The Pongola Floodplain , South Africa – Part 1 : Two-dimensional hydrodynamic modelling in support of an environmental flows assessment

e Pongola Floodplain in the Makhathini Flats is an area of low topographic relief between the 1973-commissioned Jozini Dam, and the Usuthu River which borders Mozambique. e oodplain system is characterised by a complex mosaic of meandering river channels, levees, and oodplains interspersed with pans (or depressions) and wetlands. e landmark 1982 study of the oodplain, Man and the Pongolo Floodplain, suggested a pattern of ows to ‘maintain the oodplain’ based on socio-ecological criteria. Since 1998, however, annual releases have been primarily targeted at the needs of recession agriculture and inundation of the oodplain in the Ndumu Reserve. No releases have been speci cally aimed at maintaining the oodplain ecosystem and the services it delivers to support the livelihoods of local communities. In 2013, the Department of Water and Sanitation commissioned an Ecological Reserve study of the Usuthu/Mhlatuze Water Management Area, which incorporates the Pongola Floodplain. is paper describes two-dimensional hydrodynamic modelling using RMA2 to inform this ow assessment. Four computational Pongola Floodplain models have been developed since 1979, including cell-based, oneand two-dimensional approaches. e RMA2 model is based on existing topographical, hydrological and hydraulic information, and was calibrated and veri ed for the period 2008 to 2010 using water-level data from the local hydrometric monitoring network. Generally, good replications have been achieved in terms of peaks, rising and recession limbs, recession of ponded pan water-levels, and lowow river stages. e RMA2 modelling represents an advancement of previous hydrodynamic studies of the oodplain and contributes to an improved understanding of its hydraulic behaviour. Model application was for the 15-year period 1990 to 2004, and simulations included naturalised, present management (2014), and 7 potential dam operational scenarios. e results were post-processed for analyses in the DRIFT DSS, described in the companion paper.


INTRODUCTION AND BACKGROUND
e Pongola Floodplain is an area of low topographic relief in the Makhathini Flats, northern KwaZulu-Natal, bordered by the Usuthu River and Mozambique to the north, the Lebombo Mountains and Swaziland to the west, and the Indian Ocean to the east (Figs 1 and 2).From Jozini Dam (Fig. 3) the Pongola River ows in a north-easterly direction to its con uence with the Usuthu River.e river's average longitudinal gradient is 0.039% over a total length of approximately (~) 139 km and, being characteristic of a low-gradient channel, displays a meandering planform across the ats, as illustrated in Fig. 4.
e oodplain is interspersed with large depressions or 'pans' of varying size and permanence, and, whereas some pans are also fed by tributaries, most are dependent on the Pongola River for the bulk of their water supply.People have lived on the high dry ground of the Makhatini Flats for hundreds of years and are heavily dependent on its resources, including water, ood recession agriculture, grazing for livestock, sh, wood, wild vegetables, fruit, reeds and grasses (Heeg and Breen, 1982;DWS, 2015a).e rst comprehensive document describing the manyfaceted aspects of the Pongola Floodplain was Man and the Pongolo Floodplain, by Heeg and Breen (1982), and remains a landmark account over 3 decades later.is document is a synthesis of contributions to a workshop held in 1979, and covers the following aspects: general description (including geology, climate, vegetation and human links to the oodplain); hydrology; water quality; the ecosystem; humans and the oodplain; impact of development and development options; and conservation and the cost thereof.is account not only provides a comprehensive compilation of knowledge from the late 1970's, but also carried this through to a suggested pattern of ows to 'maintain the oodplain through the removal of accumulated wastes, stimulation of sh migration and spawning; submergence of marginal vegetation for a su ciently long period to allow assimilation into the aquatic system and the provision of ood irrigation to cultivated lands on the oodplain.'It is worth noting that the controlled ooding regime suggested by Heeg and Breen (1982) pre-dated (South African) instream ow requirements for river maintenance by a decade, which was rst addressed nationally in the late 1980s (King and Louw, 1998).Of concern is that, 36 years later, there are no operational releases speci cally targeted at maintaining the oodplain ecosystem and its services which support the livelihoods of local communities.A Preliminary Ecological Reserve using the Desktop Model (Hughes and Münster, 731 2000) was undertaken by DWAF (2000), and provided for an allocation of 223 × 10 6 m 3 /a for an Ecological State C river.( e Ecological Reserve is a provision made through the National Water Act (RSA, 1998) for the protection of water resources through an allocation of water quantity and quality to the environment.)Between 1998 and 2012, annual October releases peaking between 450 and 700 m 3 /s (average daily) were made regularly at the end of the dry season (Fig. 5), primarily to meet the needs of recession oodplain agriculture and, ostensibly at the same time, to inundate the oodplain in the Ndumu Reserve near the Pongola-Usuthu Con uence. is timing is asynchronous with natural ooding patterns, where the highest volumes generally occurred in January/February.ese and other issues are discussed in the article 'Pongolapoort Dam: development steeped in controversy' by Van Vuuren (2009).
ere have been no high ow releases since late-2014, due to falling dam levels.Heeg and Breen (1982) recognised the need for 'the construction of a hydraulic model of the system which will establish relationships between river ow and ood levels, and will provide the means for testing the e ects of this and other engineering alternatives for the optimisation of the use of available water resources.'A relatively recent study of the Pongola Floodplain by Lankford et al. (2010) is entitled 'Hydrological modelling of water allocation, ecosystem services and poverty alleviation in the Pongola Floodplain, South Africa.' eir hydrological modelling involved the use of measured 'natural river regime' ows and the development of relationships between discharge and ooded area from the previous studies of Phélines et al. (1973), Heeg andBreen (1982) and Basson et al. (2006).e latter, and other historic hydrodynamic models of the oodplain are discussed next.

