The importance of ecosystems in storing and sequestering carbon is increasingly recognized given the threat of climate change and the rapid human-induced rise in atmospheric CO₂ concentrations (Portela et al. 2008). Because these processes are influenced by different ecological processes in different regions, local carbon models have been developed in ARIES for Madagascar as well as three ecologically distinct areas in the United States: Southeast Arizona and Northern Sonora’s San Pedro River Watershed, Western Washington State, and the Rocky Mountains of Colorado and Wyoming. In the ARIES system, the areas of carbon sequestered in vegetation and soils are designated as sources of the ecosystem service; the areas of potential stored carbon release, due to fire, land use change, deforestation, or other vegetation and soil disturbances, are sinks. By subtracting the stored carbon release (estimated based on historical or modeled land-use change or fire) from carbon sequestration in a region of interest, we can compute the carbon available to offset anthropogenic emissions. Greenhouse gas emitters can be conceptualized as the beneficiaries of carbon sequestration and storage. By mapping levels of carbon sequestration, stored carbon release, and anthropogenic emissions in a common unit (tonnes C/yr), we can fully describe regional carbon balances – the level of a region’s net release or uptake of atmospheric CO₂. Please see the ARIES modeling guide for full documentation and references for these models.

Carbon source: The term “carbon sink” is used in the scientific and popular media to describe locations that sequester or store carbon (e.g., forests, soils, grasslands, the ocean, or underground geologic formations). This definition aligns with the ARIES carbon “source” models, since sources are areas where an ecosystem service carrier, in this case, carbon absorption capacity, is generated. Ecosystems provide carbon sequestration capacity, which can be used by greenhouse gas emitters to “offset” their emissions.

Carbon sink: Carbon sinks in the ARIES framework are areas of potential stored carbon release, due to fire, land use change, deforestation, or other vegetation and soil disturbances. These sinks leave less carbon sequestration capacity available for offsetting anthropogenic emissions. By overlaying historical or projected land-use change or fire data, we can estimate actual loss of carbon storage.

Carbon beneficiaries: In ARIES, greenhouse gas emitters are defined as the users of carbon sequestration and storage. By releasing CO₂ into the atmosphere, emitters rely on ecosystems to absorb and store carbon in order to avoid even larger rises in atmospheric CO₂ than are currently seen.

Carbon flow: Since carbon dioxide is a well-mixed atmospheric gas, the benefits of carbon sequestration and storage can be enjoyed by any human beneficiary on Earth, regardless of location. As such, no flow model is necessary for carbon sequestration and storage. However, for a given region, it is possible to calculate the differential between carbon uptake by ecosystems (sequestration minus release of stored carbon) and anthropogenic carbon release. This information can show whether that region has a negative or positive carbon balance, i.e., whether its emissions are greater or less than the amount of carbon sequestered.

Model structure and assumptions

Case studies 1. Western Washington State 2. San Pedro River, Arizona and Sonora 3. Madagascar 4. Rocky Mountains

