Generating daily land surface temperature at Landsat resolution by fusing Landsat and MODIS data

Land surface temperature (LST) is a crucial parameter in investigating environmental, ecological processes and climate change at various scales, and is also valuable in the studies of evapotranspiration, soil moisture conditions, surface energy balance, and urban heat islands. These studies require thermal infrared (TIR) images at both high temporal and spatial resolution to retrieve LST. However, currently, no single satellite sensors can deliver TIR data at both high temporal and spatial resolution. Thus, various algorithms/models have been developed to enhance the spatial or the temporal resolution of TIR data, but rare of those can enhance both spatial and temporal details. This paper presents a new data fusion algorithm for producing Landsat-like LST data by blending daily MODIS and periodic Landsat TM datasets. The original Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) was improved and modified for predicting thermal radiance and LST data by considering annual temperature cycle (ATC) and urban thermal landscape heterogeneity. The technique of linear spectral mixture analysis was employed to relate the Landsat radiance with the MODIS one, so that the temporal changes in radiance can be incorporated in the fusion model. This paper details the theoretical basis and the implementation procedures of the proposed data fusion algorithm, Spatio-temporal Adaptive Data Fusion Algorithm for Temperature mapping (SADFAT). A case study was conducted that predicted LSTs of five dates in 2005 from July to October in Los Angeles County, California. The results indicate that the prediction accuracy for the whole study area ranged from 1.3 K to 2 K. Like existing spatio-temporal data fusion models, the SADFAT method has a limitation in predicting LST changes that were not recorded in the MODIS and/or Landsat pixels due to the model assumption.