Geochemical scaling potential simulations of natural organic matter complexation with metal ions in cooling water at Eskom power generation plants in South Africa

e modi ed database in the pH, redox equilibrium calculations code (PHREEQC) with a Tipping and Hurley database (T_H.DAT) coupled with the Windermere’s humic acid model (WHAM) was used to simulate scale formation potential in cooling water circuitry, at Eskom power generating stations in South Africa. is study reports a semi-empirical simulative approach in which organic matter fractions, metals and anions in raw and cooling water were used as modelling experimental inputs. By using the saturation index pro les of Ca2+/ Mg2+ with fulvic acid in a modi ed Tipping and Hurley (T_H.DAT) database, fulvate complex species such as CaFulvate, MgFulvate, and geochemical modelling predictions, mineral phases that potentially precipitate are discussed. Speciation calculations showed that the increase in fulvic acid levels decreased saturation indices of scaling metal phases due to reduced levels of Ca2+ and Mg2+ in the water. Furthermore, if the concentrations of fulvic acid are known, semi-empirical calculations using the geochemical PHREEQC code with a modi ed T_H.DAT are possible. Consequently, mineral phase equilibria outputs may give an indication of how the pH and temperature is to be manipulated to optimally predict and control the incidence of scaling.


INTRODUCTION
Eskom, a power-generating company in South Africa, draws water for cooling turbines mainly from surface water systems.
is type of water has high levels of dissolved solids, mainly organic and inorganic compounds (Pather, 2004). e interaction of the various chemical components may produce ions, compounds in solid or liquid phases and complexes. Comprehensive studies have shown that when such water goes through industrial pipes, solid mineral phases, known as scale, form (Zhang et al., 2013, Rosenberg et al., 2012, which compromises water ow e ciency. One challenge, therefore, is the fact that there are dynamic chemical interactions of the water components that lead to changing phases because of changing physico-chemical parameters (such as conductivity, alkalinity and pH) (Brunner, 2014). Current research has shown that natural organic matter (NOM) in uences chemical interactions of certain species in natural water. Studies on NOM range from its characterization and fractionation to complexing capacity and degradation. Research has revealed that NOM is a complex matrix with transphilic, hydrophobic and hydrophilic components (Filloux et al., 2012;Zhou et al., 2014).
Evidence indicates that cooling water constituents change as it is recycled over and over, which in turn a ects the stability of the NOM-metal complexes and consequently results in changed solubility of the salts present (Hoch et al., 2000, Tipping, 2002. What is not clear is how the fractions of NOM in water and their complexation reaction capacities for Ca 2+ and Mg 2+ in uence pipe scaling. is study sought to provide some answers to this and related questions through geochemical modelling and speciation simulations. One shortcoming, however, is that most geochemical codes do not have comprehensive databases that describe complexation reactions, particularly those that describe NOM fractions. e current study describes (i) the modi cation of the Tipping and Hurley database with the Windermere Humic Acid Model (WHAM) database (T_H.DAT) to incorporate humate and fulvate equilibrium de nitions with Ca and Mg ions, and (ii) speciation calculations generating saturation indices of scaling mineral phases.
Successful modi cation of the T_H.DAT was achieved by coding log k values for Mg and Ca equilibrium reactions with predominant NOM fractions (fulvates and humates) into it. Stability de nitions for these reactions were obtained from risbergfa13.vdb, shmgeneric10.vdb, genericha08.nic and genericfa08.nic databases in Visual MINTEQ version 3.1.1. All these databases are based on the NICA-Donnan model (NIC) (Milne et al., 2003, Koopal et al., 2005, Christensen et al., 1998 and the Stockholm humic model (SHM) (Gustafsson, 2001;Gustafsson and Berggren, 2005;Pourret et al., 2007). To obtain the desired outputs, the modi ed T_H.DAT was set as the default database in the PHREEQC program and therea er used for speciation simulations. e simulation mixture comprised of scaling cations (Ca 2+ and Mg 2+ ) and anions (SO 4 2-, Cl -, NO 3 -) de ned as SOLUTION_SPECIES. e carbonate concentration was considered as alkalinity measured as HCO 3 which was also de ned in the SOLUTION_SPECIES block. Fulvate-2 and Humate-2, previously de ned as SOLUTION_MASTER_ SPECIES in the Tipping and Hurley's WHAM model (2003), generated MgFulvate, CaHumate and MgHumate in addition to other mineral phases. Saturation indices (SI) were calculated using adjusted PHREEQC database (F_H.DAT) to show mineral dissolution or precipitation.

