Scholarlink

<p>Water in agriculture is central to food security, particularly in the face of rising challenges of climate change, population growth and resource scarcity. Decisions about irrigation timing, level, and location are growingly steered by application of machine learning (ML) models to incorporate meteorological, soil and crop and management data. Yet, the nature and obscurity of such models restricts their use among farmers and policymakers. The paper examines the application of two well-known explainable artificial intelligence (XAI) methods SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) to ML-based decisions on irrigation. study train and test different ML models (Random Forest, XGBoost, LightGBM, and Neural Networks) on real-world datasets obtained in the United States Department of Agriculture (USDA), National Oceanic and Atmospheric Administration (NOAA), and European Space Agency (ESA) Sentinel-2 satellite (2018–2024) to predict irrigation demand in the major agricultural areas of the United States. study then use SHAP and LIME to come up with local and global explainability of model predictions making actionable insights of irrigation behavior possible. study show that SHAP gives more robust, theoretically motivated explanations, especially in modeling higher-order feature interactions whereas LIME is faster and offer localizing to instances. study present a new hybrid explanation framework, which borrows the virtues of both methods to increase transparency and trust. The paper closes with the planning of policies, validation of the model, and in combination with the XAI inclusion in the decision support system to ensure sustainable management of water.</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

Interpreting Irrigation Decisions: Explainable AI Using SHAP and LIME in Agricultural Water Management Models

  • 期刊论文

  • 发表日期：

    2026年1月5日

查看全文
  • Shah Mehmood Wagan,
  • Sidra Sidra

摘要

Water in agriculture is central to food security, particularly in the face of rising challenges of climate change, population growth and resource scarcity. Decisions about irrigation timing, level, and location are growingly steered by application of machine learning (ML) models to incorporate meteorological, soil and crop and management data. Yet, the nature and obscurity of such models restricts their use among farmers and policymakers. The paper examines the application of two well-known explainable artificial intelligence (XAI) methods SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) to ML-based decisions on irrigation. study train and test different ML models (Random Forest, XGBoost, LightGBM, and Neural Networks) on real-world datasets obtained in the United States Department of Agriculture (USDA), National Oceanic and Atmospheric Administration (NOAA), and European Space Agency (ESA) Sentinel-2 satellite (2018–2024) to predict irrigation demand in the major agricultural areas of the United States. study then use SHAP and LIME to come up with local and global explainability of model predictions making actionable insights of irrigation behavior possible. study show that SHAP gives more robust, theoretically motivated explanations, especially in modeling higher-order feature interactions whereas LIME is faster and offer localizing to instances. study present a new hybrid explanation framework, which borrows the virtues of both methods to increase transparency and trust. The paper closes with the planning of policies, validation of the model, and in combination with the XAI inclusion in the decision support system to ensure sustainable management of water.