefore represent areas of highest probability for wetland 2 development. The low-lying areas are watercourses per the National Water Act (No. 36 of 1998) and include amongst others rivers, springs, natural areas in which water flows regularly or intermittently, wetlands, lakes and dams (DWAF 2005). It follows that although wetlands are most likely to develop within these low-lying areas, features other than wetlands are also present. It is also possible that many of the watercourse areas do not contain any wetlands at all as wetland development does not only require the presence of low-lying areas, but also depends of other factors such as mean annual precipitation, slope and soil depth for example. The aim was therefore to map the valley floor areas, but to limit the valley floor areas mapped to those that are considered to be wetland as subjectively identified from imagery. Depending on a number of factors the areas mapped in the final output may or may not be wetland, and considering the fact that it is a desktop approach using remote sensed data, there is no certainty as to the likelihood of the mapped areas actually being wetland or not. The output of the mapping process is therefore referred to as a 'wetland probability map'.
There are a number of tools available in Geographical Information Systems (GISs) with which to identify and map the valley floor landscape unit from DEMs. Probably the best known is the Topographical Position Index (TPI) tool (Jenness 2013). The TPI tool determines landscape position by comparing differences in elevation of a cell with the average elevation of surrounding cells. The degree to which elevation and slope of the target and surrounding cells differ are used to assign cells to different landscape position categories. For example, cells that are significantly higher than the surrounding cells and that are located on a flat surface are likely to be ridgetops or hilltops, whereas cells that are on a flat location and at or near the bottom are most probably valley floors. The number of values surrounding the target cell considered is determined by the user by specifying the size of a moving window where the elevation of the target cell (the cell in middle of the window) is compared with the elevation of the cells within the remainder of the moving window.
The TPI tool was used during initial attempts at mapping valley floor areas and although the outputs were subjectively considered to be not satisfactory, they did represent a minimum level of improvement on previous attempts at mapping wetlands. Perceived inaccuracies of the TPI tool relate to mapping wetland occurrence and extent. Relief can change over very short distances and areas of different relief require different TPI tool parameters, specifically different moving window sizes. Suitable parameters are determined by trial and error and the area of which the valley floors were satisfactory mapped for a specific set of parameters are determined as a post mapping exercise. New parameters need to be set for the 'not so accurately mapped' areas which are only able to accurately map certain areas, whilst new parameters will have to be determined for the remainder of the areas for which the current parameters do not apply. To map wetlands in different landscape positions within the same area will also require different parameters, specifically different moving window sizes. It follows that a set of parameters will map some of the wetlands within an area while a new set of parameters are required to map other wetlands of the same area. Similarly, the wetlands of other areas will again require new parameters. Because the accuracy with which the valley floor areas are mapped can only be determined after the tool finished a run, the TPI tool cannot be applied in a systematic manner to consistently map wetlands of all landscape positions in all areas and was therefore considered not suitable for mapping the valley floors of an area the size of South Africa. In addition to the latter the output of the TPI tool is a single continuous feature which in reality may consist of different HGM units. This is undesirable as the mapped feature (valley floors) will at a later stage have to be typed per the Classification System for Wetlands and other Aquatic Ecosystems in South Africa ). This will be problematic if wetlands of potentially different HGM unit are not mapped as separate features.
While the TPI tool maps all terrain units such as ridges, upper slopes, middle slopes, flat slopes, lower slopes and valleys, other tools are dedicated to mapping valley floor areas only, for example the 'Extract Valleys' tools in Whitebox GIS which attempts to map the 'wetness' of an area based on the upstream contributing area and slope. Other tools are also available such as the Topographical Wetness Index (TWI) which is available in both SAGA and Whitebox GIS.
Unlike the TPI tool many of the above-mentioned tools are not able to map the extent of the valley floor areas resulting in the identified valley floor being mapped as a single line rather than as an area or polygon. Another limitation is that some of the tools, similarly to the TPI tool, require different parameters to accurately map the valley floors of different areas, or different wetlands within the same area, due to differences in topography for example. Similarly, the areas accurately mapped can only be identified at the end of the run. Applying these tools for mapping wetland extent therefore require continuous adjustment to accurately account for different areas and different wetland ecosystems, particularly when applying them at a country-wide scale. Their application is therefore mostly a 'trial and error' approach which inevitably becomes very time consuming and cumbersome.
It was subjectively concluded that none of the investigated tools were deemed suitable for improving NWM4 by way of modelling. Requirements for a new method for creating an improved wetland map included the ability for it to:  map the extent of valley floor areas;  reflect the contiguity of inland wetland ecosystems;  be applied to all regions of the study area which in this case is South Africa;  be rapid while still producing results suitable for 1:50 000 country-wide application; and  have minimum data and budget requirements.
Considering the limitation of the TPI and other existing tools, as well as the above-stated requirements, modelling the extent of wetlands from a country-wide DEM was considered the most suitable alternative approach for creating a new and improved wetland map.

Aim
The aim of this part of the study was to develop a method for creating an improved wetland map for South Africa that will satisfy the above-stated requirements.

Introduction
Of the different tools and methods considered, an approach of combining outputs of the 'Percentile filter' tool of Whitebox GIS (Lindsay 2014) and flow accumulation maps of ArcGIS 10.5. 1 (ESRI 19991 (ESRI -2017 were considered to be most suitable at achieving the stated objectives. Whitebox GIS was used to create percentile filter maps as this or a similar tool is not available in ArcGIS. Whitebox GIS also offers DEM preparation tools that are superior to those of ArcGIS (Section 5.4.5).
Implementation of the method required the following data: A number of python scripts were developed to facilitate the mapping process. Python 2.7 (Van Rossum 1995) is the preferred ArcGIS scripting language that allows for automation of GIS related tasks. Python 2.7 is automatically installed with every ArcGIS installation and is therefore easily accessible and freely available to all ArcGIS users. In addition to providing access to the standard ArcGIS tools, Python 2.7 also provides additional functionality without which automating certain aspects of the mapping process would not have been possible such as branching (if-then-else statement), the use of lists and dictionaries. The scripts are run from the standard ArcMap and ArcCatalog interfaces and automate many of the required processes such as creating the mapping regions (Section 5.4.3). The scripts also automate the process of converting user defined mapping parameters (Section 5.4.6) to the wetland probability map. It is important to note that the mapping process per se is not automated, but that the scripts merely automate some of the geoprocessing tasks required for creating the wetland probability map. The mapping process can be summarised as follows:  South Africa as the study area was subdivided to account for computer processing limitations (Section 5.4.2);  The divisions were further subdivided into mapping regions (Section 5.4.3) which serve as the mapping units for which percentile filter and flow accumulation thresholds were determined (Section 5.4.3);  Percentile filter maps of varying moving window sizes were created for each of the divisions (Section 5.4.4);  Flow accumulation maps were created for each of the divisions (Section 5.4.5); and  For each of the divisions: o the mapping regions were displayed over the SPOT 2013 imagery; o percentile filter and flow accumulation mapping thresholds were determined for each of the mapping regions (Section 5.4.6); o wetlands not sufficiently accounted for by the percentile filter and/or flow accumulation maps were captured manually and included as ancillary data (Section 5.4.7); and o the wetlands, as informed by the percentile filter and flow accumulation thresholds and ancillary data, were mapped (Section 5.4.8).
The method used to create the wetland probability map is therefore not an automated process. It can be considered to be similar to onscreen mapping. However, what distinguishes it is that with standard onscreen mapping each wetland identified from imagery is mapped individually, whereas the modelled approach simultaneously maps all wetlands identified from imagery within a mapping region using an 'overall best fit' approach. The user determines a set of mapping thresholds which when applied to the percentile filter and flow accumulation maps gives the best approximation of wetland extent on valley floor areas within the mapping region.

Subdividing the study area
To overcome computational limitation imposed by the size of the study area the latter was subdivided according to existing provincial boundaries. The provincial boundaries used are those per the South African Municipal Demarcation Board (MDB). Due to its size the Northern Cape Province was further subdivided into four regions, these being the Northern Cape East (NC_E), Northern Cape North (NC_N), Northern Cape South (NC_S) and Northern Cape West (NC_W) (Figure 5.2). The division boundaries of the Northern Cape Province follow those of secondary catchment areas (refer to Table 5.1 for an explanation of the abbreviations).

Creating the mapping regions
Because no single set of mapping thresholds are suitable for all areas of a division, the latter were further subdivided into 'mapping regions'. The mapping regions are a first approximation of areas considered to be homogenous in terms of the variables that were subjectively considered to be most influential at determining wetland development such as relief, mean annual precipitation and generalised geology. They therefore represent areas in which a certain set of parameters also known as the mapping thresholds should theoretically be able to consistently map the wetlands of the valley floor areas.

Relief
Relief was determined by creating a grid of 1 000 hectare sized hexagons for each of the divisions after which the standard deviation of elevating for each hexagon was calculated using the ArcGIS 'Zonal Statistics as Table' tool. Elevation data were derived from the 30 m DEM. The standard deviation of elevation is considered to be a quantitative expression of the relief of the terrain underlying each hexagon ( Figure 5.3 A). The map was subsequently simplified by grouping the standard deviation values. Grouping was achieved by displaying the output of the Zonal Statistics tool in ArcMap with symbology set to quantity, five classes and natural breaks (Jenks) as the classification method. The indicated Jenks break values were used to define the borders for the five categories according to which they were grouped. The output is therefore a map with five zones of relief ( Figure 5.3 B).

Generalised geology
The geological data used is a simplified version of the 1:1 million Geological Data of the Council for Geoscience's RSA Geological Dataset (CGS 1997) ( Figure 5.4).

Mean Annual Precipitation (MAP)
While relief and the generalised geology relate to the extent to which the percentile filter maps will be able to consistently map valley floor areas within different mapping regions, the mapping regions also need to account for the environmental variables that influence the probability of the mapped valley floor areas actually being wetland. There may, for example, be differences in the likelihood of wetlands occurring within regions of similar relief but which are characterised by differences in MAP. As wetlands are predominantly a function of rainfall, the MAP is expected to be an important modifier that needs to be accounted for . MAP data were sourced from the Water Research Commission (Lynch & Schulze 2000) and was categorised into 100 mm intervals ( Figure 5.5). The output is therefore a mean annual precipitation map with zones of 100 mm interval.

Figure 5.5: Mean annual precipitation (MAP) of the South Africa mapped at 100 mm intervals which was used to inform creation of the mapping regions.
Mapping regions for each division were obtained by creating a union of the relief (the 5 category zones), generalised geology and MAP at the 100 mm interval zones using the ArcGIS 'Union' tool ( Figure 5.13). Mapping regions smaller than 40 km 2 were dissolved into neighbouring features using the ArcGIS 'Eliminate' tool with the Eliminate by border parameter checked. Mapping regions were added during the process of assigning mapping thresholds by manually subdividing mapping regions to account for situations where the wetlands in a mapping region could not be adequately accounted for by any set of percentile filter and flow accumulation map thresholds.

Percentile filter maps
Percentile filter maps were used to map the broader valley floor areas where typically floodplain and channelled and unchannelled valley-bottom wetlands would occur.
Percentile filter maps were created using the 'Percentile filter' tool of Whitebox GIS (Lindsay 2014). As the name of the tool suggests it performs a percentile filter analysis on a raster image. The tool expresses the value of each cell as a percentile (0% -100%) of the range of values within a moving window. Because cell values of a DEM represent elevation, the percentile value assigned to each target cell (the cell in the centre of the moving window) is therefore a quantitative expression of its elevation relative to the elevation of the surrounding cells, specifically the cells covered by the moving window. Once the percentile for a target cell has been calculated, for example target cell A of Figure 5.6, the tool moves the target as well as the moving window to the neighbouring cell. Another example is target cell B and moving window B of Figure 5.6 respectively. As was for target cell A, a percentile value is calculated for target cell B by expressing its value as a percentage of the range of values of all the other cells that are covered by the moving window at its new location. This process is repeated until a percentile value has been calculated for all cells.