Hydrodynamic models of the Pongola Floodplain
Over the past 45 years, since the commissioning of Jozini Dam, four computational models have been developed to simulate the hydrodynamic behaviour of the downstream Pongola Floodplain.ese include the one-dimensional (1d) models of Pitman and Weiss (1979); Department of Water A airs (1987);and Beck and Basson (2003), and the two-dimensional (2d) version of Basson et al. (2006).
e Pitman and Weiss (1979) 1d cell-based model had previously been successfully applied for simulating inundation behaviour in other oodplains.Limited data were available to calibrate the model for the Pongola Floodplain, however, and predictions indicated that a peak discharge of 690 m 3 /s (100 × 10 6 m 3 ) lled the downstream pans, whereas a lower peak of 345 m 3 /s (50 × 10 6 m 3 ) did not.e next model developed, by the South African Department of Water A airs in 1987,

733
was also cell-based, with stage-storage functions and weir connections between cells.e model was essentially steadystate incorporating Manning's formulation for ow resistance.Whereas the 1979 model excluded a section of the (more con ned) river immediately downstream of Jozini Dam, the entire extent of the oodplain to the Usuthu River con uence was included in the 1987 model.Historic dam releases and resulting pan levels were used for calibration, but predictions tended to underestimate peaks and overestimate associated lag times by up to a few days.A 1d hydrodynamic model of the Pongola River and Floodplain (using Mike 11: 1d hydraulic, water quality and sediment transport modelling so ware) was developed 16 years later by Beck and Basson (2003), under the auspices of a South African Water Research Commission Project.e model was parameterised with cross-sections at ~500 m intervals along the river, and hydraulic controls (channels and weirs) provided the connectivity between the active channel and major pans, and between adjacent pans.e simulated ood peak was overestimated by ~0.5 m, and travelled through the system too fast by roughly half a day.
Basson et al. ( 2006) followed-up their 2003 study with linked 1d/2d models (using Mike 11 and 21C) for the upper and lower oodplain, respectively.Initial setup involved the use of a curvilinear grid with higher spatial resolution covering the river channel.Simulations encountered instabilities that could not be resolved, however, and a rectilinear grid was ultimately used, with the upstream 18 km of the system modelled one-dimensionally.e grid size applied for the 2d analysis was 20 m laterally by 50 m longitudinally.e model was parameterised with topographical data from two sources: digitised cross-sections and contours from 1930s and 1950s maps, and bathymetric surveys during a dam release in 2004.
e October 1986 release was used for calibration, and model performance was checked against measured data associated with releases in 1986, 2002 and 2005.Computational time for an event was ~24 h.Basson et al. (2006) quoted a predictive accuracy of ~0.5 m for pan water levels, and less than 1 day for the timing of peaks.Water-level plots, however, indicate some substantially higher di erences of up to ~1.0 m.Discrepancies were attributed to measurement errors, and possible geomorphological changes are mentioned concerning a 2.5 m di erence in peak water level for the Msenyeni Pan in 1986.e calibrated model was also used for simulating hydrodynamic behaviour in response to di erent operational scenarios.ese included di erent hydrograph peaks, volumes and shapes, as well as varying initial pan water levels.Key ndings were the importance of peak duration and volume on pan inundation and (peak) discharge at the Mozambique border; the minor in uence of initial pan levels on the e ectiveness of largevolume releases; and the widespread ooding associated with extreme events.Also noted was the sensitivity of model results to topography, with a vertical accuracy of ~0.3 m suggested for future detailed surveys.
e present investigation was initiated in response to the need for improved hydrodynamic information in support of a socio-environmental ow assessment for the Pongola Floodplain, described in Part 2 of this set of papers (Brown et al., 2018).e ow assessment was commissioned by the Department of Water and Sanitation (DWS) as part of a basinwide assessment of the Ecological Reserve.Concomitant with rapid advances in computing technology over the past few decades has been the development of multi-dimensional hydrodynamic models.For spatially extensive, topographically and hydraulically complex systems, such as the Pongola Floodplain, a model with advanced functionality is required.Appraisal of available 2d models, both commercial and freeware, led to the selection of RMA2 for this study.

