Large area boreal forest investigations using ERS INSAR

Multi-temporal InSAR data from a boreal forest area in Sweden is analyzed in order to estimate stem volume using coherence and backscatter. A model-based regression is performed using 21 forest stands and tested on another 21 forest stands, 2 to 14 ha in size, from an area with accurate in situ stem volume data varying from 8 to 335 cu m/ha. The model approach is discussed and compared with other approaches found in the literature. Results from the different pairs are combined to give a best stem volume estimate. The accuracy in terms of RMSE for standwise estimated stem volume corrected for sampling errors is 10 cu m3/ha. Evaluation at plot level (20m dimneter - 216 plots) showed an RMSE of 55 m/ha. The best pairs are characterized by below zero temperature and snow on the ground. For the large area (4325 km2) 166 Swedish National Forest Inventory (NFI), plots were used as reference resulting in an RMSE of 71 cu m/ha, i.e. 30 percent worse than the reference area located up to 50 km away. The plot based accuracy estimates illustrate effects of the limited resolution of the coherence estimate and the variability of the forest and stresses the need for evaluations over forest stands. By averaging over larger areas we obtain an accuracy of 30 cu m/ha or better for areas 150 sq km or greater. We conclude that InSAR can for areas above 2 ha provide forest stem volume estimates of similar order of accuracy as ground data. (Author)