MODIS/LST Product Validation for Mixed Pixels at Linzhi of Tibet

Southeastern part of Tibet is featured with complicated terrain and diverse land cover types. Validation of MODIS/LST product (1-km spatial resolution) in this region is faced with mixed pixel issue. Point-based LST measurements cannot represent the pixel well. To obtain ground LST measurements at pixel scale, traditional method usually depends on the high spatial resolution of thermal images, such as Aster and TM. However, these data are often unavailable due to persistent cloud cover and long repeat cycle. Therefore, a new simple method called area-weighted average (AWA) method is proposed, in which land cover map at high spatial resolution is combined with multi-site field observations to model the hypothetical observations at moderate pixel scale. The assumption of AWA method is that field observations can be shared within the same land cover. The AWA method is applied and analyzed on the case of Linzhi (with an area of 20 km²) which locates in southeastern part of Tibet. First, 5 field stations are set up on 5 typical land covers: Grassland, farmland, floodplain, forest at sunny slope, and forest at shadowy slope. The upward and downward long-wave radiations are measured simultaneously. Then the land cover map at 30 m spatial resolution is derived from TM image using maximum likelihood classification method. For every 1-km MODIS pixel, the fraction of each typical land cover is calculated, and the radiation at MODIS pixel-scale is estimated through area-weighted averaging. The broadband emissivity is calculated using linear combination of narrowband emissivity of MODIS band 31 and 32. Finally, LST at MODIS pixel-scale can be calculated based on Stefan-Boltzmann law. The AWA method is used for validating daily product of MODIS/LST from Terra and Aqua platforms on 10 June 2013 (LST at night is used because it changes slowly both in temporal and spatial domain). Results show that the RMSE of MODIS/LST is below 1.4 K (n=30) when applying the AWA method. If a point-based measurement is used to directly represent a MODIS pixel, the RMSE is more than 2.2 K (n=10), showing a tendency of over-estimation. The error of Aqua LST is slightly greater than that of Terra LST, probably due to a larger sensor view zenith angle during overpass. Superiority of the AWA method is more noticeable for pixels with high land cover heterogeneity and gentle terrain. The difference in LST between satellite and field observations can be decreased from 3 K to 1 K. However, for pixels with homogeneous land covers or with very tough terrains, the advantage of AWA method is limited. To further improve the AWA method, terrain adjustment should be taken into account when extrapolating point-based measurements to the same land cover but from another region, because the slope and aspect will influence the surface energy balance process even when the land cover stays the same. Results also indicate MODIS/LST data at nighttime in Linzhi Area are accurate, which are very meaningful considering the low density of meteorological stations in this area.