An End-to-End Deep Learning Approach for High-Accuracy Brightness Temperature Calibration of Satellite Microwave Sounders: A Case Study of the FY-3F MWTS-III
Year: 2026
Type: article
Abstract: High-accuracy brightness temperature calibration of spaceborne microwave radiometers is fundamental to quantitative remote sensing. Radiometric calibration is jointly affected by factors such as the instrument thermal state, calibration reference responses, and scan geometry. Characterizing the nonlinear coupling among these factors and improving brightness temperature calibration within the existing operational calibration framework therefore remain topics requiring further investigation. This study proposes an end-to-end deep learning calibration method for the Microwave Temperature Sounder-III (MWTS-III) onboard Fengyun-3F (FY-3F), driven by actual instrument observations and supervised using simulated brightness temperatures (BTs) as references. The method employs a multilayer perceptron (MLP) that jointly uses multichannel raw Earth-view counts, warm-minus-cold calibration reference counts, instrument temperatures, and scan angles to estimate BTs for individual channels. The training labels are generated using the Advanced Radiative Transfer Modeling System (ARMS) with the fifth-generation reanalysis from the European Centre for Medium-Range Weather Forecasts (ERA5). Evaluation using independent on-orbit data shows that, compared with operationally calibrated BTs, the model-predicted BTs agree more closely with the simulated references, with reduced scan-dependent differences in all target channels and reduced latitude-dependent differences in most channels. Ablation experiments using ascending-orbit data demonstrate that cross-channel information further improves agreement between the model outputs and the simulated references. Comparisons based on simultaneous nadir overpasses (SNOs) with the Advanced Technology Microwave Sounder (ATMS) onboard NOAA-20 indicate improved intersatellite consistency of the model-predicted BTs in some channels. Within the validated range of scenes and instrument states, the method can reduce reliance on contemporaneous background data and pixel-by-pixel forward calculations during BT generation. It provides a complementary approach for parallel quality assessment of operational BTs and a basis for future research on historical observation reprocessing and near-real-time auxiliary calibration.
Source: Remote Sensing
DOI: 10.3390/rs18193400
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,258
OpenAlex ID: W7220372937