Monitoring the temporal evolution of surface water distribution is crucial to support surface water management and conservation, and could also effectively contribute to the achievement of Sustainable Development Goal 6 (SDG 6) ‘Clean Water and Sanitation’ at the regional level. Despite its importance, there is a lack of an operational method for determining surface water extent that also shows the interannual variability in Ethiopia. We assess Gradient Tree Boosting (GTB), Support Vector Machines (SVM), and Random Forest (RF) running on the Google Earth Engine (GEE) using Landsat for surface water monitoring at four sites in Ethiopia from 1986 to 2023. The results show that GTB, RF, and SVM have excellent classification accuracies, with overall, producer, and user accuracies consistently above 90%. GTB slightly outperforms the other two machine learning methods. The estimated water cover for our study sites shows a high degree of agreement with a benchmark dataset from the Joint Research Center (JRC), as indicated by coefficient of determination (R2) > 0.9 and root mean square percentage error (RMSPE) < 1%. The surface water dynamics of the four study sites depict a long-term increasing trend from 1986 to 2023, characterized by notable inter-annual variability. We identify the locations of this variability by analyzing the frequency of water occurrence over time and find that 84–94% are permanent water bodies, with the remaining water surface area changing over time. Mann–Kendall trend analysis does not confirm a general pattern over time for the four sites, suggesting that local site characteristics, water management and anthropogenic impacts are superimposed on the likely effects of climate change. Therefore, our results provide spatiotemporal information for surface water monitoring to support water resource management and policy in Ethiopia. This could also effectively contribute to the sustainable use and achievement of SDG 6 at the regional level.
Tesfaye et al. (Wed,) studied this question.