TWO DIMENSIONAL HYDRODYNAMIC MODELLING USING RMA2
Background RMA2 is a 2d, depth-averaged, hydrodynamic model using nite elements, and is based on implicit solutions of the fully non-linear shallow water equations.It was developed by Norton et al. (1973) of Resource Management Associates, under contract with the United States Army Corps of Engineers (USACE; Wurbs, 1994).e model has been extended over the past 4 decades, and a version, together with pre-(CFGEN -ConFig GENerator) and post-processors which are part of the TABS numerical modelling system, is maintained by the Waterways Experiment Station (WES) Hydraulics Laboratory (Donnel, 2011).A commercial version, with licensing, is also available through Resource Modelling Associates (King, 2017) that includes active updates.Pre-and post-processing so ware for RMA2 includes RMAGEN (RMA geometry GENerator) and RMAPLT (RMA PLoT), which are for developing network geometry les and facilitating the display of results from the RMA suite of models, respectively.
RMA2 was one of the rst multi-dimensional models widely used for modelling riverine and estuarine applications, and is a rst-generation hydrodynamics engine.Over the past 3 decades, many new computational engines have been developed, although earlier models such as RMA2 still receive frequent use.According to Jones (2011), the main drawbacks of the early computational engines are numerical instability, particularly when the application involves substantial wetting and drying and relatively long run times.In this study, the King (2014) version of RMA2 was applied to the Pongola Floodplain, which is characterised by extreme wetting and drying of an extensive oodplain (~13,000 ha) that includes a well-de ned active channel (Figs 4 and 6).e oodplain contains numerous pans with ~150 identi ed by La Hausse (1987).ese are generally connected to the Pongola River through small tributary and paleo channels that breach levees adjacent to the active channel, as illustrated in Fig.

734
and evidenced in the 1969 photographs in Fig. 8. e pans are isolated from surface ow in the river during the low ow (winter) season, with their water levels falling due to evapotranspirative losses.e hydrodynamic modelling required simulation over a large discharge range, associated with rapid changes in ow resulting from managed releases at Jozini Dam under present day (PD) operation.Furthermore, simulations were needed for long periods of at least a decade.More commonly, multi-dimensional hydrodynamic models are used to simulate behaviour over much shorter periods, such as hours or days, these being generally associated with isolated hydrological events., 2011). Figure 4 shows a section the oodplain where the active and paleo channels, raised levees, oodplain and the MandlaNkuzi Pan are clearly discernible.

Discharge and stage records
Available discharge and stage records were obtained from the DWS for hydrometric stations along the Pongola River and at pans, and are listed in Table 1.Discharge at Station W4H013 is accurately gauged at a compound sharp-crested weir (Fig. 9), which has been calibrated up to ows of ~850 m 3 /s using an Acoustic Doppler Pro ler (Le Roux, 2008).At the remaining 6 stations (examples of which are illustrated in Figs 9 and 10), local water levels are recorded using data loggers, and these are converted to elevations relative to mean sea level.An exception is at the Ndumu Station (W4H009), where levels are relative to the (local) gauge datum.Records were also obtained for 2 stations along the Usuthu River from the Swaziland Department of Water A airs.

Model setup
e 2d hydrodynamic model extends from the Jozini Dam wall to the Pongola-Usuthu con uence at the South Africa-Mozambique Border (Fig. 1).It includes all oodplain areas outlined in Fig. 1 that are directly inundated by Pongola River ows. e modelled area therefore excludes the Msunduzi and Shokwe Pans, with the latter associated with ooding along the Usuthu Floodplain (Fig. 11).
Inundation of the lower Pongola Floodplain (including its pans and wetlands) in the Ndumu Reserve, which is a Ramsar site, is associated not only with ows in the Pongola River, but also with ows from the Usuthu River. is relationship is illustrated in Fig. 12, which is a plot of recorded stage levels     13), and are frequently inundated for only a few days each year during the October release. is impact is expected, since the backwater in uence of the Usuthu River, that ameliorates the e ect of the dam, reduces with increased distance upstream of the Pongola-Usuthu con uence.Consequently, the future ecological status of the oodplain in the Ndumu Reserve depends not only on Pongola River ows, but on future water resource developments along the Usuthu River in Swaziland -which were not addressed in this study.

Topography
e national DEM was used for the Pongola Floodplain.Its spatial resolution allows mapping of the Pongola River by its lower elevation, as illustrated in Fig. 4. e 2004 bathymetric survey of the active channel, however, provides superior accuracy for the channel bed level, and this, together with measurements of channel width, was used to characterise the longitudinal bed topography of the river.

Finite-element mesh and topographic elevations
e nite-element mesh (e.g.Fig. 14) used was developed using a combination of purpose-coded so ware and geographic information systems.e procedure used was as follows: • e river thalweg was digitised using the national DEM, producing a smoothened polyline containing 1 294 vertices.
• Channel bed widths were measured using satellite photography, and varied from 50 m immediately downstream of the Jozini Dam wall to 15 m at the Pongola-Usuthu River con uence.Trapezoidal channel crosssections were applied, with maximum bank slopes of 45° and 7.5 m widths.e river bed, banks and adjacent levees were included as quadrilateral elements, whereas triangular elements were used to characterise the highly variable oodplain topography (e.g.Fig. 14).Bespoke so ware automated the construction of quadrilateral elements (and nodes) based on the channel's planform.

