Submitted to: International Geoscience and Remote Sensing Symposium Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 5/17/2000
Publication Date: 7/24/2000
Citation: Interpretive Summary: Community-level vegetation type differentiation is an important application of remote sensing in monitoring semiarid environments. It has proved difficult to discriminate more than two or three community types with confidence when using coarse spatial resolution data from whiskbroom sensors such as the NOAA (National Oceanic and Atmospheric Administration) AVHRR (Advanced Very High Resolution Radiometer), even when multi-temporal datasets are used. Large classification errors have been attributed to subpixel scale topographic and soil background variations, lack of in flight calibration, inaccurate reflectance retrieval and poor registration. While these are important factors, anisotropy in surface reflectance, as a result of the bidirectional reflectance distribution function (BRDF), also causes severe perturbations in the signal and yet this is rarely accounted for adequately. Recent years have seen a wider acknowledgement of this fundamental aspect of optical remote sensing, although it is usually approached as a problem. The extent of the improvement in community type differentiation may be considerable when a BRDF model is used to separate isotropic and anisotropic components. This potential was tested using data from the AVHRR over semiarid regions in Xilingol League, Inner Mongolia Autonomous Region, P. R. China and New Mexico, U.S.
Technical Abstract: Degradation of semiarid grasslands is often assessed in terms of the spatial distributions of community types since it is generally agreed that sustained climatic or anthropogenic disturbance leads to encroachment of shrubs and annuals into grass and perennial-dominated ecosystems. The AVHRR directional signal is tested in community type differentiation over discontinuous but statistically homogeneous semiarid canopies using a linear semiempirical BRDF model (isotropic, LiSparseMODIS geometric-optical and RossThin volume scattering kernels). The model was inverted against reflectance data derived from AVHRR HRPT data from 21 orbits over a 17-day period over Inner Mongolia and 17 orbits over a 25-day period over New Mexico at the peak of the growing season and the end of dry season, respectively. The SMAC algorithm was used to estimate surface reflectances. Cloud screening was affected by temporal consistency tests. Training sites were defined for both study regions and signature data were extracted from visible and near-infrared surface reflectance estimates; isotropic reflectance (modeled with nadir viewing and sun at zenith); anisotropic parameters describing geometric-optical and volume scattering phenomena; and maximum NDVI composites from PM orbits. When BRDF effects are taken into account, distributions of community types in visible near-infrared space are much more reasonable. The directional signal captured in the AVHRR-derived anisotropic BRDF model parameters shows extraordinary potential for community type differentiation.