Search

Forest stem volume information is needed in planning of sustainable forestry, mapping of exploitable forest resources, carbon balance studies and many other environmental applications. Radar sensors offer an efficient and weather-independent means for stem volume mapping. The radar dataset consisted of an ALOS/PALSAR dual-polarised scene from September 2008 and four TerraSAR-X spotlight scenes from February-March and July-August 2009. Ground data consisted of plot-wise data and stand-wise data. Regression models were developed with stand-wise training data where the stem volume varied between 0 and 390 m3/ha. The best 3-predictor model - 2 ALOS-PALSAR amplitudes and the phase difference between HH and VV data in a TerraSAR-X scene - produced an RMSE of 46 m3/ha (R2 = 0.7) when evaluated against stands not used in the model training. Stem volume estimation with plot-wise ground data produced lower estimation accuracies. The main reason was most likely mis-registration between the oppositelooking ALOS/PALSAR and TerraSAR-X scenes, which was caused by canopy height and which was not corrected by ortho-rectification with terrain elevation model. In future work, similar look directions should be used when no canopy surface model is available for orthorectification. A filtering approach was developed for using the stand-wise stem-volume model in areas with no forest stand information.