Modeling sources of uncertainty in satellite-based estimates of leaf area index using a scene simulation model

Effects associated with the directional reflectance properties of land surfaces, solar and sensor geometries, atmospheric turbidity, sub-pixel heterogeneity, and sensor response and calibration interact in a complex manner to produce significant uncertainty in satellite-based estimates of land surface parameters such as leaf area index (LAI). Because all of these effects are present in varying degrees within any remotely sensed image, the degree of uncertainty introduced by each to satellite-based estimates of surface biophysical variables is difficult to deconvolve and quantify in individual images, or even in extensive time series of images. To allow systematic examination of these effects, the authors have implemented a model that simulates remotely sensed imagery from realistic ground scenes for a suite of common sensors such as Landsat TM and AVHRR based on first principles models of the remote sensing data acquisition process. The authors present a preliminary analysis of the interaction between the scale of variation in vegetation density (LAI) and measurements of simulated normalized difference vegetation index (NDVI) from satellites. The results show that the statistical relationship between NDVI and LAI is dependent on the relationship between the spatial scale of variation of LAI on the ground versus sensor spatial resolution