Makerere University researchers have developed a new deep learning framework that could transform flood preparedness in regions with limited hydro-meteorological data. The study, published in the Journal of Hydrology, introduces Time-Varying Dynamic Model Averaging (TV-DMA), a method that improves river flow prediction by dynamically adjusting model weights based on predictive errors.
The innovation is particularly relevant for tropical catchments such as the Semliki and Tochi rivers, which have experienced devastating floods in recent years. With more accurate forecasts, communities will receive timely alerts, reducing the risk of loss of life and property.
The framework also strengthens water resource management. Reliable river flow predictions can guide planning for irrigation, drinking water supply, and hydropower, helping communities adapt to droughts and erratic rainfall. Because the approach is computationally efficient, local institutions can adopt modern forecasting tools without requiring extensive datasets or costly infrastructure.
Lead researcher Angel I. Tumushabe, alongside Seith N. Mugume of Makerere University and Johanna Sörensen of Lund University, emphasized that the framework is a practical solution for East African communities. “Our approach provides a strong basis for developing flood forecasting and early warning systems in regions with limited data,” the authors noted.
The study underscores Uganda’s growing role in advancing climate resilience research. By bridging cutting-edge science with community needs, Makerere University continues to contribute to sustainable development and disaster risk reduction across Africa.
Below is the attached publications.