Simulating the Effects of Climate Change on Fire Regimes in Arctic Biomes: Implications for Caribou and Moose Habitat

Wildfire is the primary ecological driver of succession in the boreal forest and may become increasingly important within tundra ecosystems as the Arctic warms. Migratory barren-ground caribou (Rangifer tarandus granti) rely heavily on terricolous lichens to sustain them through the winter months. Lichens preferred by caribou can take 50 or more years to recover after being consumed by wildfires. We simulated effects of climate change on the fire regime within the winter range of one of the largest caribou herds in the world, the Western Arctic Herd, to assess how their forage may be impacted. We forecast that the total area burned (AB) in the near term (2008–2053) will be 0–30% greater than during our historic reference period (1950–2007) depending on the climate model (CGCM3.1 or ECHAM5) considered. Further into the future (i.e., 2054–2099), we forecast AB to increase 25–53% more than during our reference period. In contrast to the entire study area, which contains both tundra and boreal forest habitats, we forecast that the amount of AB in tundra alone will increase (0–61%) in the near term. Simulated high-quality caribou winter range, as indexed by tundra and spruce habitat that had not burned in 50 years, decreased modestly (,6%) in the near term over the entire study area. Simulated changes were more dramatic within the herd's core winter range, with declines in high-quality caribou winter range approaching 30%. Conversely, moose habitat was projected to increase by 19–64% within the core winter range in the near term. The simulated declines in the quantity of core winter range in the future due to larger and more frequent fires could impact caribou abundance through decreased nutritional performance and/or apparent competition with moose. These impacts would likely be detrimental to the subsistence users that rely on this resource. Additionally, changes in the fire regime and decreases in caribou abundance could amplify feedback mechanisms, such as decreasing albedo, by facilitating shrub growth that may hasten climate-driven changes to the composition and structure of vegetation communities in the low Arctic. Attribution License, which permits restricted use, distribution, and reproduction in any medium, provided the original author and sources are credited.