Figure 5.6: The percentile filter tool analyses a raster on a cell-by-cell basis as illustrated by target cells A and B and the accompanying moving windows which follow the target cell. The value of the target cell is expressed as a percentage of the value of all cells that are covered by the moving window.
The size of the moving window is adjustable, which implies that the size of the neighbourhood against which the value of the target cell is being evaluated can be changed. This feature is of particular use and allows for the percentile filter tool to be applied to areas of different relief. For example, a small moving window of 9x9 cells was found to be appropriate for mapping valley floors in areas of high relief while a larger moving window of (81x81 cells was found to be more suitable in areas with a relative flat topography or low relief.
In Figure 5.7 A the 9x9 moving window size is unable to detect the valley-floor areas as the entire window is located within the valley-floor as opposed to the 81x81 window of Figure 5.7 B which, because the window extends beyond the valley-floor, is able to detect it. Both the 9x9 and 81x81 windows in the high relief areas of image A and image B are able to detect the valley floors. However, because of the different window sizes, it will require different percentile values or percentile thresholds to display them. For example, the 10 th percentile of the 81x81 window of B will represent a larger portion of the valley floor than the 10 th percentile of the 9x9 window of A because the former includes a larger range of numbers of which most are of higher elevation than the latter. For mapping purposes percentile filter maps with moving window sizes of 9x9, 15x15, 21x21, 31x31, 51x51 and 81x81 were created for each of the divisions. These are referred to as the base percentile filter maps.
The extent of area mapped by a percentile filter map is controlled by specifying a percentile threshold. A threshold of 10%, for example, means that only the cells of which their elevation is 10% or less of the elevation of all cells in the moving window will be mapped, i.e. the cells at the lower end of the elevation (the valley floors). Similarly, a percentile threshold of 50% means that all cells with a value that is halfway between the cell with lowest elevation and the cell with highest elevation will be mapped.

Flow accumulation maps
The flow accumulation maps were used to map the narrower wetland systems that typically connect the broader floodplain and valley-bottom wetlands (those mapped by the percentile filter maps) as well as those on the adjacent slopes. These are typically rivers or seeps that occur along channels or within the valley-floor landscape unit, but can also be narrow occurrences of other HGM units.
Creating flow accumulation maps has become standard practice in hydrological modelling processes. To create flow accumulation maps the DEM first needs to be hydrologically corrected, a process also referred to as DEM pre-processing which involves removing sinks from the DEM. In addition to the standard fill tools, Whitebox GIS also includes tools to create hydrologically corrected DEMs by way of breaching the sinks instead of filling them as the standard ArcGIS 'Fill' tool does. Whereas the standard fill method simply raises the elevation within depressions, breaching 'carves' a channel through the sinks. The advantage of this approach is clearly visible in (Figure 5.8, blue lines) where the outcome of the flow accumulation derived from a DEM pre-processed with the 'breach' tool results in lines which reflect a 'natural' path that water would follow in the landscape. In contrast, the result of flow accumulation derived from a traditional fill operation ( Figure 5.8, red lines) result in illogical and unnatural straight lines running through the fills.
DEM pre-processing was done by first subjecting the DEM to the 'Breach Depressions' tool of Whitebox GIS, of which the output was used as input for the 'Breach Depressions (Fast)' tool (also from Whitebox GIS). The output was subsequently exported to ArcGIS and used as input to create flow direction maps using the 'Flow Direction' tool. The latter informed the flow accumulation maps which was created using the ArcGIS 'Flow Accumulation' tool.

Determining mapping thresholds
All the mapping regions of a division occur as separate features within the mapping region attribute table.
The attribute table contains fields in which the percentile filter and the flow accumulation threshold considered to best map the wetlands of the mapping regions are recorded.

Percentile filter thresholds
It follows from Section 5.4.4 that it is highly unlikely that a single percentile filter map will be suitable to accurately map the valley floor areas for all mapping regions of a division. The user therefore needs to make a judgement on the appropriate moving percentile base map and percentile threshold for different mapping regions.
In many instances the accuracy of the base percentile filter maps can be improved by iteratively expanding and shrinking cells or vice versa. The process of creating the modified percentile maps is an iterative one which is not done as a separate task but as part of the process for determining which percentile filter maps and thresholds are best for mapping the visually identified wetlands of a mapping region. Percentile filter thresholds were determined by first subjectively deciding on a base percentile filter map and a percentile threshold which was considered to be suitable for the mapping region. Depending on whether the chosen percentile threshold was found to accurately represent the identified wetlands or not, different base maps with the same or other threshold, or a different threshold using the same base map were tested. It may become evident at this time that the accuracy of the base map for a given threshold may be improved with some modification. The base percentile filter map was modified in such instances.
Modification involves reclassifying the base map so that only values (percentiles) that satisfy the subjectively chosen threshold are retained while the remainder of the cells are discarded and assigned 'NoData' (A of Figure 5.9). The remaining cells may then at the discretion of the user be repeatedly expanded and/or shrunk so that the end result is improved representation of the identifiable wetlands (B of Figure 5.9). The benefits of such modifications are clearly visible from Figure 5.9 where A is a base percentile filter map (81x81 moving window) with a threshold of 30 percentile, while Figure B is the same percentile filter map (81x81 moving window) which was also reclassified to the same percentile threshold (30) after which it was modified by first shrinking it with one cell, then expanding it by five cells and then again shrinking it by 8 cells. Subsequent to having created the different modified percentile filter maps, the process of assigning percentile thresholds to a mapping region involved: i. testing different percentile filter base maps and thresholds; ii.
creating modified percentile filter maps where the base maps are found to be unsuitable; or iii.
determining which of the already created modified percentile filter maps are best at representing wetland occurrence and extent What is referred to as the percentile threshold for a mapping region is therefore actually a map which is the culmination of a process involving identifying a suitable base percentile filter map, reclassifying it to the determined percentile threshold and applying the modification per the above explanation.
It follows from the above that the percentile filter tool provides an almost endless number of permutations that can be created by: i. using different moving window sizes (the base maps); ii. applying different thresholds to the base maps; and iii. applying a number of additional modifications to the reclassified base maps It is mostly the modified percentile filter maps that were used to inform mapping and not the base percentile filter maps. The latter were only used for mapping areas where the modified maps did not increase mapping accuracy. The process of creating the modified percentile filter maps is simplified and automated by a script developed specifically for this purpose. The output is a new map with a unique name that incorporates all these modifications and it is the new map that is recorded in the appropriate field of the mapping region attribute. The name of the map is entered to inform the mapping tool to use this modified map to map the wetlands of the valley floor areas for the mapping region; Section 5.4.8). The neighbouring mapping region may require the use of a different combination of base map, threshold and modifications. In such instances, as for the previous mapping region, the script was used to apply the reclassification and modifications to the base map and to again create a new modified percentile filter map with a different name. This newly created modified percentile filter map was then listed as the map to inform wetland mapping for that mapping region. In addition to creating the modified percentile filter maps, the script also creates a text file which lists the base percentile filter, the threshold which is used for reclassifying the base map and the expand and shrink operations used to create the modified percentile filter map. This text file is automatically saved to disk for record keeping purposes.
For many mapping regions more than one modified percentile filter map is required to account for all the wetlands visually identified from the image. This is common for mapping regions that contain wetlands of very dissimilar shape, specifically where they contain large and flat wetlands while also containing wetlands that are long and narrow. All modified percentile filter maps necessary to map all of these wetlands can be entered into the mapping region attribute table separated by an underscore.

Flow accumulation thresholds
Determining the mapping threshold for the flow accumulation maps followed a similar process whereby the most appropriate threshold was determined by trial and error on account of what was subjectively considered to give the best representation of the visually identifiable wetlands. However, unlike the percentile filter maps, the flow accumulation maps were not modified. The thresholds value represents the number of cells of which their surface water will flow towards a particular cell. It follows that if the flow accumulation threshold is set to 1 then all but the outer cells at the top of the catchment of the flow accumulation map will be mapped as they have the surface water of at least 1 cell flowing towards it. A threshold of 1 000 means that only cells which receive surface water from 1 000 cells or more will be mapped. Flow accumulation thresholds were recorded in the appropriate field of the mapping regions attribute table.
While determining the most appropriate maps and thresholds for each mapping region, the most appropriate mapping relationship between the percentile filter and flow accumulation features were also determined. As for the percentile filter and flow accumulation thresholds, the attribute table of the mapping regions also contains a field in which the user can specify the mapping relation between the percentile filter and flow accumulation features. If the option to use multipart flow accumulation features is enabled, then the flow accumulation lines included in the final wetland probability map will be based on a multipart feature selection, i.e. all sections that are connected to any part of the flow accumulation features that intersects a percentile filter features will be selected and will be mapped Figure 5.10 A). However, if the option is not enabled then the flow accumulation features will be considered to be singlepart, i.e. only those sections of the flow accumulation feature that intersects the percentile filter feature will be selected and mapped, while those that do not intersect any percentile filter feature will not be selected and therefore not be included in the final wetland probability map ( Figure  5.10 B). For the purpose of creating the wetland probability map, this option was enabled, i.e. flow accumulation lines were multipart.

Figure 5.10: Comparison of the output for Option 3 with the option to base selection of the flow accumulation features on it being multipart (A) and singlepart (B).
In some instances, it may be found that no combination of percentile filter and/or flow accumulation maps are able to consistently map all wetlands within the mapping region. In such instances the mapping region was manually subdivided and separate percentile filter and flow accumulation thresholds are assigned for the two newly created mapping regions.

Incorporating ancillary data into the output probability map
In addition to having wetland presence and extent predicted through the percentile filter and flow accumulation maps, wetlands which cannot be adequately accounted for using these tools can be captured by way of manual on-screen digitizing. The script that performs the mapping (Figure 5.23) allows for such ancillary data to be listed as inputs to supplement the percentile filter and flow accumulation data. Similarly, vector and/or raster wetland data from other existing sources can be added to the list of ancillary data to be included in the final output. Although available, limited use was made of this option as it contradicts the rapid mapping objective of the method. Ancillary data can be incorporated during the mapping process according to the following options (setting for both vector and raster ancillary data can be set individually for each mapping region):  Map all the vector ancillary data irrespective of whether they intersect the percentile filter or flow accumulation features or not. This was the default option used during the mapping process (A of Figure 5.11);  Map only the vector ancillary data that intersects the percentile filter or flow accumulation features while excluding the ancillary features that do not intersect the percentile filter or flow accumulation features (B of Figure 5.11);  Map all the raster ancillary data irrespective of whether they intersect the percentile filter or flow accumulation features or not; and  Map only the raster ancillary data that intersects the percentile filter or flow accumulation features while excluding the ancillary features that do not intersect the percentile filter or flow accumulation features.

