B31F-08
Using Multiple Soil Carbon Maps Facilitates Better Comparisons with Large Scale Modeled Outputs
Wednesday, 16 December 2015: 09:45
2008 (Moscone West)
Kristofer D Johnson, USDA Forest Service Northern Research Statiuon, Newtown Square, PA, United States, David V D'Amore, USDA Forest Service, Vallejo, CA, United States, Neal J. Pastick, Stinger Ghaffarian Technologies Sioux Falls, Sioux Falls, SD, United States, Helene Genet, University of Alaska Fairbanks, Fairbanks, AK, United States, Umakant Mishra, Argonne National Laboratory, Environmental Science, Argonne, IL, United States, Bruce K Wylie, USGS, EROS Data Center, Baltimore, MD, United States and Norman B Bliss, ASRC InuTeq, Sioux Falls, SD, United States
Abstract:
The choice of method applied for mapping the soil carbon is an important source of uncertainty when comparing observed soil carbon stocks to modeled outputs. Large scale soil mapping often relies on non-random and opportunistically collected soils data to make predictions over remote areas where few observations are available for independent validation. Addressing model choice and non-random sampling is problematic when models use the data for the calibration and validation of historical outputs. One potential way to address this uncertainty is to compare the modeled outputs to a range of soil carbon observations from different soil carbon maps that are more likely to capture the true soil carbon value than one map alone. The current analysis demonstrates this approach in Alaska, which despite suffering from a non-random sample, still has one of the richest datasets among the northern circumpolar regions. The outputs from 11 ESMs (from the 5th Climate Model Intercomparison Project) and the Dynamic Organic Soil version of the Terrestrial Ecosystem Model (DOS-TEM) were compared to 4 different soil carbon maps. In the most detailed comparison, DOS-TEM simulated total profile soil carbon stocks that were within the range of the 4 maps for 18 of 23 Alaskan ecosystems, whereas the results fell within the 95% confidence interval of only 8 when compared to just one commonly used soil carbon map (NCSCDv2). At the ecoregion level, the range of soil carbon map estimates overlapped the range of ESM outputs in every ecoregion, although the mean value of the soil carbon maps was between 17% (Southern Interior) and 63% (Arctic) higher than the mean of the ESM outputs. For the whole state of Alaska, the DOS-TEM output and 3 of the 11 ESM outputs fell within the range of the 4 soil carbon map estimates. However, when compared to only one map and its 95% confidence interval (NCSCDv2), the DOS-TEM result fell outside the interval and only two ESM’s fell within the observed interval. Overall, these results challenge how we view the accuracy of soil carbon reference data and our interpretation of model output validity. The example in Alaska also gives a reference for understanding similar issues when using soil carbon maps for comparison within other data-sparse northern circumpolar regions.