INTRODUCTION
Climate change impacts on the habitats of arctic land mammals are predicted to be severe (Lawler et al. 2009) and have already been implicated in the decline of caribou (Rangifer tarandus) populations across the Arctic (Vors and Boyce 2009).Mechanisms by which climate change may negatively impact caribou include deeper snows, increased incidence of icing (rainon-snow) events, increased human disturbance, increased insect harassment, changes to habitat quality and quantity, and increased frequency of wildfire on winter ranges (Jefferies et al. 1992, Walsh et al. 1992, Griffith et al. 2002, Putkonen and Roe 2003, Johnson et al. 2005, Rupp et al. 2006, Tews et al. 2007, Joly et al. 2011).However, the effects of climate change on caribou populations are likely not uniform due to inherent complexities of the climate system (e.g., Zalatan 2008, Joly et al. 2011).For example, increased summer temperatures and longer growing seasons may prove to be nutritionally beneficial to caribou during the summer months (Jefferies et al. 1992, Griffith et al. 2002, Tews et al. 2007).
Tundra covers over 5,000,000 km 2 of the Arctic (Walker et al. 2005).Additionally, vast areas of tundra are interspersed with boreal forest in the tundra-boreal forest interface that stretches around the globe north of the continuous boreal forest biome (Callaghan et al. 2002).The effects of climate change are already apparent (IPCC 2007), especially in the Arctic (Callaghan et al. 2004).Mean annual temperatures have increased by 2-38C in the region in recent decades, with larger increases apparent during the winter months (Stafford et al. 2000, Hinzman et al. 2005).Positive feedback mechanisms, including melting snow and sea ice, increasing shrubs, and greater prevalence of wildfires will likely amplify these changes via decreased albedo (Overpeck et al. 1997, Chapin et al. 2000, Qu and Hall 2006, Balshi et al. 2009, Euskirchen et al. 2009).
Climate warming in the Arctic could have several important consequences for the fire regime and caribou winter range.Increased area burned, in both boreal forest (Duffy et al. 2005, Balshi et al. 2009) and tundra (Joly et al. 2009b, Hu et al. 2010a), is associated with warmer and drier summers.Thus, it is predicted that fire will increase in both of these biomes under climate warming scenarios (Callaghan et al. 2004, Flannigan et al. 2000, Higuera et al. 2008, Balshi et al. 2009).Wildfire is the dominant driver of change to ecosystem structure and function in the boreal forest (Payette 1992, Chapin et al. 2006) but wildfires are less frequent and smaller in the tundra biome where lichens represent a large portion of the biomass and species diversity (Wein 1976, Holt et al. 2006).However, the Seward Peninsula and Noatak River valley regions of Alaska (Fig. 1) exhibit greater fire frequency than other areas within the tundra biome (Racine et al. 1985, Racine et al. 2004, Joly et al. 2009b, Hu et al. 2010a).While lichens are highly flammable, greater fire frequency in these regions is likely due to climate conditions rather than lichen abundance because the Noatak region has lower lichen abundance than the Seward Peninsula, and graminoids provide fine fuels that increase the potential for reburning in tundra much sooner than feather mosses of the boreal forests (Cronan and Jandt 2008, Jandt et al. 2008, Joly et al. 2009b).
Wildfires destroy terricolous fruticose lichens, a staple of the winter diet of migratory caribou (Klein 1982, Bernhardt et al. 2011).Lichens are slow to recover after being burned, with primary caribou forage species (e.g., Cladina rangiferina, C. stellaris) often taking several decades or more to return to pre-fire levels (Holt et al. 2008, Jandt et al. 2008, Klein and Shulski 2009, Joly et al. 2010, Collins et al. 2011).Caribou avoid burned areas in the boreal forest and tundra during the winter on a time scale coincident with lichen recovery rates (Joly et al. 2003, 2007a, 2010, Collins et al. 2011).Increases in fire frequency may limit the total available habitat for wintering caribou that is old enough to support high levels of preferred lichen species and tempers the argument that fire is not important to caribou because they can simply utilize alternative ranges (see review in Rupp et al. 2006).Further, Rupp et al.'s (2006) conclusions were likely conservative because they did not address the direct detrimental impacts climate change would have on lichens through warming and drying nor the indirect impacts on lichens by enhancing growing conditions for vascular plants such as shrubs and trees (see review by Joly et al. 2009a).
Warmer and longer growing seasons induced by climate change will likely degrade permafrost, deepen the active layer (i.e., the top layer of soil that thaws during summer) and provide vascular plants a competitive advantage over lichens (Chapin et al. 1995, Walker et al. 2006).Increases in shrub abundance have already been detected in the sub-Arctic and Arctic (Sturm et al. 2001, Joly et al. 2007c, Forbes et al. 2010).Shrubs may further inhibit re-generation of lichens by trapping snow, increasing leaf litter, competing for resources, and closing the canopy (Joly et al., 2009a(Joly et al., , C ˇabraji c et al. 2010)).
Burn severity is an important factor that determines the successional trajectory after a fire (Racine et al. 2004, Johnstone et al. 2010, Bernhardt et al. 2011).In more severe burns, stand self-replacement (e.g., birch [Betula spp.] forest returning to birch forest), which is common in Arctic ecosystems, is not automatic.An increase in shrubs, including fire-adapted dwarf birch (Betula spp.), relative to pre-fire conditions can follow fire, especially in moderate severity burns (de Groot and Wein 1999, Bret-Harte et al. 2001, Racine et al. 2004, Joly et al. 2010).This process could lead to an amplifying feedback where more shrubs lead to increasingly severe fires (Higuera et al. 2009, Hu et al. 2010a, Xue et al. 2010).Similarly, greater burn severity in the boreal forest could alter the black spruce (Picea mariana) self-replacement trajectory by consuming semi-serotinous seeds and reducing organic layers which results in an early successional deciduous phase (Johnstone et al. 2010).This would also be disadvantageous for overwintering caribou as they avoid these habitat types (Joly et al. 2010, Collins et al. 2011, Joly 2011).Likewise, additional shrub cover, such as Salix spp., could attract herbivores associated with boreal forests, such as moose (Alces alces; Bryant andReichardt 1992, Joly et al. 2010).
Unlike caribou in winter, moose select for areas with abundant early seral stage habitats (Weixelman et al. 1998, Maier et al. 2005).Thus, increased fire due to climate change may benefit this species.It has been hypothesized that more abundant moose may allow for increased wolf (Canis lupus) densities (i.e., a numeric response; James et al. 2004).Wolves are the main predator of caribou during winter months; thus, increases in moose densities associated with climate change-induced higher fire frequency may indirectly lead to increased caribou predation by wolves (Latham et al. 2011, Robinson et al. 2012).This would not impact neonatal survival of our study caribou herd, the Western Arctic Herd (WAH), as their calving ground is north of the Brooks Range and far (circa 150 km) from the existing tree line.
Our goals were to identify the impacts of climate change on the fire regime in tundradominated landscapes of northwest Alaska.Using our simulations, we projected the amount and quality of caribou winter range available under climate warming scenarios.Additionally, we quantified how the amount of preferred moose habitat changed because this may have additional indirect impacts on caribou.We hypothesized that the fire regime in northwest Alaska in the next 50-100 years would be more spatially extensive (i.e., greater area burned [AB]) as compared to the current  regime.We hypothesized that the potential increase in AB in northwest Alaska would lead to decreased quality and quantity of caribou winter range as determined by the amount of available habitat over 50 years old (Joly et al. 2010, Collins et al. 2011), and to an increase of high-quality moose habitat as indexed by the amount of 10 to 30-year old deciduous vegetation (Weixelman et al. 1998, Maier et al. 2005).