Producing the wetland probability map
The wetland probability map can be created for a division once the mapping thresholds of all mapping regions have been determined and captured. This is accomplished by combining the percentile filter and flow accumulation data per the thresholds as captured in the mapping region attribute table, as well as all indicated ancillary data. The process of combining these data sources is facilitated by an ArcGIS python script which creates the predicted wetlands maps for each division. Processes performed by the Python Script are the following and do not necessarily have a set order: A. The tool loops through the mapping regions one-by-one and performs the following tasks for each mapping region before moving on to the next: a) The tool reads the percentile filer and flow accumulation thresholds from the mapping regions attribute table. b) The mapping region is buffered by 150 meters. All subsequent clipping and selections are per the buffered extent of the mapping region to allow for seamless integration of features of neighbouring mapping regions. c) The flow accumulation map is clipped to the buffered mapping region extent. d) The clipped flow accumulation map is reclassified according to the flow accumulation threshold whereby only flow accumulation cells with value larger or equal to the threshold are retained (assigned value 1) while the rest are discarded and assigned value 'NoData'. e) The flow accumulation raster is converted to a line feature. f) The modified percentile filter maps per the attribute table per the attribute table are clipped to the buffered mapping region extent. g) Raster ancillary data, if listed, are clipped to the buffered mapping region extent. h) Vector ancillary data, if listed, are clipped to the buffered mapping region extent. i) The flow accumulation lines are buffered by a standard 10 meters (top left of Figure 5.12).
j) The mapping relation instruction is applied, i.e. percentile filter and flow accumulation data are selected and retained or discarded depending on whether they intersect each other or not. The option to have flow accumulation be treated as multipart or singlepart is applied during this process.
B. The tool starts the process of combining the features of the different mapping regions. k) The flow accumulation polygons of all mapping regions are merged. l) Overlapping features in the merged flow accumulation output are dissolved to establish a smooth transition of the features over mapping region boundaries. m) The percentile filter rasters are 'merged' to a single raster. n) The merged percentile filter rasters are modified to fill voids (donut holes) that may be present. o) The raster percentile filter data are converted to vector format. p) All vector ancillary data are merged. q) All raster ancillary data are 'merged' and converted to vector format. r) Depending on the ancillary data, setting all or only the ancillary features that intersect either the percentile or flow accumulation, data are selected. s) The ancillary data are merged with the vector percentile filter data. t) The buffered flow accumulation features and the percentile filter features are smoothed. u) All features obtained from the flow accumulation maps are selected and assigned a value 'ST' in the attribute table. v) All features obtained from the percentile filter maps are selected and assigned a value 'VB' in the attribute table. w) The percentile filter features are used as erase feature to remove portions of the buffered flow accumulation features that overlap with the percentile filter features ( Figure 5.12 B). x) The remaining flow accumulation features are merged with the percentile filter features to create a single continuous wetland probability layer for the division. y) A simplification and cleaning process are applied to remove voids (donut holes) and other undesirable features to produce a single seamless wetland probability layer for the division ( Figure  5.12 C, blue features). To create a single seamless wetland probability map for the study area the maps of all divisions are merged and then dissolved on account of the field containing the values indicating the source of the feature, i.e. 'ST' for features derived from the flow accumulation maps or 'VB' for features derived from the percentile filter maps.

Results
The result of subdividing the study area and creating the mapping regions are presented in Figure 5.13, while Table 5.1 provides a summary of the total number of mapping regions per division.  A visual comparison was done as a first estimation of whether the DEM-based approach provides any improvement on NWM4. A comparison of the area covered by each as well as a quantitative comparison was also done to gain more insight into improvements, if any, offered by the DEM-based approach.
Comparisons were limited to mesic areas of high relief in the eastern regions of South Africa as both the DEM-based and the imagery-based method used for creating NWM4 are expected to perform optimally in such areas (Figure 5.14). The visual comparison (Figure 5.15;27° 48' 39.17''S;29° 11' 29.88''E) revealed significant differences between the two maps. The contiguity and extent of valley-floor wetlands can be estimated from the sinuous features in the landscape of Figure 5.15 A. It can be seen that NWM4 has poor representation of the valley floor wetlands with most of the NWM4 features being farm dams (red features in Figure 5.15 B). The wetland probability map shows a marked increase in representation of wetland extent for this area (cyan in Figure 5.15 B). To obtain a more quantitative estimation of the observed differences the area (km 2 ) covered by the different maps were compared (cyan square of Figure 5.14 and Figure 5.16). A visual comparison of the areas cover by the respective maps is presented in Figure 5.16 while Table 5.2 presents a quantitative comparison. (Table 5.2) was done for features located within the area indicated by the cyan square. According to Table 5.2 the DEM-based method resulted in a 201% increase in wetland area compared to NWM4.

Figure 5.16: Differences between the wetland probability map (A) and National Wetland Map version 4 (NWM4) (B) are clearly visible when comparing divisions Gauteng (GT), Mpumalanga (MP), Free State (FS) and KwaZulu-Natal (KZ). A comparison of area
To gain more insight as to whether the DEM-based approach offers better accuracy of wetland presence, the ability of each to indicate actual wetland occurrence was tested. A total of 93 points of confirmed wetland occurrence were collected along a route of 150 km by driving along a road in the north eastern Free State (points per Figure 5.15). The number of points that intersect the wetland probability map as well as NWM4 were subsequently determined and expressed as a percentage of the total number of points (

Wetland probability map NWM4 Buffer (m) Cells (n)
Intersect Percentage  Intersect  Percentage  0  0  31  34  2  2  30  1  51  55  2  2  60  2  67  72  5  5  90  3  74  80  8  9  120  4  78  84  9  10  150  5  81  87  13  14  180  6  82  88  16  17  210  7  87  94  18  19 The results suggest that the wetland probability map offers a significant wetland presence accuracy advantage over NWM4. Most notable from Table 5.3 is the fact the wetland probability map intersects more none-buffered points (31 points; 34%) than 210 m buffered points intersected by NWM4 (18 points, 19%). These results would suggest that the wetland probability map provides improved coverage for wetlands in this and other areas of similar environmental characteristics. However, one must remain conscious of the fact that these findings are not the result of an accuracy assessment, but of a comparative analysis, i.e. if all of the Free State Province was mapped as wetlands then 100% of the wetland probability map would have intersected the reference points, which is an obvious false representation of accuracy. A proper accuracy assessment should in addition to 'confirmed wetland' reference points also include 'nonwetland' reference points so that both omission and commission errors are reported and considered.
A visual comparison of accuracy in other regions was less favourable for the DEM-based approach. Initial indications are that the DEM-based approach had low accuracy in arid regions as well as regions of low relief. Regions that are characterised by these constrains are expected to have lowest accuracy. The eastern region of the Northern Cape Province (division NC_E) is arid but also of high relief. Figure 5.17 (30° 39' 03.41''S; 25° 16' 17.07''E) shows modelled wetlands features (cyan) as well as wetlands from NWM4 (red). As for Figure 5.15, the NWM4 features accounts mostly for farm dams whereas the wetland probability map (cyan features) includes the farm dams but also the watercourses that drain to and from the farm dams. However, unlike Figure 5.15 where the wetland probability map covers areas with high certainty of them being wetland, this is not the case for the features of this area. While the wetland probability features of Figure 5.17 cover some of the watercourses, numerous watercourses not covered by the map are evident (indicated by the blue arrows). There is high certainty that the areas covered by the features of Figure 5.17 are watercourses, however, there is also high uncertainty as to whether they represent actual wetlands or not. The same applies to the watercourses indicated by the blue arrows. Of the indicators for identifying wetlands it is only the vegetation indicator that can be detected with some degree of reliability from imagery. The uncertainty whether the valley floor areas are wetlands or not, therefore stems from the fact that these features are without vegetation. The fact that some of the watercourses are mapped and others are not is the result of a conscious decision to indicate selected areas of the watercourses as possible wetland, while excluding other areas from this possibility. In this case that rationale was that there is uncertainty all around as to whether any of the valley floor areas are actually wetland, but that the majority of the lower parts of the valley floor areas have the highest likelihood of being wetland, and were therefore mapped. It is important to note that exclusion of the remainder of the valley floor areas, whether rightly or wrongly so, is a result of conscious decisions taken by the user and not a failure on account of the method, as these areas would have been mapped if they were considered to be wetland. Different users, depending on what they consider to be wetland from the imagery, will therefore produce different results and it is in instances like this where the expert knowledge of the user is key to producing an accurate map. A common feature of the Northern Cape Province is what is commonly referred to as 'washes' which are low lying areas towards which surface water flows during high rainfall events (black arrow of Figure 5.18). It is clear from Figure 5.18 that these areas are water affected, but it is not clear whether they are wetlands or not as they are similar to the valley floor areas of Figure 5.17, and are often without vegetation. A challenge experienced during the mapping process is that when deciding that an area indicated by the blue arrow is a wetland, is to then motivate why any one of the other valley floor areas are not wetland since these, similarly to the valley floor indicated by the blue arrow, are equally likely/unlikely to satisfy the indicators and criteria for identifying and delineating wetlands (all valley floors of Figure 5.18). Because it is a desktop-based method, assumptions had to be made. When following this rationale then all of the visually identifiable valley floors should or should not be mapped as wetland. This conundrum applies to all valley floor areas in arid regions and highlights the subjective nature of the mapping process. However, this limitation of mapping wetlands in arid regions does not only apply to the DEM-based method but also to other method applied previously as is evident from the NWM4 features (red features of Figure 5.18). The DEM-based approach, however, will allow for the mapping of these areas should it be established that they are wetland or if the objective is to map all watercourses, whereas an imagery-based approach will most probably fail.

Figure 5.18: Because there is no vegetation in the valley floor areas, assumptions have to be made on whether they represent wetland or not. Rightly or wrongly so, these washes (areas indicated by all areas) were not considered to be wetland. All areas indicated by the yellow outline were mapped as wetland for the wetland probability map.
It is subjectively concluded from the above assessments that areas for which low accuracies are expected are those of low MAP (MAP < 500 mm) and low relief (standard deviation of elevation less than 10 m). While low relief relates to difficulties at accurately mapping the valley floors and wetland extent, low MAP relates to difficulties at identifying and recognizing wetlands from imagery. Ratings of mapping confidence were assigned to areas on account of MAP and relief ( Figure 5.19). The categories of confidence rating are relative, implying that a high confidence rating does not imply that the wetlands of these areas are necessarily accurately mapped, but rather that they are considered to be more accurately mapped than regions with a lower confidence rating. A high confidence rating therefore refers to mapping accuracy within the constraints and limitation of the DEM-based method as discussed in Section 5.6.
The confidence rating per Figure 5.19 is a very simplistic representation and does not account for factors other than MAP and relief. Regions of low MAP and sparse vegetation were found to be more problematic than regions of similar MAP but which are better vegetated. Wetlands located within the grasslands of the low confidence regions are therefore expected to be better mapped then e.g. areas of desert or Karoo in the same low confidence region. The converse is also true, i.e. certain vegetation types in the high confidence region are not conducive to the DEM-based approach and wetlands of these areas will be mapped at lower accuracy than the wetlands of other vegetation types. The confidence ratings also do not account for other environmental drivers such as seasonality of rainfall. Differences in mapping accuracy were also reported for the high relief (> 10 standard deviation) category. In some regions, wetlands at the lower end of this category were found to be reasonably well mapped while wetlands at the upper end of this category (i.e. areas of very high relief) were found to be poorly mapped.
Criticism of the existing national wetland map (NWM4) includes amongst others the lack of contiguity amongst and within wetland ecosystems. The DEM-based method was designed to address these shortcomings and the results indicate that the method has to some extent succeeded therein. The isolated and disconnected nature of features in the NWM4 Map (red in A & B of Figure 5.19) is clearly visible as opposed to the modelled wetland probability map (blue), which to a large extent overcomes this defect (B).

