By Valérie Laurent
Abstract:
The aim of this study is to retrieve pixel-based canopy cover and crown diameter for a boreal mountain forest located in the Swiss National Park to help evaluating the quality of the protection provided by the forest against natural hazards. The methodology involved the use of a radiative forest canopy reflectance model – the geometric optical model of Li and Strahler, high resolution ROSIS, and multi-angular CHRIS/PROBA data.
The ROSIS image was classified in 5 classes (sunlit background, sunlit canopy, shadow, riverbed and road). This classification was used to spatially unmix the nadir CHRIS image and retrieve the signatures of the sunlit background, sunlit canopy and shadow endmembers. The 5 CHRIS angular acquisitions were then spectrally unmixed using those 3 endmembers to provide images of the proportion of sunlit background for each viewing angle. Finally, the Li-Strahler model was inverted over the proportion images using the stand and tree parameters, viewing and solar angles, slope and aspect data. Since the field measurements could not be obtained simultaneously, the canopy cover and crown diameter reference data was computed, by using LiDAR-derived tree data.
The backward viewing angles did not provide realistic results for a sufficient number of pixels and were discarded from the study. The analysis showed that the proportion of sunlit background was underestimated for the nadir viewing angle and overestimated for the forward scatter viewing angles. As a consequence, both the canopy cover and the crown diameter were overestimated for the nadir viewing angle and underestimated for the forward scatter viewing angles. The correlation between the model inversion output and the reference data were very weak. The combination of the inversion outputs for the individual viewing angles did not provide better results. The results were better for the canopy cover (R-squares between 0.01 and 0.13) than for the crown diameter (R-squares between 0.01 and 0.05). The nadir estimate provided the best results for the canopy cover. No best estimate could be distinguished for the crown diameter.
The poor quality of the results might be due to the presence of small trees in the understory, the assumption that the relative illumination and viewing geometry of the ROSIS and CHRIS nadir images were identical, the assumption of Lambertian behaviour of the canopy endmember for the spectral unmixing, the choice of an ellipsoid crown shape and the use of a random distribution to characterize the repartition of the trees over the scene for the model inversion.
Further research should first compare the model outputs to the field data. Then, different crown shapes may be used for the model inversion. It may also be useful to input the pixel tree density. To solve for the non-Lambertian behaviour of the canopy endmember, one could try to simulate the canopy signature for the zenith angles corresponding to the CHRIS acquisition geometry. Finally, to be able to evaluate the quality of the protection provided by the forest, the study area has to be extended to the whole forested area. This should involve the collection of protection, meteorological and soil data.
Keywords:
Conifer forest, canopy cover, crown diameter, hazard protection, multi-angular data, CHRIS/PROBA, Li-Strahler BRF model