A surrogate model-based framework for national-scale assessment of climate change impacts and vulnerabilities: A review with applications in the agricultural sector

Climate change poses serious threats to agricultural productivity and environmental sustainability, highlighting the need for scientific, evidence-based quantitative assessments. Process-based models provide scientific reliability in simulating crop growth, hydrological processes, and greenhouse gas emissions; however, their application to high resolution, national scale assessments across multiple climate scenarios demands substantial computational resources, thereby limiting the timeliness of policy analysis and decision making. To address this challenge, this study outlines a surrogate modeling framework that links process-based models with machine learning techniques. The concept and classification of surrogate models are reviewed, including data-driven models, meta-models, and hybrid and multi fidelity models. In addition, recent applications are examined to identify the purposes for which surrogate modeling techniques have been applied in combination with process-based models. Based on these insights, a national-scale assessment framework is proposed that integrates large scale input-output data generation, sensitivity analysis, dimensionality reduction, and uncertainty quantification techniques. The framework also utilizes outputs from multiple General Circulation Models (GCMs), thereby reducing assessment uncertainty and improving the reliability of results. This review suggests that combining the scientific reliability of process-based models with the efficiency of surrogate models provides a promising and scalable framework for future climate impact and vulnerability assessments, which could serve as a valuable tool for adaptation planning and effective national policy making in the agricultural sector.