Forest parameter estimation from dual-frequency polarimetric SAR

Abstract:
Dual-frequency polarimetric SAR data is prepared in five scenarios – L-band, X-band, layer stack, additive, and multiplicative fusion – and subjected to a Random Forest regression analysis for the large-scale estimation of forest parameters like tree type, count, and height, crown area, base height and volume etc. along four transects in the Bavarian Forest National Park, Germany. Half a million tree polygons serve as reliable reference. The prediction performance is evaluated per transect, over all transects, and as transfer model. The simple layer stack of Kennaugh elements delivers best predictions within the transects, whereas the additive fusion convinces with superior transferability.