The Murchison Bay is an extension of Lake Victoria situated in the south-east of Kampala which lies between latitudes 00˚10'00"N - 00˚30'00"N and longitudes 32˚35'00'E - 32˚50'00"E with average elevation of 1224 m above sea level. The Bay which is divided into Inner and Outer Bay covers an area of about 60 km². The Inner Murchison Bay is comparatively a semi-enclosed small water body with an area of 18.4 km² and length of 5.6 km off the main part of Lake Victoria. The Murchison Bay watershed covers an average combined catchment area of 282 km² which consists of about 20% swamps and 80% upland [7] [10] with the largest catchment drainage inlet being the Nakivubo channel which passes through Kampala City. This study covers the largest watershed of the Murchison Bay that drains through the Nakivubo Channel (Figure 1). It covers an area of about 40.9 km². The Nakivubo channel runs from the central populated Kampala district passing through slums, industrial settlements and markets before discharging its water to the Lake Victoria at Murchison Bay. The water in the channel is a mixture of secondary effluents from NWSC sewage treatment at Bugolobi within Kampala city and heavily polluted untreated wastewater from other parts of the city. There used to be a predominant papyrus wetland filtration system but it has been drained and turned into agricultural land or developed for commercial, industrial or residential purposes.

The catchment is within the equatorial belt, and has a moist sub-humid climate. It receives a bi-seasonal rainfall in the periods of March to May and September to November [25] . Mean monthly precipitation ranges from 24 mm in January to about 154

mm in October with mean precipitation of about 1293 mm/year.

The soils of the Nakivubo swamp area are alluvial and lacustrine sands, silts, and clays overlying granite gneisses. Alluvial soils within the swamp range from semi-liquid organic material in the very upper layers of the emergent vegetation zones, through reddish ferruginous loams to clays [25] .

The input data used in the study included the Digital Elevation Model (DEM) of Central Uganda, land use-cover map of the Murchison Bay catchment, soil and hydro- meteorological data of the study area.

The model uses readily available input data such as Digital Elevation Model (DEM), land use data, soil data and climatic data as described in Section 2.2. In this study, modelling of the hydrological process was carried out using the extension of SWAT for ArcGIS software called ArcSWAT [31] .

ArcSWATv2012.10.1.18 was downloaded from the website for the model (http://swat.tamu.edu/software/arcswat/) and installed in ArcGISv10.1. In the model setup, the first step was to delineate the catchment using the DEM into several connected sub-basins. The sub-basins were further divided into smaller units called hydrologic response units (HRUs). HRUs are lumped land areas within the sub-basin that are comprised of unique land cover, soil, slope and management combinations [21] . A total of 23 sub-basins and 247 HRUs were created. The HRUs were created by defining the thresholds of land use over sub-basin area at 5%, soil class over land use area at 5% and slope class over soil area at 5% using the multiple HRUs definition. The model was run on daily time step for a period of 16 years from 1995 to 2010 with a warm up period of two years.

Sensitivity analysis was performed to choose the most sensitive flow parameters [32] [33] [34] [35] that influence the catchment represented by SWAT to be used for calibration. This was achieved using the global sensitivity approach in semi-automated Sequential Uncertainty Fitting (SUFI2) algorithm. The global sensitivity analysis method takes into consideration, the sensitivity of one parameter relative to the other in order to give their statistical significances [36] . The t-statistics and p-values of the parameters were used to rank to the different parameters considered to influence flow and the final selection done based on the significance of the ranked values. Table 2 shows stream

flow parameters that were tested for their sensitivity. These are useful in estimating the amount of flow from a catchment [37] .

Calibration was accomplished by comparing the output of the SWAT model with the observed data at the same conditions [19] [38] . For calibration and validation, the semi-automated Sequential Uncertainty Fitting (SUFI-2) calibration method within the SWAT Calibration and Uncertainty Procedures (SWAT-CUP) was used. SWAT-CUP version 5.1.6.2 was used. SWAT-CUP [16] is a stand-alone calibration program developed for SWAT that operates on Latin Hypercupe sampling procedures. Due to lack of observed data for sediment and nutrients, the model was calibrated and validated only for streamflow. The model was calibrated with observed daily discharge data for the period of 1997 to 2002 and validated for the period of 2004 to 2008 since there was no observed data for 2003. Though, there were gaps in the observed data, the challenge was addressed by writing it in a format suggested by [36] that SWAT-CUP tool can read.

