Modelling and Retrieval of Forest Parameters from Synthetic Aperture Radar Data

Licentiate thesis, 2012

Currently, one of the most uncertain factors in the global carbon cycle models lies in the terrestrial carbon stock, mainly forests. The available methods for global forest resource mapping provide only rough estimates of biomass, the most relevant practical quantity related to carbon stock. Spaceborne synthetic aperture radar (SAR) is a tool potentially suitable for global forest monitoring. As an active microwave sensor, SAR has the advantage of being independent of weather and external illumination. Spaceborne SAR can be designed for different frequencies and with resolutions as low as a few metres. Moreover, SAR systems operating at frequencies below L-band show good sensitivity to biomass. A spaceborne solution introduces also the possibility of frequent acquisitions, which is beneficial in applications such as detection of unlawful clear-cutting, storm damages, and forest fires. In the first paper, a new biomass retrieval model for boreal forest using polarimetric, airborne P-band SAR backscatter is presented. The model is based on two main SAR quantities: the HV backscatter gamma nought and the HH/VV backscatter ratio, together with a topographic correction. Data from the two airborne experiments BioSAR 2007 and BioSAR 2008, performed in two distinct test sites Remningstorp and Krycklan, were used for this study. The model was compared to other, previously published models in a set of tests. In one of the tests, the models were evaluated across sites, i.e. training was done with data from one test site, and the models were validated using data from the other test site. Stand-wise root-mean-square errors of 40-59 tons/ha, or 22-32% of the mean biomass were observed for across-site validation. In the second paper, a forward model for extended covariance matrix prediction for boreal forest in P-band SAR is presented. Data from BioSAR 2007 campaign were used for model derivation. The model is able to predict backscatter at HH, HV, and VV, together with the complex correlation between HH and VV, and complex correlation coefficients for three interferometric pairs (one for each polarisation). The forward model builds on a physical model and linear regression of BioSAR 2007 data. The model is further developed in the third paper. In the fourth paper, a tropical forest scenario is added, derived from the data acquired within the TropiSAR 2009 experiment. In the fifth paper, spaceborne SAR is used to delineate wind-thrown trees and clear-cuts during a controlled experiment conducted in the test site of Remningstorp in 2009. Data from three satellites were used: ALOS PALSAR (L-band), RADARSAT-2 (C-band), and TerraSAR-X (X-band). The detection capabilities vary for the different satellites due to different resolutions, and also due to different scattering properties. It is observed, that TerraSAR-X is suitable for storm damage detection due to its high resolution. ALOS PALSAR is suitable for detection of clear-cuts due to its sensitivity to biomass.