Land cover type and biomass classification using AirSAR data for evaluation of monitoring scenarios in the Colombian Amazon

Abstract

The potential role of a spaceborne SAR component within a dedicated global monitoring system for tropical rain forest areas is investigated. Use is made of NASA's airborne radar system AirSAR, which acquired C-band, L-band, and P-band polarimetric data of a colonization area located at the edge of the Colombian Amazon. Classification accuracy for primary forest, secondary forest, recently deforested areas, and pastures are studied to determine optimal wave parameter combinations, using an extensive database of 778 plots. Kappa statistics are used to compare results for different combinations. The relevance of polarimetry and the effect of speckle level are studied by incorporating the multilook pdf's of polarimetric phase differences and the polarimetric correlations. Kolmogorov-Smirnov tests of fit well confirm the agreement of theoretical pdf's used and experimental observations. In addition, possibilities for biomass estimation are studied using detailed vegetation structure measurements of bush-invaded grasslands (5 plots), secondary forest (10 plots), and primary forest (13 plots). Accuracy for land cover-type classification over 90% can only be obtained when two frequency bands are combined. L-band with HV polarization and P-band show the best possibilities for biomass estimation. After land cover-type classification, eight biomass classes can be differentiated at a high level of confidence. The results clearly indicate how SAR systems may be designed to accurately monitor processes of deforestation, land and forest degradation, and secondary forest regrowth. The effect of Faraday rotation on P-band data collected from spaceborne SAR is also taken into consideration.