Study area and wildlife
The study area is the range of the WAH and the entire Seward Peninsula of northwest Alaska (Fig. 1).The approximately 377,000 caribou in the WAH can be found distributed over some 360,000 km 2 (1.05 caribou/km 2 ; Dau 2007).Overwintering caribou can be found throughout this range, though the areas of recent concentrated use have been in the Nulato Hills and Seward Peninsula (Joly et al. 2007a).While the region is dominated by arctic tundra, especially these areas of concentrated use, there are large expanses of tundra-boreal forest interface, boreal forest, alpine tundra, mountainous terrain and wetlands (Joly et al. 2010).Moose densities range from very low (;0.04/km 2 ) in the northwest portion of the study area to very high (;3/km 2 ) in southeast, but are low (;0.12/km 2 ) over much of the study area (Alaska Department of Fish and Game [ADFG] 2008).The higher densities of moose correspond with expansive riparian complexes formed by the confluence of the large Yukon and Koyukuk Rivers.

Alaska Frame-Based Ecosystem Code (ALFRESCO)
The planet's climate system is inherently complex, including how the climate impacts fire and vegetation at a local-scale.We used Alaska Frame-Based Ecosystem Code (ALFRESCO) to explore the interactions and feedbacks among fire, climate, vegetation, and caribou and moose habitat in northwest Alaska.ALFRESCO is a spatially-explicit cellular automata model that simulates fire and vegetation successional dynamics in Alaska at a 1-km spatial resolution on a 1-year time step (Rupp et al. 2002).For the first time, we modified the model to incorporate the tundra habitats of northwest Alaska (see Vegetation section below).ALFRESCO models the relationship between climate (i.e., monthly average temperature and total precipitation) and total annual area burned (i.e., the areal extent of fire on the landscape) rather than explicitly modelling fire behavior.We used a generalized boosting model (GBM) similar to that used by Hu et al. (2010b) to determine the effect of climate on the likelihood of cell ignition.
Annual fire occurrence, driven by climate, vegetation type and time since last fire, was simulated stochastically (Rupp et al. 2000b).The ignition of any given cell (pixel) is determined by comparing a randomly generated number against the flammability value of that cell.Fire may spread from an ignited cell to any of its eight surrounding first-order neighbor cells.The burn algorithm in ALFRESCO employs a recursive cellular automaton approach, so fire spread depends on the flammability of cells in the firstorder neighborhood and any effects of natural firebreaks including non-vegetated mountain slopes and large water bodies, which do not burn.The flammability coefficient is tied to vegetation type, so ALFRESCO allows for changes in flammability that occur through succession (Chapin et al. 2003).There are different fire regimes for different ecoregions (Joly et al. 2009b), therefore we subdivided our study area into 4 subregions (Interior, Seward Peninsula, North Slope and Yukon Lowlands) and assigned each a different relative flammability and probability of ignition.These values were determined by comparing model output to observed data during the calibration phase.Additional information regarding ALFRESCO can be found in Rupp et al. 2000aRupp et al. , 2000bRupp et al. , 2002Rupp et al. , 2006.

Vegetation
Our version of ALFRESCO reclassified the 2001 National Land Cover Database (NLCD; www.mrlc.gov;accessed 15 December 2011) vegetation classification into three vegetation types: tundra, spruce, and deciduous vegetation.We categorized non-vegetated, non-flammable NLCD cover types (e.g., open water, perennial ice) as non-vegetated.Our tundra class contained dwarf scrub, grassland, sedge, lichen and moss NLCD cover types.We categorized NLCD evergreen forests and woody wetlands (spruce bogs) as our spruce class.Our deciduous class contained NLCD deciduous and mixed forests categories.The shrub/scrub NLCD class was classified as either tundra or deciduous based on elevation, aspect and growing season temperature.White and black spruce have been differentiated using this technique in previous work (Rupp et al. 2002), but they were combined for our simulations.Succession, occurring as either a transition from deciduous or spruce forest to early successional deciduous vegetation or the self-replacement of tundra (see Discussion), is initiated exclusively by fire (Rupp et al. 2000b).Conceptually, the deciduous vegetation type is an early successional stage of spruce forest.An exception to this successional trajectory can occur when repeated burning and/or climatic conditions preclude the transition from deciduous to spruce (Rupp et al. 2000a).We based transitional ages for deciduous to spruce succession from existing literature (Fig. 2; Viereck et al. 1986, van Cleve et al. 1991, Rupp et al. 2006, Kurkowski et al. 2008) and observational data from the Joint Fire Science Program (JFSP; Duffy 2006).Following fire in the boreal forest, flammability is low and stays low for several decades (Racine et al. 1985, 2004, Jandt et al. 2008, Joly et al. 2010).Vascular tundra vegetation quickly re-sprouts after a fire and can be difficult to differentiate from nearby unburned tundra, with the exception of the absence of caribou forage lichens, after just a couple of years (Racine et al. 1985, 2004, Jandt et al. 2008, Joly et al. 2010).Thus, in a departure from previous modelling efforts, we developed a novel, separate function for the flammability of tundra over time which reflects the speed with which fine fuels accumulate within the tundra (Fig. 2).Successional transitions were determined stochastically in ALFRES-CO (Rupp et al. 2002).