Carbon sequestration source models. Although carbon sequestration data are available globally at 1 km² resolution, we developed simple Bayesian network models that include the influences on carbon sequestration (e.g., vegetation, soils, climate). Existing datasets can be used in ARIES to provide mean values to use in training finer-grained models, allowing estimation of carbon sequestration changes in scenarios or for up-scaled modeling of carbon sequestration where higher resolution input data are available. Based on the literature (e.g., Iverson et al. 1994, Eade and Moran 1996, Gaston et al. 1998, Chan et al. 2006, Naidoo and Ricketts 2006, Egoh et al. 2008, Wundscher et al. 2008, Auch 2010, Wendland et al. 2010, Tallis et al. 2013) and discussions with regional experts, we set carbon sequestration as a function of vegetation density and sequestration rate, two intermediate variables created to keep conditional probability tables tractable (Marcot et al. 2006). We set sequestration rate as a function of soil C:N ratio and the difference between mean summer high and winter low (in Madagascar and Western Washington), and as a function of land cover, vegetation type, and actual evapotranspiration (in Orange County). We set vegetation density as a function of hardwood:softwood ratio, percent tree canopy cover, and successional stage (in Western Washington), and percent tree canopy cover and forest degradation status (in Madagascar). For the San Pedro, Orange County, Rocky Mountain, and Vermont agricultural carbon models, we used a collapsed number of variables, removing the intermediate nodes for vegetation density and sequestration rate. For the San Pedro model, we estimate sequestration as a function of vegetation type, percent tree canopy cover, and mean annual precipitation. We added a node for bark beetle kill in the Rocky Mountain model, where outbreaks of pine and spruce beetles have killed large numbers of trees, lowering sequestration in forests with a substantial number of dead trees (Coops and Wulder 2010). For the Orange County model, we used the above noted variables as input nodes to sequestration rate, then combined sequestration rate with percent tree canopy cover to estimate annual vegetation and soil carbon sequestration. Actual evapotranspiration (AET) has been found to have a strong relationship with primary productivity, and therefore carbon sequestration (Lieth and Box 1972, Elegene et al. 1989, Metherell et al. 1993). This is especially true in water-limited regions such as semi-arid biomes, as with the Orange County case study (Claudio et al. 2006, Fuentes et al., 2006). Vegetation type can help to predict the quantity of vegetation sequestration and storage capacities from expected biomass for certain plant species (Kirby and Potvin 2007). In the Vermont model, we estimated sequestration as a function of vegetation carbon storage (itself a function of mean annual precipitation, vegetation type, and the difference between mean summer high and winter low) and soil C:N ratio (Liu et al. 2010). We used Jenks natural breaks to discretize summer high-winter low, soil C:N ratio, and actual evapotranspiration. We used equal intervals to discretize vegetation and soil carbon sequestration, hardwood:softwood ratio, and percent tree canopy cover. We based prior probabilities for the models on either the actual distribution of regional data (where we have these datasets), expert opinion (where consensus by experts was possible), or uninformed priors (where there was true uncertainty and a lack of consensus by experts). We filled out conditional probability tables by setting extremes set at both ends (i.e., “pegging the corners,” Marcot et al. 2006) and interpolating intermediate values. Where possible we used expert opinion about which variables are most influential, and which should have the greatest influence on the contingent probability tables, and what the general level of uncertainty was for that system (i.e., how wide to set the distribution of values across discrete states). All else being equal, we set vegetation density at its highest values at greater percent tree canopy cover, later successional stages, more softwoods, and no forest degradation (where applicable). We set sequestration rate with its highest values at higher C:N ratios, higher actual evapotranspiration, lower differences between mean summer high and winter low temperatures, and land cover and vegetation types with greater biomass (where applicable). We set sequestration to its greatest values at high levels of vegetation density and sequestration rate.