•
e extent of oodplain was delineated using maximum recorded stages from the hydrometric stations (refer to Table 1) and the oodplain topography.Meshing of the oodplains was computed in QGIS (Quantum GIS; www.qgis.org) using the 'Triangle' so ware developed by Shewchuk (undated), available as the Basemesh Plugin for QGIS (Vetsch et al., 2014).e meshing so ware produces conforming Delaunay triangles based on the polygon model boundary ( oodplain and levee), breaklines used to align mesh segments, holes within the mesh where elements are not required (e,g.elevated topographical features that are not ooded), conforming vertices (Steiner points) and restrictions on maximum element areas.
e Basemesh so ware provides output as geographic information system shape and text les.
• So ware was also developed to merge quadrilateral channel elements with oodplain triangulation, and assigns elevations to all nodes in the mesh (bed elevations were assigned from the bathymetric survey; levee and oodplain elevations were assigned from the DEM).• Lastly, the nite-element mesh was written in text format for nal pre-processing using RMAGEN.

Floodplain wetting and drying and maximum retention levels in the pans
e 'marshing' feature in RMA2 was successfully used to model wetting and drying of the oodplain associated with ooding.Using this, when water levels fall below the ground surface, ow occurs in the 'low porosity groundwater zone'.Pans become isolated from surface ow in the river when water levels drop below invert levels -which are the hydraulic controls on river-pan connectivity.Maximum (pan) retention levels (MRL) are de ned by the point of (dis)connection.Concrete weirs were constructed downstream of the Banzi and Nyamithi Pans, and although the latter structure, which was constructed in 1983 (Whittington et al., 2013) is still intact, the Banzi Weir is breached.Associated with this, is the 'Lower Usuthu Breach', where this river broke its southern bank diverting ows through the Banzi Pan and into the Pongola River (Fig. 15).Due to international implications, studies have investigated possible causes for the breach and its remediation (Wadeson, 2006;Anderson, 2009;Basson, 2011and SALOMON LDA, 2010and 2011).As at August 2017, the Lower Usuthu Breach continues to divert ows through the Banzi Pan; a Google Earth image dated 21/08/2017 shows entire diversion of a low ow. Figure 16 shows the incised active channels and riparian forest downstream of the Banzi Pan.A key nding of the geomorphological scoping study of the Lower Usuthu Breach (Wadeson, 2006) is the naturally unstable characteristic of the Usuthu River.Frequent channel change was evidenced from paleo channels, but upstream catchment conditions were seen to be responsible for accelerated instability.

Recent changes in the hydraulic behaviour of the oodplain in the Ndumu
e in uence of Usuthu River ows on the Pongola Floodplain is therefore even greater than pre-breach conditions.For this study, however, insu cient topographic data were  available during model setup to include the Usuthu River from its breach position to the Pongola con uence.

