Reconstruction of MODIS daily land surface temperature under clouds

Abstract:

Land surface temperature (LST), generally defined as the skin temperature of the Earth's surface, controls the process of evapotranspiration, surface energy balance, soil moisture change and climate change. Moderate Resolution Imaging Spectrometer (MODIS) is equipped with 1km resolution thermal sensor andcapable of observing the earth surface at least once per day.

Thermal infrared bands cannot penetrate cloud, which means we cannot get consistency drought monitoring condition at one area. However, the cloudy-sky conditions represent more than half of the actual day-to-day weather around the global. In this study, we developed an LST filled model based on the assumption that under good weather condition, LST difference between two nearby pixels are similar among the closest 8 days. We used all the valid pixels covered by a 9*9 window to reconstruct the gap LST. Each valid pixel is assigned a weight which is determined by the spatial distance and the spectral similarity. This model is applied in the Middle-East of China including Gansu, Ningxia, Shaanxi province. The terrain is complicated in this area including plain and hill. The MODIS daily LST product (MOD11A3) from 2000 to 2004 is tested. Almost all the gap pixels are filled, and the terrain information is reconstructed well and smoothly. We masked two areas in order to validate the model, one located in the plain, another located in the hill. The correlation coefficient is greater than 0.8, even up to 0.92 in a few days. We also used ground measured day maximum and mean surface temperature to valid our model.  Although both the temporal and spatial scale are different between ground measured temperature and MODIS LST, they agreed well in all the stations. This LST filled model is operational because it only needs LST and reflectance, and does not need other auxiliary information such as climate factors. We will apply this model to more regions in the future.