Modeling flood dynamics in the Sudd wetlands using Google Earth Engine and AI models optimized with metaheuristics in a changing climate

Abstract

ABSTRACT This study proposes a reliable, automated methodology to delineate the spatial extent and variability of the Sudd wetlands under changing climatic conditions. Due to limited in situ measurements and ground observations, the approach primarily relies on remotely sensed data processed through the Google Earth Engine platform. As with other optical sensors, cloud contamination posed a significant challenge to data accessibility. In the absence of ground-observed data, NDWI-derived water maps demonstrated moderate to high consistency when validated against historical and time-lapse imagery from Google Earth for specific seasons. Metaheuristic algorithms were employed to optimize the hyperparameters of the predictive artificial intelligence models, resulting in improved performance and accuracy. Unlike traditional grid or random search methods, metaheuristics such as genetic algorithms and particle swarm optimization leverage adaptive, iterative strategies to explore complex problems efficiently. These algorithms balance global exploration and local exploitation, enabling the identification of near-optimal configurations that enhance model generalization and reduce overfitting. Optimal threshold values were applied to generate a time series of water maps for the Sudd region spanning 2025–2099. The scenarios were based on five climate projections and three scenarios per model. Findings indicate climate change exerts a detrimental influence on the hydrological dynamics of the Sudd wetlands.