Mapping of Taiga Forest units using AIRSAR data and/or opticaldata, and retrieval of forest parameters

摘要

A maximum a posteriori Bayesian classifier is used to perform asupervised classification of multifrequency, polarimetric, airborne, SARobservations of boreal forests from the Bonanza Creek ExperimentalForest, near Fairbanks, Alaska, into six categories: 1) white spruce; 2)black spruce; 3) balsam poplar; 4) alder; 5) treeless areas; and 6) openwater. Tree classification accuracy is highest (86%) using L- and C-bandfully polarimetric combined on a date where the forest just recoveredfrom river flooding. The SAR map compares favorably with a vegetationmap obtained from digitized aerial infra-red photos. C-band frequencyand HV-polarization are, respectively, the most useful frequency andpolarization for mapping tree types using SAR. Combination of multi-dateSAR observations does not improve classification accuracy, and SAR dataacquired on different dates, under different environmental conditions,yield classification accuracies 16% to 41% lower. Single-frequency,single-polarization, SAR data show limited mapping capability.Multispectral SPOT observations of the same area on a single date yielda classification accuracy of 78%. Combining optical and SAR data isuseful for identifying tree species, independent of ground truthverification, using biomass estimates from SAR, at L-bandHV-polarization, NDVI from SPOT red and infra-red radiances, and anunsupervised segmentation map of the SAR data