An interpretable machine learning approach based on SHAP, Sobol and LIME values for precise estimation of daily...

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Increasing water scarcity and climate variability necessitate precise agricultural irrigation management.
• Accurate estimation of crop coefficients (Kc) is crucial for determining crop water needs, especially in arid and semi-arid regions.
• Conventional Kc estimation methods may not capture local climatic variations.

Purpose of the Study:

• To predict the daily crop coefficient (Kc) for soybean using machine learning models.
• To evaluate the interpretability and physical consistency of these models.
• To provide a robust framework for improving Kc estimation and supporting sustainable irrigation.

Main Methods:

• Four machine learning models (Extra Tree, XGBoost, Random Forest, CatBoost) were employed.
• Models were trained on meteorological data from Suhaj Governorate, Egypt (1979-2014).
• Interpretability was assessed using SHapley Additive exPlanations (SHAP), Sobol sensitivity analysis, and Local Interpretable Model-agnostic Explanations (LIME).

Main Results:

• The Extra Tree (ET) model achieved the highest accuracy (r=0.96, NSE=0.93, RMSE=0.05, MAE=0.02).
• XGBoost and Random Forest models also demonstrated high performance.
• Antecedent crop coefficient and solar radiation were identified as key influential variables by SHAP and Sobol analyses.

Conclusions:

• Interpretable machine learning models enhance the accuracy and reliability of daily Kc estimation.
• The study highlights the importance of aligning model predictions with physical processes for robust agricultural management.
• The proposed framework supports sustainable irrigation practices and climate-resilient agriculture.