Optimising water allocation with hybrid machine learning–water evaluation and planning under climate and population pressure in Upper Awash–Akaki, Ethiopia

ABSTRACT This study presents a hybrid approach combining machine learning (ML) and the Water Evaluation and Planning (WEAP) model to optimise water allocation in the Upper Awash–Akaki catchment, Ethiopia, under projected climate and population pressures. Population growth (2.5–3.5% annually) is expected to increase demand from 6.3 million in 2025 to nearly 39 million by 2075. Climate scenarios (SSP4.5 and SSP8.5) project temperature increases of 0.9 to 1.6 °C and precipitation changes ranging from +15.5% to −6.3%, intensifying hydrological stress. Hydrological impacts include a 46.9 mm decline in baseflow, reduced soil moisture (−124.9 to +16.4 mm), and evapotranspiration losses up to 1,010 mm, leading to unmet water demand exceeding 1.58 billion m3. Integrating ML improved WEAP's forecasting, with Random Forest outperforming Long Short-Term Memory (LSTM) (MAE: 0.41 vs. 0.46). SHapley Additive exPlanations (SHAP) analysis identified lagged population and unmet demand as key predictors. Using Non-dominated Sorting Genetic Algorithm II (NSGA-II), two Pareto-optimal strategies emerged: one prioritising equity (Gini index 0.22), the other economic efficiency (9% gain), highlighting allocation trade-offs. The framework, supported by SHAP interpretation and IoT-based monitoring, offers a replicable tool for adaptive, climate-resilient water governance in data-scarce and rapidly urbanising basins.