Figure 5.20: The Digital Elevation Model (DEM)-based approach (blue features) was to a large extent able to overcome the poor contiguity of wetlands mapped in the National Wetland Map version 4 (NWM4) (red features).
A comparison of spatial accuracy also allowed for the comparison of other technical aspects. Manually mapped polygons, as is required for mapping wetlands, typically suffer from a number of mapping errors such as gaps between features, unintended overlaps, switchbacks, knots and loops. It is evident from casual observations of the wetland probability map that the automated process has to a large extent negated the occurrence of such errors in the wetland probability map.
An objective when developing the DEM-based approach was for it to be able to rapidly map extensive areas with minimum data, skill and cost requirements. Imagery and DEMs were freely available while it took a single person on average approximately 7-9 working days to determine and capture the mapping thresholds for all mapping regions of a division, excluding the time for data preparation such as creating the percentile filter and flow accumulation maps. Although it is acknowledged that the results have suffered from the fact that all mapping was done by a single user, it was possible for a single person to map the entire South Africa within reasonable time at cost. Development of the Python scripts did require skills not common amongst most whom in future will be mapping wetlands. However, once the Python Scripts are made available the mapping of wetlands following the DEM-based approach will require minimum GIS skills of the user. An added advantage of this benefit is that users who have detailed knowledge of wetlands in a particular region will in future be able to easily improve on the results of the mapping exercise using the same method, something that was not possible with previous mapping method.
Although the percentile filter and flow accumulation features are combined to create a single seamless wetland probability map, these remain as individual features within the final output ( Figure 5.20). The fact that they are mapped as separate features allows for the possibility of automating HGM unit typing of these features ) which would not have been possible if all percentile filter and flow accumulation features were dissolved into a single feature. The fact that the features are mapped individually is considered to be a distinct advantage for the initial creation of a wetland map that will allow for the assessment of other wetlands attributes related to wetland function and wetland health, which in turn are required to prioritize wetlands. Wetland priority is a key factor to be included in tools and systems that inform wetland conservation and management decisions such as provincial biodiversity plans. Wetlands are a key component of provincial biodiversity plans. For most provinces, the wetland component of their biodiversity plans is informed by the NFEPA of which the spatial component is informed by NWM4. If the finding that the NWM4 accounts for less than 50 % of the wetlands associated with the valley floor landscape unit is accurate, then it means that priority wetlands were identified without having considered half of the wetlands. NFEPA priority wetlands may not be the priority wetlands, implying that the Critical Biodiversity Areas (CBAs) that are selected on account of important wetlands may have been incorrectly identified as such. The wetland probability map may to some extent contribute towards the spatial component of future wetland priority assessments and therefore contribute towards improved confidence of such processes.
Wetlands are defined as areas which in 'normal circumstances support or would support vegetation typically adapted to life in saturated soil' (RSA 1998:9). Methods employed during previous attempts at creating a national wetland map used multispectral imagery with which to detect wetlands. Only wetlands that displayed sufficient reflectance were mapped which, amongst others, resulted in the poor contiguity of wetlands per NWM4. It follows that the wetlands of which the hydrology has been altered and therefore no longer supports hydrophytic vegetation, but which under normal circumstances would, were not mapped and are therefore not included in NWM4. The DEM-based method assumes such areas to be wetland and is therefore more compliant with mapping wetlands as per the National Water Act (No. 36 of 1998). It is also more suitable for mapping original wetland extent which if combined with wetlands identified from multispectral imagery allows for determining wetland loss.

Image limitations
In-field identification and delineation of wetlands are informed by the indicators and criteria as contained in the field procedure for the identification and delineation of wetlands and riparian areas (DWAF 2005). Of the indicators it is only the soil wetness indicator that is considered to be diagnostic 3 . Although soil maps can be included as ancillary data, the method relies mostly on the visual observation of wetlands from remote-sensed images. This is considered to be problematic, especially so in arid regions where neither the soils nor the vegetation of valley floor areas satisfies any of the wetland indicators.
The ability to identify wetlands from imagery differs from one region to the next as areas with different vegetation structure, moisture regime, topography, et cetera, present different conditions for wetland development and has different reflectance properties. The extent to which vegetation expresses the presence of a wetland therefore differs between mapping regions so that the accuracy with which wetlands are mapped will differ, even if mapped by the same user. Similarly, areas that are frequently inundated within a mapping region are more readily identified as wetland as opposed to areas in the same mapping region that are inundated less frequently. Depending on the quality of the imagery, on-site disturbances and experience of the user, the drier wetlands may or may not be identified as wetland ( Figure 5.21). A certain amount of error is therefore an inherent part of the mapping process.
Deriving wetlands from imagery should ideally be informed by images from different years and seasons to account for environmental and temporal variability. All of the mapping performed was informed by identifying wetlands from satellite imagery of a single season of (2013; Section 5.4.1). It is can be seen from Figure 5.22 that the three years preceding 2013 received mostly above average annual rainfall (indicated by the red line), with the preceding year received below average rainfall (South African Weather Service 2016).
The dry year preceding the imagery used for mapping is not considered to be ideal and may contribute towards omission errors.

DEM limitations
The ability of the percentile filter tool to distinguish valley floors from surrounding slopes diminishes in areas of low relief. The accuracy with which wetland presence and boundaries could be predicted is therefore assumed to be low in flat areas, and more so where these occur in arid regions. Increased accuracy is expected in areas of higher relief and rainfall ( Figure 5.19).
Another limitation of the DEM-based approach is that it does not perform well at mapping very large areas of valley-floor landscape units. The full extent of such areas is often not fully mapped and they are more prone to having voids (donut holes) in the final output.

Other technical limitations
The mapping tool ( Figure 5.23) offers various options which, in general, are expected to improve the output. They may, however, have unintended consequences which in some instances may result in mapping errors.
The option to fill voids (parameter 'Fill voids (donut holes) that occur within a single feature' in Figure 5.23) will fill all voids located within a feature ( Figure 5.24 B). Often the filling of voids is desirable as many of them are the result of the percentile filter and ancillary maps not covering the full extent of the visually identified wetland, especially in extensive wetland systems. However, this option may result in nonwetland areas being filled and therefore incorrectly mapped as wetland ( Figure 5.25 B).

Figure 5.25: Comparison of results with the 'Fill voids (donut holes) that occur within a single feature' option not enabled (A) as opposed to it being enabled (B).
Another limitation of this option is that voids that contain islands of isolated features which are not considered to be empty and are therefore not filled ( Figure 5.26). Contrary to the example of Figure 5.25 such areas of wetland are incorrectly mapped as non-wetland.

Figure 5.27: A indicates all voids before filling compared to B which indicates the voids after filling. Voids that contain features are not filled thereby resulting in wetlands areas being incorrectly mapped as non-wetland (B).
Although the method is aimed at mapping wetlands which are primarily associated with the valley-floor landscape position, it inevitably predicted wetlands of other HGM units, specifically depression wetlands. For the purpose of creating a single wetland layer for South Africa containing wetlands of all HGM units, the latter is considered to be more of an inconvenience as opposed to an error. Ideally the output of a wetland probability map would be able to predict different HGM units, or alternatively an output containing only wetlands that are associated with the valley floor-landscape unit. The use of the percentile filter tool, however, predicts valley-floor-related HGM units such as channelled and unchannelled valley-bottom wetlands, but also includes HGM units found on plains, such as depression wetlands. Further investigations are required to determine whether depressions can be excluded.
As mentioned previously the mapping thresholds represent a best average for all wetlands within a mapping region. Although this allows for the rapid assessment of wetlands, it does imply that wetland occurrence and extent (boundary) of almost none of the wetlands will be 100% spatially accurate. The final output will therefore always to a certain degree be an approximation of wetland occurrence and extent.
It is important to keep in mind that the DEM-based method only maps wetlands of the valley floor landscape position. To create a national wetland map that includes wetlands of all HGM units the wetlands of other landscape units will need to be obtained from alternative sources.

Recommendations for improvement
The 10 m buffer applied is a subjectively chosen width that was considered to represent a suitable average for wetlands of the flow accumulation maps. Wetlands located in the upper catchments are typically narrower than those of the lowlands. The possibility of applying a stream-order analysis to quantitatively inform buffer allocations should be investigated.
Inclusion of a wetland probability map determined by way of e.g. Bayesian statistics or logistic regression  will greatly contribute towards and should ideally be available during mapping. The use of multiple imagery for different years and seasons as well as images derived from other sensors will also allow for improved detection of wetlands to inform the process of setting mapping thresholds. Data from non-imagery related sources to confirm wetland presence or absence should also be considered, such as relevé data from the national vegetation database where vegetation and soil data can confirm the presence or absence of wetland ecosystems.
Although initial testing did not suggest that DEMs of courser resolution (e.g. 90 m and 500 m) provide any advantage over and above those provided by the 30 m DEM, the ability of coarser DEMs to possibly overcome the limitation of mapping extensive wetland systems requires further investigation.
The method relies heavily on the ability of the user to visually identify wetlands from imagery using the vegetation indicator. Vegetation, however is absent in most parts of the arid regions. Research of the valley floor areas of the arid regions to better understand their ecology will greatly improve decision making as to whether these areas should be mapped as wetlands or watercourses.

Conclusion
Development of the method and its application were done simultaneously. Time to collect alternative data sources for indicating wetland absence and presence was therefore limited. However, in spite of this and other limitations it is concluded that the method does show promise for improving on existing wetland spatial data, more so for some regions than others. Although it was not possible to do a detailed accuracy assessment, the wetland probability map is considered to represent an improvement on the existing NWM4 in mesic regions of high relief. The worth of the results in arid regions, especially in regions of low relief, will need to be reviewed. In the absence of a detailed accuracy assessment the use of the results from such areas to inform conservation and management decisions is not recommended. Areas of which the results are subjectively considered to be unsuitable for general use are those from regions that receive on average less than 500 mm rainfall per year and / or that have a standard deviation in elevation of 10 meters or less.
For the remaining areas the results are considered to be an improvement on existing alternatives. Although the accuracy with which wetland extent was mapped in these areas could be improved, the results do indicate wetland presence where many of the alternatives do not. From this perspective the results of these regions are considered to be useful to inform decision making at various levels for areas of high confidence ( Figure 5.19).
The accuracy of the outputs is very much dependant on the effort. Depending on time constraints the user can decide how much effort to increase accuracy that can be afforded. Such efforts will include more percentile filter base maps of different DEM resolutions and different moving window sizes, of DEMs with improved horizontal and vertical accuracy, spending more time testing different modification options to arrive at the best modified percentile filter map, using other environmental variables to inform the mapping regions, obtaining other sources of known wetland location to use as 'training features' during the process of determining mapping thresholds, obtaining more imagery as well as imagery from other sensors and manual subdivision of mapping regions. All of these, along with an improved understanding of what constitutes wetland may result in greatly improved mapping results. Although many of the above stated recommendation will improve accuracy, many of them will result in severe time penalties and the user will need to determine the trade-off between increased accuracy and mapping effort.
Irrespective of the accuracy of the wetland probability map, the logic and rationale on which it is founded is considered to be sound. Mapping errors are mostly the result of DEM resolution, horizontal accuracy of the DEM, time constraints, insufficient scientific data to distinguish wetlands from watercourses in the arid regions, etc. Although these have resulted in unsatisfactory mapping accuracy in the regions mentioned above, future improvements to these current limitations may still prove the approach to be useful in such areas.
Although the wetland probability map is at present being viewed as being the final output of a wetland mapping exercise, its use to support other mapping processing should be considered. Multispectral approaches often use masks to 'filter' out errors of spectral confusion. A map of valley floor areas can be used to distinguish between wetland reflectance values that are likely to be wetland (i.e. those within the valley floor areas), and those that are unlikely to be wetland (i.e. this not in valley floor areas). The method can therefore in future be used to create landscape data to support and inform other mapping process. This chapter provides information on the comparison between fine-scale mapped wetlands and modelled wetlands for a number of areas in South Africa. At least three country-wide modelled wetland datasets were compared for four catchments, one district and three metropolitan municipalities. Recommendations are provided on the use of these datasets for the National Wetland Map version 5 (NWM5).

Introduction
Omission errors in NWM5 remain a concern and are difficult to quantify at a country-wide scale. Where reference datasets exist, a comparison of agreement, omission and commission errors can be done to evaluate the quality and completeness of a NWM. At a finer scale, and where funding for fieldwork can be made available, the level of agreement, omission and commission can be assessed for a corridor along transects, similar to those done by Werner (2004Werner ( -2005. Supplementation of a NWM with modelled wetlands data has been considered previously (Chapter 1), and likewise, the supplementation of NWM5 with the wetlands probability map data, modelled at a country-wide scale, has also been considered and reported here.
This chapter has two objectives: i) to evaluate the agreement, omission and commission errors of NWM5 in comparison to reference data; and ii) to assess whether wetland data modelled at a country-wide scale could be used to address omission errors of NWM5 without introducing more commission errors.

