Abstract:
Water scarcity is a persistent challenge in semi-arid regions like Karamoja in northeastern
Uganda, where effective water resource management is critical for agricultural sustainability.
This study presents a decision-support model for the Kobebe Dam catchment that integrates
rainfall prediction, runoff estimation, and dynamic water allocation to optimize water
availability and usage. A machine learning approach was employed to forecast rainfall using
33 years of historical weather data from NOAA and NASA, including variables such as
temperature, humidity, wind speed, and pressure. Among the three models tested Random
Forest, XGBoost, and Linear Regression the Random Forest Regressor demonstrated superior
performance, achieving an RMSE of 0.00575 mm, MAE of 0.00244 mm, and an R² of 0.99998.
Its high accuracy provided reliable monthly rainfall forecasts, including a predicted 35 mm for
May 2025. Using the Soil Conservation Service Curve Number (SCS-CN) method, rainfall
predictions were converted into runoff estimates. Soil and land use data sourced from FAO and
local hydrological assessments produced a weighted average Curve Number of 69.33. For May
2025, the runoff depth was estimated at 0.89 mm, corresponding to a total runoff volume of
approximately 9.72 million litres flowing into Kobebe Dam—highlighting its potential for
meaningful rainwater harvesting. A flexible water allocation algorithm was then applied to
simulate various distribution strategies based on available runoff, crop water requirements, and
management priorities. Initial tests used demonstration ratios of 80%, 15%, and 5% to validate
the algorithm’s capacity to model diverse allocation scenarios. These ratios are not fixed but
are intended to explore “what-if” situations, allowing iterative evaluation of water distribution
outcomes under changing conditions. This adaptability supports the identification of optimal
allocation strategies that ensure sustainable water use and continuous crop production,
ultimately enhancing water security in the region.
