ObjectiveLand surface temperature (LST) is a key parameter for analyzing and simulating land surface processes from local to global scales, which plays an important role in the processes of energy exchange between land surface and atmosphere. As the latest achievement of the series of Landsat missions, Landsat-8 has two thermal infrared channels compared with previous Landsat version, and has higher spatial resolution compared with other sensors such as AVHRR or MODIS, it has greater advantages when it comes to the use of land surface temperature retrieval. This paper uses three different algorithms including the improved mono-window algorithms proposed by Wang (2015), the improved single channel algorithm proposed by Cristóbal (2018) and the split window algorithm proposed by Jiménez-Muñoz (2014), respectively to retrieve land surface temperature from Landsat-8 data in Beijing region, and chose MODIS LST product for cross validation in order to assess the accuracy and applicability of Landsat-8 used for LST retrieval, and provide a reference for the subsequent study of Landsat-8 LST retrieval.

MethodThis paper determines and estimates 3 important parameters used for LST retrieval: atmospheric effective temperature, land surface emissivity, and water vapor contents. Sensitivity and differential analysis were conducted after getting the LST result by 3 algorithms. Comparative analysis methods between the retrieval result and MOD11A1 product with the closest satellite passing time to Landsat-8 include the comparison of average temperature between retrieval result and MODIS temperature product by different administrative regions and different ground features, as well as choosing the center region of Miyun Reservoir, which has lower scale effect and little temperature variation to compare the temperature difference between the above two.

ResultThe general average temperature difference between the three algorithms was less than 1K, and the temperature difference between the improved single channel algorithm proposed by Cristóbal (2018) and the other two algorithms was the smallest, and the temperature difference between the improved mono-window algorithms proposed by Wang (2015) and the other two algorithms was the largest. The result of Landsat-8 LST retrieval was generally higher than MODIS temperature product, the mean temperature difference was 1.3 K. By selecting the center area of Miyun Reservoir to compare the results between Landsat-8 LST retrieval and the temperature products, the total average difference between the three algorithm results and the MODIS temperature products was 1.373 K.

ConclusionThe retrieval results have an overall satisfying accuracy, among which, the split window algorithm proposed by Jiménez-Muñoz (2014) has the best sensitivity analysis result, and has the closest result to MODIS temperature product. The results of Landsat-8 retrieval are consistent with the distribution of MODIS temperature products in the overall LST distribution, but Landsat-8 can better distinguish the temperature differences between different types of small feature and have a greater advantage when it comes to the use of accurate retrieval of land surface temperature because of the higher resolution.