Inventory and cartography of forest variables derived from LiDAR data: comparison of methods

The most common method to estimate forest variables to a large or small scale is the forest inventory based on field sampling. Currently, remote sensing techniques offer a range of possibilities in forest resources estimation; this is the case of LiDAR (Light Detection And Ranging) that allows the characterization forest structure in three-dimensions. We analyzed the relationship between LiDAR and field data to estimate forest variables such as: basal area (AB), total biomass (BT), crown cover (COB) and timber volume (VOL) through four methods: 1) multiple linear regression, 2) non-linear regression, 3) ratio estimators and 4) traditional forest inventory (stratified sampling). Total estimates derived from the ratio estimator were within the 95% confidence interval calculated by traditional inventory for AB, BT and VOL; this estimator showed the closest values and precision to those obtained by traditional forest inventory. In general, estimates through non-linear models were the most optimistic compared to the traditional forest inventory. Our results indicated a good relationship (R-2 > 0.50) between LiDAR metrics and field data, particularly the percentiles of height and rates of return on a defined height. From the linear models fit we generated maps for each of the forest variables analyzed.