MATERIALS AND METHODS
e concentrations of the metal ions were determined by inductively coupled plasma optical emission spectrometry (ICP-OES) (Spetro Arcos, with a Cetac ASX-520 autosampler) and the anions in the samples by ion chromatography, IC (ICS -1500 model) using the Dionex IonPac AS914 (Analytical 4 × 250 mm) column, and the Dionex IonPac AG14 (Guard 4 × 50 mm) column. e LC-OCD analyses were carried out on DOC-Labor LCOCD instrument (Mode 8, Version of 2012-08-27). Fluorescence EEM measurements were conducted using the Aqualog uorescence instrument (Horiba, New Jersey, USA). e speciation and saturation index simulation results were generated by modelling so ware, PHREEQC, with an interactive interphase, and a modi ed Tipping and Hurley (T_H) database.

Sampling and sample preparation
Water samples were collected in 1 000 mL sampling bottles from 4 sampling sites (A: raw, B: clari er, C: inlet and D: outlet cooling water) (Fig. 1) in the cooling water processes at Kriel and Lethabo power stations. e samples were transported to the laboratory and immediately refrigerated at 4°C.

pH and conductivity analysis
e pH and conductivity measurements of all samples were done using pH/conductivity meter Model 3540 (Jenway Inc, Essex, CM6 3LB, England) and calibrations performed with calibrating standard solutions (pH = 4.01 and 7.00 at 25°C). e accuracy of the pH measurements was ± 0.05 pH units. Triplicate readings of each analysis were taken and then the average value used for the analysis.

ICP-OES and ion chromatography analyses
All the samples were ltered through 0.45 µm syringe lters before analysis. e anions in the samples were analysed by ion chromatography (IC). A mixture of 3.5 mM sodium carbonate (Na 2 CO 3 ) and 1.0 mM sodium bicarbonate (NaHCO 3 ) was used as the eluent.

LC-OCD analysis
Dissolved organic carbon (DOC) was determined chromatographically using the DOC-labor LC-OCD technique and correlated with the absorbance at 254 nm for speci c ultraviolet absorbance (SUVA). Using this technique, a mobile phase was delivered with an HPLC pump (S−100, Knauer, Berlin, Germany) at a ow rate of 1.1 mL·min −1 to an autosampler (MLE, Dresden, Germany, 1 mL injection volume) and the chromatographic column (a weak polymethacrylate cation exchange column, 250mm_ 20 mm, TSK HW 50S, 3000 theoretical plates, Toso, Japan). Before chromatographic separations, water samples were passed through an in-built 0.45 µm membrane lter to remove particulate organic matter. Systematically, the organic carbon detector comes a er the non-destructive, xed wavelength UV-detection (UVD 254 nm,Knauer,Berlin,Germany). is LC-OCD procedure was described by Huber

Figure 1
The cooling water process at a typical power generating station at Eskom, South Africa.  Gilmore et al., 2003) et al. (2011). e procedure includes acidi cation of the water samples at the inlet of the OCD (at a ow rate of 0.2 mL·min −1 ) to convert carbonates to carbonic acid. is initial stripping process removes all inorganic carbon. Subsequent steps involving analysis of organic carbon were done by wet oxidation. e oxygen required to oxidize DOC comes from water by a process called radiolysis of water at 185 nm. is way, most compounds including uoro-organic compounds, were oxidized. e organic carbon detector (OCD) calibration was based on potassium hydrogen phthalate. In this case, the carbon mass was used to calibrate the OCD and its extinction coe cient, ε = 1.683 × 10 −3 L·mol −1 ·cm −1 , used to calibrate the UVD (Huber et al., 2011).