Methods
NWM5 was evaluated against three fine-scale wetland reference datasets, mapped by Wetland Consulting Services Pty Ltd (WCS) for study areas in the Free State (FS) and Mpumalanga (MP) provinces (Table 6.1, see also Appendix H). The comparison was done with a near-final draft of NWM5, version 5.4, which was available on 30 November 2017. No edits to these areas were done between this version and the release of NWM5 for these areas.
The datasets were compared using ArcGIS 10. 3 (ESRI 19993 (ESRI -2014 and an overlay procedure with all the data in the Albers Equal Area coordinate system for South Africa which least distorts surface area. The statistics were summarised in two ways: a) the extent of the wetlands mapped for each dataset was listed in hectares (ha) as well as the percentage (%) of wetlands relative to the surface area of each study area. These statistics provide a general overview of the extent of wetlands mapped, but do not indicate the agreement or disagreement between the datasets; and b) the percentage agreement, omission and commission errors of NWM5.4 relative to the reference dataset and expressed as percentage of the surface area of each study area. Three wetland datasets modelled at a country-wide extent were subsequently considered for supplementation of NWM5:  Wetlands modelled for South Africa by ;  The Global Inundation Extent from Multi-Satellites (GIEMS) (Fluet-Chouinard et al. 2015); and  The wetland probability map (see Chapter 5).
Although other probable wetland mapping datasets have been produced for the KwaZulu-Natal Province, these were not considered for inclusion in the NWM5 . For example, time and budget constraints prohibited the typing of the extent of wetlands modelled for the province (Hiestermann & Rivers-Moore 2015) from inclusion and comparison to fine-scale inland wetland datasets.
A comparison was done between the three datasets which had modelled inland wetlands at a national scale and eight fine-scale wetland datasets from across the country (Table 6.2). This included inland wetlands from the three study areas listed in Table 6.1 which were compared to the wetland probability map. In addition, a baseline of NWM5.4, which was available by 30 November 2017 and included fine-scale data for the three metropolitan municipalities as well as the Cape Winelands District (CWD), was used in a comparison against the wetland probability map, as well as , and the GIEMS data (Fluet-Chouinard et al. 2015). The datasets in Table 6.2 vary in completeness and accuracy. Most were captured at a desktop scale without fieldwork verification and edits. In some instances, such as the City of Tshwane, the data capturing focused on what was attainable within a restricted timeframe and budget. Therefore, many of the reference datasets represent only a portion of wetlands that exist in these catchments, although with high confidence. For most of the municipalities, the latest boundaries from the Municipal Demarcation Board have been used, except for the City of Tshwane where the 2009 boundary was used, because the reference data was mapped to the 2009 boundary. The method of comparison and resulting statistics between the modelled wetlands datasets and the reference datasets, were similar to those used for the NWM5.

NWM5 comparison to reference datasets
All three reference datasets showed that more than 10% of the surface area of the study areas in the Free State and Mpumalanga (grassland biomes), were mapped as wetlands (Table 6.3). For two of the three study areas, the NWM showed an underrepresentation of mapped wetland area by > 5% of the surface area of the study area, whereas in the Vaal River study area there was only a 3% difference in wetlands mapped. NWM5.4 showed a low percentage of agreement relative to the surface areas for the Free State Study Area (1.1%) compared to the Orange River and Vaal River study areas (8-10%) (Figure 6.1). The omission errors of the Olifants River Study Area were the highest (17%) followed by the Free State Study Area (9%) and the Vaal River Study Area (6%). This means that between 30% of the Vaal River, 70% of the Olifants River and 90% of the Free State Study Areas' wetlands of the reference dataset are not represented in NWM5.4. Commission errors, on the other hand, were generally low (< 3%), ranging from the highest for the Vaal River (3%), followed by the Olifants River (1.3%) and the Free State (0.2%) study areas.

Comparisons between modelled wetland datasets and reference data
The extent of wetlands relative to the surface area of the study areas varied from 2% in the arid biome to 25% in some parts of the grassland biome (Table 6.4). On average, 10% of the surface area of most study areas was wetland. The modelled wetland data from  and the wetland probability map (Chapter 5) predicted the surface areas of the study areas to be between 1% and 12% wetland, and on average 6%, whereas the GIEMS data, with an average spatial resolution of a pixel 450 m, ranged from 11 to 28% and on average predicted 11% of the study areas to be wetland.   The comparison between the modelled data from  and the reference data showed a low percentage of agreement (< 2.7%) relative to the surface area of the catchment (Figure 6.2). This accounts for less than 24% of the extent of wetlands mapped by the reference dataset. Omission errors ranged from 3.7 to 8.8% of the study areas, or between 76% and 90% of the wetlands mapped in the reference dataset. The amount of commission errors was generally low (< 2%). The GIEMS data showed a low agreement with the reference datasets (≤ 2%) relative to the surface area of the study areas when compared to the modelled wetlands data of   (Figure 6.3). Even though there is a low agreement, in the City of Cape Town Metropolitan Municipality, it accounts for 43% of the extent of the wetlands mapped by the reference dataset. In the Gauteng municipalities, it accounts however only for < 3.5% of the extent of wetlands mapped by the reference data, but increases to 33% for the Cape Winelands District. The omission errors ranged from 2.6% to 11% of the surface area of the study area, accounting for between 57% and 97% of the extent of wetlands mapped by the reference data. The amount of commission errors was generally low (< 6.5%) except for the City of Cape Town Metropolitan Municipality where 25% of the surface area of the municipality was predicted as wetlands. After visually inspecting these areas it was concluded that they largely coincided with the coastline and the Cape Flats and are unlikely to be wetlands. When the wetland probability map was compared to the reference data, it showed a low agreement (< 8.9%) with the reference data relative to the surface area of the study areas (Figure 6.4). This accounted for less than 58% of the extent of wetlands mapped by the reference datasets. Omission errors ranged from 1% to 18% of the surface area of the study areas, which accounts for between 42% and 74% of the extent of wetlands mapped by the study areas. The Olifants River catchment showed the highest omission error (18%), whereas the omission errors in the other study areas were lower (< 8%). Commission errors were mostly low (< 6.4%) for all study areas.

Discussion and recommendations
The assessment of the representation of NWM5.4 relative to three reference datasets, showed agreement between 8% and 10% for the Orange and Vaal River study areas, relative to the surface area of the study areas, which accounts for only between 30% and 64% of the extent of wetlands mapped by the reference datasets. Wetlands in the Free State Study Area were poorly represented, since NWM5.4 mapped only 10% of the wetlands which were mapped by the reference dataset. The omission errors ranged from 6% to 17% of the surface area of the study areas, which accounted for up to 89% of the extent of wetlands mapped in the reference dataset. The amount of commission errors was generally low (≤ 1.3%) compared to the surface area of the study areas.
Of the three datasets that modelled wetlands for the country, the wetland probability map appeared to perform best: for three of the eight reference study areas, the agreement ranged from 6% to 9% while the maximum commission error was < 6.3% of the surface area of the study areas. All three modelled datasets showed, however, that a large percentage (average 70.4%) of the extent of wetlands mapped in the reference datasets have not been included in the prediction, raising a concern about the capability of the models to predict true presence of wetlands. The  model had the least commission errors (< 2.2% of the surface area of the study areas), compared to the GIEMS data and the wetland probability map. Yet, in Section 4.1 of this report, we indicated the type of commission errors which were present in the NWM4, as a result of the inclusion of modelled data.
The wetland probability map which models watercourse probability would likely address shortcomings of the NWM5 and can be selectively used to supplement NWM5. However, the limited number of reference data available for comparison, yielded a narrow view and understanding of where the data could be valuable, as opposed to adding a level of error that is unacceptably high. It is clear from the statistics of the reference datasets that the extent of wetland area varies across regions and that reference data or an accuracy assessment would be required to determine the completeness of representation of the NWM across the country. Visual inspection in some parts of Gauteng did indicate that the wetland probability map predicted valley-bottom systems which were omitted, however, the wetland probability map overestimated the extent of these systems. The results will be presented to the Wetland Ecosystems Classification Committee (WECC) at the end of January 2018, and the final decision to include the modelled data into NWM5 for the assessment of inland aquatic ecosystems in the National Biodiversity Assessment of 2018 (NBA 2018), will be documented in an update of this report, as well as the NBA 2018 technical reports. The intention is to recommend no modelled wetlands data to be included in NWM5. Nevertheless, these modelled data can be used as a guide for the capturing of wetlands in NWM6. The data should be made widely available by the South African National Biodiversity Institute (SANBI) to interested parties, to use in combination with NWM5 for an improved inland wetlands dataset. This chapter provides an overview of the motivation for a South African Inventory of Inland Aquatic Ecosystems (SAIIAE) and provides details for the appropriate scale and uses of the associated data. A summary of the recommendations of the SAIIAE data is provided, as well as those pertaining to the rivers and inland wetlands.

Introduction
A wide range of strategic, regulatory and reporting processes require a comprehensive inventory of inland aquatic ecosystems, including the National Biodiversity Assessment (NBA), State of Environment reporting, reporting on international conventions (such as the Ramsar Convention, the United Nations Convention on Biological Diversity (UNCBD) and the Sustainable Development Goals (SDGs), systematic conservation planning, Environmental Impact Assessments (EIAs) and Strategic Environmental Assessments (SEAs). An inventory of inland aquatic ecosystems is, therefore, not only just necessary, but also supported by a number of legislative Acts, such as the South African National To date, inventory of inland aquatic ecosystems in South Africa has had a strong focus on the river ecosystems as well as the National Wetland Map (NWM). The capabilities of Geographical Information Systems (GISs) enabled the accelerated capturing, ease of updating, efficient storage and wide dissemination of these datasets to multiple stakeholders. Building on previous efforts, this document provides an overview of the creation of the first South African Inventory of Inland Aquatic Ecosystems (SAIIAE) which includes additional data on the pressures and condition of inland aquatic ecosystem types.
This SAIIAE version and document lists key decisions made between 2015 and 2018 by the inland aquatic team, in agreement with the inland aquatic reference committee of the NBA 2018, as well as the River and Wetland Ecosystem Classification Committees (RECC; WECC) with regards to the mapping of the ecosystem types as well as proposed improvements for future updates and versions. It has become quite clear that the representation of both rivers and inland wetlands, which are small, fine-scale objects, poses a challenge at a country-wide scale. The maps serve multiple purposes at various scales. Considering that the datasets will be used in a variety of applications, we have, therefore, attempted to detail the origins and shortcomings of the SAIIAE as best as possible. In conclusion, we provide guidance on the appropriate use of the products, as well as future recommendations, in this chapter.

Key highlights, findings and knowledge gaps
Key highlights  A South African Inventory of Inland Aquatic Ecosystems (SAIIAE) has been established as a collection of data layers representing the extent of river and inland wetland ecosystem types, as well as pressures on these systems;  The extent of inland wetland ecosystems has been increased by 123% compared to the National Freshwater Ecosystems Priority Areas (NFEPA) wetlands;  Eight unique freshwater lakes (limnetic wetlands) have been identified; and  A confidence map of inland wetlands guides users to appropriate use of the map.

Key findings
 Inland aquatic ecosystems are represented by a river lines dataset as well as polygons of river and inland wetland types in NWM5. Both datasets should be used to represent inland aquatic ecosystems;  NWM5 represents nearly 4 million hectares (ha) of aquatic ecosystems which cover 3.3% of the surface area of South Africa. These include: o Inland wetlands which constitutes > 2.6 million ha or 2.2% of the surface area of South Africa; o > 1 million ha of river channels; and o 201 381 ha of estuarine ecosystems (< 0.2% of the surface area of South Africa).  The rivers dataset represents 200 955 km of river length of which 164 018 km (82%) is situated within South Africa. The majority of rivers in South Africa (90%) have a river ecosystem type assigned; 10% are ephemeral and episodic systems within the arid Northern Cape Province and < 0.1% coincide within an estuary. Mainstream rivers constitute 76 830 km (47%) of the total length of the South African rivers, and tributaries 87 188 km (53%);  Eight freshwater lakes (limnetic wetlands) have been identified where the depth of the water-level at low tide is > 2 m. These systems constitute 13 376 ha of which Lake Sibayi is the largest (Table 7.1); and  Artificial wetlands have been mapped as a separate layer, totalling almost 600 000 ha.

