Research On Multi-source Satellite Remote Sensing Inversion Of Atmospheric Temperature And Humidity In The Arctic Region

The acquisition of real-time atmospheric temperature and humidity profiles in the arctic region is of great significance to arctic climate and scientific research.In terms of atmospheric temperature and relative humidity profiles inversion,clouds are regarded as pollution factors in satellite observation radiation,and the Arctic is a region with more clouds in the world and is covered by clouds all year round.Therefore,even if the infrared hyperspectral detector can provide high-precision detection,the detection will be affected by the cloud.The clear sky pixels selected by the traditional cloud detection method still have clouds,which affects the accuracy of the inversion model.Although the spectral resolution of microwave detector is low,it can penetrate clouds,and the sensitivity of microwave band to clouds is much lower.Therefore,this paper proposes to use artificial neural network algorithm(ANN)to retrieve atmospheric temperature and relative humidity profiles from FY-3D/ Vertical Atmospheric Sounding Systems(VASS)multi-source remote sensing observations in the clear sky area of the Arctic.The specific research contents are as follows:(1)Neural network(NN)retrieval models were constructed using infrared hyperspectral data.In order to compare the improvement of the accuracy of the FY-3D/VASS multi-source remote sensing retrieval model,we firstly studied the method of atmospheric temperature and humidity profiles retrieval in the Arctic region using HIRAS data alone,and analyzed the accuracy of the retrieval results.In the research,we focused on the clear sky detection of HIRAS pixels,and the method of selecting suitable channels from 2287 channels of HIRAS for retrieving atmospheric temperature and humidity profiles.In order to evaluate the performance of HIRAS in retrieving the temperature and humidity profiles in polar regions,the AIRS data was also used to establish neural networks using the same method,and the retrieval accuracy of the two infrared hyperspectral instruments was compared.(2)Neural network retrieval models were constructed using FY-3D / VASS multi-source remote sensing data.This paper further studied the addition of microwave data on the basis of infrared materials to solve problems that retrieval accuracy in the lower troposphere due to errors in clear sky detection.The matching method between different observational data is mainly studied,and the accuracy of the neural network retrieval models established from multisource remote sensing data is evaluated from multiple perspectives.The results show that: HIRAS data is generally comparable to AIRS in the performance of retrieving atmospheric temperature and humidity under clear-sky condition.Compared with radiosonde observations(RAOBs),root mean square error(RMSE)of results retrieved form HIRAS data is 1.54 K and 16.86% in the warm season,1.89 K and 18.35 % in cold season.The accuracy of retrievals in the middle and lower troposphere still needs to be improved.Compared with the RAOBs,RMSE of temperature retrieved form FY-3D/VASS decreased by 0.11 K and0.3 K and relative humidity RMSE decreased by 0.36 % and 1.86 % in the warm and cold season respectively;The accuracy of NNs constructed from FY-3D / VASS multi-source remote sensing data was effectively improved below 600 h Pa,especially on land where cloud detection is difficult.