Background on the USGS assessment
Future scenarios of LULC change in the Great Plains of the U.S. have been developed as part of a national carbon sequestration assessment. The assessment, required by the U.S. Congress (Energy Independence and Security Act of 2007) and conducted by the USGS, used a methodology that linked ecosystem carbon models to separate models of wildfires and LULC changes, and produced spatially and temporally explicit carbon stock and flux estimates (Zhu et al. 2010). Future potential LULC change was based on a set of scenarios from three United Nations Intergovernmental Panel on Climate Change (IPCC) Special Report on Emission Scenarios (SRES) (Nakicenovic and Swart 2000): A2 (emphasizes economic development with a regional focus), A1B (emphasizes economic development with a global orientation), and B1 (emphasizes environmental sustainability with a global orientation) (Zhu et al. 2011).
To develop LULC change scenarios that were logically consistent with SRES storylines, Sleeter and others (2012) took projected national LULC change data from the Integrated Model to Assess the Global Environment (IMAGE) (Strengers et al. 2004) and allocated them to U.S. Environmental Protection Agency (EPA) Level III ecoregions based on land-use histories from the USGS Land Cover Trends project (Loveland et al. 2002), as well as expert knowledge. The LULC classification scheme of the scenarios closely followed the National Land Cover Database (NLCD) (Vogelmann et al. 2001; Homer et al. 2007) and includes broad classes such as cropland, hay/pasture, development, grassland, herbaceous wetlands and woody wetlands.
The USGS used a probabilistic LULC model, FOREcasting SCEnarios of land-use change (FORE-SCE) to distribute future regional LULC change on the landscape for each LULC change scenario (Sohl and Sayler 2008; Sohl et al. 2012). The allocation was based on probabilities of occurrence determined by present-day LULC associations with biophysical and socioeconomic characteristics of the landscape, such as slope, elevation, soil carbon, climate and distance to roads and cities. Model outputs are maps of LULC change generated yearly from 2006 to 2050 at a spatial resolution of 250 m.
Carbon dynamics and GHG fluxes between the land and the atmosphere under LULC change, climate change, and land management scenarios were estimated using three ecosystem models: the Erosion-Deposition-Carbon Model (EDCM) (Liu et al. 2003), the CENTURY model (Parton et al. 1987, 1993), and a spreadsheet model (Zhu et al. 2010). Using datasets for LULC change, simulations of areas burned by wildland fires, agricultural land management, climate, and other biophysical data, the three models were run for each of three IPCC-SRES scenarios and for one Global Climate Model (MIROC 3.2-medres) (Zhu et al. 2011). In this study we utilize outputs from one model — EDCM — because of its approach to modeling soil carbon dynamics.
The Great Plains region of the U.S. is characterized by a continental climate with cold winters and warm summers, and a moisture gradient that increases from west to east, from less than 380 millimeters (mm) of rain annually to 640 mm. Historically the Great Plains was dominated by native prairie grassland communities defined by this rainfall gradient. In the west, short-grass prairie occurs in the rain shadow of the Rocky Mountains, while mixed-grass prairie occurs in the central Great Plains and tall-grass prairie occurs in the wetter eastern region. Deep, black soils (Mollisols) are dominant in the Great Plains and contain high organic matter in the more humid eastern region. Given that these soils are highly productive, the prairie grasslands are among the largest farming and ranching areas in the world. With agricultural conversion, the short-, mixed- and tall-grass prairies now correspond to the western rangelands, the wheat (Triticum spp.) belt and the corn/soybean (Glycine max) regions, respectively (Commission for Environmental Cooperation et al. 2008; Encyclopædia Britannica Online 2013). The Great Plains region has been divided into three EPA level II ecoregions, which are each subdivided into four to eight Level III ecoregions (Omernik 1987, 1995) (Fig. 1). Level II ecoregions include Temperate Prairies to the north, West-Central Semi-Arid Prairies to the west, and South Central Prairies and Southern Texas Plains (South Central Semi-Arid Prairies, from here forward) to the south.
Each level II ecoregion contains several distinct freshwater wetland ecosystems. In the Temperate Prairies, the prairie pothole system, depressional wetlands formed by glaciers during the Pleistocene era, dominates the Northern Glaciated Plains and Lake Agassiz Plain level III ecoregions, which also contain some peatlands. Also in the Temperate Prairies, the eastern floodplain forests are common wetland types in the Central Irregular Plains and Western Cornbelt Plains level III ecoregions. In the West-Central Semi-Arid Prairies, the Nebraska Sand Hills level III ecoregion contains abundant depressional wetlands, while the Montana Valley and Foothill Prairie, Northwestern Glaciated Plains and Northwestern Great Plains level III ecoregions all support western riparian woodland and shrubland. The South Central Semi-Arid Prairies contain western riparian woodland in the Central Great Plains and Central Oklahoma/Texas Plains level III ecoregions, while playas (seasonal depressional wetlands) are common in the Southwestern Tablelands and Western High Plains level III ecoregions (NatureServe 2011; LANDFIRE 2012). While these wetland types vary by vegetation communities, hydrology and geomorphology, they may also be described and mapped in carbon-relevant terms.
