| Autor: |
Liu, Shubo, Li, Yinan, Zhou, Wu, Jin, Xv, Yang, Xiaojiao, Dou, Haofeng, Li, Hao |
| Zdroj: |
IEEE Transactions on Geoscience and Remote Sensing; 2024, Vol. 62 Issue: 1 p1-13, 13p |
| Abstrakt: |
Sea surface temperature (SST) and wind speed (SSWS) are two significant parameters in the coupled ocean-atmosphere system. Nowadays, spaceborne microwave radiometers are the main approaches to measuring global SST and SSWS with high coverage and high accuracy. For conventional processing of passive microwave data, radiometer calibration is conducted to convert the radiometer raw output voltage data into the top-of-atmosphere (TOA) brightness temperatures (TBs). After calibration, sea surface parameters are retrieved from TBs using statistically or physically based retrieval methods. Considering the complex procedures of instrument calibration and retrieval, a novel estimation method based on deep learning is proposed to obtain SST and SSWS directly from radiometer voltages. A comprehensive matchup dataset, including the data measured by the scanning microwave radiometer (SMR) onboard the Chinese HY-2B satellite, the European reanalysis 5 (ERA5) products, and the WindSat measurements, is used to develop and test the deep learning models. Validation against ERA5 and WindSat products indicates that the deep learning models perform well. To compare with traditional methods, we also retrieve SST and SSWS from TB data of SMR using a statistical regression retrieval algorithm. The comparison suggests that the deep learning models provide results close to and sometimes better than the regression algorithm. Furthermore, the feature importance of deep learning models and the dependence of their performances on sea state are analyzed. In this article, it is demonstrated that the deep learning method is a reliable and feasible tool for SST and SSWS estimation from the radiometer voltages with high accuracy, which can simplify the data processing procedure and improve processing efficiency. |
| Databáze: |
Supplemental Index |
| Externí odkaz: |
|
