Karahan et al. (2026) Evaluating Reference Evapotranspiration and Soil Moisture as Predictors for Machine Learning-Based Actual Evapotranspiration Estimation: Model Performance and Explainability

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Atmosphere
• Year: 2026
• Date: 2026-09-27
• Authors: Halil Karahan, Devrim Alkaya
• DOI: 10.3390/atmos17100941

Research Groups

Not specified in the provided text.

Short Summary

This study investigated the impact of reference evapotranspiration (ET0) and soil moisture (SM) availability on actual evapotranspiration (ETa) prediction performance and predictor importance using various machine learning models. It found that ET0 is the dominant predictor, but when unavailable, global solar radiation (Rs) assumes the primary role, with SM providing crucial complementary information.

Objective

• To investigate how the availability of reference evapotranspiration (ET0) and soil moisture (SM) affects actual evapotranspiration (ETa) prediction performance and predictor importance using Random Forest (RF), Bagging Trees (BT), Least Squares Boosting (LSBoost), Generalized Additive Models (GAM), and Multiple Linear Regression (MLR).

Study Configuration

• Spatial Scale: Not specified in the provided text.
• Temporal Scale: Daily (inferred from RMSE units of mm day⁻¹).

Methodology and Data

• Models used: Random Forest (RF), Bagging Trees (BT), Least Squares Boosting (LSBoost), Generalized Additive Models (GAM), Multiple Linear Regression (MLR).
• Data sources: Predictors evaluated include global solar radiation (Rs), land surface temperature (LST), normalized difference vegetation index (NDVI), soil moisture (SM), and reference evapotranspiration (ET0). The specific source types (e.g., satellite, observation, reanalysis) for these variables are not provided.
• Model interpretability: Assessed using SHAP, Permutation Feature Importance (PFI), Partial Dependence Plot (PDP), and temporal signed-SHAP analyses.
• SM-ablation experiment: Quantified the incremental predictive contribution of soil moisture.

Main Results

• In Scenario I (with ET0 available), ET0 was the dominant predictor. Random Forest (RF) and Bagging Trees (BT) achieved the highest independent-test performance (R² = 0.875; RMSE = 0.398–0.399 mm day⁻¹).
• Excluding ET0 (Scenario II) reduced predictive performance, with RF, LSBoost, and BT achieving R² values of approximately 0.80. The predictor-importance structure shifted primarily toward Rs, followed by SM, LST, and NDVI.
• Removal of SM further reduced performance in both scenarios, with a substantially greater deterioration when ET0 was unavailable: R² decreased by 0.025–0.044 in Scenario I and by 0.051–0.099 in Scenario II.
• Under the SM-excluded Scenario I configuration, R² values ranged from 0.820 to 0.832, indicating that ETa prediction performance depends strongly on predictor information content, not solely on model structure.
• Overall, ET0 provides substantial atmospheric-demand information for ETa prediction. When ET0 is unavailable, Rs assumes the dominant predictive role, and SM provides particularly important complementary information.

Contributions

• Demonstrates the critical importance of considering predictor composition alongside model structure when developing and comparing ETa estimation approaches.
• Quantifies the specific impact of reference evapotranspiration (ET0) and soil moisture (SM) availability on ETa prediction performance and the relative importance of other predictors.
• Utilizes advanced explainability analyses (SHAP, PFI, PDP) to provide insights into the behavior and decision-making processes of various machine learning models for ETa prediction.

Funding

Not specified in the provided text.

Citation

@article{Karahan2026Evaluating,
  author = {Karahan, Halil and Alkaya, Devrim},
  title = {Evaluating Reference Evapotranspiration and Soil Moisture as Predictors for Machine Learning-Based Actual Evapotranspiration Estimation: Model Performance and Explainability},
  journal = {Atmosphere},
  year = {2026},
  doi = {10.3390/atmos17100941},
  url = {https://doi.org/10.3390/atmos17100941}
}