Fluorescence excitation-emission matrices (FEEM)
Fluorescence emission excitation matrices (FEEMs) were collected using a Horiba AquaLog Spectrometer (Jvon, New Jersey, USA). No pre-treatment of the samples was applied. Maintaining a common pH between samples ensures that the uorescence characteristics of the acidic functional groups in humic molecules remain constant. For this reason, all the samples were adjusted to a pH of 7. Collection of intensity values was at 10 nm increments within excitation-emission ranges of 250-450 nm and 300-600 nm, respectively. Scan rate was set to 600 nm·min −1 , slit width was set to 10 nm and photomultiplier tube voltage was set to 775 V. Type 1 water at pH of 7 was used as a blank solution. e uorescence spectrometer used a xenon lamp excitation source and the emission at longer wavelengths was detected at 0.3 nm steps. Further to this, inner-lter correction and Rayleigh scattering was applied to the data.
In this study, uorescence was carried out under dilute sample conditions, which follows the equation: (1) where I O is the exciting light intensity, Abs OV is the number of quanta absorbed in the overlap region (OV) and ØF is the ratio of the quanta emitted to the quanta absorbed. e spectra obtained were corrected for instrumental and sample dependent spectral anomalies such as the inner-lter e ects associated with sample absorbance of excitation and emission (Rayleigh scattering). Appropriately, spectral data can be compared to standard data and regions in literature and qualitative and quantitative NOM details will be drawn. Figure 2 shows some literature-based spectral regions using de ned excitation and emission wavelength boundaries (Gilmore et al., 2003).
Peaks at shorter excitation wavelengths (280 nm) and longer emission wavelengths (> 380 nm) are associated with humic acid-like compounds. is NOM-related spectral information lacks capabilities of quantifying multiple broad-shaped EEM peaks. Chen et al. (2003) postulated that integration beneath EEMs within selected regions would represent cumulative uorescence responses of NOM with similar properties. is technique shows that the volume Ф i beneath a region said to be i of an EEM be calculated by Eq. 2: (2) where dλ em is the excitation wavelength interval (taken as 10 nm), dλ em is the emission wavelength interval (taken as 0.5 nm), and I(λ ex λ em ) is the uorescence intensity at each excitation-emission wavelength pair. e values of Ф i can be normalized to a DOC concentration of 1 mg·L −1 for comparison of EEMs from di erent sources of NOM (Zhou et al., 2005). Via ParaFac modelling, di erences in the spectral slopes have been related to the proportional contribution of UV-B

709
absorption coe cients to the CDOM spectra which, as literature suggests, is dependent on the di erent CDOM sources in natural waters (Guéguen et al., 2011). is model factorizes a spectral matrix into three sub-matrices using the same number of factors in Eq. 3. erefore, the F-EEM data is broken down into a set of tri-linear terms and a residual array.
( 2 ) ��� =∑ �� �� �� + ��� � ��� i = 1,…I; j = 1,…,k = 1,…K; (3) Equation 3 is the ParaFac equation in which X ijk is the uorescence intensity of emission sample I at the j th emission wavelength and the k th excitation wavelength for ParaFac model with F number of components. e terms a, b, and c represent the concentration, emission spectra, and excitation spectra, respectively, of the di erent uorophores. e a if term is directly proportional to the concentration of the f th analyte of the i th sample, while b and c kf are scaled estimates of the emission and excitation spectra at wavelengths j and k, respectively, for the f th analyte. e ε ijk term represents any unexplained signal resulting from residual noise or un-modelled variability (Gentry-Shields et al., 2013). Evidence suggests that with the ParaFac model, four fluorescent components are identifiable: a protein-like component, soluble microbial by-product-like component, non-humic-like component and fulvic-like component (Guo et al., 2012).

Modelling using PHREEQC-Interactive
e geochemical code, PHREEQC, was used to calculate the aqueous speciation and saturation indices of di erent mineral phases. e phases included aragonite, calcite, gypsum, magnesite, anhydrite and dolomite. During the assessment of the impact of humic substances on speciation and saturation indices, PHREEQCI databases were modi ed before calculations. PHREEQC Version 3 implements many types of aqueous models, among them the following: ion-association aqueous models (the Lawrence Livermore National Laboratory model and WATEQ4F), a Pitzer speci c-ion-interaction aqueous model, and the SIT (Speci c ion Interaction eory) aqueous model (Parkhurst and Appelo, 2013). Using these aqueous models, PHREEQC is able to carry out the following simulations: ion speciation (output including saturation indices), batch-reaction modelling (including equilibrium phases), surface complexation and ion exchange, reaction (addition/removal of speci ed elements), mixing waters and changing temperature, transport modelling, advection with reactions with and without dispersion, and inverse modelling, and can calculate reactions that account for changes in composition of water along a ow path (Parkhurst and Appelo, 2013).

