Influence of Large-Scale Climate Indices on Reservoir Surface Extent Variability in West Africa

In recent years, artificial reservoirs have attracted increasing attention, not only for their essential role in mitigating hydrological and meteorological extremes but also for their vulnerability to climate variability in region like West Africa. This growing interest has been supported by advances in remote sensing, which now allow near-real-time monitoring of reservoir surface extent (RSE) dynamics. However, regional-scale research quantifying and communicating trends and variability in reservoir surface dynamics remains limited. Additionally, while the effects of large-scale climate indices on hydrological processes have been largely investigated, any study has specifically examined how RSE respond to these climate indices across West Africa. At the same time, regression-based machine learning approaches, frequently used to assess multiple teleconnections while addressing multicollinearity issues, often lack systematic evaluations of their robustness and reliability. These gaps constitute important challenges for both local and regional efforts to monitor hydroclimatic shift impacts and anticipate water resource stress under ongoing global warming. In this study, we addressed these challenges by assessing the spatiotemporal variability and trends in RSE dynamics from 1985 to 2022 for 482 reservoirs across West Africa in relation to ten Sea Surface Temperature Anomaly (SSTA) indices. We further evaluated the performance of three supervised machine learning methods, Ridge Regression, Elastic Net, and Partial Least Squares (PLS) to identify the most suitable for modeling and predicting the effects of SSTA indices on RSE. Finally, we identified the dominant oscillations influencing RSE dynamics and highlighted the regional response patterns of West African reservoirs to the ten SSTA indices. We found strong interannual and decadal variability in both the SSTA indices and RSE, underscoring a dynamic coupling between oceanic conditions and terrestrial hydrology in West Africa. Among the ten indices, the Western Mediterranean Index (WMED) shows the strongest and statistically significant upward trend (p < 0.05). At the reservoir level, 43.26% of the 485 reservoirs exhibit significant long-term trends, with 31.07% showing declines. Of the three algorithms tested, PLS delivers the best generalizability and the most stable out-of-sample predictive performance, but only when using PCA-filtered and low-collinearity predictors. WMED emerges as the most influential driver, with moderate contributions from the Atlantic modes (AMO and AMM). Finally, SSTA regression coefficients vary widely across reservoirs, with minimal spatial clustering, indicating uneven reservoir sensitivity to oceanic oscillations likely shaped by local factors.