Increasing elevation of fire in the Sierra Nevada and implications for forest change

. Fire in high-elevation forest ecosystems can have severe impacts on forest structure, function and biodiversity. Using a 105-year data set, we found increasing elevation extent of fires in the Sierra Nevada, and pose five hypotheses to explain this pattern. Beyond the recognized pattern of increasing fire frequency in the Sierra Nevada since the late 20th century, we find that the upper elevation extent of those fires has also been increasing. Factors such as fire season climate and fuel build up are recognized potential drivers of changes in fire regimes. Patterns of warming climate and increasing stand density are consistent with both the direction and magnitude of increasing elevation of wildfire. Reduction in high elevation wildfire suppression and increasing ignition frequencies may also contribute to the observed pattern. Historical biases in fire reporting are recognized, but not likely to explain the observed patterns. The four plausible mechanistic hypotheses (changes in fire management, climate, fuels, ignitions) are not mutually exclusive, and likely have synergistic interactions that may explain the observed changes. Irrespective of mechanism, the observed pattern of increasing occurrence of fire in these subalpine forests may have significant impacts on their resilience to changing climatic conditions.


INTRODUCTION
. These effects might be particularly acute in forest types where fire has been Increasing incidence of fire may have cascad-historically rare, such as high-elevation forests. ing effects on forest structure, function and If vegetation is not adapted to frequent fire, biodiversity (Bowman et al. 2009, Stephens et increasing fire activity may alter community composition, potentially driving further unex pected changes (''ecological surprises'') (Paine et al. 1998). We present evidence that fire is becoming increasingly frequent at higher eleva tions in the Sierra Nevada of California.
Fire occurrence and spread have been limited historically by sparse, relatively wet fuels in the low-productivity, high-elevation forests of the Sierra Nevada Fites-Kauf man 2006, Fites-Kaufman et al. 2007). However, empirical studies have documented increasing tree density in subalpine forests, suggesting that fuels may also be increasing (Millar et al. 2004, Dolanc et al. 2013.
Warming temperatures may also be directly reducing fuel moisture (Fried et al. 2008), while indirectly promoting fuel production by encour aging pest and pathogen attack (Raffa et al. 2008) and increasing tree mortality (van Mantgem et al. 2009, Allen et al. 2010. Ultimately, increasing fire frequency may serve as a catalyst of changing species distributions and vegetation type conver sion in these forests (Lenihan et al. 2008, Moritz and. Our aim was to determine whether there is evidence of an on-going upward shift in maxi mum elevation of fire in California's Sierra Nevada. We evaluated the maximum elevation of fires as recorded by the California Fire Perimeters Database version 12.1 (www.fire.ca. gov), a geodatabase of fires in California that dates from the early 20th century. Observing strong increases in the upper elevation extent of fire in the Sierra Nevada through the 20th century, we present five hypotheses that may explain this pattern and discuss what might constitute evidence for these explanations. Many of the mechanisms that may drive increasing fire elevation, however, lack adequate data to rigor ously test the hypotheses. In light of this inability to conduct fair tests, and recognizing these as non-exclusive hypotheses (e.g., many may be contributing to the pattern), we focus on inter preting the ramifications of the observations. We synthesize these observations with a discussion of the implications for forest structure, composi tion and management in the subalpine zone of the Sierra Nevada during the coming century.

METHODS
We downloaded fire perimeters from the California Fire Perimeter Geodatabase (version 12.1, April 2014), archived by the California Department of Forestry Fire and Resource Assessment Program (FRAP), available at http:// frap.fire.ca.gov/projects/fire_data/fire_perimeters_ index.php (Fig. 1). The time scale of the data was 1908 to 2012. We selected all fires that occurred in California's Sierra Nevada, spanning an elevation range from 10 to 4421 m and an area of 60,755 km 2 (Fig. 1).
There are several ways to evaluate potential changes in the elevation of fires through time in the Sierra Nevada. Our goal was to be thorough and test fire elevation patterns to ensure that our observation is robust to data selection. The first issue of concern is that the fire record becomes more detailed and more thorough over time. To evaluate potential specious patterns we evaluat ed fires in two sets by size, all fires and all fires greater than 202 ha (500 acres) in size. We did this to account for the hypothesis that small fires early in the record would have been less frequently detected, reported and recorded with in the dataset. Large fires, however, would be well-recorded throughout the data record. We arbitrarily chose 202 ha (500 acres) as a cut-off to represent large fire areas. A preliminary assess ment of the frequency of fires of different sizes reported in the database suggests that fires smaller than 50 ha are much less frequent earlier in the time period. However, there is no evidence of an increasing proportion of fires larger than 50 ha reported since 1950. Hence using 1970, and 202 ha, represents conservative cut-offs that likely remove potential bias that arises from differential reporting through time.
Similarly, as record-keeping has improved through time, earlier records may be inadequate. Therefore, we ran our analysis for all fires and for larger (.202 ha) fires for the entire time series, as well as from 1970 onwards, when FRAP indicat ed that record keeping became much more thorough and accurate.
We also calculated two metrics of elevation for each fire: the maximum elevation within each fire boundary and the highest elevation along the fire boundary line as recorded in the fire database. The lack of data on burn coverage within a fire v www.esajournals.org boundary could obfuscate what represents the highest elevation of a fire. An alternative approach is to use the highest elevation along the perimeter of a fire. We found, however, there was no substantive difference in patterns be tween the highest elevation within a burn or the highest point along the fire boundary (perimeter) in preliminary assessments, and the correlation between the two measures across all fires was 0.998. As a consequence, we report only the maximum elevation along the fire boundary.
We extracted these metrics for the seven fires with the highest maximum elevations for each year and tested the correlation between eleva tions and year for the single, three, five and seven highest. We evaluated several of these metrics in order to reduce the stochasticity that is observed when only the single highest fire was used to represent each year, and further explore the degree to which our observation is robust. For each subset of data we calculated the slope of the prediction, the standard error of that regression coefficient, and the p-value for each. For ease of interpretation, we also calculated the fraction of the variation explained (r 2 ) and the predicted change in elevation across the time period v www.esajournals.org evaluated. Finally, we used a contingency test to evaluate changes in the frequencies of fires at different elevation bands through time. Again, because of reporting issues, we repeated our analysis using both all fires and fires larger than 202 ha that are likely to have accurate reporting throughout the temporal record of FRAP. We used a likelihood ratio test to assess the frequency of fires in early , middle  and late (1989-2012) time periods. Time periods were selected to divide the number of observed fires roughly into three groups. Fires were then grouped by their upper elevation extents into five 500-m elevation bands ranging from 1000 m to .3000 m. Although fire frequency is increasing at low elevation, we did not include fires below 1000 m as they would contribute to significant changes in the distribution of fire frequency by elevation, without being indicative of high elevation.

