A New Strategy for Comparison of Land Surface Temperature Retrieval Methods with Landsat Remote Sensing Images Considering Regional Consistency

Landsat remote sensing images are widely used in fields such as land surface temperature retrieval, urban expansion, and urban heat islands, due to their high spatial resolution and the availability of long time series data. Long-term land surface temperature (LST) change analysis usually requires comprehensive utilization of remote sensing data from different sensors, such as Landsat 7, Landsat 8. A LST retrieval algorithm with generalization for Landsat thermal infrared image can effectively improve the reliability of long-term analysis using multi-source images. In order to evaluate the performances of different LST retrieval methods for Landsat images, a new strategy based on regional consistency was proposed in this paper so that different LST retrieval algorithms can be compared with each other without utilizing reference data from ground observation. The general hypothesis is that there is a significant positive correlation between the obtained Landsat 7 and Landsat 8 LST products with adjacent imaging time, similar imaging environment for the identical area. Firstly, the Landsat 7 and Landsat 8 image pairs from Shenzhen were selected under aforementioned constraints. Secondly, four representative LST retrieval methods, radiative transfer equation method (RTEM), image-based method (IBM), mono-window algorithm (MWA) and single-channel algorithm (SCA) were used to generate the LST products from the Landsat 7 and Landsat 8 image pairs respectively. Lastly, the correlation between the Landsat 7 and Landsat 8 LST products from different methods can be calculated as different indexes, including the goodness of fit, Pearson correlation coefficient and Euclidean distance. It is convincing that the optimal LST retrieval method should exhibit a higher regional consistency between the Landsat 7 and Landsat 8 LST product pair given the certain area. The experimental results show that the radiative transfer equation method generates the highest correlation and it is considered as a suitable option for long-term LST research with multi-source Landsat datasets.