Retrieval of canopy structural parameters from multiangle observations using an artificial neural network

This paper describes a procedure for the retrieval of canopy structural parameters (e.g. height, shape, density) from multiangle reflectance measurements using an artificial neural network (ANN). The objective is to train a neural network to learn the association of canopy structural parameters with its corresponding directional reflectance pattern. The Li-Strahler [1] geometric-optical mutual shadowing model is used to simulate the bidirectional reflectance of a canopy based on the geometry of the trees. The reflectance generated from the model is used as an input to a multilayer feed-forward neural network, with the canopy structural parameters as outputs. ANNs have great potentials to learn the relation (or any continuous function) between input patterns and desired outputs without any prior knowledge of the mapping function. Using the neural network retrieval approach, the R² between the model predicted canopy parameters and the actual parameters of density is 0.85 and 0.75 for the tree crown diameter and canopy height.