Climate data
We employed the spatially explicit, global, gridded (0.5 degree by 0.5 degree) data provided by the Climatic Research Unit (CRU; Mitchell and Jones 2005; http://badc.nerc.ac.uk/view/badc.nerc.ac.uk__ATOM__dataent_1256223773328276, accessed 15 December 2011).We extracted our study area (Fig. 1) from this comprehensive dataset for the period 1901-2000.The dataset is comprised of the monthly averages for temperature and precipitation.We down-scaled by computing the differences between monthly CRU data and corresponding calculated monthly CRU climate normals (''deltas'') for 1961-1990(Hay et al. 2000) ) using natural neighbor to 2 km 2 resolution and then added PRISM (Parameterelevation Regression on Independent Slopes Model; http://www.prism.oregonstate.edu/,accessed 15 December 2011) climate normals.PRISM pixels were further interpolated to 1 km 2 using ArcGIS (Environmental Systems Research Institute, Redlands, California, USA) nearest neighbor re-sampling.

Fire and fire sizes
The amount of AB annually in Alaska is strongly correlated with monthly weather (average temperature and total precipitation; Duffy et al. 2005, Joly et al. 2009b).We developed a GBM to quantify the impact of monthly temperature and precipitation variables on annual area burned for our study region.The GBM is a non-spatial statistical model with annual AB as the response and the corresponding year's monthly temperature and precipitation values as the explanatory variables.The utility of the GBM in this context is to provide a quantitative framework to inform the generation of spatiallyexplicit maps of flammability that account for the variation of temperature and precipitation across the simulation domain.The spatially-explicit maps of temperature and precipitation come from the PRISM downscaling of both CRU and Global Circulation Model (GCM)-based future climate scenarios.Quantifying the linkage between annual AB and monthly climate variables using this non-spatial approach is appropriate for large spatial domains where the climate signals conducive to fire are almost always sufficiently accompanied by lightning ignitions (Duffy 2006).The climate-fire relationship quantified by this point model was then used as a proxy for quantifying the relative flammability of pixels in spatially-explicit maps of annual landscape flammability.The climate-fire relationship is different for tundra and forest vegetation types; hence we used a different GBM-based functional linkage between climate and flammability for tundra and forest vegetation.
In order to calibrate our model we iteratively compared existing empirical data  to our simulated output for the same time period.For our empirical fire perimeter data, we employed a spatially-explicit historical database maintained by the Alaska Fire Service (http://fire.ak.blm.gov/;accessed 15 December 2011).Estimates of AB based on these perimeters are presumed to be over-estimates due to the existence of unburned inclusions within fire perimeters (Duffy 2006).Previous research has shown that the proportional size of inclusions increases with total fire size and fires .2000ha have approximately 5% inclusions (Eberhart and Woodard 1987).Comparable data do not exist for tundra ecosystems so we applied this factor to our analyses for northwest Alaska.The relative proportion of AB by fires of a given size is an important characteristic of fire distribution (Duffy 2006).Our analysis of the historical database for our study area revealed that most fires are small (,1000 ha) but the majority (.56%) of the v www.esajournals.orgarea burned comes from a small percentage (,5%) of fires that are much larger (.40,000 ha) than the rest.Our findings for northwest Alaska are similar to analyses conducted within the Interior region of Alaska (Kasischke et al. 2006).Therefore we focused our comparisons between historical and simulated fire size on the upper tail of the size distribution because the majority of the area burned occurs due to relatively few but large fires.

