THE SCALING CHARACTERISTICS OF REMOTELY-SENSED VARIABLES FOR SPARSELY-VEGETATED HETEROGENEOUS LANDSCAPES

With increasing interest in airborne and satellite-based sensors for m apping regional and global energy balance, there is a need to determin e the uncertainty involved in aggregating remotely-sensed variables [s urface temperature (T-k) and reflectance (rho)] and surface energy flu xes [sensible (H) and latent (lambda E) heat flux] over large areas. T his uncertainty is directly related to two factors: (1) the non-linear ity of the relation between the sensor signal and T-k, rho, H or lambd a E; and (2) the heterogeneity of the site. In this study, we compiled several remotely-sensed data sets acquired at different locations wit hin a semi-arid rangeland in Arizona, at a variety of spatial and temp oral resolutions. These data sets provided the range of data heterogen eities necessary for an extensive analysis of data aggregation. The ge neral technique to evaluate uncertainty was to compare remotely-sensed variables and energy balance components calculated in two ways: first , calculated at the pixel resolution and averaged to the coarser resol ution; and second, calculated directly at the coarse resolution by agg regating the fine-resolution data to the coarse scale. Results showed that the error in the aggregation of T-k and rho was negligible for a wide range of conditions. However, the error in aggregation of H and l ambda E was highly influenced by the heterogeneity of the site. Errors in H larger than 50% were possible under certain conditions. The cond itions associated with the largest aggregation errors in H were: sites which are composed of a mix of stable and unstable conditions; sites which have considerable variations in aerodynamic roughness, especiall y for highly unstable conditions where the difference between surface and air temperature is large; and sites which are characterized by pat ch vegetation, where the pixel resolution is less than or nearly-equal to the diameter of the vegetation 'element' (in most cases, the diame ter of the dominant vegetation type or vegetation patch). Thus, knowle dge of the surface heterogeneity is essential for minimizing error in aggregation of H and lambda E. Two schemes are presented for quantifyi ng surface heterogeneity as a first step in data aggregation. These re sults emphasized the need for caution in aggregation of energy balance components over heterogeneous landscapes with sparse or mixed vegetat ion types. (C) 1997 Elsevier Science B.V.