To evaluate performance of the model during calibration and validation, statistical measures as well as graphical representations at daily time step were used. This was employed to confirm the relationship between simulated or predicted values and observed values [1] and to verify the robustness of the model [37] . Four statistical measures were employed. They are the coefficient of determination (R²), the Nash-Sutcliffe efficiency (NSE), the percent bias (PBIAS) and the ratio of mean squared error to the standard deviation of the measured data (RSR) [26] [37] [39] . Equations (1)-(4) were used to determine NSE, PBIAS, RSR and R², respectively. Other details of these measures such as their utility and satisfactory range of values are explained by Moriasi et al. [40] .

where O_i is the observed daily discharge, S_i is the simulated daily discharge, is the average measured discharge, is the average simulated discharge and n is the number of observations.

The global sensitivity analysis of 16 flow parameters (Table 2) showed that, only seven were very sensitive to flow. The rankings of the flow parameters are presented in Table 3 while the fitted values for the most sensitive parameters are indicated in Table 4. The most sensitive parameter was the SCS runoff curve number (CN2). The curve number estimates runoff based on the relationship between precipitation, hydrologic soil group and land uses. Other researchers [41] [42] [43] [44] have also found the SCS curve number to be the most sensitive streamflow parameter in modelling hydrology in their studies. The other sensitive parameters included the Manning’s n value for the main channel (CH_N2), effective hydraulic conductivity in the main channel alluvium (CH_K2) and groundwater delay time (GW_DELAY), baseflow alpha factor (ALPHA_BF), groundwater “revap” coefficient and the soil evaporation compensation factor (ESCO). Although, the rest of the parameters were found not to be sensitive to flow in the catchment as their p-values were greater than 5%, other studies [37] [39] [45] [46] [47] have found some of them to be sensitive in their studies. This is expected as conditions such as land use and land covers, soil characteristics and climatic factors vary from one catchment to the other.

The graphical results of the model streamflow simulation performance during the

The extension (e.g., .mgt, .rte) refers to the SWAT input file where the parameter occurs. The qualifier (r__) refers to relative change in the parameter where the value from the SWAT database is multiplied by 1 plus a factor in the given range, while the qualifier (v__) refers to the substitution of a parameter by a value from the given range.

calibration and validation periods are shown in Figures 3-6. The year 2003 was not incorporated in the validation period since there were no recorded observed daily discharges in that year. In general, graphical results during calibration (Figure 3 and Figure 5) and validation (Figure 4 and Figure 6) indicated adequate calibration and validation over the range of streamflow discharge, although the calibration results showed a better match than the validation results. Statistical model streamflow discharge simulation performance results during both calibration and validation periods are presented in Table 5. Generally, the statistical results showed a satisfactory performance of the model [40] . Just as with the graphical results, the model simulated streamflow discharge better during the calibration period than the validation period. The relatively low statistical measures during the validation can be attributed to the quality of the observed data

used as there were many missing days. The results also showed that, the model slightly underestimated flow during both the calibration and validation period as indicated by PBIAS values of approximately 8% and 23%, respectively. According to Moriasi et al. [40] , a PBIAS greater than zero is an indication that the model underestimated flow. During the calibration period (1997 to 2002), the observed and simulated mean daily flows were 3.1 m³/s and 2.91 m³/s, respectively, while during the validation stage (2004- 2007), the observed and simulated mean daily flows were 1.94 m³/s and 1.50 m³/s, respectively.

Also, the underestimations could be attributed to uncertainties that exist in the catchment such as the wetland processes, quality of input data and wastewater dis- charge from point sources which could not be accounted for during the simulation. These plausible reasons including the limitations of using the SCS curve number (CN2) to compute estimate water budget componentshave also been noted by Abbaspour et al. [32] [48] and Betrie et al. [37] to be the cause of over and under prediction of flow. This observation agrees with that of Mutenyo et al. [43] and Qiu et al. [47] who recorded an underestimated streamflow discharge by the SWAT model. However, Qiu et al. [47] suggested that underestimation or overestimation [37] of streamflow discharge by the SWAT model is partly due to the use of curve number,which cannot give accurate pre- diction of runoff for days with several storms. Mutenyo et al. [43] attributed inaccurate simulation of streamflow discharge by the SWAT model to insufficient rain data. It is obvious that, if all these uncertainties are minimised, a well calibrated SWAT model can efficiently predict flow in the catchment for any management purpose.

