Modeling and Trend Analysis of Reference Evapotranspiration in Western Maharashtra, India

In this study, reference evapotranspiration (ET) was estimated over nine stations of western Maharashtra, India, using nine climate-based models, four linear regression (LR) models and four artificial neural network (ANN) models. None of the climate-based models replicated the spatial patterns of ET_o as depicted by the FAO-56 Penman-Monteith (P-M) method. The P-M method based ET_o values indicated an increase in ET_o values from the central region toward the southeast part of the study area, with comparatively higher values also observed in the northeast area. The LR and ANN model outputs showed spatial patterns closely aligning with the P-M method. Overall, all LR and ANN models provided reliable ET_o predictions. Hargreaves-Samani (H-S) and Pan Evaporation (PAN) models showed close ET_o patterns, with ET_o values of H-S model being higher (4.79-5.40 mm day^-1) than that of PAN (3.22-5.11 mm day^-1). Likewise, Jensen-Haise (J-H) and Turc models showed almost similar patterns, but ET_o values of J-H model were higher (3.02-4.01 mm day^-1) than that of Turc (1.68-1.87 mm day^-1). Also, Priestly-Taylor (P-T) and Radiation (RAD) models showed spatial patterns differing from other climate-based methods, with ET_o ranging from 3.92-4.39 mm day^-1 and 5.72-6.02 mm day^-1, respectively. Trend results using Mann-Kendall test indicated a decrease in ET_o, estimated using the P-M model, at all stations. The slopes obtained from the P-M method estimated ET_o indicated decreasing ET_o trends at all stations. The climate-based models, SCS-BC, THOR, and Hargreaves-Samani (H-S) exhibited positive slope values across most stations, indicating a gradual increase in ET_o. These findings highlighted the importance of developing region-specific ET_o estimation tools to support effective irrigation planning, drought mitigation, and sustainable water resource management under conditions of climatic variability and limited facilities available to record meteorological observations. The LR and ANN approaches demonstrated reliable capability for ET_o predictions and strong agreement with that of the P-M method, even under limited inputs.