Comparisons of Performance Using Data Assimilation and Data Fusion Approaches in Acquiring Precipitable Water Vapor: A Case Study of a Western United States of America Area

Abstract

There are two main types of methods available to obtain precipitable water vapor (PWV) with high accuracy. One is to assimilate observations into a numerical weather prediction (NWP) model, for example, the Weather Research and Forecasting (WRF) model, to improve the accuracy of meteorological parameters, and then obtain the PWV with improved accuracy. The other is the direct fusion of multi-source PWV products. Regarding the two approaches, we conduct a comparison experiment on the West Coast of the United States of America with the data from May 2018, in which the WRF data assimilation (DA) system is used to assimilate the Global Navigation Satellite System (GNSS) PWV, while the method by Zhang et al. to fuse the GNSS PWV, ERA5 PWV and MODIS (moderate-resolution imaging spectroradiometer) PWV. As a result, four groups of PWV products are generated: the assimilated GNSS PWV, the unassimilated GNSS PWV, PWV from the fusion of the GNSS PWV and ECWMF (European Centre for Medium-Range Weather Forecasts) ERA5 (ECWMF Reanalysis 5) PWV, and PWV from the fusion of the GNSS PWV, ERA5 PWV and MODIS PWV. Experiments show that the data assimilation based on the WRF model (WRFDA) and adopted fusion method can generate PWV products with similar accuracy (1.47 mm vs. 1.52 mm). Assimilating the GNSS PWV into the WRF model slightly improves the accuracy of the inverted PWV by 0.18 mm. The fusion of the MODIS PWV, GNSS PWV and ERA5 PWV results in a higher accuracy than the fusion of GNSS PWV and ERA5 PWV by a margin of 0.35 mm. In addition, the inland canyon topography appears to have an influence on the inversion accuracy of both the methods.