Model spin-up and validation
We developed 50 randomly-generated initial stand age/vegetation maps by running ALFRES-CO for 400 years retrospectively to approximate ecologically-realistic initial conditions (2007) with respect to the current distribution and composition of vegetation (Duffy 2006).This time period is approximately twice the duration of the longest reported fire return interval in interior Alaska (Yarie 1981, van Cleve et al. 1991, Chapin et al. 2003).We assembled historical climate data for the model spin-ups consistent with Barber et al. (2004).We conducted 50 different simulations from 1860-1949 using these initial age/vegetation maps and then subjected them to burning from 1950-2007, for comparison to historic, observed fire perimeter data, using the GBM-based flammability maps derived from the historical fire-climate relationship.This process resulted in the generation of a final spin-up consisting of 50 different stand-age maps across northwest Alaska (Duffy 2006).The spin-up phase therefore provided initial conditions with realistic patch size and age-class distributions generated over multiple fire cycles that are used as input for simulations run into the future.We compared the simulated and empirical annual AB from 1950-2007 for each of the 50 simulations generated by ALFRESCO (Fig. 3).We used simple linear regression to assess the relationship between the average annual AB from the 50 simulations to the historical data from 1950-2007.We selected the 5 most representative runs to seed the spin-ups for the 50 future simulations based on vegetation type and ratios between modeled and observed data for AB for the years 2004 and 2005; years with climatic conditions that facilitated unrealistically large modeled fires which could propagate landscape-level impacts throughout the simulation period.In other words, we chose runs that were most likely not v www.esajournals.orgto exhibit unrealistically large AB in warm, dry years.The most representative runs were also selected because they most closely represent realistic depictions of the current landscape based on observed fire activity.Among the stochasticity that is driven by random ignitions, burning, and patterns of vegetation succession, these maps best represented the landscape at the starting point (2007) for the future runs.
Projections with future climate scenariosWe simulated AB 92 years into the future (i.e., .This allowed us to compare the 58-year historic period , for which we have empirical data, to the next two similar length (46year) periods of 2008-2053 and 2054-2099.From the best performing GCMs for Alaska, as determined by an analysis of 15 GCMs (Walsh et al. 2008), we selected the ECHAM5 (high-end warming) and CGCM3.1 (low-end warming) GCMs and imposed the intermediate A1B emissions scenario (Nakicenovic et al. 2000, IPCC 2007) to provide a range that captures potentially high and conservative forecasts, respectively, for future climatically-driven fire activity (Walsh et al. 2008, Balshi et al. 2009).
AnalysesWe calculated annual AB for the entire study area, as well as for both the tundra and boreal forest ecoregions separately, for the 3 time periods (1950-2007, 2008-2053, and 2054-2099) using R programming language scripts (R Foundation for Statistical Computing, Vienna, Austria, www.R-project.org).Similarly, we calculated the areal extent of habitat in (1) deciduous stands 10-30 years old and (2) tundra and spruce forest 50 years old.WAH caribou utilize both tundra and boreal forest habitats during winter (Joly et al. 2007a, Joly 2011).We also calculated these extents for just the herd's core winter range (Fig. 1).

