Simulation of organic carbon loading using MIKE 11 model: a case of River Nzoia, Kenya

Lake Victoria, one of the largest fresh water lakes in the world and affects the lives of about 30 million people in Kenya, Uganda and Tanzania (Khan et al. 2011), is under threat from pollution. According to Wali et al. (2011), water hyacinth is a perennial problem in the lake due to excess nutrient loading from rivers draining into it. River Nzoia is the largest river in the Kenyan portion of Lake Victoria basin and therefore any change in water quality of the river is likely to adversely affect the lake. River Nzoia is not only of trans-boundary importance due to its large volume of water contribution to Lake Victoria and subsequently to the larger Nile river system, but also due to its role as a conduit of pollutants into Lake Victoria. Apart from non-point sources from agricultural runoff and other diffuse sources, River Nzoia also receives point sources of pollution from municipal and industrial waste. The major industries being the agro-based industries such as the Nzoia and Mumias sugar factories and Pan African paper mills. These industries contribute to deterioration of water quality in River Nzoia (Abira 2010; Twesigye et al. 2011, Akali et al. 2011). According to Penn et al. (2006), sources of biochemical oxygen demand (BOD) are readily biodegradable organic carbon (carbonaceous, CBOD) and ammonia (nitrogenous, NBOD) which cause reduction of dissolved oxygen (DO) in water bodies. The main objective of this study was to determine the pollution status of River Nzoia over two seasons. This was achieved through modelling DO and BOD in the middle reaches of the River using MIKE 11 simulation model.

The necessary input data for model simulation consist of initial values, boundary data and topographical data including river hydrographical data (river water level), river cross-sections (this was obtained by surveying the river at several sections), water quality data and flow rate. In this study, five sampling sites were used for the measurement of temperature, DO, BOD and electrical conductivity (EC). The sampling points are illustrated in Figure 1.

Data for calibration and validation were obtained from Water Resources Management Authority (WRMA) Kakamega regional office. Monthly water quality data for 2009 were used to calibrate the model while measured water quality data for the period March–April 2013 were used for validation.

Boundary conditions were defined both on the upstream and downstream side of the river. The choice of the boundary conditions depended upon the availability of the data. The boundary data used included daily discharges for the year 2009 at Webuye (upstream boundary) and rating curve at Mumias (downstream boundary). Other boundaries included inflows at Kuywa, Chwele and Khalaba tributaries.

The resistance number in MIKE 11 was found to be the most sensitive variable and thus was adjusted accordingly to take into account local variations in the topography for a specified land use. In this model, the resistance number in the form of Manning's coefficient, n, was varied until the simulated and observed discharge and water levels matched and this was found to be 0.33. This value was applied for the entire network.

In this study, observed discharge values were used to minimize error propagation from Rainfall–Runoff model (NAM model) to the hydrodynamic module and eventually the water quality module which was the main focus of the study. For this reason, the model simulation was almost a perfect match with the observed (Figure 2) with R² of 0.998.

In this module, EC was simulated as conservative pollutant and therefore, decay was set to zero, the only calibration parameter being the dispersion coefficient. For this study, dispersion factor of 10 m²/s was used for the entire river network with exponent of 1. Maximum and minimum dispersion coefficient was set at 20 and 100, respectively.

EC of January 2009 was simulated and calibrated by comparing with the observed values at four points where data was available. The results are as indicated in Figure 3. The simulated results showed a good agreement with the observed values with R² and Nash-Sutcliffe Efficiency (NSE) values of 0.867 and 0.748, respectively. The percent bias (PBIAS) value was 3.921 indicating a model under-estimation bias (simulated values are less than observed values). However, the magnitude of the bias is small (less than 10) which according to Moriasi et al. (2007) indicates a very good model performance.

Validation of the model was done using EC observed data in March and April 2013. The observed results showed a good agreement with the simulation results with R² of 0.914 and 0.995 for March and April, respectively. The NSE values were 0.806 and 0.993 for March and April, respectively. PBIAS value was −0.198 showing model over-estimation bias for April and 1.605 (under-estimation bias) for March. The results are shown in Figure 4. Simulation results for March showed that EC increased. This could be attributed to inflows from municipal waste water treatment ponds (WWTP) at Webuye and accumulation of organic and inorganic waste from rivers Chwele, Kuywa and Khalaba. The increase towards the end is likely to have been caused by the discharge from Mumias sugar factory (MSF) wastewater treatment plant for both industrial and municipal wastewater. This is supported by Melaku et al. (2007) who showed an increase in EC after discharge from wastewater treatment plant.

Results for the month of April 2013 which represented the wet season showed lower EC values compared to the drier month of March (Figure 4). Analysis of variance (ANOVA) showed significant variation between the 2 months (p < 0.05). This is attributed to dissolution of salts (ions) because of increasing discharge. The findings concurred with studies by Slaughter (2011); Shrestha & Kazama (2007) and Alam et al. (2007) which indicated that EC values decreased with increasing flow volume. There is a general decrease in EC values except after the discharge from Webuye WWTP some distance downstream because of dilution of ions due to incremental flow from Chwele, Kuywa and Khalaba rivers feeding the main Nzoia River.

The model was calibrated by running the full daily simulations for the year 2009 and comparing it with the available monthly measurements at the downstream boundary for the same duration and the results are illustrated in Figure 5. The simulated results matched well with the observed values with R² value of 0.932 (Figure 6) and NSE value of 0.920. PBIAS value was −4.660 indicating that the simulated values were higher than the observed values in most of the instances.

