Monitoring bacterial contamination of West African surface waters using Earth observation data and machine learning methods

International audience In 2021, diarrheal diseases caused approximately 434,000 deaths in sub-Saharan Africa, mainly due to water contamination by fecal pathogens such as Escherichia coli. While environmental conditions and human activities are known to influence bacterial contamination in surface waters, their respective impacts remain poorly quantified, complicating efforts to model this health risk. In this context, developing approaches to monitor contamination without relying solely on field-based analyses has become increasingly important. This study explores the potential of Earth observation (EO) data to monitor bacterial contamination in surface waters in West Africa. It investigates the relationship between E. coli concentrations and various water quality parameters, some measured in situ (water quality parameters, suspended particulate matter (SPM), etc.), and others derived from EO data (rainfall, specific humidity, etc.). These variables are integrated into eight machine learning models capturing the temporal dynamics of E. coli concentrations in two sites: the Bagré reservoir (Burkina Faso) and Kongou Lake (Niger). Using all available variables, ensemble tree-based models provided the best predictive performance for both sites. When using only EO variables, these models maintained good performance, achieving R² values of approximately 0.7 for Kaporé and 0.65 for Kongou. For both sites, the variables playing the strongest role are SPM (or NIR band) and rainfall, to which are added, for Kaporé, air humidity and NDVI. These results demonstrate the feasibility of using EO data alone to monitor E. coli contamination in West African surface waters. This approach can enable remote monitoring of microbial water quality.