Evaluating different models to predict biomass increment from multi-temporal lidar sampling and remeasured field inventory data in south-central Alaska

Abstract

We evaluated two sets of equations for their predictive abilities for estimating biomass increment using successively acquired airborne lidar and ground data collected on western lowlands of the Kenai Peninsula in south-central Alaska. The first set included three base equations for estimating biomass increment as a function of lidar metrics, and the remaining equations enhanced the three base equations by considering the hierarchical structure of the data.

It is shown that the mixed effect framework substantially improved the accuracy and precision of biomass increment prediction over the fixed effects that assume the observations are independent for the area covered by two lidar acquisitions, 5 years apart from one another. On the average, root mean square error values were reduced by 19.8% by using a plot-level random coefficient model that account for the impacts of site (biophysical factors) on biomass increment on the western Kenai Peninsula.  

Mixed effect models are effective statistical tools, but their effective application requires some sample growth data. As such, we recommend two models for estimating biomass increment on the Kenai Peninsula.  If a subsample of ground data is available to predict the plot random intercept, the enhanced model is suggested.