Model calibration and veri cation
e model was calibrated using data from the hydrometric network: the two river channel locations (viz.Lake View and Ndumu) and four pan locations (viz.Msenyeni, Tete, MandlaNkuzi and Nyamithi) -refer to Table 1; data from the remaining three gauges provided some of the boundary conditions.Measurement-based data were used as far as possible for calibration, including daily discharge time series from Station W4H013 (below Jozini Dam), and from Stations GS6 or GS16 in Swaziland.e parameter values for the following variables were determined as part of the calibration: ow resistance as a function of depth; turbulence parameters; marshing parameters; depth for element elimination/addition, and evapotranspiration.
For the above 4 pans with continuous water level recorders, MRL can be identi ed from the stage hydrographs recessions (refer to Fig. 17), and invert levels were determined as part of calibration.For the remaining pans, MRL were estimated by combing the DEM with vegetation mapping (using highresolution aerial and satellite imagery and ground-truthing).
e oodplain was delineated into 56 areas (or sites), based mainly on the presence of 30 major named pans from the literature (Phélines et al., 1973 andHeeg andBreen, 1982).
e model thus consists of the active channel and 56 adjacent and contiguous oodplain areas, most of which contain wellde ned pans (e.g. the MandlaNkuzi Pan in Fig. 14).
Flow resistance values (Manning's n) used in the model were 0.030 and 0.040 for the river channel and oodplain, respectively, and these increased tenfold for depths below 0.40 m to maximum values at ground level.Drying and wetting depths of 0.2 and 0.1 m, respectively, were applied to leveetype and certain oodplain elements to disconnect inundated oodplain from river ows when water levels fall below the ground surface.For the remaining oodplain elements, substantially higher values of 3 to 5 m were used to dampen instabilities found to result from cyclical wetting and drying associated with close-to-steady conditions.e extent to which the modelling was required to simulate episodic drying and wetting of an extensive oodplain is apparent by the inundation range, being between ~17 and 111 km 2 .e 3-year period from October 2008 to September 2010 was used for model calibration and veri cation, since it includes six events of varying magnitude: three were arti cial end-of-dry season releases of up to ~630 m 3 /s (daily average), and the remaining were wet season releases between ~50 and 120 m 3 /s.e rst year was used for calibration and the remaining to assess model performance.is 3-year period also provides reliable stage measurements compared with both prior, and more recent, times: the gauging of water levels in the oodplain has been historically challenging for the DWS, due to vandalism of equipment and the removal of xed stations for agriculture (Kempen, 2014).
Figure 17 shows comparative plots of modelled (calibration and veri cation) and measured stage hydrographs for the six hydrometric stations (river and pan).Generally, good replications have been achieved in terms of peaks, rising and recession limbs when the river and pans are connected, recession of ponded levels in the pans, and low-ow stages in the river.A constant evapotranspiration rate of 4.0 mm is shown to produce satisfactory drawdown results, being almost identical to the WR2012 (WRC, 2017) annual average of 4.1 mm for this region.
Measured stage recessions in the pans (Tete, MandlaNkuzi and Nyamithi) all indicate rises in August 2009, which are attributed to in ows from the adjacent catchments and intercepted rainfall.Flows from adjacent catchments, many of which enter the Pongola River through pans, were modelled hydrologically at a monthly time-scale.
ese estimates were of insu cient accuracy, however, to be meaningfully applied as pan in ows, and were thus speci ed as direct river inputs.
For the Tete Pan, the MRL is lower a er the October releases than following wet season inundation.A possible reason is due to increased vegetation cover during the naturally wet period, which may act to elevate the e ective invert level through higher ow resistance and obstruction of return ow.
e modelled and measured low ow stages in the Pongola River at Lake View di er by ~0.5 m for the periods December 2008 to March 2009, and a er April 2011.ese deviations are attributed mainly to temporal changes in the hydraulic behaviour of the low-ow channel.For example, preceding the March 2009 release, a discharge of ~6.3 m 3 /s resulted in an average stage of ~26.5 m, whereas following the release, ~7 m 3 /s produced a stage some 0.5 m lower.
RMA2 was found to run reasonably e ciently for the modelling (22470 mesh elements), with the 1-year calibration simulation taking roughly 3 hours.e default time-step used in the simulations was about 4 hours, which is targeted at the dry season when changes are gradual; ow is con ned to the active channel and the oodplain pans are ponded.Smaller time-steps down to 1 s were applied to facilitate convergence where necessary; convergence criteria were reasonably severe: 5 mm/s for velocity and 0.1 mm for water surface computations.
e previous studies of Phélines et al. (1973), Heeg and Breen (1982) and Basson et al. (2006) have contributed estimates of discharge required to inundate the major Pongola Pans.ese values have been used in subsequent studies such as that of Lankford et al. (2010).A compilation of these estimates, together with those from this study, is provided in Table 2. Phélines et al. (1973) and Heeg and Breen (1982) provide measurement-based estimates, whilst the more recent studies involved modelling.e estimates of Heeg and Breen (1982) are provided as ranges, since initiation of pan lling was noted not to have occurred at the lower discharge, but took place at the higher value.It is likely that discharge estimates of Heeg and Breen (1982) incorporate previous estimates of Phélines et al. (1973), although this is not clear.
Overall, values from this and the 2006 modelling study are reasonably similar, although this study indicates generally higher discharges that are closer to those suggested by Heeg and Breen (1982).Exceptions are, however, for the Sokunti and MandlaNkuzi Pans.For the latter, initiation of pan lling from this study agrees with a gauged steady release.It is worth noting that geomorphological changes have  (2006) note that for 80 km analysed, the river has narrowed by 35%, with the greatest changes closest to the dam wall.Fluvial modi cations have been brought about through dam closure combined with regular October ood releases over 2 decades, whose peaks exceed annual events (close to a 1:5 yr return period, Phélines et al., 1973).ese are likely to have altered the hydraulic behaviour, and di erences in discharge estimates over time are not unexpected.is, together with modelling uncertainties, dictates that the results of this hydrodynamic study, and the broader ow assessment (see Brown et al., 2018), should it be implemented, need to be implemented within a framework of adaptive management that involves monitoring.e study of Phélines et al. (1973) indicated that ood peaks of ~120 m 3 /s with 3-day durations would be su cient to replenish most of the pans.Heeg and Breen (1982) do not specify (peak) discharges per se, but identify the pans that require ooding at di erent times of the wet season.is RMA2 study provides the basis for estimating releases to achieve Heeg and Breen's (1982) suggestions (e.g. a release of 150 m 3 /s for 3 days has been applied, and tested, for ooding all the pans).Perhaps even more importantly, since these discharge-duration estimates existed a priori, this study allows changes in hydraulic behaviour, associated with di erent release patterns (or scenarios), to be quanti ed.

