Predicting tropical cyclone rainfall through perceptual-based identification of similar atmospheric fields

Analog methods for tropical cyclone (TC) rainfall prediction compare current and past TC-related information (e.g., atmospheric fields) to estimate rainfall. However, comparing spatially distributed data like atmospheric fields is challenging, as common metrics fail to capture multiple characteristics of the data simultaneously. For example, conventional cosine similarity excels at identifying changes in pattern similarity but falls short in capturing absolute magnitude similarities. To address this challenge, we provide a proof of concept for using a perceptual-based measure – the Structural Similarity Index (SSIM) – to identify analog TCs based on atmospheric field similarity. Our study demonstrates the effectiveness of SSIM in selecting analogous atmospheric fields for TC rainfall prediction in a basin, producing results comparable to existing studies. Compared with cosine similarity, SSIM proved more effective in selecting atmospheric field-based analogs, leading to improved basin rainfall prediction performance. This study highlights the potential of perceptual similarity metrics, such as SSIM, to better capture the complex spatial characteristics of atmospheric fields. It also illustrates how techniques from image processing can be innovatively applied in meteorology and geosciences, offering a new perspective on analyzing spatially distributed environmental data.