After Calibration, the model was run using data obtained during the months of March and April 2013 sampling exercise. The simulated values of DO are in satisfactory agreement with the observed values at selected points of the river section where measurements were undertaken with R² value of 0.964 (NSE value of 0.927) for the month of March and R² value of 0.909 (NSE value of 0.827) for the month of April. PBIAS values were −0.477 and 2.566 for March and April, respectively, which showed that the model had over-estimation bias for the month of March and an under-estimation bias for the month of April.

For the month of March, the model showed a gentle decrease in DO just below the upstream boundary which is likely as a result of the effect of municipal wastewater from Webuye Municipal sewage treatment ponds (Figure 7). The steep decrease after 30 km is due to inflows from Chwele, Kuywa and Khalaba which carries nutrients such as ammonia and nitrates which led to additional demand for oxygen. The DO decreased towards the end of the river section due to industrial and municipal wastewater effluent from MSF. The value of the DO after wastewater discharges from MSF is below the level recommended by WRMA and NEMA of 4 mg/l (which is necessary to sustain aquatic life). Simulation results for the month of April 2013 showed a similar trend with DO decreasing from 6.42 to 4.53 mg/l at the downstream boundary.

A comparison of the DO values for the months of March and April showed that in March, the DO level is below the minimum recommended value at the downstream boundary (3.72 mg/l) while in April the DO was 4.53 mg/l. The DO for March is statistically and significantly varying (p < 0.05) less than for April. Rivers Khalaba, Chwele and Kuywa contribute to the overall pollution of the river as indicated by the steeper decline in DO levels after 30 km. This is partly due to their proximity to one another along the river and therefore cannot allow the river sufficient time for self-purification. Further, these rivers pass through urban and industrial zones. River Kuywa receives effluent from Nzoia sugar factory while municipal effluent from Bungoma WWTP is discharged into River Khalaba.

The river DO levels were higher during the month April (Figure 7) and after 120 days (Figure 5) showing that the impact of wastewater discharges to River Nzoia is negligible during high flows due to dilution effect since higher runoffs bolster the river's natural self-purification capacity (Assaf & Saadeh 2008).

First order BOD decay constant (k) of 0.5 per day was used and decay temperature coefficient of 1.07 which are default values in the model. The calibrated model used in DO simulation was used and assumed to be satisfactory in modelling BOD since the performance of the model was satisfactory in predicting DO values. The results of the model when run using data for March and April 2013 were as shown in Figure 8. Regression analysis of simulated and observed values of BOD at selected locations where the measurements were undertaken showed a good agreement with R² of 0.981 for March and 0.911 for April. The NSE values were 0.962 and 0.889 for March and April, respectively. PBIAS was −4.466 showing model over-prediction and 0.394 (model under-prediction) for April and March, respectively. The BOD values for both months increased gradually from the upstream boundary up to about 9 km perhaps due to the impact of Webuye WWTP effluents. The BOD concentrations then decreased because of self-purification of the river and also due to dilution by inflows from rivers Chwele, Kuywa and Khalaba. There was, however, a steep rise towards the end of the river section attributed to the discharge from MSF WWTP which, according to Akali et al. (2011), discharged wastewater having about 300 mg/l BOD₅. The BOD value for the month of March at the downstream was above the maximum limit of 30 mg/l set out by NEMA guidelines. However, it satisfied the maximum limit criteria in the month of April suggesting that the river is healthy during high flows (wet season).

A comparison of the BOD values for the months of April and March 2013 showed that April had lower values than March with the variation being statistically significant (p < 0.05). These results are similar to those found by Liu et al. (2005) which showed a decrease in BOD with increase in discharge. Similarly, the BOD values for the month of April concurred with the findings of Mokaya et al. (2004) for River Njoro in Kenya during a similar period. The impact of MSF effluent is minimal during high flows than at low flows as indicated by sudden increase of BOD at the downstream boundary (Figure 8). This therefore suggested that MSF effluent, notwithstanding the fact that NEMA thresholds were not met, does not deteriorate the water quality of River Nzoia during high flows because the purification capacity of the river is not exceeded. However, during low flows, the impact of MSF was noticeable and therefore it is necessary for measures to be instituted towards improving the efficiency of the industrial and domestic wastewater treatment plants.

The major parameters considered were EC, DO and BOD. EC was simulated in order to determine the values of dispersion coefficient and dispersion factor used in the advection–dispersion module in MIKE 11. BOD measures the organic carbon loading in the river. River Nzoia had low values of DO (<7 mg/l) suggesting that the river water quality is deteriorating and therefore means that measures should be undertaken to curb pollution. The main contributors of low DO especially during low flow period were MSF and Webuye WWTP. BOD simulations met the maximum permissible limit of 30 mg/l set out by NEMA in all the sections of the river except at the downstream where discharge from MSF elevated the level to above 30 mg/l.

ANOVA showed significant variation in the concentration of EC, DO and BOD for wet and dry period. EC and BOD values were higher for the dry period compared to the wet period while DO was higher in the wet period compared to the dry period.

The study recommends that the government agencies such as WRMA and NEMA should device mechanisms to effectively enforce the effluent standards to ensure that industries and WWTP adhere to the maximum allowable limit for BOD and also improve their treatment efficiencies in their wastewater plants. This is likely to improve the quality of water in River Nzoia and in turn the quality of water in Lake Victoria.