RESULTS
Mean AB from our 50 ALFRESCO simulations was significantly correlated with historical, empirical data from 1950-2007 (F ¼ 13.19, P , 0.001, DF ¼ 56, r 2 ¼ 0.191).While the empirical and simulated AB means were very similar (977 versus 953 km 2 per year, respectively), annual variability was much greater for the historical than the modeled data (SE ¼ 251 versus 89, respectively).The historical data had an AB range of 0-10345 km 2 , whereas the modeled output was 501-5183 km 2 .Our attempts to restrain mean AB to historic levels came at the expense of conservative maximum AB and lower variability in the simulated data.We found that allowing the variability of simulated data to approach the actual data allowed for unrealistic maximum AB in warm years (e.g., 2005).Our model presents a compromise between realistic, yet conservative, maximum AB and variability.While the one-to-one relationship between actual and modeled mean AB was not strong, several other metrics (e.g., cumulative area burned over time, maximum fire size, and mean AB of tundra) indicated that the mean tendency of our model output performed relatively well (Figs.4-6).The strength of the correlation between of each of the 50 runs to the empirical, observed data was variable (mean r 2 ¼ 0.092, range ¼ 0.001-0.372).
The average annual AB within the study area from 1950-2007 was 953 6 89 km 2 (mean, SE).Our projections reveal that within the near term  annual AB would increase (to 1235 6 154 km 2 ) under the ECHAM5 GCM, but not for the CGCM3.1 (924 6 110 km 2 ).Total annual AB was projected to continue to increase for both CGCM3.1 (1188 6 110 km 2 ) and ECHAM5 (1460 6 154 km 2 ) in the 2054-2099 time period.These increases in AB incorporate the fire-and succession-induced changes in flammability through time.
Tundra comprised 73% of the study area; this proportion did not change throughout the study period because our model assumed burned tundra was self-replacing and did not incorporate treeline or successional changes.We did, however, allow the probability of tundra burning to vary as a function of time since last fire due to vegetation-induced changes in flammability (Fig. 2).We projected increases in the amount of tundra AB using CGCM3.1 in the latter half of the century (2054-2099; 52%), but not the near term (Fig. 7).In using ECHAM5, the amount of tundra AB increased by 61% during the near term and more than doubled in the 2054-2099 time period (Fig. 7).
The projected annual AB of spruce showed only modest increases over the coming century for either climate scenario (Table 1).However, the total amount of spruce on the landscape was  v www.esajournals.orgpredicted to decline substantially throughout the remainder of the century (Table 1).This result is due to conversion of spruce to deciduous forest (see Discussion).
AB in deciduous stands during the historic period was only 54 6 9 km 2 .Using CGCM3.1, substantial increases were detected for 2054-2099 (98 6 10 km 2 ) but not the near term (64 6 10 km 2 ).ECHAM5 projections revealed substantial increases in deciduous AB from 2008-2053 (85 6 9 k m 2 ) and 2054-2099 (110 6 9 km 2 ).The total amount of deciduous forest was projected to Fig. 6.Mean area burned of tundra as a performance metric comparing simulated (gray; n ¼ 50) and historicempirical (black line) wildfire data for the entire study area, northwest Alaska, 1950-2007. Fig. 7. Amount of tundra burned (km 2 ) in the entire study area, northwest Alaska, for the simulated historic reference period ; thin solid blue line with diamond markers) and future projections using ECHAM5 (thick solid green line with triangle markers) and CGCM3.1 (dotted red line with square markers) general circulation models (GCMs) under the A1B emissions scenario.
Potential high quality (50 years post-fire) tundra and spruce caribou winter range comprised 84% of the study area in 2007.This area was projected to decline to 82% during the near term and again to 79% during the 2054-2099 period using CGCM3.1.Larger declines were projected using ECHAM5 (79% for 2008-2053 and 75% for 2054-2099).
Quality moose habitat, expressed as deciduous cover types 10-30 years old in the simulations, during the historic period averaged 11045 6 268 km 2 .Using CGCM3.1, moose habitat was projected to increase in the near term (13606 6 301 km 2 ) and the 2054-2099 time period (14475 6 301 km 2 ) compared to the historic period.ECHAM5 projected an even larger (47%) increase (16247 6 248 km 2 ) in the near term.While lower than the near-term time period, the amount of quality moose habitat was projected to be substantially greater from 2054-2099 (14619 6 248 km 2 ) than during the historic period.
Amount of AB, caribou winter range, and moose habitat within the WAH's core winter range (Fig. 1) showed trends that were similar to those in the entire winter range (i.e., study area).AB within the core winter range was projected to substantially increase under the ECHAM5 in the near term (490 6 82 km 2 ) and for both CGCM3.1 and ECHAM5 scenarios (476 6 52 km 2 , 531 6 82 km 2 , respectively) towards the end of the century (2054-2099) as compared to the reference period (281 6 46 km 2 ), but not for CGCM3.1 in the near term (301 6 52 km 2 ).The substantial declines (15-29%) in simulated high-quality caribou winter range 50 years old were more dramatic within the herd's core winter range (Fig. 8).Simulated quality moose habitat increased substantially in the near term within the core winter range under both CGCM3.1 (2415 6 39 km 2 ) and ECHAM5 (3346 6 69 km 2 ) scenarios as compared to the reference period (2038 6 35 km 2 ).

