Research area: Lake Chad (Republic of Chad). Purpose: To identify significant remote sensing and ground-truth climate factors and their interactions and contributions in predicting remote sensing and ground-truth lake levels. A comparative analysis from 2013 to 2021 using Linear model (LM), regression tree (RT), random forest (RF), and gradient boosting regression (GBR) shows that GBR outperforms other methods for both remote sensing and ground-truth data. Ground-truth lake level regressed on ground-truth features ( \({R}^{2}\) = 71%, \(MAE\) = 0.23, \(MSE\) = 0.09, \({CV}_{MSE}\) = 0.12) outperforms that regressed on remote sensing features ( \({R}^{2}\) = 64%, \(MAE\) = 0.27, \(MSE\) = 0.11, \({CV}_{MSE}\) = 0.15). Local Interpretable Model-agnostic Explanations and SHapley Additive exPlanations based on GBR reveal that ground-truth air temperature influences the most ground-truth lake level: higher temperatures decrease predictions, while lower temperatures increase them. Remote sensing precipitation also significantly affects ground-truth lake level: higher precipitation reduces predictions, while lower amounts increase them. Air temperature emerges as the most critical factor, whether from remote sensing or ground-truth data. Precipitation and evaporation are 90% clustered, irrespective of the data source. These findings provide valuable insights for decision-makers regarding the impacts of climate change and water resource management. Further studies are necessary for validation purposes.
