Model-Assisted Forest Yield Estimation with Light Detection and Ranging

Abstract

Previous studies have demonstrated that light detection and ranging (LiDAR)-derived variables can be used to model forest yield variables, such as biomass, volume, and number of stems. However, the next step is underrepresented in the literature: estimation of forest yield with appropriate confidence intervals. It is of great importance that the procedures required for conducting forest inventory with LiDAR and the estimation precision of such procedures are sufficiently documented to enable their evaluation and implementation by land managers. In this study, we demonstrated the regression estimator, a model-assisted estimator (approximately design-unbiased), using LiDAR-derived variables for estimation of total forest yield. The LiDAR-derived variables are statistics associated with vegetation height and cover. The estimation procedure requires complete coverage of the forest with LiDAR and a random sample of precisely georeferenced field measurement plots. Regression estimation relies on sample-based ordinary least squares (OLS) regression models relating forest yield and LiDAR-derived variables. Estimation was performed using the OLS models and LiDAR-derived variables for the entire population. Regression estimates of basal area, volume, stand density, and biomass were much more precise than simple random sampling estimates (design effects were 0.25, 0.24, 0.44, and 0.27, respectively).