FY-3A/MERSI precipitable water vapor reconstruction and calibration using multi-source observation data based on a generalized regression neural network

Abstract

Precipitable water vapor (PWV) is a key element in the water and energy transfer between the surface and the atmosphere, especially in the Three-River Headwaters (TRH), which exhibits sensitive climate and hydrology. To overcome the limitations of missing data and poor accuracy of the moderate resolution spectral imager (MERSI) on-board the second-generation polar-orbiting satellite FengYun-3A (FY-3A) in the TRH region, a model for FY-3A/MERSI PWV reconstruction, based on generalized regression neural network (GRNN) using meteorological factors, normalized differential vegetation index (NDVI) and digital elevation model (DEM), is established, and then, a method for daily systematic error of MERSI PWV calibrating based on global navigation satellite system (GNSS)-derived PWV is proposed. The root mean square (RMS), standard deviation (Std), and bias values of the reconstructed PWV are 0.57, 0.52, and 0.02 mm compared with the original MERSI PWV. The RMS, Std and bias of the calibrated MERSI PWV are 2.42, 1.92 and 0.14 mm compared with GNSS-derived PWV, and the correlation coefficient between MERSI and GNSS-derived PWV is improved from 0.78 to 0.95, which is satisfactory. Compared with the original MERSI PWV, the proposed reconstruction and calibration model can fill the missing data and effectively improve the accuracy of the MERSI PWV products. The proposed method can provide a basic PWV data ensuring for hydrological and ecological research in the TRH region.