Database modi cation
e speci c database used in PHREEQC for the simulations is based on various approaches such as (i) the ion association approach (these include phreeqc. dat, wateq4f.dat, minteq.dat, llnl.dat, iso.dat and phreeqd.dat (Charlton and Parkhurst, 2011); (ii) the di usion coe cients: multicomponent di usion, speci c conductance; iii) molar volumes for calculation of density; and iv) the Pitzer speci c interaction approach -Pitzer.dat. Various databases are based on the equations that de ne activity coe cients for model solutions and reactions and are discussed in this section. For example, the Lawrence Livermore National Laboratory aqueous model (LLNL) uses the following expression (Eq. 4) for the log (base 10) of an activity (Charlton and Parkhurst, 2011): ��� =∑ �� �� �� + ��� � ��� i = 1,…I; j = 1,…,k = 1,…K; (3) where A γ is the Debye-Hückel A parameter corresponding to -dh_a, B γ is the Debye-Hückel B parameter corresponding to −dh_b, B is the Debye-Hückel B-dot parameter corresponding to -bdot, a i 0 is the hard-core diameter, which is speci c to each aqueous species, and I is the ionic strength. e LLNL_ AQUEOUS_MODEL_PARAMETERS data block de nes A γ , B γ and B as functions of temperature (Charlton and Parkhurst, 2011). e P zer database is based on the P zer activity coe cients. Equation 5 shows the expression for a Pitzer parameter, P with T as the temperature in Kelvin, and T r as the reference temperature (298.15 K). If less than 6 parameters are de ned, the unde ned parameters are assumed to be zero. When modifying a Pitzer aqueous interaction parameter, care is needed to ensure thermodynamic consistency among all the parameters. If the same type of parameter with the same set of species is rede ned, even if the order of the species is di erent, then the previous de nition is removed and replaced with the new de nition (Charlton and Parkhurst, 2011).
Carefully, therefore, the humate and fulvate thermodynamic de nitions were incorporated into a Tipping and Hurley database with a WHAM (T_H.DAT) (Tipping, 2002), which de nes the LLNL and the P zer databases. Table  1 shows the log k values for equilibrium phase equations of the reactions incorporated into the T_H.DAT database together with their stability constants, all drawn from visual MINTEQ v 3.1.1. For these reactions, the phases were de ned as SOLUTION_SPECIES in T_H.DAT. e simulation mixture comprised of scale-forming cations (Ca 2+ and Mg 2+ ) and anions (SO 4 2-, Cl -, NO 3 -) de ned as SOLUTION_SPECIES. e carbonate concentration was considered as alkalinity measured asHCO 3 which was also de ned in SOLUTION_SPECIES block. Fulvate-2 and Humate-2, have been previously de ned as SOLUTION_ MASTER_SPECIES in the Tipping and Hurley WHAM model (2003). e modi ed T_H.DAT incorporates log k values from risbergfa13.vdb, shmgeneric10.vdb, genericha08.nic and genericfa08.nic databases in visual MINTEQ version 3.1. ese values de ne the equilibrium phase equations of NOM (humic and fulvic acid) with Mg 2+ and Ca 2+ that give rise to CaFulvate, MgFulvate, CaHumate and MgHumate (Fig. 3).

710
Mineral saturation e pH, alkalinity, metal cations and anions were modelled together with organic components (fulvates) using a fulvateadjusted T_H database. e input-output geochemical execution processes on the PHREEQC code are summarised in Fig. 4. To show mineral dissolution or precipitation, saturation indices (SI) were calculated. Herein, a general reaction (Eq. 6), describes super-saturation which is the key parameter that determines precipitation. e saturation ratio (S) is described as the ratio of the ionic activity product (IAP) to the equilibrium constant (Ks) (Eq. 7).
where Ais the anion being equilibrated (A -= (SO 4 2-, Cl -, NO 3 -, CO 3 2-, PO 4 3-), B + is the metal cation, AB is the mineral phase formed and g and l are any gas or liquid phases, respectively.
A value of S < 1, indicates an under-saturated solution, S = 1 shows a solution in equilibrium with the solid, while S > 1 shows

