Nyanza gulf is in Western Kenya whose watershed lies between 0.25N - 1.00S latitudes and 34.0E - 36.0E longitudes as shown in Figure 1. It covers an approximate total drainage area of 12,300 km². The Rivers originate from East and south East highlands. They flow from up to 3005 m elevation to the outlet at the lake which is 1134 m below East-west and South East to North direction. The average areal annual rainfall distribution of up to 1400 mm on the Eastern side-1600 mm on South Eastern-South and the annual maximum and minimum temperature of the watershed area is about 30˚C and 8˚C, respectively. The catchment experience equatorial climate with two rainfall seasons; March to June (long rain season), September to November (short rain season), followed by a relatively dry season in December to February and July to August respectively. Land preparation mostly starts in January to February and cereal cropping done with onset of long rainfall seasonal apart from the plains in Nyando and Kibos River basin that benefit largely from established irrigation scheme. Land is also prepared in August-September for short rain season cereal/vegetable crops. The catchment is considerably covered with Nitisols, Fluvisols and Cambisols that are stable to erosion and leaching under minimum management. Awach-Tende and Awach-Kibuon basins are an exception dominated with phaeozems that are susceptible to leaching and Acrisols that are prone to erosion. The fishing industry and favorable weather condition have catalyzed rapid urbanization extending to the shore of the lake and uncontrolled farming practices encroaching riparian zones.

The region is covered with three weather stations but only two (2) at Kisii (−00667, +034783) and Kisumu Airport (−00100, +034750) were utilized. Kericho (−00367, +035350) station had a lot of inconsistence data. Daily climate data (rainfall, maximum and minimum temperatures, relative humidity and solar radiation was used as model input and monthly data for the Weather Generator (WGN) Parameters. There are no hydrometric stations in this region. Local detailed Soil data map derived from KENSOTER for studies of carbon stocks [16] was used. The soil data (Figure 2) contains detailed information of the 2-soil depth: top soil (0 - 30 cm) and subsoil (30 - 100 cm). Land use statistics was obtained from World resource institute-WRI (Figure 3). It contains different classes; Agricultural land use (cereals, sugar, paddy rice/sugar under irrigation, perennial crops like tea), forests (Protected forest, evergreen, deciduous and mixed), Wetland and urban residential. Agriculture is dominant in the region covering more than 60% of the total land use. Crop management was scheduled as; long season I (January-Land preparation, March-planting and fertilization, August-harvesting) and Short Season II (September-land preparation, October-planting and fertilization, December harvesting). Digital Elevation model

(DEM) was derived from Shuttle Radar Topography Mission (SRTM) at 30 m resolution.

SWAT is a river or watershed, spatial model developed to predict the impact of land management practices on water, sediment hydrodynamics, and agricultural chemical yields in large complex watersheds with varying soils, land use, management conditions and weather conditions. The model consists of the following main components: Weather, hydrology, plant growth, nutrients (Nitrogen and Phosphorous based), pesticide, bacteria and land management-SWAT version 2012 [17]. The Model is run in QGIS interface enabling integration of spatial distributed characteristics of the watershed into calculations.

This study focuses mainly on hydrologic component coupled with effects of land use management to determine sediment and nutrient concentration in water of specific rivers discharging into the Gulf. In SWAT, watershed is portioned into sub-basins with a given threshold using DEM, which is further subdivided into Hydrological Response Units (HRUs) with homogeneous soil type, land use and slope [18]. HRUs form basis for water balance calculation utilizing nearest weather station data. Watershed hydrology is calculated in two separate components: land phase and routing phase. The land phase determines the amount of water (surface and base flow), sediment, nutrient and pesticide loading into the route channel in each sub-basin [19]. These hydrologic processes are based on infiltration, percolation, evaporation, plant uptake, lateral flows and groundwater flows including snowfall and snowmelt [14]. Sediment yield is estimated based on Modified Universal Soil Loss Equation (MUSLE) which factors in the surface runoff volume, the peak runoff rate, the area of the HRU, the Universal Soil Loss Equation (USLE) soil erodibility factor, the USLE cover and management factor, the USLE support practice factor, the USLE topographic factor, and a coarse fragment factor. On other hand routing phase defines movement of water, sediments and nutrient loads from each channel network to the outlet. Channel transmission losses, evaporation, return flow etc., are adjusted for estimation of outflow from a channel which is predicted by the Muskingum method [20]. The hydrological balance is simulated by SWAT model according to the equation below [21].

S W t = S W o + ∑ i = 1 t ( R d a y − Q s u r f − E a − W s e e p − Q g w ) (1)

where: SW_t is the final soil water content (mm); SW₀ is the initial soil water content on day i (mm); R_day is the amount of precipitation on day i (mm); Q_surf is the amount of surface runoff on day i (mm); E_a is the amount of evapotranspiration (ET) on day i (mm); W_seep is the amount of water entering the vadose zone from the soil profile on day i (mm); Q_gw is the amount of return flow on day i (mm).

SWAT-Calibration Uncertainty Programs version 2012 (SWAT-CUP) was utilized for Calibration/validation, uncertainty and sensitivity analysis of the model. Investigation of sensitivity and uncertainty in stream flow and sediment concentration was done by Sequential Uncertainty fitting (SUFI-2) algorithm. Several objective functions were used to gauge model performance by: coefficient of linear correlation R², Nash-Sutcliffe Efficiency (NSE) and the coefficient of percentage biasness (PBIAS). NSE (Equation (2)) is a normalized statistic that determines the relative magnitude of the residual variance in comparison to the measured data variance indicating how well the plot of observed fits the simulated data; 1:1 [22] [23]. NSE = 1 is the optimal value, values between 0.0 and 1.0 regarded as acceptable levels of performance of the model, whereas values ≤ 0.0 indicating that mean observed value is a better predictor than simulated value (unacceptable performance of the model).

