Artificial neural networks in solving accuracy challenges of short-term rainfall forecasting in the Lake Victoria Basin

Abstract

Accurate rainfall prediction is crucial for effective water resource management, disaster risk reduction, and agricultural planning in Uganda's Lake Victoria Basin. This study focused on Kampala as a representative case to address this need. Given that agriculture in the region is predominantly rain-fed and employs 68% of Uganda’s population, reliable forecasts are vital for economic stability and food security. However, achieving accuracy is challenging in this region because rainfall patterns are complex and non-linear, exhibiting high spatial-temporal variability. These patterns are influenced by large-scale climatic phenomena, including the Inter-Tropical Convergence Zone, El Niño/La Niña events, and the Indian Ocean Dipole, which often cause traditional forecasting models to struggle. This study investigated the use of Artificial Neural Networks (ANNs) to enhance shortterm (1-10 hours lead time) rainfall forecasting using Kampala as a case study within the Lake Victoria basin. A hybrid model, combining Convolutional Neural Networks (CNNs) with Long Short-Term Memory (LSTM) networks was developed to address the region's prediction challenges. The model was trained using 80% of a dataset of hourly weather data from the Uganda National Meteorological Authority and the Trans-African HydroMeteorological Observatory (TAHMO) while 20% of the data was reserved for testing. The model outputs a quantitative prediction of precipitation (in mm) for a 10-hour horizon Additionally, the study incorporated lightning data, a parameter often overlooked in traditional models but essential for accurately forecasting the convective rainfall prevalent in the study area. The results of the study show that the proposed CNN-LSTM model significantly improves forecasting accuracy over traditional approaches. Notably, it achieved a reduction in Mean Absolute Error (MAE) of up to 95.30% against the one-dimensional CNN model and 89.28% against the standalone LSTM model when used for point predictions. The model also consistently outperformed Random Forest and Vector Autoregression (VAR) models. Consequently, integrating such machine learning models into forecasting frameworks for areas within the Lake Victoria Basin, such as Kampala can directly improve disaster management, agricultural decisions, and water resource planning

Keyword: Artificial neural networks