Model application
e model was run using discharge time series developed and provided by Aurecon (Pty) Ltd (DWS, 2015a) for naturalised and baseline conditions (Fig. 13), as well as for potential future scenarios that include all water resource demands from the dam (agricultural, inter-catchment transfers, irrigation and municipal/domestic). e latter were coupled with four di erent high-ow release patterns for the downstream oodplain.Hydrological time series simulations were based on monthly modelling using the Water Resources Yield Model (WRYM) inherited from the PRIMA IAAP 10 Study (TPTC, 2011).Naturalised monthly discharges were disaggregated for hydrodynamic modelling using historic hydrometric data from the upstream catchment.For baseline and future scenarios, MODSIM was used to simulate daily releases from Jozini Dam.
e simulated time series extends from 1951 to 2004, but this period was reduced to the most recent 15 years for hydrodynamic simulations, giving more acceptable run times of about 24 hours.

Post-processing RMA2 results for analyses in the DRIFT DSS
e standard output from a RMA2 simulation is a binary results le that may be graphically displayed and post processed using RMAPLT.e large spatial extent of the Pongola Floodplain, length of record simulated and number of time series analysed (natural, baseline and 7 scenarios), meant that it was necessary to develop so ware to automate the post-processing of results for further analysis in the DRIFT (Downstream Response to Imposed Flow Transformation) DSS (decision support system).A results le for selected oodplain nodes was created.For each of the contiguous oodplain sites (which incorporate the major pans in Table 2), site-speci c 25 m-gridded DEM data were generated.ese were combined with stage levels to compute 56 site-speci c geometric data les, with each containing tabulated relationships between stage and the following 10 parameters: • Pan/s: inundated area, average depth, area with depth range 1.0-1.5 m, and area with depth range greater than 1.0 m • Floodplain: inundated area and area with depth range 0.2-1.0m • Pan/s and oodplain: inundated volume, area, average depth and area with depth range 0.2-0.6 m Geometric relationships were then combined with daily stage time series (for selected nodes per site) to generate site-speci c time series for each of the 10 parameters listed above.Example excerpts from the results les, which are the hydrodynamic basis for further analyses in DRIFT, are given in DWS (2015b).Maximum retention levels for the pans provide the spatial delineation between pan and oodplain inundation.
e depth ranges (or classes) used above were identi ed as constituting critical (hydraulic) habitat for indicator vegetation species and/or sh guilds (see Brown et al., 2018)

Scenarios
Seven potential future water-use scenarios (Table 3) were constructed and their time series of daily releases from the Jozini Dam modelled by Aurecon (DWS, 2015a).e rst 5 scenarios (Scenarios 3 to 7) include the same existing demands (inter-catchment transfers, irrigation and municipal/ domestic), but coupled with di erent high-ow release patterns for the downstream oodplain.is allowed the impacts of di erent ow releases on the ecological condition and linked sociological use of the oodplain to be assessed (using DRIFT).
e suggested ow pattern of Heeg and Breen (1982) has been used as the basis for Scenario 4 and modi cations of it (Scenarios 5 to 7).Heeg and Breen (1982 p. 87) provided the following description of ow patterns for the Pongola Floodplain, noting that experimental changes for optimisation should follow: 1. 'Raise to ood all pans in December, hold 3 days, drop to normal river level to drain and follow by 2 days at 56 cumec ow and 4 days at 28 cumec ow. is should a ect ushing and allow sh migration.2. Raise to ood Tete [Pan], oscillate water level about this point to ood subsistence lands.Such oscillations would probably range between ooding Mthikeni at highest level and maintaining the Namanini-Bumbe-Ngodo complex at high ood level, and can probably, with the use of in atable weirs, be done on base ow and overspill alone, although some water release may be necessary.3. Raise level to ood all pans during February, hold for 5 days and return to 2 above.4. Drop to level of Namanini-Bumbe-Ngodo during March.
Oscillate about this point, raising to level of Tete [Pan] perhaps once or twice.5. Unimpeded ow April -November.' Heeg and Breen (1982) estimated this annual release at 41 × 10 6 m 3 /a, and allude to the principal of adaptive management concerning the e ect of this release pattern on the environment, agricultural and other developments.
e levels and ows suggested result in a substantially higher volume than their estimate, and there is also no clear indication what constitutes 'normal river levels', 'base' and 'unimpeded ows'.It needs to be noted that these recommendations pre-date instream ow requirements for  (Ferrar, 1989;Bruwer, 1991).
e high ow volume is 225 × 10 6 m 3 /a, calculated from the di erences between daily releases and a low ow of 5.45 m 3 /s to meet international obligations with Mozambique.Environmental water requirements (EWR) estimated using the Reserve Desktop Model (Hughes and Münster, 2000) for an Ecological State C, range between 2.4 and 6.4 m 3 /s for maintenance low ows, which are close to the values given by Hughes (2000) of 2.1 and 6.6 m 3 /s . .e release to meet international obligations accounts for 172 × 10 6 m 3 /a, whilst a constant release of 2.4 m 3 /s requires 66 × 10 6 m 3 /a.All other demands (viz.inter-catchment transfers, irrigation and domestic/municipal) account for 117 × 10 6 m 3 /a.
e hydrodynamic model was used to simulate downstream inundation for the nine hydrological scenarios in Table 3, with tributary and Usuthu River ows maintained at the historical (PD) situation.Of these scenarios, two selected ones (for clarity), together with baseline conditions (viz.essentially the October release) are plotted in Fig. 18.Example time-series plots of inundated area for the Tete EFlow (environmental ow) Site (pan and oodplain) for ve selected hydraulic parameters and three scenarios including baseline, are provided in Fig. 19.Hydraulic parameters provide drivers for ecosystem indicators and were used in the DRIFT DSS described by Brown et al. (2018).For example, couch grass (Cynodon dactylon) is optimally inundated by pan water depths in the range 1.0 to 1.5 m; ood-dependent benthic sh guilds are linked to the number of days that oodplain water depths are in the range 0.2 to 1.0 m.

