Landsat-8 Surface Temperature Reconstruction And Spatial-temporal Validation Based On STARFM And FSDAF Models

High spatial resolution/high temporal resolution Land surface temperature(LST)plays an important role in the study of fine scale of agriculture and evapotranspiration in cropland.At present,there are two main types of thermal infrared sensors to obtain surface temperature,One is high spatial resolution/low temporal resolution,such as TM,ETM+,TIRS and ASTER,with a revisit period of 16 days;the other is low spatial/high temporal resolution,such as AVHRR and MODIS,with good timeliness but a spatial resolution of 1 km.It is of great significance to combine the thermal band data obtained by the two kinds of sensors and generate the LST data with high spatial/high temporal resolution and validate the results.The study takes a full consideration of the above issues,taking the middle reaches of Heihe river basin as the research area,34 time periods of Landsat-8 data from 2013 to 2016 were fused with the corresponding MOD11A1 data to generate Landsat-8 LST data of high time sequence by Spatial and Temporal Adaptive Reflectance Fusion Model(STARFM)and Flexible Spatiotemporal DAta Fusion(FSDAF)algorithm.and we choose the Middle Heihe River Basin as a test area and many multiple scales data as the test data.The Ground validation data mainly include GB site with gobi as the underlying surface,SSW site with desert as the underlying surface,JCHM site with desert as the underlying surface,SD site with reed wetland as the underlying surface and CJZ site with corn as the underlying surface.The specific research contents and results are as follows:(1)In the study,the Landsat-8 LST data was firstly obtained by land surface temperature retrieval in Landsat-8 TIRS band 10,and the ground site data was used for validation.The results show that the SD station inversion results are greatly deviated from the measured data on the ground site,with an average deviation absolute value of 8.72 k.Other stations have a good fusion effect,with an average deviation absolute value of about 2k.(2)Using STARFM and FSDAF spatial-temporal fusion algorithm,the Landsat-8 LST data with high time sequence was generated and validated with ground site data.For STARFM algorithm,the mean deviation values of five sites(SD,GB,CJZ,SSW and JCHM)are-9.34 k,-1.84 k,-2.74 k,-2.51 k and-1.81 k.Respectively,For the FSDAF algorithm,it is-8.97 k,-1.52 k,-1.84 k,-2.16 k and-1.65 k.The predicted results of the two algorithms have a small deviation.(3)In order to further analyze the fusion effect of the two algorithms in the generation of high spatial and temporal resolution surface temperature,in the fusion experiment,the input low-resolution data not only used MOD11A1 data,but also used Landsat-8 LST resampling data(called MOD11A1 simulated data).The fusion result of MOD11A1 simulated data only has the error brought by the algorithm itself when using the simulated data,and there is no error brought by different data sources,which can better represent the advantages and disadvantages of the two algorithm.The results show that STARFM algorithm has better fusion effect and robustness when MOD11A1 is used as low-resolution data.FSDAF algorithm has better fusion effect when MOD11A1 simulated data is used.