Retrieving Land Surface Temperature From Fengyun3Thermal Infrared Imageries

Land surface temperature (LST) is a good indicator of the greenhouse effect and the energy balance of the earth’s surface, and is widely required in meteorology, hydrology, environment, ecology and many other fields. As a kind of new and important satellite data sources, the thermal infrared data of the Visible and InfraRed Radiometer (VIRR) and the Medium Resolution Spectral Imager (MERSI) on board the Fengyun3satellite have a good prospect of application to land surface temperature product inversion. Because our country’s FY3LST products have a lower accuracy compared with the LST products abroad, it is necessary to improve the existing inversion algorithm and develop a FY3LST inversion algorithm with high precision based on the channels’characteristics of these two sensors.The MODerate spectral resolution atmospheric TRANsmittance algorithm and computer model (MODTRAN) was used to simulate FY3VIRR and MERSI infrared data. The split-window algorithm was employed to build VIRR4/VTRR5LST estimation model and VIRR4/MERSI5LST estimation model based on the simulated data. The pixel emissivity was calculated respectively using the average emissivity method and the vegetation coverage cycle method based on the information of land coverage types and soil texture, and The LST estimation experiments were conducted in Jiangsu province, Guangdong province, Heilongjiang province and region of Dunhuang.The land surface temperature inversion results of the two methods were both validate by MODIS LST product and the in-situ data. It shows that compared wih the LST from VIRR4/MERSI5model, LST from VIRR4/VIRR5model has a higher correlation coefficient and a lower RMSE between MODIS LST products, which means the VIRR4/VIRR5model can obtain higher LST estimation accuracy. The correlation coefficients between the LST estimated from VIRR4/VIRR5medel with average emissity method and MODIS LST is0.869, the RMSE is1.584K, the BIAS is-0.525K; and the RMSE between this estimated LST and the in-situ data is0.638K, the BIAS is-0.619K. The correlation coefficients between the LST estimated from VIRR4/VIRR5medel with vegetation coverage cycle method and MODIS LST is0.854, the RMSE is1.510K, the BIAS is0.120K; and the RMSE between this estimated LST and the in-situ data is0.439K, the BIAS is0.378K. This indicates that both average emissivity method and vegetation coverage cycle method can obtain quite high accuracy in LST inversion, and the accuracy of LST results with vegetation coverage cycle method has a little improvement compared with that with average emissivity method. In addition, the LST estimated with average emissity method has a small negative deviation, while which estimated with vegetation coverage cycle has a positive one. According to multiple experiments, both of the deviations have a absolute value which is lower than1K.