NSE = 1 − ∑ i = 1 n ( X i o b s − X i s i m ) 2 ∑ i = 1 n ( X i o b s − X a v g ) 2 (2)

where X i o b s is the ith observation for the constituent being evaluated, X i s i m is the ith simulated value for the constituent being evaluated, X a v g is the average/mean of observed data for the constituent being evaluated, and n is the total number of observations.

PBIAS (Equation (3)) is a deviation term used to evaluate the accumulation of differences in streamflow/sediment concentration between simulated and measured data for the period of analysis. PBIAS = 0 is optimal indicating unbiasedness and larger value show more variance between simulated value and observed information. Positive value indicates model overestimation bias, and negative value indicates model underestimation bias.

PBIAS = ∑ i = 1 n ( X i o b s − X i s i m ) ∗ 100 ∑ i = 1 n ( X i o b s ) (3)

In this study, model performance for a monthly time and specific day data was judged as satisfactory if NSE > 0.50 and PBIAS = ±25% and graphical analysis that reveals a good agreement between predicted and measured hydrographs (R² = 0.6).

The calibration of a conceptual model necessitates setting the input variables to correspond optimally in mimicking measured observations thus representing the reality on of studied phenomenon. It is deliberately carried out with the purpose of defining the values or desirable ranges of the model parameters that depend broadly on the nature and specific properties of the study area. Preliminary analyses, subsequent simulation of databases combination of DEM, precipitation, crop and soil, yielded a fair default performance (NSE = 0.0922) before calibration. The impact of soil data set was most significant in the modeling of the basins for discharge and water quality. Calibration of SWAT model in the Nyanza through semi-automated approach (SUFI-2) method was performed over a period of 9 years by comparing the mean monthly measured flow rates (stream flow estimated done by floating method) and sediment concentration to simulated rates. This was performed to the 5 study rivers with monthly mean estimates of the flow rates and water quality regarding TSS concentration from 2005-2014. The following represents one of the major basins (Figure 4) in Gulf’s catchment during calibration and validation period basing on monthly discharge and concentration of sediments loads.

These figures exhibit successful model simulation of the river flow and sediment concentration variations. The model meets the performance assessment criteria as recommended for a monthly time step [23]. The calibration of individual river basins attained good model performance for flow rates with a

coefficient of NSE of the order of 0.69 to 0.81 and PBIAS of the order of 9.52 to 21.45. The performance of the model for sediment concentration was within adoptable range with NSE of 0.64 to 0.79 and PBIAS of 9.09 to 19.63. The model of the largest contributing basin, Sondu River, had coefficient of NSE 0.72, R² of 0.82 and PBIAS of 13.32 basing on monthly mean flow rates. The sediment concentration of the same basin has a coefficient NSE of 0.78, R² of 0.79 and PBIAS of 11.63 as shown in Figure 5. Calibration decreases the uncertainties associated with modelling and provides better estimates of the studied seasonal water qualities in the catchment [25]. Due to lack of consistence in either daily or monthly data available on the concentration of nutrients in these basins, manual calibration was in cooperated to adequately estimate the daily water quality flowing into the Gulf.

Validity of a model is gauged on how comparable it is to the actual situation of the study area basing on the series observations done over a defined period. Any calibration procedure of a model should be put necessarily to the control of testing its reliability and performance. Validation of SWAT model of each basin was performed from 2014-2015 (1 year) over the period of calibration (2005-2014) by comparing monthly mean flow rates, specific day sediment and nutrient concentration to measured data (e.g. Figure 6). Sondu Miriu basin. The limnology data was obtained from selective grab samples at the mouth of the rivers and water quality analysis done on specific days during Lake Victoria Comprehensive Research for Development (LAVICORD PROJECT, 2015).

Considering daily simulated results of the model in comparison with the observations, all the modeled river basins showed high performance as indicated below (Table 1). Despite of few observed water quality data for Awach-Tende river, the model response was quite successful indicating both equivalence and trends in the changes (Figure 7 and Table 2). Its probability of biasness was within good range though higher compared to the other models.

The results indicate high inflow of sediments with mean of 578 mg/L, 526 mg/L, 384 mg/L,198 mg/L, and 143 mg/L from Awach-Kibuon, Awach-Tende, Nyando, Sondu Miriu and Kibos respectively as (Figure 8). The highest loads of sediment into the gulf within the study period was found to be in April-May

during long rains followed by October-November during short rain season.

Total nitrogen and Phosphorous Loads were substantially high in April ? May as modeled in Figure 9 & Figure 10 attributed to high precipitation and agricultural activities within the entire catchment.

In modeling river discharge, sensitive values in each basin were evaluated in parameter estimation process. Table 3 shows optimum values determined by semi-automated SUFFI-2 imbedded in SWAT-CUP for calibration of Stream

flow, Sediment and nutrient loads of the 5 basins. Based on the 22 selected SWAT parameters, globalized sensitive analysis was used for identifying sensitive and important model parameters while holding ALPHA_BF at fixed value. Several iterations and simulations of each basin independently till acceptable results were realized. Eight (8) parameters i.e. CN2, CH_K2, SOL_AWC, USEL_P, NPERCO, PPERCO, SOL_K and SPEXP were found to be the most sensitive for the gulf catchment in entirety. The performance of the SWAT model for the 5 basins was good during calibration with NSE > 70, thus consistency in the data sets utilized.