Machine learning and CORDEX-Africa regional model for assessing the impact of climate change on the Gilgel Gibe Watershed, Ethiopia

Abstract

Climate change is one of the most pressing challenges of our time, profoundly impacting global water resources and sustainability. This study aimed to predict the long-term effects of climate change on the Gilgel Gibe watershed by integrating machine learning (ML) methods and climate model scenarios. Utilizing an ensemble mean of four regional climate models (RCMs) from the Coordinated Regional Climate Downscaling Experiment (CORDEX) Africa project, we forecast future climatic conditions. Although global and regional climate simulations offer valuable insights, their limitations necessitate alternative approaches, such as ML, for improved accuracy. Employing an ensemble ML model with Random Forest (RF), Extra Tree (ET), and CatBoost (CB) algorithms, we assessed various bias-correction methods using historical data from 1993 to 2009. Our results highlight the effectiveness of distribution mapping (DM) in capturing temperature variability and precipitation patterns, using the power transpiration (PT) method to represent precipitation variability. Projections indicate a decline in future precipitation under the RCP 8.5 (-32.2%) and SSP 4.5 (-88.8%) for 2024-2049, with further decreases expected for 2050-2099. Conversely, temperatures will rise under RCP 4.5 (TMAX 0.67 °C) and RCP 8.5 (TMAX 0.25 °C and TMIN 1.11 °C) in the near term, exacerbated by higher emissions under SSP 4.5 and 8.5. By leveraging an ensemble mean of four observed RCMs in an ML framework, our study successfully reproduced future Coupled Model Intercomparison Project (CMIP5) and (CMIP6) climatic datasets, with the CB model demonstrating superior performance in predicting future precipitation and temperature trends. These findings offer valuable insights for shaping future climate scenarios and informing policy decisions for the Gilgel Gibe Watershed, thereby enhancing water resource management in the basin and its environs.

Keywords: Bias correction; Climate change modeling; Climate change prediction; Machine learning; Statistical downscaling; Time series analysis.