Scholarlink

<p>Reliable quantification of reference evapotranspiration (ETo) is essential for optimizing irrigation scheduling, improving water-use efficiency, and assessing crop water demands. However, in many semi-arid regions of Morocco, the scarcity of ground-based solar radiation measurements limits the operational application of conventional approaches. This study explores the potential of machine-learning (ML) algorithms to estimate daily ETo using a combination of MODIS remote-sensing and limited meteorological variables across two representative Moroccan agro-climatic zones: a semi-arid region (Doukkala) and a sub-humid coastal region (Loukkos). Six ML models—Support Vector Regression (SVR), Random Forest, XGBoost, LightGBM, CatBoost, and Artificial Neural Network (ANN)—were evaluated under multiple input scenarios to examine the effects of excluding solar radiation (Rs) and temperature data. Ensemble-based models, particularly LightGBM, XGBoost, and CatBoost, achieved the most stable performance under Rs-free configurations, with mean coefficients of determination (R²) exceeding 0.96 and RMSE ≈ 0.36 mm day⁻¹. Feature-importance analysis identified land-surface temperature (LST) as the most influential predictor, followed by maximum temperature, Julian day, wind speed, and minimum relative humidity, highlighting the role of thermal and seasonal controls on evapotranspiration variability. Reflectance-based parameters such as albedo and vegetation indices (NDVI, EVI) provided complementary information related to surface-energy fluxes and canopy behavior. Transferability tests indicated strong robustness within the same semi-arid region (R² &gt; 0.90) but lower accuracy across contrasting climatic zones (R² ≈ 0.63), reflecting the influence of regional heterogeneity. Overall, the results demonstrate the potential of remote-sensing-driven ML frameworks for accurate and scalable ETo estimation in data-scarce environments, supporting advances in smart irrigation and precision water-management strategies in Morocco.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning estimation of reference evapotranspiration using MODIS-Derived and limited ground variables across Moroccan agro-climatic zones

  • 期刊论文

  • 发表日期：

    2026年3月23日

查看全文
  • Zaid Belarbi,
  • Yacine El Younoussi

摘要

Reliable quantification of reference evapotranspiration (ETo) is essential for optimizing irrigation scheduling, improving water-use efficiency, and assessing crop water demands. However, in many semi-arid regions of Morocco, the scarcity of ground-based solar radiation measurements limits the operational application of conventional approaches. This study explores the potential of machine-learning (ML) algorithms to estimate daily ETo using a combination of MODIS remote-sensing and limited meteorological variables across two representative Moroccan agro-climatic zones: a semi-arid region (Doukkala) and a sub-humid coastal region (Loukkos). Six ML models—Support Vector Regression (SVR), Random Forest, XGBoost, LightGBM, CatBoost, and Artificial Neural Network (ANN)—were evaluated under multiple input scenarios to examine the effects of excluding solar radiation (Rs) and temperature data. Ensemble-based models, particularly LightGBM, XGBoost, and CatBoost, achieved the most stable performance under Rs-free configurations, with mean coefficients of determination (R²) exceeding 0.96 and RMSE ≈ 0.36 mm day⁻¹. Feature-importance analysis identified land-surface temperature (LST) as the most influential predictor, followed by maximum temperature, Julian day, wind speed, and minimum relative humidity, highlighting the role of thermal and seasonal controls on evapotranspiration variability. Reflectance-based parameters such as albedo and vegetation indices (NDVI, EVI) provided complementary information related to surface-energy fluxes and canopy behavior. Transferability tests indicated strong robustness within the same semi-arid region (R² > 0.90) but lower accuracy across contrasting climatic zones (R² ≈ 0.63), reflecting the influence of regional heterogeneity. Overall, the results demonstrate the potential of remote-sensing-driven ML frameworks for accurate and scalable ETo estimation in data-scarce environments, supporting advances in smart irrigation and precision water-management strategies in Morocco.