CMIP6 multi-model ensemble projection of reference evapotranspiration using machine learning algorithms

Changes in reference crop evapotranspiration (ETo) due to climate change (CC) can severely impact food and water security, emphasizing the need for integrating ETo projections into agricultural water management strategies. In this study, ETo changes were projected for two future time slices with respect to the baseline using several machine learning techniques, incorporating minimum and maximum temperature, diurnal temperature range, and extraterrestrial radiation across Iran. Additionally, an ensemble of 10 CMIP6 Global Climate Models, downscaled by the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP-CMIP6), was employed. The X-means clustering algorithm was also exploited to classify ETo based on various characteristics, including minimum, maximum, average, skewness, and standard deviation, as well as ETo ranges of 0–5, 5–10, and greater than 10 mm d⁻¹. This clustering approach divided the study area into five distinct clusters. Apart from cluster I, where the Support Vector Machine outperformed, the Random Forest technique provided more accurate ETo predictions. The findings project an average ETo increase of 4.8 % and 5.3 % during 2030–2049, and 8.0 % and 13.3 % for 2080–2099 under SSP245 and SSP585, respectively. Geographically, the highest ETo increases are anticipated primarily in the northern and western parts of the country, predominantly within clusters I and II. Notably, the ETo rise will exceed 40 % relative to the baseline during the late century under the SSP585. Furthermore, the most significant ETo increment is expected during winter. Future projections also indicate that cluster V, which already experiences significant daily ETo peaks, will face even more ETo extremes. Given the critical importance of these regions for sustaining food and water security and preserving natural resources, the substantial rise in ETo under future CC poses a significant threat to natural sustainability in Iran. This highlights the critical necessity for adaptive strategies in agricultural water management to mitigate the adverse CC effects. In this context, the current findings can assist decision-makers in identifying hotspots and quantifying CC impacts, thereby enabling the design of crucial adaptations.