Mapping and modeling forest tree volume using forest inventory and airborne laser scanning

In this paper, we present a study on the efficiency of multi-return LIDAR (Light Detection Ranging) data in the estimation of forest stem volume over a multi-layered forest area in the Italian Alps. The goals of this paper are (1) to verify the usefulness of multi-return LIDAR data compared to single-return data in forest volume estimation and (2) to define the optimal resolution of a stem volume distribution raster map over the investigated area. To achieve these goals, raw data were segmented into a net, and different cell dimensions were investigated to maximize the relationship between the LIDAR data and the ground-truth information. Twenty predicting variables (e.g., mean height, coefficient of variation) have been extracted from multi-return LIDAR data, and a multiple linear regression analysis has been used for predicting tree stem volume. Experimental results found that the optimal resolutions of the net square cells were 40 m. The analysis indicated that in a mixed multi-layered forest, characterized by a complex vertical structure, the correct selection of the map spatial resolution and the inclusion of the secondary-return data were important factors for improving the effectiveness of the laser scanning approach in forest inventories. The experimental tests showed that the chosen model is effective for the estimation of stem volume over the analyzed area, providing good results on all the three considered validation methods.