Assessing Sampling Uncertainty in FVS Projections Using a Bootstrap Resampling Method

USDA Forest Service Proceedings RMRS-P-25. 2002 In: Crookston, Nicholas L.; Havis, Robert N., comps. 2002. Second Forest Vegetation Simulator Conference; 2002 February 12–14; Fort Collins, CO. Proc. RMRS-P-25. Ogden, UT: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station. T. F. Gregg recently retired from the USDA Forest Service, Region 6 in Portland, OR. He was the Regional Biometrician for the Forest Insects and Disease Group. S. Hummel is a Research Forester, USDA Forest Service, Pacific Northwest Research Station, P.O. Box 3890, Portland, OR 97208. Abstract—The Forest Vegetation Simulator (FVS) lets users project changes in forest stands associated with different initial conditions and silvicultural treatments. Our objective is to develop tools that help model users estimate the precision of FVS projections. A technique called bootstrap resampling (bootstrapping) allows us to approximate the sampling distribution of any variable simulated by FVS. To use the technique, the original FVS tree list is sampled repeatedly, with replacement, to build hundreds of bootstrapped tree lists. These bootstrapped tree lists are then used to make several hundred FVS projections. Each projection is thus based on a resample of the original tree list. The resulting empirical distribution provides information on the sampling uncertainty associated with the original tree list, which is important for making statistical inferences about FVS model outcome. This paper introduces a new bootstrapping program (FVSBoot) and describes its purpose and potential value.