BIOPHYSICAL FOREST TYPE CHARACTERISATION IN THE COLOMBIAN AMAZON BY AIRBORNE POLARIMETRIC SAR

摘要

Fully polarimetric C-, L- and P- band data were collected by NASA's AirSAR system at the Araracuara test site, a well-surveyed forest reserve in the centre of the Colombian Amazon. The area is characterised by a high diversity of forest types, soil types and flooding conditions. A polarimetric classification technique is used to assess AirSAR's potential for forest structural type mapping and, indirectly, forest biophysical characterisation. Field observations were made at 23 0.1 ha plots to obtain additional quantitative descriptions on forest structure and ground surface conditions, but also to assess the suitability of existing map legends for SAR mapping. It could be shown that a new type of legend leads to physically better interpretable results. A method based on iterated conditional modes is introduced and is shown to yield radar-derived maps with a high level of agreement with existing maps, as well as with the ground observations. The following results may indicate the high level of accuracy obtained: 15 classes can be differentiated, the average radar map agreement ranges from 68-94% (depending on the type of map and approach) and for only a few classes the agreement is less than 70%. The relation between physical forest structure and polarimetric signal properties is studied explicitly using polarimetric decomposition. A new method is introduced based on the decomposition of polarimetric coherence, instead of power. It is based on simple physical descriptions of the wave-object interaction. The accuracy of the complex coherence estimation derived from the complex Wishart distribution. Thus several interesting physical relations between polarimetric signal and forest structure can be revealed. The physical limitations of this technique and its relation with sample size are indicated.