Estimating Crown Volume through Implicit Surface Reconstruction from LIDAR Points for Forest Fire Simulation

Knowledge of tree crown information is critical to the modeling of forest fires. For example, FARSITE, one of the most commonly used fire simulators, uses crown volume to estimate crown fire behavior. Conventionally, tree crown volume is estimated using plot-level regression models from LIDAR (Light Detection and Ranging) data of the canopy structure. Convectional approaches, however, tend to result in rough estimations of crown volume. A more accurate method for computing single stand-level crown volume from LIDAR data would result in improving species characterization, automating tree identification from LIDAR data, and simulating fire behavior more precisely than conventional approaches. In this research, we propose a novel technique which more accurately computes individual tree crown volume from LIDAR data. First, we identify individual trees from unorganized LIDAR data points using a level set method, a shortest path algorithm and known GPS points for the stem locations. Second, we use radial basis functions (RBFs) to reconstruct implicit surfaces approximating individual tree crown shapes. These implicit surfaces, which effectively "wrap" each tree crown, are used to reconstruct a Digital Surface Model (DSM) of the canopy and to estimate individual tree crown volume.

Download : ka-ca-sc-sc-2006.pdf