Browsing by Author "Lolem, Joseph Pelerino"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Development of a Hybride Model for Prediction of River Water Quality
    (Busitema University, 2026) Lolem, Joseph Pelerino
    Water quality degradation in transboundary catchments threatens public health and water supply, yet predictive tools combining physical hydrology with data driven classification remain absent for data scarce regions like Uganda’s Malaba River catchment. This study developed, calibrated, and validated a hybrid SWAT ANN model to predict river water quality classification from simulated hydrological and pollutant outputs. SWAT was set up using CHIRPS rainfall, NASA POWER climate data, land use, soil, and DEM, calibrated (2003–2014) and validated (2015–2025) against observed streamflow, achieving satisfactory performance (calibration NSE=0.79, R²=0.85; validation NSE=0.69, R²=0.76). An ANN with three hidden layers was trained on the Washington State WQI dataset (805 clean records) to classify water quality into five CCME classes, attaining 84.47% accuracy, with sediment and turbidity as dominant drivers. Applied to SWAT outputs, the hybrid model generated 276 monthly predictions (2003–2025), predicting Fair (48.6%) and Marginal (42.0%) as dominant classes, no Poor quality, and identifying hotspot subbasins (1,6,19,24). The Long-Wet season (March–May) showed 75.4% Marginal conditions, while July recorded the best quality. External validation achieved 100% agreement with primary field data (mean WQI 76.2, Fair) and seasonal benchmarks (4/4 seasons), with a paired t test confirming no significant difference between observed and simulated streamflow (p=0.0506). The hybrid SWAT ANN framework provides a physically grounded, data driven water quality prediction tool supporting proactive management by MWE and NEMA.