Neural Network Approximation Based on ANFIS and Geographic Information System Mapping for Reliable Evapotranspiration Prediction in Khenchela, Algeria

Accurate estimation of reference evapotranspiration (ET₀) is critical for sustainable water resource management, irrigation scheduling, and climate adaptation in heterogeneous semi-arid regions. This study presents a stream-lined neural network (NN) approximation inspired by the Adaptive Neuro-Fuzzy Inference System (ANFIS) for predicting daily ET0 in Khenchela province, northeastern Algeria. Utilizing meteorological and soil data from 2000 to 2024 at 16 representative stations (Babar (1), Babar (2), Babar (3), Baghai, Bouhmama, Chechar, Djellal, El Hamma, Kais, Khenchela, Khirane, M’sara, Remila, Tamza, Taouzient, and Zaoui), sourced from the Open-Meteo Historical Weather API, the model employs inputs including air temperature, relative humidity, precipitation, wind speed, sunshine duration, terrestrial radiation, soil temperature, and soil moisture. The NN was trained to closely approximate the FAO-56 Penman-Monteith reference ET0 values computed directly by the API. Performance evaluation yielded strong agreement across stations: R² > 0.96, RMSE 0.22 - 0.46 mm/day, NSE > 0.95, RSR < 0.13, and Willmott’s index 0.88 - 0.93, with peak accuracy (R² > 0.99, RMSE < 0.24 mm/day) at high-elevation sites. Spatial patterns, mapped via GIS-based in-verse distance weighting interpolation, revealed pronounced topographic and aridity-driven variability, confirmed by Emberger and De Martonne indices. This computationally efficient NN offers a scalable surrogate for FAO-56 calculations in data-limited, heterogeneous environments, supporting precision irrigation, drought monitoring, and adaptive strategies in semi-arid North Africa and Mediterranean regions.