Clustering the Climate: A Machine Learning Approach to Microclimate Zoning and Crop Suitability

Understanding microclimates is essential for improving agricultural planning and crop suitability, particularly in geographically diverse regions like Turkey. This study introduces a machine learning-driven framework for microclimate zoning using high-resolution agro-meteorological sensor data from the TARBIL network. The data were aggregated and transformed into 'typical year' profiles for 444 agricultural stations, with feature selection optimized through a custom genetic algorithm. Unsupervised clustering methods, particularly Agglomerative Clustering, were applied to identify localized climate zones, achieving 79.21% alignment with existing Köppen-Geiger classifications. Further refinement produced a 15-zone microclimate map, revealing granular patterns not captured by traditional systems. These zones were then linked to crop suitability information, with notable improvements observed for crops such as silage maize and rye, which showed reduced intra-cluster yield variance, indicating their strong response to microclimatic differences. The proposed system represents a scalable, data-driven approach for advancing agro-climatic intelligence and supporting climate-resilient agricultural planning.