Quantifying the Model-Related Variability of Biomass Stock and Change Estimates in the Norwegian National Forest Inventory

Abstract

National Forest Inventories (NFIs) provide estimates of forest parameters for national and regional scales. Many key variables of interest, such as biomass and timber volume, cannot be measured directly in the field. Instead, models are used to predict those variables from measurements of other field variables. Therefore, the uncertainty or variability of NFI estimates results not only from selecting a sample of the population but also from uncertainties in the models used to predict the variables of interest. The aim of this study was to quantify the model-related variability of Norway spruce (Picea abies [L.] Karst) biomass stock and change estimates for the Norwegian NFI. The model-related variability of the estimates stems from uncertainty in parameter estimates of biomass models as well as residual variability and was quantified using a Monte Carlo simulation technique. Uncertainties in model parameter estimates, which are often not available for published biomass models, had considerable influence on the model-related variability of biomass stock and change estimates. The assumption that the residual variability is larger than documented for the models and the correlation of within-plot model residuals influenced the model-related variability of biomass stock change estimates much more than estimates of the biomass stock. The larger influence on the stock change resulted from the large influence of harvests on the stock change, although harvests were observed rarely on the NFI sample plots in the 5-year period that was considered. In addition, the temporal correlation between model residuals due to changes in the allometry had considerable influence on the model-related variability of the biomass stock change estimate. The allometry may, however, be assumed to be rather stable over a 5-year period. Because the effects of model-related variability of the biomass stock and change estimates were much smaller than those of the sampling-related variability, efforts to increase the precision of estimates should focus on reducing the sampling variability. If the model-related variability is to be decreased, the focus should be on the tree fractions of living branches as well as stump and roots.