EVAPOTRANSPIRATION MODELING FOR EFFICIENT WATER RESOURCE MANAGEMENT: A COMPARATIVE STUDY OF TAKAGI-SUGENO FUZZY SYSTEMs AND GENERALIZED REGRESSION NEURAL NETWORK MODELS IN SEMIARID ALGERIAN REGIONS

The objective of this study is to compare two modeling approaches, the Takagi-Sugeno Fuzzy System (TSFS) and Generalized Regression Neural Network (GRNN) models, with evapotranspiration calculated by FAO-56. The selected sites are situated in Constantine, Guelma, Mascara, Saida, Setif, Souk Ahras, Tiaret, and Tlemcen, known for their semiarid climates. The daily data registered from 2000 to 2022 include air temperature at 2m (°C), relative humidity at 2m (%), dew point at 2m (°C), precipitation (mm), surface pressure (hPa), ET0 FAO evapotranspiration (mm), vapor pressure deficit (kPa), wind speed at 10m (km/h), soil temperature from 0 to 7cm (°C), soil moisture from 0 to 7cm (m³/m³), sunshine duration (s), and terrestrial radiation (W/m²). The data were split into training (70%), validation (15%), and testing (15%) sets. To evaluate the two models, several indices were calculated, including Nash–Sutcliffe efficiency, coefficient of determination, root mean square error, mean absolute error, ratio sum ratio, and Willmott index.

The statistical results indicate that the GRNN model provides more accurate estimations of evapotranspiration compared to the TSFS model in semiarid regions. This is evidenced by a root mean square error (RMSE) of ≤ 0.285, a mean absolute error (MAE) of ≤ 0.212, a minimum coefficient of determination (R²) of 0.976, a Nash–Sutcliffe efficiency (NSE) of ≥ 0.976, a ratio sum ratio (RSR) of ≤ 0.156, and a Willmott index (WI) of > 0.882 for training, validation, and testing. In contrast, the TSFS model shows an RMSE of ≤ 0.513, an MAE of ≤ 0.405, an R² of > 0.923, an NSE of ≥ 0.965, an RSR of ≤ 0.277, and a WI of > 0.799.

The findings of this study confirm that the GRNN model is more suitable for accurately estimating evapotranspiration in semiarid regions, contributing to efficient water resource management in these areas . Future research should focus on expanding the dataset to include diverse climatic regions to enhance the models' applicability.