Length-Scale analysis of surface albedo, temperature, and normalized difference vegetation index in desert grassland

Abstract

The Jornada Experiment on the Jornada Experimental Range in southern New Mexico aims at the description of the surface energy balance of a desert grassland ecosystem. A large volume of both field and remote sensing data has been collected from 1995 to 1998. Airborne Daedalus scanner data with a spatial resolution of 4 m have been used to infer the following land surface characteristics: surface temperature, albedo, and normalized difference vegetation index (NDVI). These land surface characteristics can be used as input for land surface models. However, land surface models work with very coarse grid cells of at least 50×50 km, in contrast to high-resolution remote sensing data. Also, land surface models are generally based on nonlinear algorithms. Both restrictions lead to scale problems. One apparent question is how to scale up input remote sensing data to the much coarser resolution of the land surface model. The first step is to derive the length scale of the input land surface characteristics. The length scales of the land surface characteristics have been determined with the following two techniques: autocorrelation and wavelet analysis. Within the Jornada Experimental Range, three different sites with different vegetation characteristics were distinguished: grass, shrub, and a transition site with patches of both grass and shrub. The autocorrelation and wavelet analysis showed similar results for the shrub site. For the grass and transition site the wavelet analysis underestimated the length scale of the surface albedo and temperature. The length scale of the surface albedo was 35, 33, and 10 m for grass, transition, and shrub sites, respectively. The length scale of the surface temperature was 31, 20, and 8 m for grass, transition, and shrub sites, respectively. The length scale of the NDVI was 12, 6, and 5 m for grass, transition, and shrub sites, respectively. These small length scales could hamper the use of low-resolution remote sensing data for deriving input data for land surface models.