Hybrid ANN-PCA model for evapotranspiration estimation in an arid environment

ABSTRACT

Hybrid ANN–PCA framework can enhance evapotranspiration prediction for sustainable irrigation in arid regions.

Accurate estimation of daily reference evapotranspiration (ETo) is essential for sustainable irrigation scheduling and water resource management, particularly in arid and semi-arid regions. This study evaluates a hybrid modeling framework that integrates artificial neural networks (ANNs) with principal component analysis (PCA) to improve daily ETo estimation across three climatically distinct regions of Iraq (Mosul, Baghdad, and Basra). The analysis is based on quality-controlled daily meteorological observations for the period 2020–2024. In the first stage, standalone ANN models were developed using ten meteorological inputs, including air temperature, humidity, solar radiation, wind speed, and sunshine duration. In the second stage, PCA was applied to reduce input dimensionality and mitigate multicollinearity prior to ANN training. Results show that standalone ANN models achieved strong predictive performance, with correlation coefficients ranging from 0.91 to 0.97 and RMSE values of 0.86 mm/day (Mosul), 0.80 mm/day (Baghdad), and 1.43 mm/day (Basra). The hybrid ANN–PCA models consistently outperformed the standalone ANN models at all stations. For Mosul, RMSE decreased from 0.86 to 0.26 mm/day and mean absolute percentage error (MAPE) from 16.3 to 2.1%. In Baghdad, RMSE was reduced from 0.80 to 0.33 mm/day and MAPE from 12.8 to 4.3%, while in Basra RMSE decreased from 1.43 to 1.15 mm/day and MAPE from 20.0 to 4.3%. Overall, RMSE reductions ranged from approximately 20% in the Gulf-influenced southern region to nearly 70% in the northern Mediterranean climate. The PCA results indicate that temperature and wind speed dominate ETo variability in Mosul and Baghdad, whereas relative humidity plays a more influential role in Basra. These findings demonstrate that integrating PCA with ANN modeling substantially enhances predictive accuracy while providing new insight into the climate-dependent performance and robustness of hybrid models across contrasting hydroclimatic conditions. Unlike many previous studies limited to single locations, this work evaluates the robustness of the ANN–PCA framework across climatically contrasting environments, providing new insight into climate-dependent model performance.