Estimating forest above-ground carbon using object-based analysis of very high spatial resolution satellite images

摘要

The potentials and limitations of very high spatial resolution images for aboveground carbon (AGC) estimation are unknown and the methods are not developed. This research was designed to develop a method that predicts AGC at the individual tree level using object-based analysis of very high resolution QuickBird satellite images and in situ diameter at breast height (DBH) measurements. This study was based on the fact that, crown projected areas (CPA) are strongly correlated to DBH. Assuming that CPAs are delineated with higher accuracy, a spatial model that predicts AGC can be developed using in situ DBH measurements and allometric equations. The DBH (1.3 m) of sample coniferous and broadleaf trees was measured and converted to above ground biomass and then to carbon. The panchromatic and pan-sharpened QuickBird satellite images were processed through object-based analysis to derive the CPA of coniferous and broadleaf trees and then followed by an accuracy assessment. The developed model predicted AGC stock and linearly explained about 58 and 55% of the variances for coniferous and broadleaf trees, respectively. Errors of CPAs resulted from over- and under-segmentation, sampling errors, and allometric errors as well as uncertainties in the developed model caused by structural errors.