Canopy biophysical variables estimation from meris observations based on neural networks and radiative transfer modelling: Principles and validation

Algorithms have been developed to estimate vegetation biophysical variables (the products) from MERIS top of canopy reflectance observations. This includes the leaf area index (LAI), the fraction of photosynthetically active radiation absorbed by the canopy (fAPAR), the cover fraction (fCover), and the canopy integrated chlorophyll content (LAI.Cab).. The algorithm is based on the training of neural networks over an extensive data set representing a large variability in canopy characteristics made of radiative transfer model simulations (SAIL, PROSPECT). The architecture of the back-propagation neural network was optimized for each biophysical variable and provides good theoretical performances for fAPAR and fCover, while estimates of LAI and LAI.Cab for the denser vegetation shows a degradation of the performances due to saturation. Preliminary validation of the products confirmed the theoretical performances as compared to other satellite products (MODIS, MGVI) and ground measurements.