| Summary: | The combined use of field data and remote sensing to carry out forest inventories is a topic of current interest. One of the important challenges for its practical application is to optimize/minimize the volume of data to be used to achieve acceptable estimates. In this study, we analyzed the effect of the sampling design and density of LIDAR returns on the accuracy of basal area (AB), timber volume (V), and biomass (B), in addition to sampling estimators assisted by generalized additive models (GAM) and the Random Forest (RF) algorithm for a forest under management located in Zacualtipán, Hidalgo. There were 96 field sampling sites (400 m2), three LIDAR sampling designs, and 10 return densities. Two-phase, two-stage estimators were analyzed to estimate the total inventory. The GAM models proved to be efficient in estimating (0.76 to 0.92 of R2) forest variables at the LIDAR transect level. The RF algorithm showed acceptable goodness of fit (0.71 to 0.79 R2) for estimating variables at the study area level. The regression-assisted estimators showed good accuracy with an error of less than 6% in the inventory of the evaluated variables. It was demonstrated that transect sampling of LIDAR data is a viable alternative for the estimation of variables of forest interest in managed properties.
|
|---|
