Mapping boreal forest biomass from a SRTM and TanDEM-X based on canopy height model and Landsat spectral indices

Abstract

We propose a method for mapping above-ground biomass (AGB) (Mg ha−1) in boreal forests based predominantly on Landsat 8 images and on canopy height models (CHM) generated using interferometric synthetic aperture radar (InSAR) from the Shuttle Radar Topographic Mission (SRTM) and the TanDEM-X mission. The original SRTM digital elevation model (DEM) was corrected by modelling the respective effects of landform and land cover on its errors and then subtracted from a TanDEM-X DSM to produce a SAR CHM. Among all the landform factors, the terrain curvature had the largest effect on SRTM elevation errors, with a r2 of 0.29. The NDSI was the best predictor of the residual SRTM land cover error, with a r2 of 0.30. The final SAR CHM had a RMSE of 2.45 m, with a bias of 0.07 m, compared to a lidar-based CHM. An AGB prediction model was developed based on a combination of the SAR CHM, TanDEM-X coherence, Landsat 8 NDVI, and other vegetation indices of RVI, DVI, GRVI, EVI, LAI, GNDVI, SAVI, GVI, Brightness, Greenness, and Wetness. The best results were obtained using a Random forest regression algorithm, at the stand level, yielding a RMSE of 26 Mg ha−1 (34% of average biomass), with a r2 of 0.62. This method has the potential of creating spatially continuous biomass maps over entire biomes using only spaceborne sensors and requiring only low-intensity calibration.