Potential stored carbon release sink models. Stored carbon release can be calculated probabilistically or deterministically, though deterministic calculations are likely more accurate assuming historical or forecasted land-use and fire data are available. Deterministic estimates of carbon loss due to land-use change can be calculated by modeling carbon storage under pre- and post-change conditions. Carbon loss due to fire can be estimated, for example, by overlaying fire boundary polgyons (e.g., GeoMAC 2013) with fuel consumption coefficients (Spracklen et al. 2009) and carbon pool data (Smith et al. 2006), or by applying more advanced models (e.g., Lutes 2013). This method was demonstrated for the U.S. Pacific Northwest by Bagstad et al. (in press). We have also estimated stored carbon release probabilistically, as a function of vegetation and soil carbon storage (the sum of vegetation carbon storage and soil carbon storage) and the risk of deforestation and/or fire, with greater stored carbon release at higher risk and carbon storage levels. Soil carbon storage is influenced by slope, soil pH, soil oxygen conditions (i.e., greater storage in wetlands where anaerobic conditions inhibit respiration), vegetation density (an intermediate variable incorporating tree canopy cover and degradation status in Madagascar, tree canopy cover, and vegetation type in the San Pedro, and successional stage, tree canopy cover, and hardwood:softwood ratio in Western Washington, noted as important determinants of carbon sequestration in the Pacific Northwest by Nelson et al. 2008), and soil carbon:nitrogen ratio. A simpler model was applied in the Rocky Mountains, incorporating only annual precipitation (Derner and Schuman 2007) and soil order (Buringh 1984). The importance of these variables in influencing soil carbon dynamics has been noted by previous authors. We set vegetation carbon storage as a function of the difference between mean summer high and winter low temperature (Auch 2010) and vegetation density, with population density added as an influence in Madagascar. For the San Pedro, we set vegetation carbon storage as a function of mean annual precipitation and vegetation density, and for the Rocky Mountains we set it as a function of vegetation type and tree canopy cover (Smith et al. 2006). For the Orange County model, deforestation was not considered as an influence on stored carbon release (though it would be included in non-urban areas within the same biome), slope was dropped as an influence on soil carbon storage (since slope/aspect influence AET and other water balance measurements in chaparral and scrub ecosystems, Miller 1947, Parsons 1973, Ng and Miller 1980), and actual evapotranspiration and percent tree canopy cover were added as influences on soil carbon storage. We set vegetation carbon storage as a function of land cover, vegetation type, percent tree canopy cover, and AET for the Orange County model. The Vermont model used soil tillage and biomass removal rate as influences on agricultural stored carbon release (Gollany et al. 2010, Gonzalez-Chavez et al. 2010). This model considered soil C:N ratio, biomass residue input (Hai et al. 2010), and vegetation type as influences on soil carbon storage and vegetation type, mean annual precipitation, and the difference between mean summer high and winter low temperature. Iverson et al. (1994) and Gaston et al. (1998) provide discretization of continuous variables for slope and population density. Bosworth and Tricou (1999) and Darby et al. (2009) provide discretization for vegetation carbon storage in the Vermont carbon model. We used Jenks natural breaks to discretize soil carbon storage, summer high-winter low, vegetation and soil carbon storage, soil C:N ratio, vegetation carbon storage, fire frequency, and actual evapotranspiration. We used equal intervals to discretize hardwood:softwood ratio and percent tree canopy cover. All else being equal, we set soil carbon storage at its highest values at low or high pH, high C:N ratio, level slopes, greater vegetation density and annual precipitation, and on anoxic (i.e., wetland) soils, and vice versa. We set vegetation carbon storage at its greatest values with low differences between mean summer high and winter low temperature, high vegetation density or tree canopy cover, and low population density (in Madagascar). We set stored carbon release at its highest with greater vegetation and soil carbon storage and greater deforestation and fire risk. The outputs of the carbon sink model are either the potential stored carbon release (probabilistic calculation of stored carbon release) or vegetation and soil carbon storage (deterministic calculation of stored carbon release).

Greenhouse gas emissions use models. The beneficiaries of carbon sequestration and storage are greenhouse gas emitters who release CO₂ into the atmosphere. Spatially explicit data on greenhouse gas emissions exist for the United States. Globally, we use population density data multiplied by per capita emissions for the country or sub-national region of interest.

Carbon flow models. Since carbon dioxide is relatively quickly mixed in the atmosphere, the benefits of carbon sequestration and storage can be enjoyed by any human beneficiary on Earth, regardless of location. As such, no flow model is necessary for carbon sequestration and storage. However, for a given region, we can calculate the differential between carbon uptake by ecosystems (sequestration minus release of stored carbon) and anthropogenic carbon release. This information can be used in a flow model to show whether that region has a negative or positive carbon balance, i.e., whether its emissions are greater or less than the amount of carbon sequestered. Key outputs from the flow models include: 1. Carbon mitigation surplus: Calculated when local sequestration exceeds emissions plus atmospheric carbon sources. 2. Carbon mitigation deficit: Calculated when local emissions exceed net carbon uptake (sequestration minus stored carbon release).

References