Application of the Deep Neural Network in Retrieving the Atmospheric Temperature and Humidity Profiles from the Microwave Humidity and Temperature Sounder Onboard the Feng-Yun-3 Satellite.

Qiurui He,Zhenzhan Wang,Jiaoyang Li

doi:10.3390/s21144673

Abstract

The shallow neural network (SNN) is a popular algorithm in atmospheric parameters retrieval from microwave remote sensing. However, the deep neural network (DNN) has a stronger nonlinear mapping capability compared to SNN and has great potential for applications in microwave remote sensing. The Microwave Humidity and Temperature Sounder (Beijing, China, MWHTS) onboard the Fengyun-3 (FY-3) satellite has the ability to independently retrieve atmospheric temperature and humidity profiles. A study on the application of DNN in retrieving atmospheric temperature and humidity profiles from MWHTS was carried out. Three retrieval schemes of atmospheric parameters in microwave remote sensing based on DNN were performed in the study of bias correction of MWHTS observation and the retrieval of the atmospheric temperature and humidity profiles using MWHTS observations. The experimental results show that, compared with SNN, DNN can obtain better bias-correction results when applied to MWHTS observation, and can obtain higher retrieval accuracy of temperature and humidity profiles in all three retrieval schemes. Meanwhile, DNN shows higher stability than SNN when applied to the retrieval of temperature and humidity profiles. The comparative study of DNN and SNN applied in different atmospheric parameter retrieval schemes shows that DNN has a more superior performance.

Highlights

As the basic parameters of the atmosphere, temperature and humidity profiles play an important role in the research and applications of atmospheric science, such as numerical weather forecast, climate change research, and strong convective weather forecast and analysis [1,2,3,4]
The global reanalysis can provide a variety of atmospheric parameters with high spatial resolution and high accuracy, it suffers from a long time delay compared with the satellite observations, which cannot meet the requirements for real-time in atmospheric applications, such as numerical weather forecasting, extreme weather monitoring, etc
The datasets used in this study included the following: (1) Level 1b brightness temperatures of MWHTS onboard FY-3D from the National Satellite Meteorological Center (NSMC)—the quality of MWHTS observations at European Centre for Medium-Range Weather Forecasts (ECMWF) has been evaluated, and the detailed description of the evaluated results can see Lawrence et al [28]; and (2) European Centre for Medium Range Weather Forecasts (ECMWF) ERA-Interim reanalysis dataset obtained from the ECMWF website

Summary

Introduction

As the basic parameters of the atmosphere, temperature and humidity profiles play an important role in the research and applications of atmospheric science, such as numerical weather forecast, climate change research, and strong convective weather forecast and analysis [1,2,3,4]. The essence of the statistical retrieval algorithm is to estimate the atmospheric temperature and humidity profiles based on the statistical relationship between the atmospheric temperature and humidity parameters and the microwave observations, without involving any physical concepts [13,14,15,16]. The second retrieval scheme is based on the statistical relationship between the observed brightness temperature and the atmospheric temperature and humidity profiles. NNs are widely used to retrieve atmospheric temperature and humidity profiles using passive microwave observations, it is almost the SNNs that are extensively used, and they usually contain a hidden layer with a small number of neurons that are based on the error backward propagation algorithm.

Data and Model

Data Preprocessing

Experimental Results

Bias-Correction Results

The Retrieval Results of the 1DVAR Retrieval

The Retrieval Results of the NN-Based Retrieval Using the Observations