Comparison of Techniques for Estimation of Forest Soil Carbon

Soil organic carbon represents the largest constituent of the global C pool and carbon budgets are studied by researchers and modelers in C cycling, global climate change, and soil quality studies. Pedon and soil interpretation record databases are used with soil and ecological maps to estimate regional SOC even though these databases are rarely complete for surface litter and mineral subsurface horizons.

The first main objective of the project is to improve the ability to produce soil organic carbon estimates from existing spatial soils datasets, such as STATSGO. All records in the STATSGO Layer table that were incomplete or appeared to be incorrectly filled with a null or zero value were considered invalid. Data sorting procedures and texture lookup tables were used to identify exiting correct (valid) data entries that were used to substitute invalid records. STATSGO soil property data were grouped by soil order, MLRA, layer number, and texture to produce replacement values for all invalid data used to calculate mass SOC. Grouping criteria was specific to each variable and was based on texture designations. The resulting filled and unfilled tables were used with procedures assuming Normal and Lognormal distribution of parameters in order to analyze variation of mass SOC estimates caused by using different computation techniques.

We estimated mass SOC to 2 m in Maine and Minnesota using filled and unfilled STATSGO data tables. Up to 54% of the records in Maine and up to 80% of the records in Minnesota contained null or zero values (mostly in fields related to rock fragments) that were replaced. After filling, the database resulted in 1.5 times higher area-weighted SOC. SOC calculated using the Normal distribution assumption were 1.2 to 1.5 times higher than those using the Lognormal transformation. SOC maps using the filled tables had more logical geographic SOC distribution than those using unfilled tables.

The USDA Forest Service collects and maintains detailed inventory data for the condition and trends of all forested lands in the United States. A wide range of researchers and landowners use the resulting Forest Inventory and Analysis (FIA) database for analytical and decision making tasks. FIA data is available to the public in transformed or aggregate format in order to ensure confidentiality of data suppliers.

The second main objective of this project was to compute SOC (kg m-2) results by FIA forest type and forest type group for three depth categories (25 cm, 1 m, and 2 m) at a regional scale for the 48 contiguous United States. There were four sets of results derived from the filled STATSGO and FIA datasets for each depth class by region: (1) SOC computed by the Lognormal distribution approach for (1a) all soil orders, (1b) without Histosols; and (2) SOC computed by the Normal distribution approach for (2a) all soil orders, (2b) without Histosols.

Two spatial forest cover datasets were relevant to this project, FIA and AVHRR. We investigated the effects of FIA inventory data masking for Maine and Minnesota, such as plot coordinates rounding to the nearest 100 arc-second, and the use of 1 km resolution satellite-derived forest cover classes from AVHRR data, on SOC estimates to 2 m by forest type group. SOC estimates by soil mapping unit were derived from fixed STATSGO database tables and were computed by the Lognormal distribution approach including all soil orders.

The methods in this study can be used for a variety of ecological and resource inventory assessments and the automated procedures can be easily updated and improved for future uses. The procedures in this study point out areas that could benefit the most during future revisions of STATSGO. The resulting SOC maps are dynamic and can be rapidly redrawn using GIS whenever STATSGO spatial or tabular data undergo updating. Use of pedon data to define representative values for all properties in all STATSGO layers and correlation of STATSGO layers to soil horizons will lead to vast improvement of the STATSGO Layer table and promote its use for mass SOC estimation over large regions.