SUMMARY AND RECOMMENDATIONS
Four computational Pongola Floodplain models have been developed since 1979, including cell-based, 1d and 2d approaches.e varied successes of previous modelling studies attest to this ambitious task, given that the ~130 km 2 system is characterised by a complex mosaic of meandering river channels, levees, and oodplains interspersed with numerous pans (~150 identi ed by La Hausse, 1987) and wetlands.
The RMA2 2d model used in this study was parameterised and calibrated using existing information, including: a bathymetric survey and the national 25 m-resolution DEM; discharge records from the Pongola River downstream of Jozini Dam, and from two stations on the Usuthu River; and stage records from two river stations along the lower Pongola River and from four pans within the floodplain.Stage records from the period 2008 to 2010 were used for model calibration and verification, and verification produced generally good replications in terms of peaks, rising and recession limbs, recession of ponded pan waterlevels, and low-flow river stages.The RMA2 modelling  was subsequently used to model the hydrodynamics of the 594 km 2 Elephant Marsh Ramsar site in Malawi, described by Birkhead et al. (2017).Two-dimensional modelling of topographically and hydraulically complex oodplain systems, such as the Pongola, requires an accurate DEM.Any further sensible improvements to the modelling would require a more accurate oodplain survey, such as that provided by light detection and ranging (LiDAR).A er this study was completed, the existence of LiDAR data for the Usuthu River and adjacent wetlands and pans was noted, but post-dated its use.Improved model calibration would be achieved by monitoring stage uctuations in additional major pans, in response to regulated release patterns.is could be accomplished by using temporarily installed (inexpensive) loggers.
Model application was for the 15-year period 1990 to 2004, and simulations included naturalised, present management (2014), and 7 potential dam operational scenarios.Results were post-processed to provide tabulated daily time series for 11 hydraulic parameters for 56 contiguous EFlow sites -which incorporate pan, oodplain and combined regions.e hydraulic parameters included average depth; inundated area; inundated area with a speci c depth range; and inundated volume.
e companion paper by Brown et al. (2018) describes a holistic EFlows assessment using DRIFT, that analysed the various ow permutations (scenarios) to recommend an environmentally and socially sustainable management option for the oodplain.744 ACKNOWLEDGEMENTS e successful development of the RMA2 hydrodynamic model for the Pongola River and Floodplain was most reliant on the advice and guidance of Ian King, and his resolute commitment to make re nements to the source code to enhance its application in this study.His e orts and enthusiasm are greatly appreciated.We also acknowledge the Department of Water and Sanitation, speci cally Mark Kempen for extracting available information on the oodplain and elding numerous queries; Jane Mogaswe, Elias Nhlapo and Mangaroo Natasha for hydrological data; Beason Mwaka and Celiwe Ntuli for providing data from the Basson et al. (2006) study; the Chief Directorate National Geo-spatial Information (Department of Rural Development and Land Reform), speci cally Sue Kirschner, for DEM data and supporting documentation; the daily hydrology used in the model application was provided Anton Sparks (Aurecon, South Africa); the Swaziland Department of Water A airs, speci cally Petros Simelane for supplying observed records for gauges on the Usuthu River; and Mike Coke for providing literature from the 1970s and 1980s, photographs from the late-1960s, and background to the historic calculations for initiation of pan inundation.We also thank two anonymous reviewers for their positive and helpful critiques.

Figure 1
Figure 1Location of the Pongola River and Floodplain in northern KwaZulu-Natal showing the position of gauging stations (W4H0x and W4R0x -refer to Table1) and major pans (refer to Table2); modelled oodplain is indicated by black-outlined areas

Figure 2 Figure 3
Figure 2Major Pongola Pans and the extent of ooding in response to a release from Jozini Dam in November 1969(after Coke, 1970); names and spellings may vary from the more common ones used here; Jozini Dam was formerly Pongolapoort Dam, and prior to that, Strijdom Dam

