Comparing accuracy of satellite- and airborne remote sensing data in the retrieval of forest stand attributes

Recent advances in developing remote sensing methods for forest inventory have suggested a good potential for the use of multi-source, multi-dimensional remote sensing data. This paper describes results of the retrieval of stem volume, basal area, and mean height utilizing the following remote sensing data: Landsat TM, SPOT Pan and XS, ERS-1/2 PRI and SLC, airborne data from imaging spectrometer AISA, radar-derived forest canopy profiles (obtained with HUTSCAT), and aerial photographs. Ground truth data consist of 40 spruce-dominated and mixed stands, which have been carefully measured using about 10 relascope sample plots in each stand. Multivariate data analysis techniques were applied. The results suggested that (1) radar-derived stand profiles seemed to be the most accurate data source for forest inventory, (2) combining ERS-1/2 coherence images with Landsat TM improved estimation accuracy, (3) Landsat TM was a more accurate data source than other satellite data (SPOT Pan and XS, ERS-1/2 intensity and coherence images), (4) principal component analysis was found to be a good technique with this dataset to derive predictor variables from AISA data (30 channels with high multicollinearity), (5) higher coherence was obtained for winter images than summer images due to snow-covered ground