Explainable Agricultural Water-Use Efficiency: Integrating Frontier Analysis with Explainable AI for Environmental Decision Support

Abstract

This record contains the data, code, and results supporting the book chapter "Explainable Agricultural Water-Use Efficiency: Integrating Frontier Analysis with Explainable AI for Environmental Decision Support" by Morteza Yaqubi and Ali Maroosi, prepared for the Elsevier volume Environmental Intelligence: Artificial Intelligence for Environmental Systems, Natural Resources, and Sustainable Agriculture. What the study does. Using only official FAO dissemination data, the chapter builds a country-year panel (4,501 estimation observations, 169 countries, 1995–2023) that combines AQUASTAT agricultural water withdrawal with FAOSTAT agricultural land, employment, capital formation, and gross production value (constant 2014–2016 international dollars). An input-oriented, variable-returns-to-scale DEA frontier yields three efficiency measures, and an XGBoost model with SHAP explanations is then used to examine the conditions associated with water-use efficiency. Contents. Country-year panel, with codebook and data dictionary Retrieval scripts for FAOSTAT and AQUASTAT, with the observation-status (imputed versus reported) information for the water data Frontier estimation code and results (generic technical efficiency, water-specific subvector efficiency, directional distance efficiency, bootstrap output) Explainable-AI code and outputs (model specification, SHAP and permutation importance, stability checks across seeds, specifications, and targets) Robustness results, including the replication on the reported-only water subsample Data underlying the chapter's tables and figures Scripted pipeline with checksums, to reproduce all results from the frozen panel [add run instructions or the name of the README file] Important notes for users. A large share of the AQUASTAT annual water values are imputed by FAO (carry-forward and interpolation). The observation-status flag is retained in the panel so that users can separate reported from imputed values. The explainable-AI results describe statistical associations, not causal effects, and the learned relationships do not transfer reliably across world regions. Capital is measured by gross fixed capital formation (a flow), because FAO capital-stock data cover only a few countries. Source data and licensing. The underlying statistics are from FAOSTAT and AQUASTAT (Food and Agriculture Organization of the United Nations).