Generation of MODIS-like land surface temperatures under all-weather conditions based on a data fusion approach

Abstract

Land surface temperature (LST) is among the most important variables in monitoring land surface processes. LST is often retrieved from thermal infrared remote sensing data, which have a tradeoff between the spatial and temporal resolutions and are spatially incomplete due to cloud contamination. Land surface model (LSM) output can reflect LST under all-weather conditions, but the spatial resolution is relatively coarse. In this study, a two-step LST data fusion framework was proposed for generating MODIS-like LST (at the satellite overpass time for each day) at a 1 km spatial resolution under all-weather conditions. First, MODIS LST on clear days (i.e., all MODIS LST pixels are cloud-free) for a given study region (e.g., 80 km × 80 km) and China Land Data Assimilation System (CLDAS) LST at a spatial resolution of ~7 km × 7 km were fused using the Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM). Second, systematic biases of the fused LST estimates were corrected by MODIS LST for clear pixels on cloudy days. Results indicate that the fused LST after bias correction (fused LSTcorr) under all-weather conditions is highly consistent with in situ LST measurements made at three sites with different land cover types in the north of China, in terms of mean absolute errors of 2.20–3.08 K, root mean square errors of 2.77–3.96 K, and coefficients of determination of 0.93–0.95. The accuracy of these results is comparable to that of the MODIS LST retrievals at all testing sites. In addition, the fused LSTcorr can well reflect spatial heterogeneity and temporal variability in LST under all-weather conditions, and there is no significant difference in the accuracy of the fused LSTcorr between the clear and cloudy pixels on cloudy days. The developed approach maximizes the potential of quality MODIS LST retrievals and LSM LST output, and generates LST under all-weather conditions without using ancillary remote sensing and in situ data. The generated MODIS-like LST data under all-weather conditions are valuable in soil moisture downscaling and evapotranspiration estimation for better water resources management.