Classification of Precipitable Water Vapor Regimes Revealed by Clustering of GNSS Radio Occultation and ERA5 Data

Abstract

Precipitable water vapor (PWV) is an essential parameter for the study of weather and climate behavior. This work presents a novel regional classification of PWV regimes based on clustering techniques applied to Global Navigation Satellite System radio occultation‐derived and ERA5 reanalysis data with and without seasonal removal. Spatially and temporally averaged data sets were constructed using  ×  latitude, longitude bins and monthly temporal intervals over the period June 2006 to September 2024. Multiple clustering approaches with Euclidean and dissimilarity metrics, including hierarchical (Linkage) and non‐hierarchical ( k ‐means, k ‐medoids) algorithms were applied. Features such as mean and standard deviation of every full time series as well as wavelet‐derived characteristics of deseasonalized anomalies were used in Euclidean methods whereas the dynamic time warping procedure was applied to quantify dissimilarity. Results across data sets and methods are compared and they reveal similar regional structures which provide insights into global atmospheric moisture behavior. PWV across all algorithms and data shows strong and consistent latitudinal clustering patterns but with some differences over large land and ocean areas. In anomalies, the amount of groupings that the clustering methods are able to determine are much less and dissimilarity metrics shows more coherent meridional structure than Euclidean options. We also check the previous outcomes against ERA5 high resolution data results.