Comparison Of Retrieval Algorithms For Land Surface Temperature From Landsat-8 Imagery

Land surface temperature(LST)is an important parameter in the process of energy exchange and water vapor cycle.It is of great significance in climate change,drought monitoring,agricultural ecology,landscape ecological environment assessment,and environmental planning.Satellite thermal infrared remote sensing provides an effective way to acquire land surface temperature quickly and extensively.In recent years,with the widespread use of Landsat-8 data,a variety of LST algorithms have been proposed based on the two thermal infrared bands,but comparative studies of these algorithms have not been reported.This paper takes Xi'an as the study area,based on Landsa-8 image data,with different retrieval algorithms(GSC,MW,SW-JIM,SW-R,SW-D,SW-J and SW-Y Algorithm)to acquire LST.Firstly,analyze the characteristics of LST retrieval results of each algorithm from the spatial distribution characteristics.Secondly,use the measured ground temperature data and MODIS geothermal product data to respectively evaluate the retrieval accuracy of different algorithms,and then compare the correlation between the retrieval results of different algorithms and the LST of MODIS geothermal product.Finally,by modeling the algorithm,the sensitivity analysis of the two key parameters of the algorithm: land surface emissivity and atmospheric transmissivity is performed.Explore the advantages and disadvantages,suitability and sensitivity of different algorithms.The conclusions are as follows:(1)On the spatial scale,the overall distribution trend of the retrieval results of the different algorithm is roughly similar.The relatively high temperature areas generally correspond to populous and densely built areas is concentrated in the areas with dense urban buildings and high population density.The low temperature is mainly distributed in the Baqiao District and its southeast mountainous areas.But from the site scale,the LST of each algorithm is different.(2)Among the seven algorithms,the GSC and MW algorithms have the highest retrieval accuracy;followed by the SW-R algorithm.Little difference is detected between SW-Y algorithm and SW-JIM algorithm.SW-D algorithm has slightly lower accuracy.While LST retrieved by SWJ algorithm is generally high,which is obviously different from those of the other algorithms and proved to have the lowest accuracy.(3)The correlation analysis with MODIS geothermal product shows that LST of SW-JIM,SW-R algorithm has a good correlation with MODIS geothermal product,GSC,MW,SW-D,SW-Y algorithm is better;while the LST of SW-J algorithm has a significant difference with MODIS geothermal products.(4)With respect to sensitivity analysis shows,for the atmospheric transmittance,SW-JIM algorithm has the least influence on the atmospheric transmissivity,the sensitivity is the lowest.The sensitivity of the GSC algorithm and the SW-JIM algorithm are very small and close to each other,followed by SW-R,SW-Y,and MW,respectively.The SW-J algorithm has the highest sensitivity,while the sensitivity of the SW-D algorithm is changed by the variety of the atmospheric water vapor content.When the sensitivity of land surface emissivity is analyzed,it is found that the retrieval results of different algorithms have obvious fluctuations with the change of the land surface emissivity of the two thermal infrared bands in the three cases.(5)Combining the retrieval accuracy and sensitivity of different algorithms,it is believed that the SW-JIM algorithm has relatively high accuracy and low sensitivity,and can be used as a better algorithm for retrieval of land surface temperature from Landsat-8 data.