Estimating variation in a landscape simulation of forest structure

Abstract

Modern technology makes it easy to show how forested landscapes might change with time but it remains difficult to estimate how sampling error affects landscape simulation results. To address this problem we used two methods to project the area in late-seral forest (LSF) structure for the same 6070 hectare (ha) study site over 30 years. The site was stratified into patches by using aerial photos and data were collected on sample plots within the stratum. Our two methods differed in how the initial forest conditions in unsampled patches were attributed. Using method 1, we randomly assigned empirical plot data from sampled patches to unsampled patches within the same stratum; with method 2 we bootstrapped the plot data to identify a probability distribution of LSF structure for a sampled patch and then randomly assigned these probabilities to unsampled patches within the same stratum. Both methods used an individual tree growth model to project changes in forest structure. In the first decade, the ‘assignment’ method produced an estimate of 2489 ha of LSF structure, an amount lying at the upper tail of the projection interval created by using the bootstrapped or ‘probabilistic’ method. The probabilistic method identified the variation in LSF structure within landscape stratum, the strata most likely to contribute to LSF structure, the projected mean area in LSF structure each decade, and the variation in hectares associated with the projections. Data maps from both methods, together with projection intervals from the probabilistic method, offer enhanced ways to communicate the potential variation in landscape simulation.