Predictive modeling of terrestrial radiation for optimizing solar-powered irrigation under limited data in Adrar region (Algeria)

An accurate estimation of the radiation is a prerequisite for the net radiation balance, evapotranspiration modeling, and optimal scheduling of water extraction by solar-powered irrigation, especially in the water-scarce Saharan zone. In this study, we developed a predictive model to estimate daily terrestrial radiation at the surface of the Adrar region in Algeria. We used a training set of 25 years of data (2000–2025) from 10 stations (Adrar, Tamantit, Sidi Ahmed Timmi, Fenoughil, Zaouiet Kounta, Reggane, Gharmianou, Tittaf, Ikiss, Kassbet Lahrar) with only three features: soil temperature (0–7 cm), air temperature (2 m), and vapor pressure deficit. The robustness of the models was ensured by a time-based split. The random forest (RF), gradient boosting (GB), and extra trees (ET) tree-based models were evaluated on the generalization set. RF and ET exhibited the best performance (R = 0.92, rootean square error—RMSE = 31.85 W/m2, Nash–Sutcliffe efficiency—NSE = 0.84 for RF; R = 0.92, RMSE = 32.33 W/m2, NSE = 0.84 for ET), whereas GB exhibited poor performance (R = 0.90, RMSE = 36.67 W/m2, NSE = 0.79). Finally, we proposed a map for solar-powered irrigation optimization. In particular, we demonstrated that the southern part of the Adrar region (Reggane, Zaouiet Kounta) has high potential for solar-powered irrigation (more than 350 W/m2). This study contributes to hyper-arid agricultural regions in Algeria through water conservation and the utilization of renewable energy.