Improvement of ESTARFM and its application to fusion of Landsat-8 and MODIS Land Surface Temperature images

Land Surface Temperature (LST) is an important parameter to study the earth's surface thermal environment and the exchange process of matter and energy of earth's various systems, etc. In many studies, thermal infrared (TIR) images with both high spatial and frequent coverage were highly required to obtain LSTs. However, at current stage, there is no such kind of satellite sensor can completely meet the requirements. If a satellite sensor has high spatial resolution, then its temporal resolution is low, and vice versa. Therefore, in recent years, in order to enhance the spatial resolution and time coverage of the satellite images simultaneously, a variety of spatial and temporal fusion models or algorithms have been proposed. In this paper, we using the Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM), proposed by Zhu et al. (2010), taking Yanping District, Nanping City, Fujian Province, China as the study area, the original ESTARFM was improved by taking the following measures: considering the characteristics of the thermal images of the study area, we only used the Landsat and MODIS TIR radiance images to select the thermal similar neighbor pixels, calculate the conversion coefficients and the spectral, temporal weights. The improved ESTARFM was implemented by IDL (Interactive Data Language) programming, using three pairs of Landsat-8 and MODIS images acquired on December 1st, 2013, January 2nd, 2014 and February 3rd, 2014, respectively. The resultant fusion image was assessed and was applied to retrieve LST by using the newly revised Generalized Single-channel Method proposed by Jiménez-Muñoz and Sobrino et al. (2014), after that, the predicted LST was compared with the actually observed LST which retrieved from the actually observed Landsat-8 image not only from the qualitative but also from the quantitative perspectives. The research results showed that, by using the Landsat-8 and MYD11A1 LST images with the improved ESTARFM proposed by this study, strong agreements between the actual and synthetic LST images were achieved, thus the temporal coverage of the Landsat-8 LST data over the study area was successfully increased.