Machine Learning-Based Calibration Study of MERSI-II Atmospheric Precipitable Water Vapor Product

Existing studies have found that there is a systematic underestimation of the terrestrial atmospheric precipitable water products of the FY3D satellite. In order to correct the bias and improve the accuracy of the products, this paper takes the data from the global AERONET ground observatory and the data from the FY3D atmospheric precipitable water products as the data source and conducts the modeling according to the ground observatory as the center of the circle, 0.05° as the spatial radius, the time of the satellite transit, and half an hour before and after the time scale. Spatio-temporal matching is used to obtain the modeling data, and the Random Forest Model (RF) is used to model the data and correct the FY3D atmospheric precipitable products. The results show that the application of the RF model can correct the product bias and improve the quality of the products.