DISCUSSION
Overwintering caribou that incur the energetic expenses of migration and predation, such as the WAH, utilize lichens as their major food source (Klein 1982, Russell et al. 1993, Joly et al. 2007b).They are a critical source of carbohydrates that help these caribou survive winter until emergent forage appears in the spring (Person et al. 1980, Parker et al. 2005).Unfavorable nutritional status can reduce growth, compromise in utero development of fetuses, and have multiplier effects (i.e., large impacts, ranging out to the population level, from relatively small changes in forage quality; White 1983, Parker et al. 2005).In tundra and boreal forest, caribou forage lichens can take 50-80 years or more to return to previous abundance following wildfire (Holt et al. 2006, Joly et al. 2010, Collins et al. 2011).Thus, declines in winter forage, induced by climate change and increased wildfire, could lead to lowered nutritional status of individual animals that can translate into population-level impacts.We project that the combination of more burning in tundra and less overall spruce habitat should lead to modest (2-6%) reductions in the areal extent of quality caribou winter range, as indexed by tundra and spruce habitat .50 years old, in the near term.The declines were only slightly greater for the latter half of the century (2054-2099; 5-10%).The declines were more dramatic within the herd's core winter range (e.g., 15-30% of .50year old habitat by the latter half of the century, Fig. 8).At the current population size, reductions in the quantity of high-quality winter range of this magnitude within and adjacent to the herd's core winter range could limit the ability of WAH caribou to find alternative lichen-rich winter ranges.This would make mute the argument that fire is not a key factor for caribou population dynamics because caribou can just seek out and discover new, alternative, high-quality winter ranges.The projected levels of burning will also limit the amount of habitat to reach an age where mosses have the potential to out-compete lichens, as some have argued (see Coxson and Marsh 2001).
Smaller herd and group sizes, changes in range use, increased dispersal, and diminished reproductive output are all potential consequences of decreased habitat quality and quantity.
Our simulations project that the amount of AB in northwest Alaska could increase by up to 30% in the near term , as compared to our reference period , using an intermediate-level emissions scenario (A1B).This region is dominated by tundra, which we project will see proportionately greater (0-61%) increases in AB in the near term.We forecast that these relatively modest increases in AB will continue towards the end of the century (2054-2099).While the quicker rebuilding of fuels in tundra plays a role in our projections for greater proportions of tundra burning in the future, our simulations project that the amount of AB in spruce habitats in northwest Alaska will only modestly increase during the upcoming century.This projection seemingly runs contrary to predictions for Interior Alaska (e.g., Balshi et al. 2009).However, our simulations also project that the amount of spruce on the landscape will Fig.8. Projected changes in areal extent of quality caribou winter range (.50 years of age) solely within the core winter range of the Western Arctic Herd, northwest Alaska.The simulated historic reference period  is depicted with a thin solid blue line with diamond markers and future projections using ECHAM5 with a thick solid green line with triangle markers (lower line) and CGCM3.1 with a dotted red line with square markers (upper line).These general circulation models (GCMs) employed the A1B emissions scenario.v www.esajournals.orgdecline substantially, thus the proportion of spruce on the landscape that will burn is actually forecasted to increase in our simulations.The declines in spruce abundance we forecast are in line with other recent projections that predict major biome shifts within Alaska during the coming century (e.g., Gonzalez et al. 2010, Beck et al. 2011).However, spruce should be able to expand into tundra habitats as permafrost thaws and active layers deepen (Lloyd et al. 2003).Disturbance by caribou may also facilitate the expansion of spruce by exposing mineral soils (Tremblay and Boudreau 2011).Our model is not currently calibrated to allow for this transition and thus less spruce is predicted on the landscape.Further, less forecasted burning in the latter half of the century is likely a consequence of increased burns in the earlier half of the century leaving early seral stage spruce habitats that are less vulnerable to subsequent burning across the landscape.In other words, fuel loads are removed during fire events and, within spruce forests, these fuels can take decades to return to pre-fire levels.
Our simulations project that the total amount and AB of deciduous habitat will increase substantially, at the expense of spruce habitat, throughout the remainder of the century.Consistent with these forecasts, we projected large increases (19-64%) in high-quality moose habitat in the near term within the WAH's core winter range, which should facilitate increases in moose abundance there (Weixelman et al. 1998, Maier et al. 2005).In an example of apparent competition (Holt 1984), resident moose populations on or adjacent to caribou winter range may facilitate increased wolf abundance and therefore the potential for increased wolf predation on caribou (James et al. 2004).Fires may reduce spatial separation between caribou and wolves, which could also lead to higher predation rates (Robinson et al. 2012).Caribou are easier for wolves to take than moose due to their much smaller size.Latham et al. (2011) documented increases in wolf abundance and caribou predation in Alberta where deer (Odocoileus spp.) densities greatly expanded due to more development-induced early seral habitats.If increased fire did promote moose abundance and hence wolf predation on caribou, it would be a novel example of an indirect, detrimental impact of climate change on caribou population dynamics.The impact of this potential relationship would be greater if the WAH experiences large population declines.Further, numerous interior Alaska herds that are already small (,2000 caribou) may be most at risk if wolf populations increase, because lowdensity caribou populations are thought to be more vulnerable to predation (e.g., Dale et al. 1994).
Wildfire is difficult to accurately model within the tundra and tundra-forest interface (Balshi et al. 2009, this study).We found that compromises had to be made among maximum, mean, and variability of AB to achieve realistic simulations.One important virtue of modelling efforts is that they are highly effective at identifying knowledge gaps.In order to reduce model uncertainty in the future, we suggest that subsequent efforts develop parameters for multiple tundra types (e.g., lowland wet sedge tundra, upland dwarf shrub tundra), incorporate impacts from treeline advance and herbivory, and develop more detailed successional pathways that incorporate the potential for increasing shrub abundance.Moreover, developing models that differentiate years with little to no wildfire and those with substantial wildfire should alleviate some of the difficulties we had with balancing inter-annual variability with mean AB.On a practical side, the paucity of climate stations in the Arctic, especially inland, is a serious concern and should be addressed by incorporating new stations into the existing system.In addition, given the vast expanses of tundra in Russia and Canada, we suggest future modelling efforts analyze the impacts of climate change on tundra fire regimes in these regions.