711
a supersaturated solution, which leads to precipitation/scaling. Saturation indices (SI) are de ned as SI = log 10 (S), so that an equilibrium situation corresponds to SI = 0, and under-saturation and super-saturation shown by SI < 0 and SI > 0, respectively. A very high or very low SI does not always indicate 'extreme' conditions but may merely be a mineral that is less reactive. From a geochemistry perspective, reactive minerals are those that dissolve in water or precipitate from water during the residence time in the water (Peters, 2009). A general ow scheme on geochemical modelling is given in Fig. 4.

RESULTS AND DISCUSSION
e FEEM spectral signatures, LC-OCD, IC, ICP-OES results together with speciation simulations are presented for raw, claried and cooling water from Kriel (K) and Lethabo (L) power stations. is study focuses on raw and cooling water at the outlet point to explain potential scaling in water pipe circuitry. Using pH values (Table 2) , various temperature and NOM (fulvic acid) dependent simulations were carried out. e relatively similar raw water characteristics obtained in this study con rmed water use from the same source, the Vaal River. Fundamental characteristics of water, which were used as inputs in the PHREEQC simulations, are shown in Table 2.

Figure 4
A scheme on geochemical modelling processes. The PHREEQCI processes with a modi ed T_H database typify these processes

712
function of their molecularity (y-axis) (Fig. 5a). is diagram is a plot of the spectral absorption coe cient (SAC) to the organic carbon (OC) ratio (aromaticity) of a HS-fraction against its nominal molecular weight (M n ) (Huber et al., 2011). Accordingly, the curved boundaries de ne the region where the humic substances are characterized. e humic substances area was developed using fulvic and humic acid characteristics of water samples from di erent sources discussed in Huber et al. (2011). As reported in Huber et al. (2011), the humic acid (HA) and fulvic acid (FA) standard (II) isolates of the International Humic Substances Society (IHSS) were used to calibrate the HS molecular weights, prior to all characterizations. In the current study, raw and cooling water samples from Kriel and Lethabo power stations, were characterized. e results obtained showed that raw, clari er and cooling water (inlet and outlet) from Kriel (shown by K) and Lethabo (shown by L) consists mainly of pedogenic fulvic acids (Fig. 5a). e fulvic acid character increased in the order of E > A > G, B and C > H and F > D. e raw water data points A and E, from Kriel and Lethabo, respectively, are close to each other in the HS-diagram (Fig. 5a).
is con rms the common source of cooling water used at these Eskom power stations, i.e., the Vaal river. e FA reported in this study is in close agreement with data reported by Huber et al. (2011), regarding the characteristics on the HS-diagram FA obtained from the river Rhine, river Seine and Steinbach surface water (Huber et al., 2011). However, the FA obtained from our study di ers from the FA reported by Huber et al. (2011) on data obtained for river Westend, Lake Sosa, Lake Galilee and Caspian Sea waters. is variation in results further reveals the di erence in surface origin of the FA obtained in our study, which was predominant in the cooling water analysed in this study.

Quanti cation of NOM fractions
Natural surface waters exhibit 5 distinct peaks corresponding to LC-OCD elutes of di erent NOM fractions, namely: (i)