Knowledge gaps
 Uncertainties pertaining to the conceptual ecosystem types for river and inland wetlands implemented in the SAIIAE should be addressed. Research is required to better understand whether these conceptual ecosystem types adequately represent faunal and floral species biodiversity and functional diversity; and  Inland aquatic ecosystems in the arid region are poorly understood. In the rivers data layer these are typed as ephemeral and episodic rivers. In the NWM a high uncertainty of the hydrogeomorphic unit typing is associated with these systems. Further work should be done to better define and distinguish water courses, floodouts and washes in these regions.

Key Messages
 Baseline data related to inland aquatic ecosystems are crucial for planning, conservation and management of inland aquatic ecosystems. Currently the baseline datasets provide a poor representation of inland aquatic ecosystem types, as well as their pressures and impacts. The inland wetlands, for example, showed a 69% low confidence for representing the extent of ecosystems, with an estimated 50% omission error. Confidence and accuracies of other data layers, such as rivers and artificial wetlands, are deficient.

Priority actions
Institutional collaboration across all organisations and stakeholders for the improved understanding, mapping, conservation, monitoring and management of inland aquatic ecosystems should be established and sustained. Responsibilities related to ecosystem datasets should be listed by relevant data custodians under the South African Spatial Data Infrastructure (SASDI) Act No.54 of 2003.

 Research priorities:
o Improve understanding of the relationship between ecosystem types and species biodiversity; o Improve understanding and classification of watercourses, particularly in arid systems; o Broad regional representation of Level 2 of the Classification Systems should be informed by analysis of relevé data from the National Wetland Vegetation Database (Sieben 2015); o Improved national modelling and monitoring of inland wetlands across long-term hydrological regime cycles. This will improve our understanding of ecosystem types and their functions. Currently GIS is used for once-off prediction of extent, types and condition. Time-series should be considered in such modelling. The improvement of freely available remote sensing images at finer scale can support this; and o Develop further methods of accuracy assessment of the extent and types of inland aquatic ecosystems at various scales.  Improve of the extent of river and inland wetland ecosystem types.  Collate additional primary freshwater species data, particularly macro-invertebrate data .  Improve understanding of the association of ecosystem services and ecological infrastructure with ecosystem types (adapted from Nel & Driver 2012).

Appropriate use of the ecosystem datasets
Disclaimer of the data and report: This report was compiled and reviewed by at least 14 experts within the field of inland aquatic ecosystems prior to the approval for publication. Approval does not signify that the contents necessarily reflect the views and policies of the respective organisations, nor does mention of trade names or commercial products constitute endorsement or recommendation of use.
All datasets were amalgamated from previously created datasets prepared for a variety of purposes at multiple scales. Regardless of the attempt made to explicitly state known errors, the complete understanding of representation and accuracy cannot be provided in full. Therefore, we acknowledge that this nationally compiled dataset is not free from errors, be it spatial or attribute accuracy, errors of omission or commission, names, ecosystem types or any other error. Therefore, the authors and designers hold no responsibility in the use of the data and content of the report. The data is intended for countrywide assessment and planning, and no other uses at a regional to local scale. The data and reports can provide context, guidance and support, to inform the improved mapping of inland aquatic ecosystems at a regional to local scale.
The appropriate use of the data is, therefore, recommended as set out in Table 7.2.

Scale
Appropriate use Country-wide 1:250 000-1:1 000 000 1 km spatial resolution National biodiversity assessments and conservation planning  Table 7.3 provides a summary of the recommendations made in the Technical Report of the freshwater component of the NBA 2011 , and extended to the recommendations made in this report. The table summarises information discussed in Section 7.2 of this chapter and elaborates on a number of aspects. Recommendations include those made by the reference committee members of the inland aquatic ecosystems realm of the NBA 2018, the Ecosystem Classification Committees (ECCs) and the National Ecosystem Classification Committee (NECC).  Address uncertainties and knowledge gaps through concerted research funding. Funders should be approached to discuss the advancement of the listed research priorities; and  Improve species representation in the SAIIAE.

River network
 Maintain and update the 1:500 000 rivers dataset. Improved spatial accuracy should be obtained using best available data (e.g. hydrologically corrected digital elevation data at a high resolution), techniques and software. Human capacity and knowledge transfer in this regard is essential at Resource Quality Information Services (RQIS) and the Department of Water and Sanitation (DWS);  Produce a country-wide polygon coverage for the 1:500 000 rivers. The Department of Rural Development and Land Reform -Chief Directorate: National Geospatial Information (DRDLR:NGI) already mapped some river channels and their associated geomorphological features (e.g. flood benches). The DWS has done a hydrological and topological correction of the DRDLR:NGI data and this dataset should be considered; and  Obtain higher confidence in the 1:50 000 river coverages. Their utility continues to be constrained by several inconsistencies (e.g. river coverage densities differ) and inaccuracies (e.g. isolated and incomplete river arcs), imparted by compilation from aerial photographs (Weepener et al. 2015). Future improvements should also include applying the 1:500 000 verification and reach allocation procedures (DWAF 2006) as well as transferring its attributes (e.g. river ecosystem types) to the 1:50 000 rivers; and  Update the flow variability attributes attached to the 1:500 000 rivers dataset. This includes a revision and adding more river flow categories. Flow types for quinary catchments are important. There may also need to be a stronger focus on the water requirements of nonperennial rivers. These rivers cover a large area of South Africa. An update is needed for improving the description of river ecosystem types in the country ).

Catchment boundaries
 Develop and endorse quaternary and sub-quaternary or quinary catchment boundary GIS layers using the latest information and methods. Updates of these data layers were done for example by Schulze & Horan (2010), Weepener et al. (2012), and ; and  Develop sub-catchment planning units that take into account altitudes and where possible change in functional hydrology may occur.

River ecosystem types
 Improve and validate the landscape classification map with data. Consider different approaches such as the Level 1 ecoregions (Kleynhans et al. 2005)   Validate the river ecosystem types with species data. These subtypes are components of rivers with similar physical features and are an essential coarse-filter biodiversity surrogate. However, knowledge on whether the river types identified are associated with species representing biodiversity for persistence and evolution, remains lacking ;  Consideration of new approaches to the classification of geomorphological or longitudinal zones developed for both perennial and non-perennial river systems (e.g. Jaeger et al. 2017). This will influence the extent of the river ecosystem types. In addition, explore ecologically meaningful lumping of the geomorphological zone categories by extending the Dallas & Day (2007) macroinvertebrate analysis to include fish and vegetation data;  Map and identify riparian zones and their vegetation along rivers (both mainstems and tributaries) on a national scale. Riparian zones are not represented in the river ecosystem type layer. This data layer should be incorporated in river typing procedures. Buffers of vegetation surrounding all freshwater ecosystems, even heavily used ones, go a long way to reducing the effects of harmful land-use practices ; and  Publish a river ecosystem type map. This product should resemble the Mucina & Rutherford (2006) vegetation map, with descriptions and characteristic species attached.

River condition
 Improve skills and monitoring capacity for river ecosystem condition updates. This process is also linked to strengthening collaboration of DWS and Department of Environmental Affairs (DEA) around managing and conserving inland aquatic ecosystems;  Strengthen river inventorying and monitoring programmes. Strategic field sampling sites and metrics should be chosen and an implementation plan (including a financing plan) for monitoring these sites over time should be developed . The River Health Programme (RHP) (now the River EcoStatus Monitoring Programme [REMP]) offers an ideal platform from which to begin. These REMP sites should be reassessed, refined and supplemented according to an explicit set of criteria that also includes consideration of wetlands and other new information, e.g. the Strategic Water Source Areas (SWSA);  Explore all datasets (e.g. Reserve and Classification data) that could be used for a national assessment. Collation of these datasets should be a medium-term goal.
 The desktop Present Ecological State (PES) database must be updated with higher confidence data (i.e. provincial to local level PES data) as it becomes available. It is based on expert opinion. A confidence rating (e.g. mean rating per ecosystem type) could be applied to improve the accuracy of the data; and  Develop biodiversity indices that speak directly to biodiversity (i.e. of which macroinvertebrates would be one, and fish, riparian vegetation, diatoms, etc.). This is related to developing a comprehensive fish atlas and a solid database of macroinvertebrates for use in assessments and planning.

NWM5
(extent of inland wetlands)  Improve the representation and hydrogeomorphic units to at least a Low to Medium confidence level for 75% of the country within the next 5 years. These can be achieved through continuous desktop improvement of NWM5, supplemented with the cleaning and typing of the wetland probability map. The NWM5 confidence map should be used to identify areas where the wetland probability map should be used;  Prioritise certain areas where the NWM needs higher levels of confidence. Set a minimum target for the minimum confidence, omission and commission errors as well as minimum requirements for the mapping and in-field visits for these priority areas. Consideration should be given to the SWSAs identified in the Assessment report of inland aquatic ecosystems as threatened and poorly protected, Ramsar sites, and other areas where development pressure is high;  Continue to develop and compare the capabilities of remote sensing and GIS probability modelling to map the extent of inland aquatic ecosystems. The results should be evaluated against fine-scale datasets. Modelled data should be considered to supplement NWM5 only in data-poor areas; and  Improve the extent of artificial systems and consolidate categories in the polygon data layer. Ensure that systems represented in point files (e.g. Aquaculture and the Waste Water Treatment Works -WWTW) are fully represented in the polygon data layer. Inland wetland ecosystem types  Continue to develop and compare the capabilities of remote sensing and probability modelling to map the ecosystem types of inland aquatic ecosystems. The results should be evaluated against fine-scale datasets. Modelled data should be used only in data-poor areas.

Condition of inland wetlands
 Condition modelling of wetlands should be attended to with urgency, particularly identifying datasets which can best be used to represent existing pressures on inland aquatic ecosystems.  Assessment: only Levels 2 and 4A should be used for assessing biodiversity pattern.
 The improvements of L2 is essential using the NWVD.
 A target threshold of 20% will be retained for assessing ecosystem threat status until further research has been done on species responses and other outcomes which would inform other target thresholds for wetlands.
 Grouping of HGM units: 4 to 3 groups can be considered, but first present the results of the assessment, before the reference committee will decide.  Wetland flats in most of the provinces would be depressions, since they may receive surface contribution from adjacent seep wetlands. These are not purely groundwater driven systems. Wetland flats from the C.A.P.E. project was also desktop-typed and have a low confidence. These should however be kept as wetland flats in the NWM5.
* WRC project K5/2549, Project lead Mr D Ollis; 'Developing a refined suite of tools for assessing the Present Ecological State of wetland ecosystems.'

