Correcting land surface temperature measurements for directional emissivity over 3-D structured vegetation

The emissivity variation of the land surface is the most difficult effect to correct for when retrieving land surface temperature (LST) from satellite measurements. This is not only because of the emissivity inter-pixel variability, but also because each individual pixel is a combination of different surface types with different emissivies. For different illumination-observation geometries, this heterogeneity leads to different ensemble (scene) emissivities. The modified geometric project (MGP) model has been demonstrated to be able to simulate such effect when the surface structural characteristics are available. In this study, we built a lookup table to correct the surface emissivity variation effect in LST retrievals. The lookup table is calculated using the MGP model and the MODTRAN radiative transfer model. The MGP model, assumes that the land surface visible to the satellite sensor-is a composite of homogeneous vegetation and soil background surface types. The homogeneous or "pure" surface types and their emissivity values are adopted from Snyder's surface type classification. Our simulation procedure was designed to calculate the emissivity directional variation for multiple scenarios with different surface types, solar-view angles, tree cover fractions, and leaf area index. Analysis of the MODTRAN simulation results indicates that an error of over 1.4 K can be observed in the retrieved LST if surface emissivity directional variability is not accounted for. Several MODIS granule data were selected to evaluate the correction method. The results are compared with the current MODIS LST products.