An integrated model for generating hourly Landsat-like land surface temperatures over heterogeneous landscapes

The trade-off between spatial and temporal resolutions in remote sensing has greatly limited the availability of concurrently high spatiotemporal land surface temperature (LST) data for wide applications. Although many efforts have been made to resolve this dilemma, most have difficulties in generating diurnal fine-resolution LSTs with high spatial details for landscapes with significant heterogeneity and land cover type change. This study proposes an integrated framework to BLEnd Spatiotemporal Temperatures (termed BLEST) of Landsat, MODIS and a geostationary satellite (FY-2F) to one hour interval and 100 m resolution, where (1) a linear temperature mixing model with conversion coefficients is combined to better characterize heterogeneous landscapes and generate more accurate predictions for small and linear objects; (2) residuals are downscaled by a thin plate spline interpolator and restored to the primary fine-resolution estimations to include information about land cover type change; and (3) separate operations at annual and diurnal scales with nonlinear temperature modeling are designed to neutralize the hybrid impacts of large scale gap and land cover type change. BLEST was tested on both simulated data and actual satellite data at annual, diurnal and combined scales, and evaluations were conducted with the simulated/actual fine-resolution data, in-situ data, and with three popular fusion methods, i.e., the spatial and temporal adaptive reflectance fusion model (STARFM), the Enhanced STARFM (ESTARFM) and the spatiotemporal integrated temperature fusion model (STITFM). Results show higher accuracy by BLEST with more spatial details and pronounced temporal evolutions, particularly over heterogeneous landscapes and changing land cover types. BLEST is proposed to augment the spatiotemporal fusion system and further support diurnal dynamic studies in land surfaces.