Figure 5a
The

Figure 5b
Chromatograms of water in the CWP (Kriel and Lethabo power generating companies, South Africa), based on Fig. 5a Fig. 3b shows chromatograms corresponding to the NOM fractions analysed by LC-OCD. Table 3 summarizes the LC-OCD quantitative results, based on the raw and cooling water samples from the two power stations. ese results include the humic substances which were predominantly fulvic acid, shown in Fig. 5b. For all samples, the humic substances fraction quantities did not di er signi cantly with total dissolved organic carbon (DOC) (r 2 = 0.01677, p < 0.05, ANOVA). For example, the percentage HS concentration to the total DOC in the raw water was 88.4% and 89.5% for samples from Kriel and Lethabo power stations, respectively. Separately, a similar observation was made using cooling water, i.e., 87.8% and 83.5% for Kriel and Lethabo, respectively. However, the raw and cooling water HS fractions from the power stations di ered signi cantly (r 2 = 0.95, p < 5, ANOVA). Table 3 also shows the SUVA values corresponding to the various samples. NOM is hydrophilic (HPI) if the speci c UV absorbance at λ 254 (SUVA) value < 2 (an indication of non humic substances), hydrophobic (HPO) with a SUVA value > 4 (an indication of humic substances) and is transphilic (TPI, a mixture of both HPO and HPI) if 2 ≤ SUVA ≤ 4. In the present study, the SUVA values of the water samples generally show that the nature of humic substances is highly aromatic. is is supported by the high percentage of HS in dissolved organic carbon (DOC). Similar studies have shown that surface water from di erent sources contains a NOM matrix comprising of transphilic, hydrophobic and hydrophilic compounds (Filloux et al., 2012, Zhou et al., 2014. e percentage of humic substances reported by Zhou et al. (2014) (84%) agree with the LC-OCD results obtained in this study for the raw and cooling water processes (ranging from 80-90%).

FEEM characterization of NOM
Typical FEEM spectra of cooling water used at the Kriel power station are presented in Fig. 6. Generally, the observed excitation (EX) and emission (EM) spectra for water samples con rmed intense fulvic acid and humic acid activities. ese   Table 3. e spectral signatures, as revealed by FEEM (Fig. 6), showed intense activity of HS at λ EX 250-290/λ EM 350-500 nm, which further con rmed the predominance of fulvic acid in the NOM. e fulvic acid FEEM spectral regions have been reported elsewhere (Ahmed et al., 2014). e observed raw and cooling water characteristics of NOM using LC-OCD and FEEM (found to be predominantly FA), and the quantitative measurements of HS (established to be FA), formed the basis of modelling discussions under the geochemical sub-heading. ese include database modi cations and the use of the PHREEQC code to include FA equilibrium reactions with selected cations.

Geochemical modelling
At ambient temperature (25°C), the selected minerals (aragonite, dolomite, gypsum, calcite and magnesite) showed higher dissolution than precipitation (Fig. 7). ese results were further simulated in an extended range of temperatures and the concentration of fulvic acid. Mineral phase modeling, using F_H.DAT, showed distinct saturation patterns. Saturation indices of selected phases are discussed in this article, including: the two polymorphs of calcium carbonate, namely; aragonite (CaCO 3 ) and calcite (CaCO 3 ), anhydrite (CaSO 4 ), gypsum (CaSO 4 ·2H 2 O), dolomite CaMg(CO 3 ) 2 and magnesite (MgCO 3 ). Both alkalinity de ned as HCO 3 and sulphurcompounds in the form of SO 4 2values were calculated as averages of raw and cooling water from Lethabo power station. From Fig. 7, the saturation levels revealed variations depending on the source of water. is conforms to what is expected as the cooling water evaporates. As the water gets recycled over and over, its components concentrate. e in uence of temperature and concentration of humic substances on formation and saturation of these mineral phases are modelled in this section.

Temperature-dependence mineral simulations
Temperature simulations revealed 2 types of mineral phase occurrences: (i) mineral phases below and above the saturation lines (Fig. 8) showing under-saturation and supersaturation (calcite, aragonite, dolomite and magnesite), respectively; (ii) mineral phases showing under-saturation only (anhydrite and