Wetland carbon map
In a synthesis of literature and soils databases, Bridgham and others (2006) estimated the carbon balance of North American wetlands. To make estimates of wetland carbon pools and flux, the Bridgham team considered three categories of wetlands based upon major ecological differences that drive carbon cycling: 1) Peatlands [40 centimeters (cm) or more of surface organic matter with and without permafrost]; 2) Freshwater mineral soil (FWMS) wetlands (less than 40 cm of surface organic matter); and 3) Estuarine wetlands of three types: those dominated by herbaceous vegetation (tidal marshes), mangroves, and unvegetated areas (mud flats).
While estimates of carbon pools and flux were attributed to the land area of each wetland type in the Bridgham study described above, the data were not spatially explicit. We developed a wetland map of the Great Plains ecoregion that differentiates these carbon-relevant wetland types (Fig. 2). Our wetland map was derived from the USGS assessment baseline landcover data, which was a slightly modified version of the 1992 NLCD (Vogelmann et al. 2001) at 250 meter (m) resolution, and contained two wetland classes — herbaceous wetlands and woody wetlands. We reclassified these wetlands into new classes of herbaceous peatlands, herbaceous FWMS, woody peatlands, and woody FWMS. Peatlands were identified by using the U.S. Department of Agriculture (USDA) Soil Survey Geographic Database (SSURGO) at 250-m resolution to generate a map of the sum of the component percentages for histosols, histic modifiers, and peat and muck (Soil Survey Staff 2009a). Wetlands pixels with the percentage of peatlands more than 50 % were labeled peatlands. A non-wetland peatland class was also identified for NLCD land covers that were not wetlands, but had peatland characteristics, likely representing areas that were historically wetlands, or are wetlands that have been misclassified as something else.
Scenarios of LULC change
By analyzing the FORE-SCE LULC change model outputs, we identified wetland pixels that were converted to cropland, hay/pasture, or other LULC classes at some time between 2010 and 2050 for SRES A2, A1B and B1. Change in wetland area and proportion of wetlands converted in the 40 year time period were calculated by level III ecoregion and wetland type.
Land attribute data
Wetland pixels converted to other LULC classes in the LULC change model for SRES A2, A1B and B1 were labeled with additional biophysical and economic attributes. The additional attributes included: level II ecoregion, level III ecoregion, year of conversion, LULC class after conversion, SSURGO SOC to 20 cm depth, SSURGO land capability class, live aboveground biomass, non-irrigated cropland cash rent value, and where available, EDCM modeled outputs for SOC to 20 cm and total ecosystem carbon.
The SSURGO database (Soil Survey Staff 2009a) was used to calculate ecologically active stocks of SOC near the surface (0 to 20 cm depths). Where the SSURGO data were incomplete, gaps were filled from the State Soil Geographic (STATSGO2) database (Soil Survey Staff 2009b). Some attribute errors in horizon depths, component percentages, or missing rock fragment information were corrected with reasonable estimates, to provide the best estimate of the SOC for a given land area. The estimate of SOC for a SSURGO map unit was based on only the wetland components. Soils with a high potential for being wetland or former wetland were selected as those having a soil order of Histosols or a histic modifier in another soil order, or a classification in-lieu of texture of peat or muck. This calculation of SOC stocks represents a good estimate for general conditions in the time during which the database was developed (the 1970s to the 2000s) but does not represent a specific point in time. Details of the SOC calculation from SSURGO data are given by Bliss (2003).
The SSURGO data on land capability class were used as an indicator of suitability for cropland. Class 1 land has few limitations for cropland, and classes 2 through 4 have moderate to very severe limitations for cropland (e.g., erosion hazard), but even class 4 land may actually be used as cropland. Class 5 has limitations that are difficult to remove (e.g., wetness). Classes 6 to 8 are not suitable for cropland.
The SSURGO data are originally in vector format, and were rasterized to 30 m resolution, and resampled to the 250 m resolution used in this study. Many of the map units are much larger than the 250 m resolution pixels, and wetlands are sometimes a minor component of the landscape. Therefore, although the calculations using the pixel size and the sum of component percentages representing wetlands allow accurate estimates of potential wetland areas, there may be cases in which a particular pixel in the land cover database or FORE-SCE modeling that represents a wetland will not coincide with a SSURGO representation of a potential wetland.
Aboveground live biomass carbon
Aboveground live biomass carbon was modeled for the USGS assessment from USDA Forest Service, Forest Inventory and Analysis (FIA) biomass data (Blackard et al. 2008) and NLCD land cover data (Zhu et al. 2011). These data were used to represent standing woody biomass C stocks in woody wetlands.