To evaluate the effects of land use and cover change, the SWAT model calibrated and validated for streamflow discharge was used to simulate different land use and cover scenarios on runoff and sediment yield. Land use covers at two different years thus 1995 and 2003 were used as input variables in the SWAT model in order to compare output as a result of their differences in land covers. The simulated results of surface runoff and sediment yield in the catchment of the Murchison Bay under different land use and land cover scenarios using the SWAT model are shown in Table 6. The results showed an increase in runoff and average upland sediment yield, from the catchment except maximum upland sediment yield which reduced very slightly between the two periods. Surface runoff increased from 101 mm/yr in 1995 to 128 mm/yr in 2003 representing 26.7%. This is attributed to the drastic land cover changes that occurred in the catchment. It highlights the effect of changes in land use and cover on hydrology in the catchment. Within the period, there was a sharp decline of forestland from 31.15% in 1995 to 13.91% while built-up land shot from 26.53% to 39.09%.

The observation agrees with the assertion by Nie et al. [24] and Tang et al. [49] that, an increase in urban set-up increases impervious surfaces and declining forest lands, which accelerates surface runoff. Similarly, [50] reported that an increase in annual surface runoff increased when forest lands were converted to urban areas and decreased during when forestland gained a significant expansion. Maalim et al. [51] also made similar observation, areas dominated by urban development has higher runoff and lower runoff in forested areas. On the other hand, different studies [14] [44] [52] have reported that an expansion in forestland and grasslands lead to a decrease in runoff. Although, there is general agreement among researchers that changes in streamflow is due to different types of forestry activities, such as afforestation that may lead to lower runoff generation, others suggest the amount of precipitation [44] [52] and climate change [53] partly play a role. The increase in built-up land within the period might have possibly created impervious layers which reduced infiltration and percolation of water to the shallow aquifers that caused surface runoff to increase. According to Diyer et al. [54] , when industrial or urban development takes place, the new land use changes how water is transported and stored. The combination of constraints connected to compacted lands and impervious surfaces such as parking lots, roads, sidewalks, and roofs leads to reduced infiltration. This causes, for instance, reduced water quality, increased volume and velocity of runoff, increased occurrence and severity of floods, and loss of storage capacity and runoff water in natural vegetation.

Surface Rrunoff is the carrier of all other components such as sediment, nutrient, pesticides, bacteria, agricultural waste, heavy metals, industrial solid and liquid waste that undesirably affects water quality. As surface runoff increases, it is evident that all the aforementioned components will correspondingly increase depending on the activities

taking place in the catchment. However, the simulated results showed there was not much change in upland sediment yield over the period with increasing runoff. Upland sediment yield rather reduced by at least 0.50% between the two periods. This could still be attributed to the dynamics of land cover that happened at that time. Nevertheless, average upland sediment yield was about 20.50% higher in 2003 than that of 1995. This means that, some sub-basins within the catchment might have been exposed to massive degradation such as construction works and intensive agricultural activities than other sub-basins which resulted in higher sediment yield as runoff increased.

At a rate of 26.54 tons/ha/yr and 26.40 tons/ha/yr which translates to about 108,557 tons/yr and 107,984 tons/yr respectively, the sediment yield estimated in the catchment is greater than the results reported by Kimwaga et al. [55] . In the Simiyu catchment of the Lake Victoria basin, which is much larger than the Murchison Bay catchment, [55] reported a modelled estimate of sediment yield of 98,467 tons/yr though the actual measured sediment yield was 2,075,144 tons/yr. The discrepancy was attributed to the data used to run the model.

The average estimates of sediment yield from upland and runoff from the landscape as well as possible mass loadings from point sources in the Murchison Bay catchment is a clear indication of the deterioration level of water quality that ends up into the Lake Victoria. This is supported by [10] situation analysis report that increasing human population and all their associated economic and social activities have accelerated the rate of delivery of nutrients and caused eutrophication in Lake Victoria. However, it is anticipated that the values of the components estimated have changed tremendously as the catchment is now condensed with more settlements, industries and increased in population of Kampala city from 1.2 million in 2002 to 1.5 million in 2014 [28] .