Retrieval And Validation Of Land Surface Temperature From AMSR2 Data

As one of the important parameters at the interface between the earth‘s surface and the atmosphere, land surface temperature(LST) has important research significance in climate, hydrology, ecology and biogeochemical and many research fields.At present, remote sensing has been widely used to retrieve LST. Although the current research for LST retrieval from the thermal infrared(TIR) remote sensing is in the mature stage, TIR has poor ability in overcoming the atmospheric influences. Compared with the TIR remote sensing, passive microwave(MW) remote sensing is less affected by weather, and has the ability to obtain the ?all-weather‘ LST. In the process of LST retrieval based on MW remote sensing, many input surface or atmospheric parameters can be provided by optical remote sensing. Therefore we combine the optical remote sensing with the MW remote sensing to synergeticly retrieve the LST.This study adopts two ways to retrieve LST based on the GCOM-W1 AMSR2 data. Firstly, this study improves the latest MW LST retrieval algorithm based on the radiative transfer model. In this method, the radiative transfer model is simplified by using the same method, but the effect of atmospheric radiation on brightness temperature is not ignored. An optimal time window strategy is developed pixel by pixel in thirty days for the step, and the AMSR2 LST of different channels based on the optimal time window are integrated by using the Bayesian weighting model. Secondly, this paper improves the polarization ratio method which is based on AMSR-E data for tropical rainforest area. The uniform surface pixels are selected as training samples, and the MODIS vegetation index product is used to classify the land, then the relationship between the polarization ratio and the horizontal polarization channel emissivity at each frequency is constructed by using the stepwise regression method, finally the optimal channel is selected as the best channel to retrieve LST.LST retrieved based on AMSR2 data is evaluated and validated from three ways. First, taking the entire Chinese landmass as the study area, MODIS LST is used as true value to validate the AMSR2 LST. The results show that the correlation between the AMSR2 LST based on radiative transfer method and MODIS LST is relatively high, with R2 greater than 0.865. The root mean squared errors(RMSEs) in four seasons at daytime are in the range of 3.1K~4.2K, and the accuracy is higher at nighttime, with RMSEs are in the range of 2.9K~3.4K. The accuracy of AMSR2 LST retrieved based on the polarization ratio method is relatively lower, because the surface emissivity is sensitive to soil moisture.Second, the in-situ measured LST in the Heihe River Basin is used to validate the AMSR2 LST. When the LST measured at the ground site is directly used to validate the AMSR2 LST, the existence of the scale mismatch lower the accuracy of AMSR2 LST. Therefore, two methods are used to up-scale the in-situ measured LST. The first one is the area weighted average based on land cover types, and the second one is introducing the MODIS LST which provide higher spatial resolution as a ?bridge‘ between the in-situ measured LST and the AMSR2 LST, and use the in-situ measured LST to calibrate the MODIS LST. Compared with the up-scaled in-situ measured LST, the accuracies of AMSR2 LST by two ways are improved. In most cases, the error between AMSR2 LST based on radiative transfer method and the in-situ measured LST after calibration is smaller than that of the area weighted average, the RMSEs are located between 3.1K~5.8K at daytime and 1.7~3.9K at nighttime, respectively. While the error between AMSR2 LST based on polarization ratio method and the in-situ measured LST after calibration is larger than that of the area weighted average.Finally, the near surface air temperatures measured at meteorological stations distributed in the entire Chinese landmass is used to evaluate the AMSR2 LST. The results show that the correlation between AMSR2 LST based on radiative transfer method and air temperature is higher than that of the AMSR2 LST based on polarization ratio method.