Continued warming could transform Greater Yellowstone fire regimes by mid-21st century

Our combined statistical models predicted fire and characterized extreme fire years in the Northern Rockies more effectively than has been done previously (Fig. 1 and SI Text). Temperature, precipitation, and moisture deficit were all important for predicting fire occurrence and number; given that a fire occurred, cumulative water year moisture deficit was key for estimating fire size distributions (SI Text). The models explained 83% of interannual variation in large fire occurrence and 73% of the variation in burned area for the calibration dataset. For the model estimation period in the Northern Rockies, 33% of fires and 66% of burned area occurred during the two largest fire years (1988 and 1994); 30% of fires and 45% of burned area in our simulation also occurred during those two years (SI Text and Fig. S3). High fire activity always occurred during warmer-than-average years, and this pattern was clear within the GYE (Fig. S4). The relationship between temperature and fire was nonlinear, and only ∼0.5 °C above the 1961–1990 average spring and summer temperature (March–August) distinguished large-fire years from most other years in the GYE (Fig. S4). Warmer-than-average temperatures were a necessary but not sufficient condition for predicting extreme fire years; in addition to temperature, moisture deficit and summer precipitation (July–August) also influenced fire during warm years (Fig. S6). Our methods captured broad-scale topographic variability in climate (e.g., cooler temperatures at higher elevations), which is necessary for regional fire projections. The probabilistic models enabled robust estimates of fire response to changing climate, including extreme fire years, by running large numbers of simulations (SI Text).

Fire simulations for three downscaled GCM scenarios for the recent historical period (1950–1990) in the GYE produced realistic interannual variation in area burned and no directional trends (Fig. 2). Predicted area burned was <10,000 ha (<1.25% of the GYE) in most years and agreed well with observed variability; however, no simulated fire years were as extreme as 1988 before the first decade of the 21st Century. This result is not unexpected; 1988-scale events in the GYE are necessarily very rare (otherwise historical fire rotations would be less than the observed >120 y). In addition, our hydrologic simulations cannot incorporate extremes in day-to-day variability in wind speed because downscaled wind fields are not yet available. Because extreme dry conditions in the recent historical record resulted from climate (high temperature and low precipitation) and weather (high winds), the simulated dry extremes in moisture deficit using climatological winds are likely underestimated.

Post-1990 simulations projected substantial increases in GYE fire (Fig. 2). Notably, by 2020 projected increases in burned area over a 1951–1990 reference period were detectable with 95% confidence over 100% of simulations using one-sided Mann–Whitney tests. Fire seasons comparable to 1988 became more frequent, with between one (for NCAR) and five (for GFDL) such events occurring between 2011 and 2050. After 2050, all models predicted that annual area burned would exceed 100,000 ha during most years. Years with no large fires—common in the historical record—became extremely rare by 2050 and did not occur in 1,000 simulations after 2050 for either the CNRM or GFDL scenarios and occurred only three times in the 1,000 NCAR simulations. This increase in average annual “background” area burned in years without 1988-scale events is also apparent in observed fires for the last two decades (Fig. S5). By about 2075, potential annual area burned regularly exceeds the signature 1988 event, often by a large margin. By then, forest ecosystems in the GYE, and the fuels and fire responses they support, would likely have been transformed by changing climate and disturbance and moved outside the bounds of the modeled climate–fire relationship.

Fire rotations (FRs), which exceed 120 y throughout most of the GYE before 1990, are in transition to shorter periods across all scenarios by the first decades of the 21st century (Fig. 3). Downscaled climate predictions from all three GCMs were qualitatively similar but differed somewhat with respect to midcentury timing of fire-regime shifts. For 2005–2034, FRs begin to fall below 60 y in portions of the landscape for all scenarios, although the rate of change varies; differences in precipitation among scenarios have greater influence on fire early on, whereas effects of high temperatures dominate late in the century. Spatial patterns are similar across scenarios, with FRs decreasing sooner and to a greater extent in the northwestern areas of the GYE. By midcentury, FRs are <20 y throughout the majority of the GYE, except for small portions to the southeast (Fig. 3). By end of century, climatic conditions associated with large fires are the norm (Fig. S2), and FRs <10 y are projected across the GYE for all three GCMs (Fig. 3); however, by the time such short fire rotations could be driven by climate, fuel limitations would be expected to constrain fire occurrence and area burned.

Our findings indicate that GYE forests are rapidly approaching a threshold beyond which fire occurrence and extent are likely to change the ecosystem qualitatively. Recent history shows that a shift in spring and summer temperature (March–August) of just over 0.5 °C above the 1961–1990 average distinguished extreme fire years from most others in the Northern Rockies, and increases in average spring and summer temperatures predicted by 2099 are ∼4.5–5.5 °C for the GCM scenarios explored here (Fig. 2 and Fig. S2). The projected changes in temperatures and fire are not consistent with persistence of the suite of conifer species that have dominated the Yellowstone landscape throughout the Holocene. Rather, the projected climate–fire regime is consistent with lower montane woodland or nonforest vegetation and implies a shift from a climate-limited to a fuel-limited system (24) by midcentury. Climate conditions conducive to large fires in the current system would occur in most years, but less biomass would be available to burn because recovery times between successive fires would be shorter. Thus, we might expect novel fire–climate–vegetation relationships, suggesting that the GYE is approaching a tipping point (i.e., critical change in temperature and moisture deficit) that could be exceeded by mid-21st century.

