Modelling and Retrieval of Forest Parameters from Synthetic Aperture Radar Data

Doctoral thesis, 2014

Frequent, high-resolution mapping of national and global forest resources is needed for improved climate modelling, degradation and deforestation detection, natural disaster management, as well as commercial forestry. Synthetic aperture radar (SAR) is an active radio- or microwave-frequency imaging sensor, which can be optimised to fit specific needs through the choice of the centre frequency. In particular, P-band SAR, with wavelengths around 70 cm, is a promising tool for biomass mapping due to the high sensitivity to tree trunks, whereas X-band SAR, with wavelengths around 3 cm and larger available bandwidths, is a promising tool for high-resolution mapping of forest canopies. Papers A and B summarise the results obtained within the feasibility study for the European satellite BIOMASS, which is planned to become the first spaceborne P-band SAR system. In Paper A, a forward model relating relevant forest and system parameters to SAR observables is presented and evaluated. In Paper B, a new model for biomass estimation is proposed, in which the significant influence of topographic and moisture variations is treated using empirical corrections. The new model can be used with the same model parameters in two boreal test sites in Sweden, separated by 720 km, with a root-mean-square error (RMSE) of 22{33% of the mean biomass. In Papers C, D, and E, X-band SAR data acquired with the twin-satellite, singlepass interferometric system TanDEM-X are studied. Using the principles of acrosstrack interferometry, the position of the scattering centre is estimated from the phase difference between two SAR images. With a high-resolution digital terrain model, the interferometric data are ground-corrected, and the elevation of the scattering centre above ground is determined. In Paper C, boreal forest biomass is estimated for one test site in Sweden from ground-corrected TanDEM-X data using three models with tree canopies represented by a random volume, but with different assumptions of the ground component. The best results, with an averaged RMSE of 16%, are obtained with a model accounting for canopy gaps. Based on this observation, a two-level model (TLM) is introduced, in which forest is modelled as two discrete scattering levels: ground and vegetation, the latter with gaps. In Paper D, it is shown that TLM inversion of single-polarised, ground-corrected TanDEM-X data can provide forest height and canopy density estimates, with RMSE values below 10% for a boreal test site in Sweden. In Paper E, biomass is estimated from the inverted TLM parameters, with an RMSE in the interval 12{19% for eighteen acquisitions from two boreal test sites in Sweden.