Introduction and purpose
The National Wetland Map 4 (NWM4) resulted from the modelling of wetland types from the extent of wetlands in the NWM3 for the National Biodiversity Assessment of 2011 (NBA 2011) and National Freshwater Priority Areas Atlas (NFEPA)   . A flow diagram of the integration of the datasets up to NWM3 is available on SANBI's Biodiversity Advisor site ( Figure B1). A number of errors in the NWM4 became evident subsequent to the publishing of the data and user's working with the data. These included some key errors:  The inclusion of the predicted wetlands introduced a large number of commission errors (although not indicated in Figure B1, this dataset was included);  Artificial wetlands such as tailing impoundments were classified as natural; and  Polygons around dams were initially thought to be natural palustrine type of wetlands however later considered to rather be dams and should be merged with the polygon of the dam.
In addition a number of projects indicated that the NWM4 represented < 54% of the wetlands found at fine-scale level .
To address the errors and shortcomings of the NBA 2018, Heidi van Deventer and Jeanne Nel (CSIR) decided that it is better to start from scratch and create two separate products for the NBA 2018: i. A fine-scale dataset which will include the NGI data and any fine-scale data that was mapped through heads-up digitising from orthophotos. ii.
A dataset of predicted wetland extent and possibly types.
The typing of the Freshwater Ecosystems will be done according to the seven hydrogeomorphic units of the Classification System ).
In addition to the extent and types of the wetlands, the landscape unit (landforms) will be updated to improve on previous limitations (Van Deventer et al. 2014).
Sub-versions of the National Wetland Map 5 (NWM5) as well as associated data in the National Freshwater Inventory (NFI) will be made available during the course of the project on the Freshwater ecosystems page, to allow all users to provide feedback and suggestions to Heidi Van Deventer (HvDeventer@csir.co.za) for improvement. The freshwater page can be accessed via http://gsdi.geoportal.csir.co.za/projects/national-biodiversity-assessment-of-2018.
The aim of involving stakeholders during the progress of the work allows the best quality is delivered for the NBA 2018. The planned update is listed in Table B1 with the planned work flow of the project in Figure B2. Called feature class National_Wetland_Map_5_2 The data will from here on be split into provincial datasets for attending to slivers, checking names and assigning HGM units to a polygon.
The fine-scale datasets will be unioned as per Section 3 below.

5.3.
This will include the following datasets: Working for wetlands*  Selected peatlands data*  Estuaries functional zones*  Springs and thermal springs* At provincial level the data will be first cleaned as a separate dataset.
The fine-scale data obtained from other sources will be cleaned as separate datasets. The two datasets will be unioned and the Level 1 and Level 4A fields of the National_Wetland_Map_5 updated.

5.4.
Cleaned NWM5 for desktop and fieldwork validation.
The predicted wetlands and landscape features (L3) will be issued as separate feature datasets to be validated with NWM5.4.

5.5.
Cleaned NWM5 including desktop and fieldwork validation. To be completed 31 March 2017 5.6. A possible integration between the fine-scale wetlands of NWM5.5 and the wetland extent and modelled data *The CSIR and SANBI's data audit reports will document the full details and references of these datasets.

B1. Step 1: Pre-processing and preparation of the DRDLR:NGI 2006 and 2016 data as a base layer.
a. Both NGI 2006 and NGI 2016 data were imported to a geodatabase file as feature classes in the projection Albers Equal Area for South Africa and topologically cleaned of overlapping polygons. Table B2 indicate field names used in the feature class. b. The two feature datasets were unioned into NGI_2006_2016_AEA which had four fields (Table B2). The records were exploded to ensure that no multi-part polygons were present, however the records remained at 613 296 polygons (which includes slivers). c. The hectares were calculated and a frequency done on the NGI2006_FT and NGI2016_FT fields to assess how many contradictions exists between NGI2006 and NGI2016. The results are tabled in Tables B7-9 with matrices indicating similarities between the two datasets (grey cells) as well as differences. Differences were addressed in the steps below. d. Two new fields were added NGI0616_FT and NGI0616_NAME (Table B1) and the feature types (FT) from various fields consolidated to these fields. Similarly, names were consolidated to the NGI0616_NAME field. The following steps were taken to integrate the feature type classes through a sequential process of elimination to NGI0616_FT while names were integrated to NGI0616_NAME through each selection step: i. First, all feature types classes which were similar between 2006 and 2016 (grey cells in

B3. Step 2: Inclusion of other national datasets
A first level integration was done between the NGI_2006_2016_AEA_D2 feature class and a number of national datasets. A union was run with the NGI data and the following polygon feature classes:  The Working for Wetlands data captured as polygon data between 2004 and 2007 and available from SANBI's Biodiversity Advisor website (BGIS), was also included.
 The NGI springs data were buffered with 1 m to indicate possible seep wetlands.  Dr Althea Grundling and Joseph Mulders (Prime Africa) provided Heidi van Deventer with an intermediate peat point and some polygon *.kml files dated to end July 2016 as part of the National Peatland Database (WRC project K5/2346). The polygons were unioned into the other polygons, while the points will be used to guide potential locations of wetlands where C > 2 (organic soils).
The output feature class was called NWM52. The estuaries were then erased from this file (NWM52_erase_estuaries) and then the estuaries were unioned with NWM52_erase_estuaries to create NWM52_with_estuaries.
 The Estuarine Functional Zone polygons were received from Lara van Niekerk (CSIR) on 26 July 2016. These were cleaned topologically. The file is still being updated for the NBA 2018, and the capturing of micro-estuaries may have to be included in future updates of the NWM5.
The NWM52_with_estuaries layer was renamed the National Wetland Map 5.2 feature class in the geodatabase. All records were selected and exploded to ensure no multipart polygons exist.
A number of fields and associated subtypes were added for the purpose of standardising the National Wetland Map 5.2 to the levels of the Classification System  and to allow recording of ancillary data (Table B3).  Next the Working for wetlands data was used to identify CS_L1 as 'Inland -natural' and specific hydrogeomorphic units to CS_L4A, and some to Level 6 (Table 4).
o Where the NGI already indicated the wetland as artificial, the CS_L1 was kept at 'Artificial'.
o Existing categories of NGI were not overwritten by the W4W categories, particularly the artificial wetlands, and river and depression wetlands. For example, where the extent of a river channel was mapped in the NGI as a river HGM unit, but classified as an HGM unit other than 'River' in the Working for Wetlands data, the CS_L4A was kept as 'River'.

B3. Step 3: Integration of the provincial datasets
Following the compilation of the National Wetland Map 5.2, the data will be clipped to each province and allocated to a number of people to refine and validate (Table B5)  A number of people are also involved in the validation of the oxbow points and waterfall points in August 2016 (Table B6). Fine-scale data received from institutions is listed in the situation assessment and data audit reports  being compiled by the CSIR (Heidi van Deventer) and SANBI (Namhla Mbona). These reports are available at the respective institutions. Fine-scale data received will be topologically cleaned, cross-walked and then unioned with National Wetland Map 5.2 to create National Wetland Map 5.3 version. SANBI has secured funding to appoint a number of freshwater ecologists to undertake desktop and/or fieldwork validation of the data. Figure B2 shows how a number of processes will run parallel in the project and the approximate deadlines for the project. The files were imported into ArcGIS 10.3 as provincial feature classes in a feature dataset. The provinces were then merged into three feature data classes (Table C1). The intention was to union the three feature classes into a single national hydrological dataset, however owing to topological problems, the data errors had to be fixed first. The issues are therefore reported to the data custodian for improvement as per the SASDI Act.
(A) Overlapping polygons A total amount of 3 961 overlapping polygons have been detected in the data (3 rd column of Table C1; Figure C1). A large percentage of the polygons were overlaps of polygons with identical geometry and the duplicates were therefore removed, leaving 831 polygons which overlapped (last column of Table C1). These have been subsequently cleaned and edited by the CSIR and a clean version of this dataset will be sent to NGI in Pretoria.     polygons, extend or correct the boundary, HGM unit (Level 4A of the classification system) or condition, or merge the sliver with the adjacent polygon but do not delete it.
 Fill up the gaps in a polygon. In some instances, islands in a wetland were excluded or deleted, however it is part of the functional wetland unit and must be included and merged to the HGM unit. Should an inner polygon be a dam inside a river, capture the polygon and classify it accordingly.
 Wetlands should be captured at a scale between 1:500 and 1:2 000. One may browse through an area and attempt to detect wetlands at a scale between 1:2 000 and 1:5 000, depending on the region and size of the wetlands. The wetland can be captured roughly at these scales, however zoom into the wetland to reshape, cut and edit the wetland at a scale below 1:2 000.
 Always complete the following classes: o Identify Level 1 of the Classification System for Inland wetlands (CS_L1) of the polygon as any of the three types listed below. Codes are available from the dropdown option in the attribute table as per Table D1.
o Level 2 of the Classification System for Inland wetlands (CS_L2) will be unioned / integrated at a later stage and should not be completed by the data editor.
o Level 3 of the Classification System for Inland wetlands (CS_L3) will be guided by the slope categories. The L3 class will be assigned based on the majority overlap of the wetland with a slope, plain, bench or valley floor.
o Most importantly, ALWAYS capture the wetland type of the Level 4A or Hydreomorphic (HGM) unit as one of 7 types. In this field, artificial wetlands and estuaries are included to avoid these wetlands being classified as one of the 7 HGM units. Should you not be able to identify the wetland type, list it as 'Unknown'. The codes and descriptions of the HGM units are as per Table D1.
 Always add the date of the image used for capturing or desktop validation of the extent or type of the wetland. Where multi-season images are available, a comparison should be made between the images, the maximum extent of the wetland map, and the respective date of the images recorded. In this regard, the most recent imagery doesn't necessarily indicate the widest extent of the wetland.
 Always add details on the editing and/or validation in the following fields: o Data_editor: Person's full name who edited the polygon o Edit_date: Day-month-year of the date on which the polygon was edited o Val_type: the type of validation done for the polygon. Code options are listed in Table D1.
o Val_person: Person's full name who validated the polygon o Val_date: Day-month-year of the date on which the polygon was validated  Use the topographical maps to confirm the name of the wetland and correct spelling where necessary to the field [CS_NAME]. All names have been checked by Heidi van Deventer from 3 to 5 August 2016. However, a second check is always valuable. Some errors crept in where the name 'Augrabies' was assigned to a multipart polygon. All such errors should all be corrected. Do not edit any of the previous fields with names at all. It is also important to attend to the following naming conventions and other issues: o The name of the wetland should be consistently spelt in Sentence case.
o English names are usually spelt as two separate words with both starting with a capital letter, e.g. Aston Dam, Orange River, Jackson Reservoir, Heywood Dam, Albert's Falls Dam or Zoo Lake.
o Afrikaans names are usually spelt as one word. Where more than one word, only the first letter of the first word is capitalised and not the others, except where it is a name of a person. Examples of the correct spelling are Blesbokspruit, Stormsrivier, Oranjerivier, Bokdam, Seekoevlei, Hans se leegte, Gert Josop se vlei, Geelpan.
o Rivers change names at confluences or along the longitudinal extent. Be careful not to merge polygons of rivers with different names.
 All data for the NBA 2018 project will be in Albers Equal Area WGS84. Do not change this coordinate system. Please work in this coordinate system. Data will also be made available in Geographic WGS84 for external people to validate the data and ease the compatibility with QGIS and other open source software systems.
 Unknown and Unspecified is used in each field to identify the option: o Unspecified -the polygon has not been attended to yet by anyone o Unknown -the data capturer needs guidance and confirmation from the freshwater ecologist whether the polygon should be deleted or identification of the HGM unit. No subtypes were given here. This field should be populated later with the regional setting used in the NBA2011. CS_L3 The field should be populated based on the landform which fills the majority of the surface area of the polygon. The polygons should not be split. Day-month-year of the date on which the polygon was validated *CS refer to classification system ) **Hydrogeomorphic (HGM) units

