Forest Height Reconstruction From Ariborne Data: The Inversion Algorithm and its Experimental Validation

Kurzfassung

In this thesis has been developed and tested an algorithm for the forest height reconstruction. This project goes toward the Kyoto protocol request of realizing an inventory for the forest biomass all over the Planet. As proved by allometric relation the forest height is an adequate forest biomass estimator. Currently several techniques are proposed for the forest height reconstruction. Models have been developed for extracting this information by a single image SAR. Another technique considers the SAR interferometry in high frequency (e.g. X band) exploiting the low penetration of the high frequency waves in the canopy, although is inevitable the forest height underestimation that becomes smaller with the increasing of the canopy density. Consequently the idea is to use more powerful SAR technology for carrying out further information about the scene. The SAR polarimetric interferometry applies this thought for performing the tree stages height inversion that is able to reconstruct the forest height. In this thesis is followed another technique to increase the scene observables. In the multi-baseline approach more SAR images of the same area are considered. Initially was necessary to develop a scattering model, considering further acquisitions of the forest area. The model used considers a random uniform volume over a ground surface. Following was developed the algorithm that performs the constrained minimisation of the distance between the predicted and the observable coherencies. The constrain makes the minimisation more stable however are requested a priori information about the scene (information that usually is not difficult to derive with an optical image). The algorithm inversion capability was tested using a sensitivity analysis. It has been revealed a very helpful instrument for understanding the parameters setting in the constrained minimisation and for comprehending the algorithm limits. The main inversion difficulty (strong underestimation) was found when the ground scattering field contribution is too low, this could happen where the wave attenuation through the forest is significant. For this reason was proposed to increase, with a POLinSAR method, the ground scattering value. A coherence optimisation has been chosen for this goal, and was performed using a innovative algorithm (Numerical Radius). Following the inversion was tested, successfully, with simulated data (M. Williams & S. Cloude) of a forest of 10 meters height. In that case, when the ground contribution is not very low the improvement obtained using the optimisation coherence respect a single polarisation is only less noisy result. Finally the algorithm was tested over a real scenario with experimental airborne data, from the Microwaves and Radar Institute of the German Aerospace Centre, the E-SAR system. The inversions were performed using a single or the optimum polarisations. Both the techniques were able to recover the mean forest height (except for few cases using single polarisation), however, in all the inversions optimum reconstruction seemed less noisy. The algorithm was able, especially with the optimum polarisation, to follow the profile of the forest areas. A difficulty remains with the very tall forest because of the strong wave attenuation through the canopy. Anyhow, as expected, with the optimum coherence the forest height underestimation was reduced dramatically. The algorithm obtain more reasonable results with the optimum polarisation principally for the ground contribution increasing in the scattered field. although should be significant also the less noisy amplitude coherencies value that brings less noise in the forest height reconstruction. In conclusion we have shown that a remote sensing forest height extraction is suitable using a multi-baseline approach in order to carry out the necessary observables from the SAR scenes. The 2-Layer model used in this algorithm development has been demonstrated very useful for our goals, because of the simplicity as well as the accuracy of the prediction. Finally the forest height reconstruction has a significant improvement by the use of the coherence polarisation.