Informed Similarity Transfer: A Scientific Machine Learning Approach for Meteorological Data Imputation

Against the backdrop of global climate change, the availability of complete and reliable meteorological data is critical for understanding environmental dynamics and informing policy decisions. However, missing records are prevalent in large-scale meteorological datasets due to harsh weather conditions, equipment malfunctions, and other operational constraints. This study proposes the Informed Similarity Transfer (IST) framework to address these gaps through a scientific machine learning approach. By leveraging spatial proximity, land cover similarity, and elevation differences, IST reconstructs incomplete datasets, preserving both temporal patterns and spatial coherence. The results demonstrate that IST effectively enhances data integrity, making it a valuable tool for supporting accurate environmental modeling and analysis.