Mapping of floodplain inundation with multi-frequency polarimetric SAR: Use of a tree-based model

The ability of synthetic aperture radar to detect flooding beneath a plant canopy has been demonstrated for a wide variety of herbaceous and woody vegetation types by Hess et al. (1990). The potential exists to conduct accurate regional surveys of seasonal flooding events, independent of cloud cover, for input to hydrological and biogeochemical models. On a smaller scale, accurate wetlands delineation would be of great value to managers of forest, wildlife, and fisheries resources. In order to realize the potential for operational mapping of inundation, robust classification methods are needed which do nor require repeated optimization by scene or date. The accuracy of traditional parametric classifiers (e.g., maximum likelihood) is likely to be compromised by scene to-scene variability in absolute sensor calibration, dielectric constant of soil and vegetation, phenological state of vegetation, and vegetation community structure. Knowledge-based expert systems offer an alternative approach using classification rules derived from a spectral knowledge base (Wharton 1989). The authors illustrate how tree-based models can be used to construct classification rules from a microwave spectral database and to select optimal sets of sensor parameters for mapping of inundation in the floodplain and estuary of the Altamaha River, Georgia.