Wednesday, August 14, 2019
Exhibit Hall, Kentucky International Convention Center
Ge Pu, ESF-ERE, SUNY-ESF, Syracuse, NY, John L. Campbell, Northern Research Station, USDA Forest Service, Durham, NH, John Drake, Sustainable Resources Management, SUNY College of Environmental Science and Forestry, Syracuse, NY, Mark B. Green, Northern Research Station, US Forest Service and Ruth D. Yanai, Forest and Natural Resources Management, SUNY College of Environmental Science and Forestry, Syracuse, NY
Background/Question/Methods
In ecosystem studies, where using replication to establish confidence is often not an option, Monte Carlo sampling is a powerful tool for propagating uncertainty. However, it is easy to make mistakes in applying Monte Carlo sampling to establish confidence intervals. We present two case studies. First, ecosystem estimates of forest biomass and nutrient contents depend on regression relationships between tree diameter and tree biomass and also on measurements of tissue chemistry, both of which have uncertainty. These sources are usually overlooked, while natural variability is estimated via sampling error. Second, ecosystem mass balances are important in estimating denudation rates and detecting changes such as those associated with forest disturbance. However, stream loads are commonly reported without fully accounting for uncertainty in the estimates, which makes it difficult to evaluate the significance of findings or to guide efforts to improve the efficiency of ecological monitoring programs. To illustrate the use of Monte Carlo sampling to propagate uncertainty sources, we used data from the Hubbard Brook Experimental Forest, New Hampshire, USA, where forest biomass and stream budgets have been reported for small headwater catchments since 1965.
Results/Conclusions
For forest biomass, allometric equations describing the relationship between diameter and mass are based on a destructive sample, which is not perfectly descriptive of the population of inference. We show that the estimate for large numbers of trees varies very little across Monte Carlo simulations when errors are applied independently for each tree. This is incorrect. The uncertainty in the model should be sampled only once for each iteration; if the model underestimates the average tree biomass, it should be underestimated for all the trees at once. For stream export of nutrients, sources of uncertainty include stage height-discharge relationship, watershed area, analytical chemistry, and gap filling in both discharge and stream chemistry. Some sources of error, such as the watershed area, apply to the entire time series, while others, such as the precision of laboratory analysis, introduce errors that are independent for each observation. We demonstrated that applying uncertainty in watershed area independently for each day of the record resulted in unreasonably small confidence intervals. It is important to conduct random sampling at the appropriate points in the simulation. To apply systematic errors independently to each observation results in an underestimate of the true uncertainty.