Figure 7
Saturation index levels of selected mineral assemblages using a modi ed F_H.DAT on the PHREEQCI code. Anhydrite, aragonite, calcite, dolomite, gypsum and magnesite represent selected mineral phases.  . e saturation indices of minerals were calculated from the F_H.DAT. Calcite (CaCO 3 ) was found to have higher saturation indices than aragonite (CaCO 3 ). For example, in almost all (except raw water samples) calcite showed supersaturation at lower temperatures than aragonite. A very speci c illustration is the precipitation of calcite and aragonite in the Kriel clari er (K-clari er) and Kriel cooling water outlet (K-CWO), from 15°C and 30°C, respectively (Fig. 8 Calcite,and Fig. 8 Aragonite). It is important to mention that crystals of CaCO 3 exist in di erent polymorphs, i.e., calcite, aragonite and vaterite. Calcite is the most stable of the three while vaterite is the least stable. Ideally, they form under di erent conditions, for the same amounts of inputs (Wada et al., 2013). Aragonite is a high-pressure phase which changes to calcite at ambient conditions. Its precipitation is dependent on pH, temperature, super-saturation and CO 2 content. e relatively high degree of under-saturation of gypsum (CaSO 4 ·2H 2 O) and anhydrite form (CaSO 4 ) compared to other mineral phases is likely to be due to the low concentration of sulphate ions in the cooling and raw water samples (Fig. 8  gypsum; Fig. 8 anhydrite). Gypsum and anhydrite dissolution were apparent in most of the water samples, thus contributing free Ca 2+ that enhances calcite precipitation, a phenomenon called common ion driven precipitation (CIDP). We noted that calcium concentrations are controlled by calcite and aragonite dissolution. Furthermore, the dynamics of Ca complexation and dissolution govern its pH-dependent equilibria (Pokrovsky et al., 2009). Pokrovsky et al. (2009 provided a detailed argument on the dissolution kinetics and precipitation of dolomite and magnesite, in addition to calcite, as a function of pH, temperature and pressure. ey noted that the rate constants of dissolution of these minerals increase with increase in pH and temperature. is agrees with the temperature-dependent simulation results in this study.

Speciation simulations based on NOM (fulvic acid)
e concentrations of fulvic acid were determined to be in the range 3.5 to 81.5 mg·L −1 ,viz, (i)) Lethabo had concentrations as follows: raw 4.022 mg·L −1 , clari er 45.902 mg·L −1 , CWI 25.830 mg·L −1 and CWO 38.775 mg·L −1 , compared to (ii) Kriel's raw 5.934 mg·L −1 , clari er 81.422 mg·L −1 , CWI 47.023 mg·L −1 and CWO 49.290 mg·L −1 (Table 3). Raw water fulvics were compared with those from the clari er as well as those from the cooling water tower. As expected, the results showed that the fulvics from the cooling tower were found to be less than a tenth of those from the clari er. is may be attributed to the fact that recycled water volumes keep decreasing due to the evaporative processes in the cooling water chambers and this concentrates the humic substances (Pather, 2004). Various exponents of the experimental fulvic acid concentrations were inputted to the PHREEQC SOLUTION_SPECIES using the F_H.DAT and the results are presented in Figs 9 and 10.
How available and reliable are the thermodynamic and equilibrium de nitions of NOM in the existing databases? Undoubtedly, no single modi cation would suit all organic compounds. Huber et al. (2012) agree with this. In their simulations, they chose to pick thermodynamic de nitions of volatile fatty acids (acetate, propionate, butyrate and lactate) from MINTEQ.V4.DAT and included them in the PHREEQC code with interactive interface (PHREEQCI) database (Huber et al., 2012). erefore, to answer the question, the F_H.DAT was constructed using available thermodynamic de nitions for the Ca and Mg-fulvate equilibrium phases in Visual MINTEQ.v3.1.1. is work drew up speci c species-oriented modi cations on the premise that variability of both the physico-chemical conditions and the wide range of compounds in cooling water play a critical role in data generation.
Based on the expanded fulvic acid simulations from our study, the e ect of fulvate ions as de ned in the EQUILIBRIUM_PHASE and SOLUTION_MASTER_SPECIES in the F_H.DAT was investigated. Results show that by changing the fulvic acid concentrations in cooling water processes, one can alter scaling indices of mineral phases signi cantly (Figs 9 and 10). ese gures are extracts of the raw and cooling water results at Kriel power station. is reduction may be attributed to the formation of the de ned complexes between Ca/Mg and fulvic acid, viz, MgFulvate and CaFulvate.

