A hybrid AquaCrop–XGBoost framework for climate-smart irrigation and crop yield resilience under CMIP6 scenarios in the Lake Tana Basin, Ethiopia

This is the authors' abstract. We don't add key points for this paper.

Climate variability and increasing atmospheric evaporative demand are reducing crop water productivity and yield stability in semi-humid agricultural systems, requiring improved irrigation strategies. This study integrates field experiments, the AquaCrop model, and machine learning to evaluate drip and furrow irrigation systems for tomato production in the Lake Tana Basin, Ethiopia. Field experiments conducted during 2023–2024 were used to calibrate and validate AquaCrop, which showed good agreement with observed tomato yield during calibration (R 2 = 0.868 for drip and 0.848 for furrow irrigation) and validation (R 2 = 0.833 and 0.811, respectively). Basin-scale yield prediction was performed using 72 spatial samples and environmental predictors, with multiple machine learning algorithms evaluated. XGBoost achieved the highest predictive performance (R 2 = 0.86; RMSE = 0.64 t ha −1 ) and was selected for spatial yield simulation under CMIP6 climate scenarios. Results showed that drip irrigation substantially improved tomato productivity, producing higher mean yields (5.48–6.23 t ha −1 ) than furrow irrigation (2.10 t ha −1 ) and increasing water productivity (0.75 vs. 0.21 kg m −3 ). Under SSP2–4.5 and SSP5–8.5 scenarios, drip irrigation maintained greater yield stability, lower variability (CV = 2.7–2.9%), and higher climate resilience (CRI = 0.97–1.05), whereas furrow irrigation showed greater variability and reduced resilience. The integrated AquaCrop–machine learning framework enables basin-scale assessment of irrigation performance and climate change impacts under heterogeneous conditions. Efficient irrigation management is therefore critical for sustaining tomato productivity, water-use efficiency, and climate resilience under future climate variability.