Our results are consistent with predictions for increased temperature-driven fire activity in other regions (20, 21, 23, 42, 43) but also suggest that future climate–fire dynamics may differ from those observed in the past. Numerous studies demonstrate that climate and fire have varied substantially in the GYE throughout the past 10,000 y. During the early Holocene, increased summer temperature was associated with increased fire frequency on the central plateau in Yellowstone National Park, and the fire return interval declined to 75–100 y for several millennia (44). During the past 1,300 y, large fires on the central plateau occurred in decades including the driest years, and widespread burning (>10,000 ha) occurred at 150- to 300-y intervals (45). During the past 2,650 y in forest–steppe vegetation of northern Yellowstone, fire intervals averaged 75 y, but fire-free intervals of several centuries were detected twice (35). Thus, climate and fire regimes have varied considerably in the past. Nonetheless, our models indicate that, by the middle of the 21st century, forests of the GYE may regularly experience climatic conditions that exceed the most extreme years in the instrumental and paleoecological record. Our models also predicted geographic differences in the timing of predicted transitions in fire regimes, notably an earlier onset of shorter fire rotations in the northwestern portions of the region. Other studies have noted spatial variation in precipitation and fire regimes in the GYE (e.g., refs. 34 and 46–49), and geographic differences in fire regimes are likely to be observed in the future.

Uncertainty in predicted climate–fire relationships increases as climate conditions move outside the historical range of variability, and predictions for the latter 21st century must be interpreted with care. Our fire models will become increasingly poor guides as current and future climates—and the fuel conditions that support current climate–fire relationships—diverge. This is true to some extent for any statistical model given the lack of recent historical analogs and the currently highly nonlinear climate–fire relationship in the study area, and predicted fire activity (especially beyond 2050) must be interpreted in the context of key assumptions. First, the statistical models of fire occurrence and area burned are based primarily on fires in conifer forests (which dominate the vegetated area) given present-day climate and vegetation; predicted future climate diverges substantially from that used for model development. Second, the nature of feedbacks from vegetation to future fire activity is a critical unknown. Area burned cannot increase indefinitely because changing fuels will eventually limit fire occurrence (50). This feedback is not included in our models but will become an increasingly important influence on fire in the latter half of the century. Following severe fire, subsequent fire activity is reduced for some time because fuels are sparse. Younger forests will burn readily in extreme fire weather, as observed during the 1988 Yellowstone fires when regenerating saplings in Pinus contorta stands <10 y old carried crown fire (4, 51). However, fuel quantity will diminish as the time for vegetation recovery between fires declines. Third, our study used average conditions within 12-km × 12-km grid cells and cannot predict variation in fire dynamics (e.g., effects of local topography or subgrid cell FRs) at finer spatial scales. Finally, our study did not consider variation in fire severity, and we have not considered potential interactions with other disturbances, such as outbreaks of native bark beetles (Dendroctonae) (52, 53).

Anticipating vegetation shifts in the GYE under altered climate–fire regimes is complex and beyond the scope of this study. However, fire can catalyze rapid shifts in plant communities when climate conditions change sufficiently to favor postfire establishment of different species (e.g., ref. 54), and such dynamics have already been observed elsewhere (55–57). Warming temperatures alone are expected to cause declines in suitable habitat for some conifers now common in the GYE, even if precipitation also increases (58–60). In the GYE, projected changes in temperature reach the historical differences in temperature between subalpine forest (with an historical fire rotation >100 y) and lower montane forest (with an historical fire rotation <30 y) by the end of the century. Following fire, tree species that currently occupy lower elevations in the GYE (e.g., Pseudotsuga menziesii, Populus tremuloides) could potentially expand upslope on suitable substrate, as documented for andesite substrates in the GYE during the early Holocene (61). Although the paleoecological record has revealed remarkable resilience of GYE vegetation to a wide range of climate and fire regimes (e.g., refs. 44 and 61), our study suggests that future changes in climate and fire rotations could preclude conifer regeneration and convert currently forested areas to woodland or nonforest. Future research should prioritize understanding potential postfire successional trajectories.

Continued warming could completely transform GYE fire regimes by the mid-21st century, with profound consequences for many species and for ecosystem services including aesthetics, hydrology, and carbon storage. The conditions associated with extreme fire seasons are expected to become much more frequent, with fire occurrence and area burned exceeding that observed in the historical record or reconstructed from paleoproxy records for the past 10,000 y. Even in years without extreme fire events, average annual area burned is projected to increase, and years with no large fires—common until recently—are projected to become increasingly rare. The timing and spatial location of such changes varied somewhat among the three GCMs used in this study, but the models converged by the latter part of the century. The magnitude of predicted increases in fire occurrence and area burned suggests that there is a real likelihood of Yellowstone's forests being converted to nonforest vegetation during the mid-21st century because reduced fire intervals would likely preclude postfire tree regeneration. A change in dominant vegetation would also cause the GYE to shift from a climate- to a fuel-limited fire regime (24). We suggest that the climate–fire system is a tipping element that may qualitatively change the flora, fauna, and ecosystem processes in this landscape and could be indicative of similar changes in other subalpine or boreal forests.

We thank A. Keyser, K. Lubetkin, T. Das, F. Munoz-Arriola, and D. Lettenmaier for assistance with data processing, visualization, hydrologic simulations, and historical gridded climate data. We acknowledge the Program for Climate Model Diagnosis and Intercomparison and the World Climate Research Program's Working Group on Coupled Modeling for their roles in making available the World Climate Research Program Coupled Model Intercomparison Project phase 3 multimodel dataset, supported by the US Department of Energy. We thank C. D. Allen, F. S. Chapin, D. C. Donato, C. Whitlock, and M. D. Dettinger for comments that improved this manuscript. This research was funded by Joint Fire Science Program Project 09-3-01-47, US Forest Service Southern Research Station Joint Venture Agreement 07-CA-11330143-129, and National Oceanic and Atmospheric Administration Grant NA10OAR4310217.