| dc.description.abstract | Agriculture is a crucial economic activity in Uganda, with maize serving as a priority crop for both food security and income generation. Even though over 80% of households in Lake Kyoga Basin are cultivating maize, production remains below the national target. Maize production is affected by several biophysical drivers. In response, many farmers have adopted various land management practices to cope with these changes, although these adaptations have been constrained by low land productivity, resulting in persistently low maize yields. This study examined the effects of crop production drivers of climate, soil characteristics, and land management practices on maize yield in the Lake Kyoga Basin using Machine Learning (ML) models, specifically Random Forests(RF) and Lasso Regression, as well as the process-based Decision Support System for Agro technology Transfer (DSSAT) Ceres model. Multiple Linear Regression was also employed to evaluate the relationship between crop production drivers and maize yield. The study results showed that there was a significant relationship between maize yields and crop production drivers, particularly with area (0.000) and rainfall (0.041) and a negative relationship between soil pH, phosphorus and maize yields. However, soil nitrogen, potassium, organic carbon, temperature, and land management practices positively influenced maize yields. In model comparison, Lasso performed better than RF in training datasets (78% vs. 75%), but RF outperformed after cross-validation (68% vs. 65%) while the DSSAT model showed a 55% yield prediction accuracy. Both ML models recommended increasing the production area, and highlighted the importance of pesticide application at 96% and organic fertilizer application at 79% for highland ranges of Mbale district. For Kyoga plains in Pallisa district, soil health was important with nitrogen inorganic fertilizer application important at 70% and soil pH at 81.4%. DSSAT recommended improving soil organic carbon in highland ranges and nitrogen in the plains. Predictive analysis for 2025-2040 under SSP245 and SSP585 scenarios, projected a decline in maize yields in Mbale by 13.9% and 13.18%, respectively when subjected to projected increase in rainfall variability, while increased maize yield was projected under increased temperature. In contrast, maize yield increases of 19.3% and 15.3% for SSP245 and SSP585, respectively were projected in Pallisa, attributed to temperature increase and adoption of CSA practices. Rainfall variability was the biggest driver of maize yield variability between 2025 and 2040. This study recommends the implementation of effective land management practices in the various agroecological zones of the Kyoga Basin in order to enhance land productivity and enhance maize production. | en_US |
