Empirical observations offer improved estimates of forest floor carbon content across in the United States

Tuesday, 16 December 2014
Charles Hobart Perry1, Grant M Domke1, Brian F Walters1, James E Smith2 and Christopher W Woodall1, (1)Northern Research Station, Saint Paul, MN, United States, (2)Northern Research Station, Durham, NH, United States
The Forest Inventory and Analysis (FIA) program of the United States Forest Service reports official estimates of national forest floor carbon (FFC) stocks and stock change to national and international parties, the US Environmental Protection Agency (USEPA) and the United Nations Framework Convention on Climate Change (UNFCCC), respectively. These estimates of national FFC stocks are derived from plot-level predictions of FFC density. We suspect the models used to predict plot-level FFC density are less than ideal for several reasons: (a) they are based upon local studies that may not reflect FFC dynamics at the national scale, (b) they are relatively insensitive to climate change, and (c) they reduce the natural variability of the data leading to misplaced confidence in the estimates. However, FIA has measured forest floor attributes since 2001 on a systematic 1/16th subset of a nation-wide array of inventory plots (7 800 of 125 000 plots). Here we address the efficacy of replacing plot-level model predictions with empirical observations of FFC density while assessing the impact of imputing FFC density values to the full plot network on national stock estimates. First, using an equivalence testing framework, we found model predictions of FFC density to differ significantly from the observations in all regions and forest types; the mean difference across all plots was 21 percent (1.81 Mg·ha-1). Furthermore, the model predictions were biased towards the lower end of extant FFC density observations, underestimating it while greatly truncating the range relative to the observations. Second, the optimal imputation approach (k-Nearest Neighbor, k-NN) resulted in values that were equivalent to observations of FFC density across a range of simulated missingness and maintained the high variability seen in the observations. We used the k-NN approach to impute FFC density values to the 94 percent of FIA inventory plots without soil measurements. Third, using the imputed values of FFC density our preliminary estimate of FFC stocks in the US is 6 333 Tg (±106 Tg). This estimate is 29.9 percent larger (1 457 Tg) than current model-based estimates. Our research yields more representative estimates of both plot-level FFC densities and national FFC stocks that will be submitted to the USEPA and UNFCCC as part of the 2015 estimates.