Methods Our approach combined a dynamic ecosystem model to project future C stocks under different climate scenarios and fire regimes. To identify how great a change in climate and fire regime would be required to shift vegetation from C source to C sink, we ran the ecosystem model CENTURY version 4.5 (Parton et al. 1987, Smithwick et al. 2009b) aspatially for the dominant vegetation communities in the GYE given a large fire event in 1988, and a range of estimated fire-return intervals and current and future climate conditions. Based on our previous work (Kashian et al. 2006, Kashian et al., in prep), we
identified general patterns of fire regime and forest regeneration pathways across the region. Our goal was to focus on critical drivers that would be likely to result in observable and representative change across the landscape. We concluded that changes in C stocks would be most significant for transitions of forest to non-forest (rather than forest to forest only). Other studies have shown substantial differences in C stocks with stand age up to about 100 years, but less difference among conifer forest types (Bradford et al. 2008). Thus, our current modeling was focused on lodgepole pine, a representative forest type in the region. The model has to-date been additionally parameterized for warm-dry conifer (primarily Douglas fir (Pseudotsuga menziesii) forests in the GYE) and grasslands in the Lamar Valley; as validation data of C stocks in this ecosystem (Donato et al., in prep) become available, we will incorporate these vegetation types into our approach. However, to capture variation in recovery in lodgepole pine, we modeled two recovery pathways: fast (high pre-fire serotiny, more prevalent at elevations < 2400 m) and slow (low pre-fire serotiny, characteristic of elevations > 2400 m; Schoennagel et al. 2003). We expect that the slow recovery pathway will be representative of other vegetation types that lack serotinous cones and are likely to regenerate more slowly, for example, Douglas-fir or spruce-fir forests. All fires were prescribed to be highseverity, stand-replacing events. To estimate current and future climate conditions, we used historical climate data and general circulation model (GCM) runs downscaled to the North American Land Data Assimilation system 1/8-degree latitude/longitude grid (12 x 12 km resolution). We used three AR4 GCMs (CCSM 3.0, CNRM CM 3.0, and GFDL CM 2.1) forced with the Intergovernmental Panel on Climate Change’s (IPCC) Third Assessment Report: Special Report on Emissions Scenarios (SRES) A2 emissions pathway to generate a set of plausible climate futures for the western USA. The three GCMs used here are among a larger group that were assessed to adequately represent important aspects of western North American climate, including seasonality of temperature and precipitation and multiyear variability in sea surface temperatures (Daniel Cayan et al., unpublished). This particular subset of models was chosen because daily values for important variables such as temperature and precipitation were available for each GCM run, and these were required to force the hydrologic simulations used. The A2 emissions scenarios have been a frequent focus for impact assessment work because they were thought to represent a plausible high-end emissions scenario. However, for much of the past decade, emissions
and atmospheric concentrations of greenhouse gases have exceeded the range of commonly used IPCC emissions scenarios, especially SRES A2. Consequently, given current and past emissions, the long lead times necessary to reduce future emissions, and the long atmospheric residence times of many greenhouse gases, climate projections using the SRES A2 CO2 trajectory can no longer be considered a plausible representation of the future, nor representative of a “high” emissions scenario, but were used here given their availability. Because current atmospheric concentrations exceed those represented in the SRES A2 scenarios, the climate scenarios used to derive our results can be considered conservative. GCM temperature and precipitation fields were downscaled using the constructed analogs method with bias correction (Maurer and Hidalgo 2008). Gridded historical climate data (temperature, precipitation, radiation, and wind speed) were obtained from Dr. Lettenmaier at the University of Washington and Dr. Maurer at the University of Santa Clara (Hamlet and Lettenmaier 2005). For the simulations of lodgepole pine forest, we used climate data from the grid centered on the Yellowstone Lake climate station, which is centrally located in the GYE and surrounded by lodgepole pine forest. Productivity, mortality, and post-fire recovery were parameterized in CENTURY for lodgepole pine and warm-dry conifer trees based on empirical data (Tinker and Knight 2000, Pearson et al. 1987, Ryan and Waring 1992, Stump and Binkley 1993, Smithwick et al. 2009a, Kashian et al., in prep) and previous modeling efforts (Kashian et al. 2006, Smithwick et al. 2009b). The model was run in “savanna” mode, allowing for grass and tree competition for water and nutrients. For all simulations, we assumed a C3 grass parameterization available in CENTURY. Grass represented a small proportion of C stocks in mature stands but was a large and transient component of total C stocks for several years following fire. These large, transient pulses of post-fire grass were likely overestimated and future modeling efforts will be increasingly focused on grass dynamics in early post-fire years. The fire-return intervals used in the CENTURY and landscape C modeling are based on understanding of the canopy seed bank and its influence on postfire regeneration. Specifically, Turner et al. (2007) demonstrated that lodgepole pine saplings are producing cones (including a few serotinous cones) by 15 years of age. Cone production begins at about the same age or even later in other conifer species of the GYE, and recent fires
that have burned young conifer forests (< 30 years) show minimal tree regeneration (Romme and Turner, personal observations). To encapsulate this rapid but variable trend in development of a canopy seed bank, we used a 30-year fire interval as a conservative estimate of the minimum FRI that would be followed by a very high likelihood of reforestation. If fire recurs at < 60 year intervals, seeds are present but in moderate quantities. By stand age of > 90 years, lodgepole pine trees are generally producing substantial numbers of cones. Although stands are likely to regenerate at different rates following a stand-replacing fire due to patterns in fire severity and pre-fire levels of serotiny (Turner et al. 2004), cone production is not limiting by age 90 years, and even initially sparse stands experience infilling (personal observations; Kashian et al. 2005, 2006). Empirical work along chronosequences of > 77 stands in the GYE (largely in Yellowstone National Park) indicated that most differences in N and C stocks occur at stand ages < 100 years (Smithwick et al. 2009a, Kashian et al., in prep). Using these parameterizations for vegetation, climate, and fire, we performed a model experiment using a 4 x 4 x 2 factorial design in which we considered four climate scenarios (historical plus the three GCMs), four fire-event pathways (no fires after 1988, a fire 90 years after the 1988 event, a fire 60 years after the 1988 event, and fires every 30 years after the 1988 event), and two recovery pathways (fast or slow). Model output included live and dead pools (large wood, branches, leaves, coarse roots, fine roots), as well as active, slow, and passive pools in surface and soil, and relevant ecosystem processes such as respiration and decomposition. The time needed for forest C recovery following fire under both current and future climate scenarios was determined by comparing the time to recovery of pre-1988 C stocks (average of 1950–1987) of mature forest stands to that of both future periods (1970–2099) or averaged across the post-1988 simulation period (1989–2100). Total ecosystem C stocks varied little (< 10 percent) among future climate scenarios for a given fire-event pathway and were therefore averaged for the purposes of demonstrating the large differences in C stocks forecast among fire scenarios. Similarly, fast versus slow regeneration had a much smaller effect on total ecosystem C stocks than the timing of individual fire events. For simplicity, only fast recovery pathways from the CENTURY model are shown here.