Figure 4 Figure 5
Figure 4 Satellite image (GE, August 2013) of the Pongola Floodplain draped over the national DEM, showing the well-de ned meandering active channel (~15 m wide at this location), the MandlaNkuzi Pan and patchwork of agricultural elds in the oodplain (between the pan and channel).For spatial perspectives, refer to Figs 10 and 11 It is included in the well-known Surfacewater Modelling Systems (SMS) suite, and a selection of recent applications include Yin et al. (2010), Sammany and Moustafa (2011), Lee and Julien (2012), Han (2014), Fulton and Wagner (2014), Akl (2016), Tonyes et al. (2017) and Birkhead et al. (2017). 7

Figure 6
Figure 6Pongola River, riparian vegetation and agricultural elds in the adjacent oodplain(photo G Marneweck, November 2014) Accurate topographical data are essential to the development of a 2d hydrodynamic model, and two available data sources were used.e rst of these was from the bathymetric longitudinal survey of the Pongola River bed carried out by the Department of Water and Sanitation (DWS) for the Basson et al. (2006) study.ese were sourced directly from the DWS, with the bathymetric portion extracted from the dataset which also included oodplain topography.Whereas the Basson et al. (2006) study used digitised cross-sections and contours from topographical maps of the 1930s and 1950s, this study used the national 25 m-resolution digital elevation model (DEM) available from National Geo-spatial Information (NGI).e standard error of the DEM is quoted as 1.2 m, and 2.5 m in atter areas (NGI

Figure 7 Figure 8
Figure 7Levee separating the active channel in the foreground and oodplain pan beyond(photo G Marneweck, November 2014)

Figure 9
Figure 9 Top: discharge gauging station W4H013 located downstream of the Jozini Dam Wall (refer to Fig. 3) (DWAF, 2008); middle: stage gauge W4H010 attached to bridge pier at Lake View (photo M Kempen, undated); bottom: water level gauge W4H009 (refer to Fig. 11) in an active channel at Ndumu Reserve (photo M Kempen, undated) Figure 10 Stage gauges at the Tete (top) and MandlaNkuzi (bottom) Pans, and the boats used to access them (photos G Marneweck, November 2014 and M Kempen (inset, MandlaNkuzi Pan), undated (DWA, 2012))

Figure 12
Figure 12Daily stage time series monitored at Lake View (W4H010) and Ndumu (W4H009) over the periodAugust 2009 to March 2011, showing   the in uence of the Usuthu River ows in the downstream Ndumu record e boundary conditions used in the model include: • Daily discharge time series at the upstream Pongola River boundary, representing naturalised, PD and future scenario conditions • A rating (or stage-discharge) relationship at the Usuthu River boundary, immediately downstream of the Pongola-Usuthu con uence • Daily discharge time series at the upstream Usuthu River boundary • Daily discharge time series from tributaries owing into the Pongola Floodplain • Evapotranspiration from open water surfaces e elevation di erence over the modelled area is ~50 m, and the slope adjustment method in RMA2 was used to compute an initial (restart) condition from which transient (unsteady) simulations commence.
Reserve not incorporated in the model setup e oodplain in the Ndumu Reserve is characterised by pans, extensive wetlands and riparian forest.Major pans connected to the Usuthu River are the Shokwe and Banzi, whereas those adjacent to the Pongola River are the Polwe, Nyamithi, Bakabaka and Ndwanini.Numerous smaller pans were mapped by La Hausse (1987).It needs to be re-emphasized that the Ndumu Wetlands and Pans (excluding Shokwe) respond

Figure 14
Figure 14 Finite element mesh for a section of the Pongola Floodplain shown superimposed on a satellite image (GE, August 2013) which is draped over the national 25 m DEM.The projected Coordinate Reference System is Hartebeeshoek94/Lo33.

Figure 15
Figure 15 Satellite image (GE, August 2013) of the northern Ndumu Reserve showing the Lower Usuthu Breach, Banzi Weir Breach (photo G Marneweck, November 2014), dewatered section of the Usuthu River and return path into the Usuthu River 739

Figure 17
Figure 17Daily stage time series for the six gauged locations along the Pongola River and in the pans(measured -black markers; modelled -blue-shaded lines) Figure 18Daily discharge time series for baseline conditions and two scenarios for the period August to March (note, the scenarios have the same events in December and February; Scenario 6 has a 600 m 3 /s peak release in October but no events in January and March)

Figure 19
Figure 19Daily inundated area time series from 1994 to 2004 for the Tete EFlow Site (pan and oodplain) for various depth ranges and 3 scenarios (baseline and Scenarios 4 and 6 -refer to Fig.18); ' oodplain' excludes the 'pan'

TABLE 1 Hydrometric stations along the Pongola Floodplain
Stage: relative to mean sea level

TABLE 2 Discharges required to inundate major pans of the Pongola Floodplain
* ese are also dependant on Usuthu River ows; the topographical data is too coarse in the lower Pongola to provide reasonable estimates http://dx.doi.org/10.4314/wsa.v44i4.21Available on website http://www.wrc.org.zaISSN 1816-7950 (Online) = Water SA Vol.44 No. 4 October 2018 Published under a Creative Commons Attribution Licence 741 taken place since the dam was commissioned: Basson et al.