EDCM SOC and total ecosystem carbon outputs
We used the EDCM top 20 cm SOC output variable to track change in wetland SOC after conversion to cropland or hay/pasture. We used the EDCM total ecosystem carbon (TEC) output variable to track change in woody biomass C plus SOC after conversion to cropland or hay/pasture (Liu et al. 2003, 2012b; Zhu et al. 2011). EDCM adopts a structure of multiple soil layers to account for SOC dynamics in the whole soil profile and to track the impact of soil erosion and deposition. For the USGS assessment, EDCM was run with a 10 × 10 systematic subsample factor to ensure adequate time for processing, generating statistics, and calibrating the estimates. Consequently results are based on a systematic sample of 1 % of the total pixels (Liu et al. 2012a).
USDA cash rents for non-irrigated agriculture
This variable represents an indicator of agricultural land value that is collected in a consistent format across the Great Plains region. The USDA Cash Rents Survey provides the basis for county estimates of the current year’s cash rent paid for irrigated cropland, non-irrigated cropland, and pasture (USDA National Agricultural Statistics Service 2012b; USDA National Agricultural Statistics Service 2012c). We used the non-irrigated cropland data for this study because it was a more extensive dataset and available more consistently at the county level than the irrigated cropland dataset.
Carbon stocks impacted
We explored geographical patterns of wetland carbon stocks with the potential for conversion by calculating average SOC and woody biomass C for wetlands converted in the A1B scenario. Average carbon stocks were calculated by wetland type, level III ecoregion, non-irrigated land rent value, and land capability class. We applied this analysis to wetlands converted in the A1B scenario, as this provided the largest dataset of wetlands across a greater extent compared to the other scenarios.
Scenarios of avoided wetland loss
We conducted a sensitivity analysis to understand how the area and distribution of land set aside for wetlands can influence the effectiveness of climate change mitigation opportunities. Knowing that land area reserved for carbon sequestration and enhancement is limited by competing demands for agriculture, development, and other uses, we developed and analyzed multiple avoided loss scenarios that assigned incremental areal and spatial allocation of land for wetland carbon avoided loss.
To conduct the sensitivity analysis, we generated two sets of avoided loss scenarios that were superimposed on the SRES LULC change scenarios A1B and A2. In each case of the sensitivity analysis we assumed that a certain proportion and criteria of wetlands converted in the SRES LULC change scenarios is prevented from conversion in the avoided loss scenario. The two scenario sets included the carbon scenarios based on SOC and woody biomass C values, and the economic scenarios based on land rent and land suitability values. In the B1, environmental scenario, relatively few wetlands were lost, and so we did not consider avoided loss mitigation in this case. The avoided loss scenarios apply to all wetlands converted in the Great Plains LULC change scenarios A1B and A2 from years 2010 to 2050. Avoided loss of SOC and woody biomass C were calculated for each scenario. Scenarios are described below.
1) Wetlands with the highest SOC stock: avoidance of a) 10 % and b) 25 % of all wetlands with the highest SOC in the top 20 cm; 2) Wetlands with the lowest SOC stock: avoidance of a) 10 % and b) 25 % of all wetlands with the lowest SOC in the top 20 cm; 3) Wetlands with the highest woody biomass C stock: avoidance of a) 10 % and b) 25 % of all wetlands with the highest woody biomass C; 4) Wetlands with the lowest woody biomass C stock: avoidance of a) 10 % and b) 25 % of all wetlands with the lowest woody C.
1) Wetlands with the highest cropland rent value: avoidance of a) 10 % and b) 25 % of all wetlands with the largest non-irrigated land rent value and a land capability class of 1 or 2; 2) Wetlands with the lowest cropland rent value: avoidance of a) 10 % and b) 25 % of all wetlands with the lowest non-irrigated land rent value and a land capability class greater than 2.
Wetlands in the economic scenarios were divided based on this land capability reclassification to generate subgroups with relatively equal sample numbers. Using the selected wetlands in each scenario, the average and total sum of wetland SOC and woody biomass C were calculated for each level III ecoregion.
Case studies of potential carbon loss
As a case study, we modeled potential carbon loss from wetland conversion for herbaceous FWMS wetlands and woody FWMS wetlands in the Temperate Prairies only because these groups were subject to the most conversion in the Great Plains. Given that one percent of pixels were included in the EDCM model run, a subset of wetland pixels were characterized with the modeled SOC and TEC outputs. To track changes over time in carbon stocks after conversion, we identified wetland pixels with EDCM outputs that were converted at some point between 2010 and 2020 and remained in the converted land use until 2050 by the LULC change model. For these pixels we calculated the mean and variance of the differences in SOC and TEC (for woody wetland pixels) over 30 years and used them for the analysis.