CONCLUSION
Climate change is predicted to impact caribou in many ways.Decreased forage accessibility during winter, either from icing or increased snow depths, may have a stronger impact than increased summer forage biomass (Tews et al. 2007).Although wildfires occur during the summer, they negatively impact caribou winter range, which will only exacerbate forage accessibility issues.Tundra has potential to re-burn much more quickly than boreal forests, so warmer summer conditions could lead to addi-tional fires (Joly et al. 2009b).Increased warming and burning will also likely facilitate increases in the abundance of shrubs and trees in the tundra (Rupp et al. 2000a, Sturm et al. 2001, Joly et al. 2009a, Forbes et al. 2010, Beck et al. 2011) and thus increase the extent and severity of fires (Higuera et al. 2009, Hu et al. 2010a).Since our models did not account for potentially substantial amounts of tundra being converted or the direct, negative impacts of warmer and drier conditions on lichen growth, and that forage lichens can take longer than 50 years to recover post-fire in some areas, we believe that our results are likely an underestimate of potential changes.Due to edaphic conditions, barriers, time lags and other factors, we do not expect vast expanses of tundra being converted to forests during our study period (i.e., the next 100 years), though larger changes should be expected in the tundra-shrub transition zone (Rupp et al. 2000a, Rupp et al. 2001, Chapin et al. 2005, Lloyd 2005).
The amount of increased wildfire on caribou winter ranges we simulated may intensify discussions of the need for fire suppression/ management plans for conservation.These plans should incorporate traditional ecological knowledge, co-management input, and logistical realities (Beverly and Qamanirjuaq Caribou Management Board 1994, Urquhart 1996, Klein et al. 1999), as well as scientific information, including our results, and the needs of other species of interest.These plans will likely have to be tailored to individual herds and updated regularly to take into account rapidly changing conditions.For the WAH, our forecasts for modest increases in AB over the herd's entire range suggest that any fire suppression efforts be focused on its core winter range.Expansion of moose range in Alaska and other northern regions may prove to be a conservation boon as this species faces drastic climate change-related declines in more southerly latitudes (e.g., Rempel 2011).

ACKNOWLEDGMENTS
Funding for this project was generously provided by the National Park Service (Arctic Network Inventory and Monitoring Program and Gates of the Arctic National Park and Preserve), the Bureau of Land Management (Central Yukon Field Office), and the Alaska Climate Science Center (Cooperative Agree-ment Number G10AC00588 from the U.S. Geological Survey).The contents of this paper are solely the responsibility of the authors and do not necessarily represent the official views of the NPS, BLM or USGS.Mark Olson, Anna Springsteen, Tim Glaser, Matt Leonawicz, and Dustin Rice (all of Scenarios Network for Alaska and Arctic Planning) and Paul Atkinson (National Park Service) provided technical expertise and assistance to make our modelling efforts possible.

Fig. 1 .
Fig. 1.Study area (outlined by thick black line) in northwest Alaska.Tundra-dominated areas are shaded brown and forest-dominated areas are in green.The range of the Western Arctic Caribou Herd is contained within the domain and is depicted with hatching, while the herd's core winter range is cross-hatched (courtesy of the Alaska Department of Fish and Game).

Fig. 2 .
Fig. 2. Modeled flammability of pixels as a function of time since fire in northwest Alaska (boreal forest depicted with a dotted gray line and tundra with a solid black line).

Fig. 3 .
Fig. 3.A comparison of annual area burned (km 2 ) between observed (black bars) and the mean of 50 simulated ALFRESCO runs (red bars) for the entire study area, northwest Alaska, 1950-2007.

Table 1 .
Total area (km 2 ) and area burned (mean 6 SE) of various vegetation types (spruce forest, deciduous forest, tundra) in the study area, northwest Alaska, for the simulated historic record and using two climate scenarios to project into the future.Tundra was self-replacing in our model, thus its total remained unchanged.