Understanding and accurately predicting reference evapotranspiration (ET₀) is critical for effective water management, particularly in semi-arid regions facing increasing climatic variability. This study integrates climatic trend analysis, and machine learning models to improve ET₀ predictions in four key cereal-producing regions of Algeria: Oum El Bouaghi, Setif, Sidi Bel Abbes, and Tiaret. Using AgERA5 data from 1979 to 2021, climatic trends were assessed with the Mann–Kendall test and Sen's slope estimator. Significant increases in minimum and maximum temperatures were observed, with the steepest trends in Sidi Bel Abbes (0.034°C/year) and Oum El Bouaghi (0.029°C/year). Precipitation showed the largest decline in Sétif (-1.071 mm/year), while relative humidity (H%) decreased most sharply in Tiaret (-0.036%/year).Machine learning models, including Linear Regression (LASSO), k-Nearest Neighbors (KNN), Random Forest (RF), Adaptive Boosting (AdaBoost), Gradient Boosting Machine (GBM), and Support Vector Machine (SVM) were evaluated under various input scenarios. Results indicate SVM achieving near-perfect predictions in Sidi Bel Abbes (R2 = 0.999, RMSE = 2.007 mm/month, MAE = 1.724 mm/month)and Tiaret (R2 = 0.977, RMSE = 8.6 mm/month, MAE = 7.337 mm/month), under scenario 8 (S8), and scenario 7 (S7), respectively. In Oum El Bouaghi, the Random forest and Adaboost models performed best under S7 (R2 = 0.976, RMSE = 7.082 mm/month, MAE = 5.552 mm/month), and S8 (R2 = 0.976, RMSE = 7.036 mm/month, MAE = 6.5 mm/month), respectively. While KNN showed strong performance in minimal-input scenarios (scenario5), in Setif(R2 = 0.977, RMSE = 5.45 mm/month, MAE = 3.729 mm/month). LASSO regression demonstrated high accuracy in Tiaret (R2 = 0.982, RMSE = 7.683 mm/month, MAE = 7.415 mm/month), when mean temperature and relative humidity were included. Climate projections under the SSP2-4.5 scenario using CMIP6 multi-model ensemble data suggest a significant increase in ET₀ across all studied regions, with SVM and Linear Regression models projecting the highest values by 2070. These results underscore the importance of Tmax, Tmean, and H% as key predictors of ET₀ in semi-arid climates.The findings provide critical insights for developing sustainable irrigation and water management strategies, contributing to enhanced agricultural resilience and food security in regions increasingly affected by climate change.

Trends in climatic variables and machine learning based reference evapotranspiration predictions in key cereal producing regions of Algeria

期刊论文
发表日期：

2025年6月3日

查看全文

Tarek Bouregaa

摘要

Understanding and accurately predicting reference evapotranspiration (ET₀) is critical for effective water management, particularly in semi-arid regions facing increasing climatic variability. This study integrates climatic trend analysis, and machine learning models to improve ET₀ predictions in four key cereal-producing regions of Algeria: Oum El Bouaghi, Setif, Sidi Bel Abbes, and Tiaret. Using AgERA5 data from 1979 to 2021, climatic trends were assessed with the Mann–Kendall test and Sen's slope estimator. Significant increases in minimum and maximum temperatures were observed, with the steepest trends in Sidi Bel Abbes (0.034°C/year) and Oum El Bouaghi (0.029°C/year). Precipitation showed the largest decline in Sétif (-1.071 mm/year), while relative humidity (H%) decreased most sharply in Tiaret (-0.036%/year).Machine learning models, including Linear Regression (LASSO), k-Nearest Neighbors (KNN), Random Forest (RF), Adaptive Boosting (AdaBoost), Gradient Boosting Machine (GBM), and Support Vector Machine (SVM) were evaluated under various input scenarios. Results indicate SVM achieving near-perfect predictions in Sidi Bel Abbes (R² = 0.999, RMSE = 2.007 mm/month, MAE = 1.724 mm/month)and Tiaret (R² = 0.977, RMSE = 8.6 mm/month, MAE = 7.337 mm/month), under scenario 8 (S8), and scenario 7 (S7), respectively. In Oum El Bouaghi, the Random forest and Adaboost models performed best under S7 (R² = 0.976, RMSE = 7.082 mm/month, MAE = 5.552 mm/month), and S8 (R² = 0.976, RMSE = 7.036 mm/month, MAE = 6.5 mm/month), respectively. While KNN showed strong performance in minimal-input scenarios (scenario5), in Setif(R² = 0.977, RMSE = 5.45 mm/month, MAE = 3.729 mm/month). LASSO regression demonstrated high accuracy in Tiaret (R² = 0.982, RMSE = 7.683 mm/month, MAE = 7.415 mm/month), when mean temperature and relative humidity were included. Climate projections under the SSP2-4.5 scenario using CMIP6 multi-model ensemble data suggest a significant increase in ET₀ across all studied regions, with SVM and Linear Regression models projecting the highest values by 2070. These results underscore the importance of Tmax, Tmean, and H% as key predictors of ET₀ in semi-arid climates.The findings provide critical insights for developing sustainable irrigation and water management strategies, contributing to enhanced agricultural resilience and food security in regions increasingly affected by climate change.