A Spatial PWV Retrieval Model Over Land for GCOM-W/AMSR2 Using Neural Network Method: A Case in the Western United States

Precipitable water vapor (PWV) is an important and active part of the atmosphere. As known, microwave PWV retrieval well applied for ocean but challenging over land. In this article, we creatively established a spatial model land using back-propagation (BP) neural network to combine high precision ground-based Global Navigation Satellite System (GNSS) data continuity satellite-borne data. Three-year from 167 GNSS stations located in western America were utilized training network, 40 untrained sites selected as test set. Root-mean-square error (RMSE) set can reach 3.90 mm 3.88 ascending (As) descending (De) orbits. Then, analyzed influence cover types on result, found that area with single large-scale continuous type had higher accuracy. contrast, diversified lower precision. Furthermore, after fully considering impact type, performed improved 61 grasslands, results 8 showed RMSE could 3.41 3.31 De Compared previously, accuracy been by about 13%. We think due stable physical properties (such emissivity) type.