Scale control modelling studies
In this article, the composition of scale in pipes, with physicochemical conditions determined, is given in Table 4. e experimental results shown in Table 4 were obtained from   analysis of a scale sample obtained from cooling water pipes at Lethabo. Accordingly, the results showed varied concentrations of elements in the order of Zn > Al > Ca > Mg for the 4 metal ions under consideration. e second step was to accurately determine the content of scaling metals and other trace elements in water. e fundamental objective in the current context is to keep Ca/Mg ions in solution (complexed form). erefore, it was important to include organic compounds in the PHREEQC databases and simulate scale formation under di erent conditions (shown in the methodology section). In that section, a typical illustration of precipitative and complexation processes has been discussed using the context of fulvic acid interactions with metal cations. Table 5 shows potential scaling potentials of selected mineral phases of Ca and Mg in water; aragonite (CaCO 3 ), artinite (MgCO 3 :Mg(OH) 2 :3H 2 O), brucite (Mg(OH) 2 ), calcite (CaCO 3 ), dolomite (CaMg(CO 3 ) 2 ), dolomite(d) CaMg(CO 3 ) 2 , huntite (CaMg 3 (CO 3 ) 4 ), hydromagnesite (Mg 5 (CO 3 ) 4 (OH) 2 :4H 2 O) and magnesite (MgCO 3 ). From this study, the interaction between calcium and carbonate ions is de ned to in uence scaling in water pipes. e saturation indices of aragonite and calcite (polymorphs of CaCO 3 ) are positive throughout the recycle water processes (in raw and in cooling water). From the study, these polymorphs remained relatively stable throughout the recycle processes. e third step is to consider the current state Ca binding to fulvic acid, in raw water compared to recycled cooling water. For instance, the molality values of CaFulvate are lower (viz; 4.6 × 10 −13 and 1.5 × 10 −11 ) in raw water at Lethabo and Kriel, respectively.
is complex is 100-fold higher in cooling water than in raw water (Table 6). A similar scenario is experienced with MgFulvate. is phenomenon explains the fact that cooling recycled volumes have reduced precipitation potentials for Mg and Ca. is leads to the conclusion that manipulation of physicochemical parameters enhances complexation of Ca and Mg, which inhibits complexation of non-scaling metals while reducing the components that oxidise, which is critical in scale control.   Fulvate-2 5.5 × 10 -6 5.0 × 10 -5 8.1 × 10 -6 1.1 × 10 -4 HFulvate-5.3 × 10 -11 7.4 × 10 -10 1.5 × 10 -11 2.8 × 10 -10 CaFulvate 4.6 × 10 -13 1.4 × 10 -11 1.1 × 10 -12 1.3 × 10 -10 MgFulvate 6.7 × 10 -12 6.1 × 10 -10 1.2 × 10 -11 6.8 × 10 -10 Given the results in Tables 5 and 6, we can conclude that complexation between Ca/Mg cations and fulvates could be linked to saturation indices in explaining the scaling potential. Saturation index results show that only the values in raw water samples are slightly di erent. However, the fact is that the incorporation of the equilibrium equations for Ca and Mg with fulvates into the PHREEQC databases, could o er a special link that explains simultaneous complexation and precipitation reactions in fresh waters used for industrial purposes. Modelling the formation of these discrete chemical complexes showed a reduction in the concentration of free Ca and Mg ions in solution, hence decreased scaling potential. In addition, the study has shown that the concentrations of fulvic acid and temperatures may be manipulated, and using the geochemical code PHREEQC, the incidence of scaling could be predicted and controlled.

CONCLUSIONS
e HS-diagram revealed the predominance of the fulvic acid fraction of NOM in raw and cooling water. For all samples, the fulvic acid fraction did not di er signi cantly relative to total dissolved organic carbon. In addition, the spectral EEM signatures, showed intense activity of HS at λ EX 250-290/λ EM 350-500 nm, which con rmed the predominance of fulvic acid in NOM. Using the PHREEQC code, the Tipping and Hurley database (T_H.DAT) was successfully modi ed by including Mg and Ca complexation equilibrium equations with the fulvate ion. is was achieved by incorporating Fulvate-2 complexation equilibria in the SOLUTION_MASTER_SPECIES block and de ning it as SOLUTION_SPECIES. Both the humates and fulvates were incorporated in the T_H.DAT database correctly and were able to generate outputs on the PHREEQC code. As a result, new complexation the Ca-fulvate and Mg-fulvate mineral phase outputs were obtained and related to the scaling potential of Ca/ Mg. Overall, the entire study has shown that if the concentrations of organics (such as fulvic acid) are known, the conditions that favour their complexation with Ca/Mg could be manipulated appropriately to control the incidence of scaling in pipe circuitry.