D3. Steps to take in editing provincial datasets
The following steps should be taken sequentially in the editing of the provincial datasets extracted from the National Wetland Map 5_2 feature class.
The slope should be visually displayed as three categories: 0-1°, 1-3 ° and > 3 °. This can be used to inform the HGM unit. The specific threshold for the cut-off between slope and bench/plain has not been tested thoroughly and is therefore merely a given as a range to guide you. 4. When Merging polygons, first select the polygons, then zoom to the selected and ensure it is not a multi-part polygon, then press merge. If it is a multi-part polygon, then explode the data first, before restarting the selection and merging process. a. Do not merge polygons where other Levels of the classification system separate the polygon based on substratum or vegetation. 5. If you are allocated a province with estuaries, attend to the estuaries first. Where small polygons occur adjacent to the estuaries polygons, merge them into the estuary. Field [ESTUARIES_Name] show the estuary name and field [ESTUARIES] which polygons are Estuary Functional Zones. Zoom to the estuary, start editing, select all the polygons that are classified as 'estuary' as well as small adjacent polygons, and merge the polygons into one. 6. Visit each peatland point and see if you can recognise the full extent of the wetland, capture the necessary and associated fields. 7. Edit the remaining polygons by going consistently through the provinces. Use the 1:10 000 gridlines of the NGI to guide you. Browse for example from top left to right first, and then down and then right to left again till the tile is completed. 8. Look for slivers and inconsistent naming and merge these polygons. Figure D1, for example, shows slivers on the edge of a pan. In 2006 it was called 'Bitterput se pan', while in 2016 it was called 'Bitterfonteinpan'. All the polygons were selected and merged into a single polygon called 'Bitterfonteinpan'. 9. The location of floodplain wetlands can be indicated by 3 features: the presence of floodbanks in river areas from the NGI field, or the location of oxbow rivers and highly sinuous river lines. Oxbow rivers were captured as points and should be displayed in your screen. Sections of the 1:500 000 river lines which are highly sinuous were also extracted and should also be viewed. 10. River channels should be captured as separate polygons from the adjacent wetland. Only a selection of the highest river stream order numbers ( Figure D2) for each province (orders 7-4, maybe 3) should be mapped, not orders 1 and 2. The data is available in the geodatabase, see field ORDER. 11. Wetlands adjacent to river channels, could be either floodplain areas, CVB wetlands or seeps. If there is no evidence of a floodplain or seep, it could likely be CVB wetlands, for example, polygons adjacent to the Mathwaring River ( Figure D3) should be classified as CVB wetlands. 12. Small polygons adjacent to a dam should be merged with the dam. 13. Polygons on top of other polygons should be cleaned with the Clip option. 14. Many errors crept in during the dissolve operation and now a large majority of polygons in across the country appears to be Depressions but are in fact artificial wetlands. The images should be used to see whether they are not in fact ART wetlands and all of these correctly classified. 15. A number of ancillary datasets should be used to guide the mapping and classification: a. The 50 cm colour orthos of the NGI is available from the ArcGIS online service. Display this in the background.
b. Waste Water Treatment Works (WWTWs) and Water Treatment Works (WTWs) data is available to see where artificial wetlands are. Capture these as polygons and classify L1 as Inland -Artificial, and L4A as ART. c. The slope classes: depressions, flats and floodplain wetlands would usually form on low slope. Valley-bottom and seep classes on slopes (> 1 or 3° of slope). Rivers could occur on both. d. The 1:500 000 rivers can be used to identify river channels. These should be captured as separate river wetland types. If a channel is visible between wetland polygons, it could either be a floodplain wetland or a channelled valley-bottom wetland. Use the oxbows and slope < 1° for floodplain wetlands and the slope > 1° or 3° for channelled valley-bottom wetlands. e. The springs data should be used to identify seep wetlands in the landscape. The springs' data was buffered with 1 m and included as polygons in the National_Wetland_Map_5_2 feature class under field [NGI_springs] to guide you to confirm the location of this wetland. If one is unable to recognise the wetland, label is as the default 'UNKNOWN'. If you have mapped it, though can't identify the wetland type, classify it as 'UNSPECIFIED'. f. It is advised to look at the thermal springs point data, visit each point in the province, and if it is possible to detect or recognise the thermal spring feature, capture it as a polygon, and classify it as a SEEP under L4A. g. The Working for Wetlands data was also integrated in the National_Wetland_Map_5_2 feature class showing the extent and HGM unit of the wetland. The field [WETLAND_NO] shows the code of the wetland rehabilitated, with [W4W indicating the HGM unit that was identified during fieldwork. The wetland should be modified to include the adjacent polygons and the L4A category kept as indicated in the Working for Wetlands field. h. We are still expecting the Aquaculture farms' data too, and this will have to be captured and classified as artificial wetlands and Aquaculture ponds at Level 6.
Please sign below to indicate that you understood the principles and rules and agree to follow these during your work on the NBA 2018.   Looking at the timeframe and resources available, we prioritised certain areas to be mapped in this revision of the NWM. The selected district municipalities were mapped on desktop by junior data captures. The wetland datasets will be reviewed by SANBI according to the criteria listed in Table F3.

APPENDIX E: DATASETS USED FOR THE COMPILATION OF NATIONAL WETLAND MAP 5 (NWM5)
Following that it will be passed to the wetland specialist appointed for the area to be further reviewed according to the same criteria listed in Table F3 with additions of any other comments. Consultants can add important systems that have been obviously missed and also correct the typing if time is available. The wetland datasets captured by the various data capturers for priority districts in South Africa, will be reviewed according to a number of criteria (Table F2). Historical and current imagery in Google Earth, the National Geospatial Information's 50 cm colour orthophotos available online and SPOT imagery from January 2012) will be used to assess the extent and hydrogeomorphic units of wetlands in a systematic manner. Hydrogeomorphic units will be reviewed visually against existing fine scale and with also the use of ancillary data (e.g. contours, DEM) as specified in Van Deventer (2016). Attached is a shapefile containing points of areas with spatial errors to be fixed (filename) areas with over mapping, areas with under mapping and areas to be verified in field. Also please attach a shapefile for wetland typing errors.

F2. Review by wetland specialist
This section is for the data review by wetland specialist. The datasets will be submitted for each area to the specialist assigned. The project has tight timeframes and not many days within each contract. Specialist should strike a balance between comments to be sent back to data captures and the quick issues that can be fixed.

CHECK PRESENCE/ABSENCE
 Check that wetland areas that have been omitted and areas incorrectly identified as wetlands.  Check for any NWM5.2 polygons that were mistakenly deleted by the new mapping, verify if they are wetlands and, if so, include into the new mapping.  Delete polygons marked delete (Column NWM5.2_L4A) by the data capturers if you agree with them. Areas of deletion should be captured as points shapefile, so that in future iterations of the wetland map they are not to be captured again.  Use fine-scale wetlands data available for the District Municipality to check for any polygons that were missed by the new mapping, verify if they are wetlands and, if so, include into the new mapping.  Use all available fine-scale wetlands data, specifically the artificial wetlands, a) to check if any new mapping inadvertently mapped a known dam as natural wetland and b) to verify mapped dams -allocate these as high confidence (Table 4).  The NGI (for certain years) data has been used to build the NWM 5.2 as documented on Van Deventer (2018). This geodatabase has been used as the base layer in the desktop mapping process. In the NGI data it seems some years are more accurate than the other for specific areas; therefore, it can still be used in some reviews. Use NGI dataset, specifically the perennial and non-perennial pans, a) to check if any known depressions were inadvertently deleted in the new mapping and b) to verify any corresponding mapped polygons to be depression HGM unit -allocate these a high confidence to the x category of confidence.  Make use of the available site-specific delineation of what mapped by wetland specialists to a) align boundary of new mapping and b) adjust any corresponding mapped polygons to be the same HGM unit as the Working for Wetlands (included in the NWM5.2) mappingallocate these a high confidence of the x category of confidence.

Confidence level Description High
Wetland delineation reviewed by at least one wetland specialist and either groundtruthed or verified using existing high confidence datasets.

Moderate
Mapping outputs reviewed by at least one wetland specialist. Low Mapping outputs not reviewed by an expert.
 Dam and reservoirs (1:50 000 scale dataset) (DWS 2015a);  National dams register -a spread sheet with details of all registered dams with build data and dimensions (DWS 2015b).

Summary:
As part of the development of NWM 5.2 a draft artificial water bodies dataset was built using the Chief Directorate: National Geospatial Information (CD:NGI) hydrological features 2016. All nonnatural feature types were included in this feature class and were checked for consistency and then integrated with the national 1:50 000 dams map (DWS 2015). This formed the foundational layer for the development of the artificial waterbodies dataset. Due to the large number of features in the dataset, we focussed on identifying and correcting 'gross errors' including large features that were incorrectly included in the artificial waterbodies dataset and features that were incorrectly mapped or were features that were missing from the dataset. The correction of errors continued throughout the development of the dataset using a wide range of data sources and personal communication with regional experts.

Foundational data
As part of the process of updating the National Wetland Map 5.2, the following topographical features, identified as being man made or artificial in the CD:NGI 2016 hydrological features were selected and exported to a new feature class: water treatment plant, sewage works, fish farms, reservoirs, dams. The DWS 1:50 000 dams data were added to the new feature class for artificial water bodies. All features and attributes were retained.

Error checking
a. Very large features in the foundational dataset which consisted of the top 100 in terms of extent were manually inspected, and a number of natural wetland or dry river channel areas in the Northern Cape were found to be incorrectly classified as dams. These features were removed.
b. Zonal statistics tables were then calculated for all artificial water bodies features using percentage water occurrence (Pekel et al. 2016) and percentage 1990 water and percentage 2014 water (GTI 2016). The premise was that if a feature had a low percentage of surface water occurrence it was likely to be a miss-classified wetland rather than an artificial water body. All features with low water occurrence (less than 10%) and extent of over 25 ha were visually inspected using 50 cm colour orthophotography (ARCGIS online, CD:NGI, circa 2013) and satellite imagery (SPOT-5 and SPOT-6 country mosaics in false colour circa 2006,2009,2016). In total 350 individual artificial water body features were inspected and corrected where required. The artificial wetlands class from the national wetlands database used in the NBA 2011  was then compared to the dataset using the ARCGIS Erase command large NBA 2011 wetlands that did not overlap with the artificial wetland dataset were identified and inspected visually, and a number of large features greater than 25 ha in extent) were added.

Adding dam build dates
To make the dataset useful in land cover change analyses the build date of the dam or facility was added where ever this could be determined. The main source of information for this was the Department of Water and Sanitation (DWS) national dams register that includes information such as build date, latitude and longitude, dimensions etc.
The dataset was converted to a point shapefile (event theme) using the latitude and longitude in the spread sheet. No indication of datum was provided so for older features there is a risk of a datum shift of up to 300m due to use of Cape datum in the original dataset. The points were then linked to the artificial waterbodies database using the Nearest command in ArcGIS. The dam extent from the DWS database was compared to the Geographical Information System (GIS) extent and where these were different the dam was visually inspected and corrected where necessary. All dams with very low water occurrence percentages were inspected and corrected were necessary. To align with the habitat modification time series data, dams were assigned pre-1990 or post-1990 build dates based on the available information. Dams for which no build date was noted in the register were visually inspected, and using a combination of the global surface water dataset, the national water features split data, and topographic maps dams likely to have been built post 1990 were identified. This process also exposed numerous errors spatial errors and classification errors in the dataset which were corrected as part of the process.
A simple vector geo-database resulted, with the attribute Type [dam, sewage works, fish farm, reservoir, water treatment] and BuildDate [pre-1990, and post-1990, unknown]. This polygon feature was re-projected to the Universal Transverse Mercator (UTM) 35N coordinate system and converted to a 30m raster, snapped to the habitat modification raster, and then combined with the habitat modification map. Artificial water bodies of unknown build date were assumed to be built pre-1990, and were reclassified as such.

Final product
The final map contains just over 200 000 features, covering an area of 597 324 ha, with dams making up the bulk of the artificial water bodies in South Africa (   VB probability data overestimates wetland extent in comparison with fine-scale data by 167%.  The VB probability map has poor spatial accuracy in this area, with only 17.1% of the VB probability map coinciding with the fine-scale data, and roughly 82.9% of wetland area being potentially non-wetland.
 54.4% of fine-scale wetland area is not mapped by VB probability data.  The VB probability map is flawed in terms of wetland extent and spatial accuracy and should not be used for this particular area.  VB probability data overestimates wetland extent in comparison with the digitised NWM 5.4 by only 10.5%.

Cape Winelands
 However, the VB probability map has poor spatial accuracy in this area, with only 37.3% of the VB probability map coinciding with the digitised data, and roughly 62.7% of wetland area being potentially non-wetland.
 58.8% of fine-scale wetland area is not mapped by VB probability data.  The VB probability map is relatively accurate in terms of wetland extent, however spatial accuracy is flawed.
 The VB probability map should not be used for this particular area.