Linear correction method for improved atmospheric vertical profile retrieval based on ground-based microwave radiometer

Abstract

Abstract The back-propagation neural network (BPNN) is the most commonly used retrieval algorithm for microwave radiometers. Few researchers have attempted specifically to enhance training set quality, which markedly affects retrieval results and can minimize error and uncertainty in simulated brightness temperatures (BTs) in the BPNN. A local BPNN retrieval and correction method were established in this study using radiosonde data, BTs calculated from the radiosonde data, and a monochromatic radiative transfer model (February 2012 to August 2017) in Harbin. The correlation between simulated and observed BTs was improved after correction. The results were analyzed using three sets of comparisons before and after correction: (i) total root mean square errors and total mean absolute errors; (ii) root mean square errors and mean absolute errors in three layers; and (iii) root mean square errors and mean absolute errors under clear days and cloudy days. The results of this study contribute to the theoretical development of microwave remote sensing of atmospheric temperature and humidity.