View in EDS

A Neural Network Based Approach for Retrieving Atmospheric Temperature Profiles from the Satellite Microwave Temperature Sounder (MWTS) Measurements.

Saved in:

Bibliographic Details
Title:	A Neural Network Based Approach for Retrieving Atmospheric Temperature Profiles from the Satellite Microwave Temperature Sounder (MWTS) Measurements.
Authors:	Filei, A. A.¹ (AUTHOR) andreyvm-61@mail.ru, Andreev, A. I.¹ (AUTHOR)
Source:	Russian Meteorology & Hydrology. May2026, Vol. 51 Issue 5, p392-402. 11p.
Abstract:	The study presents a neural network based approach for retrieving atmospheric temperature profiles from measurements of the Microwave Temperature Sounder (MWTS) onboard satellites of the FengYun series. The retrieval algorithm is implemented as a fully connected feedforward neural network. The training data include the MWTS channel brightness temperatures simulated with the RTTOV fast radiative transfer model and the corresponding temperature profiles from the ECMWF ERA5 reanalysis. The performance of the proposed method was evaluated against the data of radiosonde observations over the Russian Far East for the summer and winter seasons of 2025. The results have demonstrated that the root-mean-square error (RMSE) of temperature retrieval does not exceed 3.5 K in the near-surface layer in summer and 5.5 K in winter while remaining below 3 K throughout the troposphere and lower stratosphere. In addition, the results obtained from the MWTS data using the developed method were compared with those obtained from the AMSU-A radiometer based on the physical 1D-Var algorithm. It was found that in summer, the deviation of temperature retrieved from the MWTS data was smaller than the one from the AMSU-A data: by approximately 1 K near the surface and by 0.5 K in the mid-troposphere. An additional analysis of errors relative to radiosonde data was carried out for coastal and mountainous stations as well as for winter temperature inversion conditions, which make the greatest contribution to an increase in retrieval errors. [ABSTRACT FROM AUTHOR]
Database:	Energy & Power Source

Description
Abstract:	The study presents a neural network based approach for retrieving atmospheric temperature profiles from measurements of the Microwave Temperature Sounder (MWTS) onboard satellites of the FengYun series. The retrieval algorithm is implemented as a fully connected feedforward neural network. The training data include the MWTS channel brightness temperatures simulated with the RTTOV fast radiative transfer model and the corresponding temperature profiles from the ECMWF ERA5 reanalysis. The performance of the proposed method was evaluated against the data of radiosonde observations over the Russian Far East for the summer and winter seasons of 2025. The results have demonstrated that the root-mean-square error (RMSE) of temperature retrieval does not exceed 3.5 K in the near-surface layer in summer and 5.5 K in winter while remaining below 3 K throughout the troposphere and lower stratosphere. In addition, the results obtained from the MWTS data using the developed method were compared with those obtained from the AMSU-A radiometer based on the physical 1D-Var algorithm. It was found that in summer, the deviation of temperature retrieved from the MWTS data was smaller than the one from the AMSU-A data: by approximately 1 K near the surface and by 0.5 K in the mid-troposphere. An additional analysis of errors relative to radiosonde data was carried out for coastal and mountainous stations as well as for winter temperature inversion conditions, which make the greatest contribution to an increase in retrieval errors. [ABSTRACT FROM AUTHOR]
ISSN:	10683739
DOI:	10.3103/S106837392605002X