National forest mapping of stem volume using TanDEM-X

In this study, we have used a nation-wide collection of 518 TanDEM-X images, an ALS based DTM, and 7-10 m sample plots from the Swedish National Forest Inventory (NFI) in order to create a nation-wide pixel product of stem volume. The overall prediction accuracy was about 24%-25% at the stand-level, depending on the region that was used. However, the accuracy varied spatially depending on weather and forest conditions, and improved predictions can possibly be achieved by using scene-wise models with additional care taken to e.g. temperature and Euclidian distance to the reference data.