Ubuntu Flood AI: A Community-Centered Framework for Integrating Traditional Knowledge and Machine Learning in Flood Early Warning Systems

Flood early warning systems (EWS) in data-scarce regions like Ethiopia often face significant challenges, including limited sensor coverage, model transferability issues, and low community trust. This paper presents UbuntuFlood AI, a novel, integrative flood monitoring and prediction framework designed for the Ribb River watershed in the Amhara Region. The system addresses these challenges by fusing multi-source satellite data (water indices, precipitation) with locallysourced Traditional Knowledge (TK), including seasonal calendars, animal behavior indicators, and river observation protocols. UbuntuFlood AI features a modular software architecture built with Streamlit, which includes a data processing engine, a machine learning-based flood predictor, and a TK integration module. The model calculates a dynamic, weighted flood risk score by combining real-time water and vegetation indices such as the Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI), and Normalized Difference Vegetation Index (NDVI) values, rainfall patterns, seasonal baselines, and community-based assessments. A detailed case study of the July 2025 flood event demonstrates the system's capability to provide community alerts. By embedding cultural context and community participation into its core, UbuntuFlood AI enhances not only predictive accuracy but also fosters local ownership, trust, and resilience, offering a scalable and sustainable model for communitycentered disaster risk reduction. Keywords: Flood Early Warning, Traditional Knowledge, Machine Learning, Data Integration