|
The estimation of vegetation traits, which is essential to characterize the health of trees from remote sensing data, presents several challenges in urban environments, due to the topography of 3D buildings and associated shading, the spectral diversity of materials, or the variety of urban morphology. Moreover, the difficulty to estimate the vegetation traits increases with the decrease of spatial resolution, mixed pixels including information on trees and their environment. The objective of this study is to estimate the influence of tree-endogenous (chlorophyll, LAI...) and tree-exogenous (urban form, tree distance to buildings, street orientation, solar angles, material types...) factors on the reflectance of Sentinel-2 pixels (10/20 m resolution). For this, a sensitivity analysis was carried out with the DART 3D radiative transfer model. First, a design of experiments was built using 15 variables describing the trees and their environment. Four urban 3D scenes that were elaborated based on the Local Climate Zone (LCZ) typology. For each of these urban 3D scenes, 3000 simulations were generated. Then, Sobol indices were computed to estimate the influence of each factor on the Sentinel-2 reflectance, more specifically on the ten spectral bands and eight vegetation indices correlated to vegetation traits. These experiments were conducted on isolated and aligned trees. In addition, the influence of the geo-registration uncertainty of the Sentinel-2 products was assessed in comparing the results obtained using a single tree-centered pixel with those using pixels offset from the tree. Results showed that Sentinel-2 data at 10 m resolution, NDVI et ARVI indices are the most relevant for the estimation of vegetation traits both for isolated and aligned trees, especially in LCZ five and eight, and in using a single tree-centered pixel approach.
|
