Estimation of forest parameters using 3D satellite data

Abstract: Accurate data about the forest are needed both in climate research and forest management planning. This thesis focuses on using different types of three-dimensional (3D) satellite data sources to accurately estimate forest variables, primarily above-ground biomass (AGB) and tree height (H). Different satellite sensors have been used and compared, and depending on the sensors, the acquired satellite images have been processed in different ways to create 3D surface models. The three main processing alternatives have been stereogrammetry, radargrammetry and interferometry. The surface models were used together with digital terrain models from airborne laser scanning, to take the difference between the two in order to create canopy height models. The models have been built and evaluated on data from two test sites: Krycklan located in northern Sweden (Lat. 64°N, Long. 19°E) and Remningstorp in southern Sweden (Lat. 58°N, Long. 13°E). The included studies have shown that 3D satellite data are efficient to use for accurate estimations of AGB and H at stand-level. Moreover, the optical stereogrammetric models can play a role in the boreal region, but to obtain accurate estimations they are dependent on rather high resolution, along-track data with pixel sizes smaller than 10 m. Radargrammetry applies stereogrammetry to radar images, which can be taken regardless of atmospheric conditions, time of day, or season. Data from TerraSAR X were used and AGB and H could be estimated with 22.9% and 7.7% root mean square error (RMSE), respectively, at stand-level. Interferometry was applied to data from the TanDEM-X mission and this technique was superior to stereogrammetric and radargrammetric techniques, where AGB and H could be estimated with 14.4% and 4.1% RMSE, respectively, at stand-level. The first three studies in this thesis used empirical approaches and in contrast to this, the last two studies employed the two-level model (TLM), which has been developed with semi-empirical modeling based on simplified physical reasoning. The TLM also used TanDEM-X data and this showed potential for even more accurate predictions of forest related variables. To conclude, this thesis shows the potential of 3D satellite data as a highly reliable and a widely applicable remote sensing data source for estimation of forest parameters.