Abstract
In the current context of climate change, and in particular the increasing intensity and frequency of droughts and fires, Mediterranean ecosystems are becoming increasingly vulnerable. Satellite imagery is increasingly used to monitor the evolution of functional features in such habitats. The arrival of hyperspectral sensors (such as DESIS, PRISMA, EnMAP) and the preparation of new missions (CHIME, SBG, BIODIVERSITY) should improve the estimation of these features thanks to a finer spectral resolution. The SENTHYMED project studies the contribution of hyperspectral images at different spatial resolutions (AVIRIS-Next Generation, PRISMA and DESIS) compared with the Sentinel-2 multispectral sensor in assessing functional traits of Mediterranean forests, in particular leaf pigment concentrations and leaf water and dry matter contents, using inversion methods based on data simulated by the DART radiative transfer model. It follows on from the HyperMED project. Two Mediterranean forests in the south of France (Pic Saint Loup and Puéchabon), mainly composed of pubescent oaks and holm oaks, with more or less closed canopies, are being studied. Several field campaigns carried out between April and October 2021 have made it possible to build up a database of spectral (reflectance and transmittance at leaf, trunk, undergrowth and canopy scales), biophysical-chemical (pigment, water and dry matter content of leaves) and structural (Leaf Area Index (LAI)) measurements. LiDAR drone acquisitions were also carried out to describe horizontal and vertical forest heterogeneity. All these data are used to simulate forests using DART. LiDAR point clouds are first converted into Plant Area Density (PAD) voxel matrices using AMAPVox [1]. Pytools4dart [2] is then used to build 3D models of the sites, manage DART parameterization and generate spectral reflectance images at canopy scale. These are then compared with the various hyperspectral acquisitions to assess the accuracy of the modeling using conventional spectral metrics. Ultimately, the aim is to use physical inversion to derive functional forest traits from hyperspectral images and assess their accuracy in relation to field data, based on Miraglio et al. (2022) [3]. The work presented here is therefore an essential pre-step in ensuring the reliability of simulations according to the degree of accuracy of the 3D model of the reconstructed forest.