RESULTS
We found significant positive regression coef ficients for 'year' on maximum fire perimeter elevation for all but two of our 16 test permuta tions (Table 1, Fig. 2). These strongly positive relationships resulted in a predicted increase in maximum fire elevation of over 700 m for fires generally (Table 1). Restricting the analysis to later in the time series, or to larger fires did not, in general alter the conclusion that the maximum elevation of fire in the Sierra Nevada is increas ing.
We also found strong supportive evidence of differential fractions of fires at different upper elevations across early , middle  and late (1989-2012) time periods (n ¼ 4545; df ¼ 12; chi-square ¼ 455.1; p , 0.0001; Table 2). Similar results were produced when restricting the data to fires greater than 202 ha ( Table 2). The probability of fires at elevations above 3000 m has increased through time with 30 of 1534 fires (1.95%; 1.36-2.78% CI) occurring since 1989 above 3000 m, while just 7 of 1531 (0.46%; 0.22-0.94% CI) recorded fires prior to 1950 were above this elevation (Table 2).

DISCUSSION
The important trend in our analysis is the increasing frequency of wildfire at higher eleva tions in the Sierra Nevada. Historically fire was rare in subalpine forests in California, with an expected fire return interval in excess of 100 years (Caprio andLineback 2002, Safford andvan de Water 2014). The observed trend suggests a fundamental change in the disturbance regime for these forests that may affect the structure, composition and function of subalpine forests. Our data are observational and cannot provide a means of identifying the potential mechanisms that may be driving this change. Below we propose five hypotheses as to what may be v www.esajournals.org driving the increase in fire incidence at high elevations in the Sierra Nevada. They are not mutually exclusive; each could contribute to observed patterns.
1. Fire management strategies.-A cessation of fire suppression at higher elevations, along with a shift toward focusing management of wildfire risk at lower elevations may result in a pattern of increasing fire elevation. Since the beginning of the 20th century, federal and state land manage ment agencies have generally sought to extin guish wildfire in the western United States. Accumulated evidence of the ecological roles played by fire in western US forests led to the adoption of ''wildland fire use'' (WFU) policies by federal land management agencies by the late 20th century (Stephens and Ruth 2005). These policies permit the management of fire for natural resource benefits in instances where threats to human life, infrastructure, and resourc es are minimal, particularly in some higher elevation forests (mostly .1800 m). In the Sierra Nevada, WFU-type policies have been the rule on many National Park Service lands since approximately 1970, and the US Forest Service adopted similar policies in the mid-1990s for some large high-elevation wilderness areas in the southern part of the range (Stephens and Sugihara 2006). Mallek et al. (2013) found that although more WFU is practiced in higher elevation forests, even limited application of fire suppression in these long fire-return interval ecosystems can significantly limit high-elevation v www.esajournals.org Fig. 3. Average August minimum temperatures for the Sierra Nevada bioregion (Fig. 1) within two elevation boundaries. Open circles capture average annualized daily minimum temperatures between 2000 and 2250 m in elevation within the Sierra Nevada (n ¼ 111, r 2 ¼ 0.11, p ¼ 0.0005). Closed circles represent average minimum temperatures between 2500 and 2750 m elevation (n ¼ 111, r 2 ¼ 0.07, p ¼ 0.004). Minimum temperature values for August were extracted from downscaled climate data used with the California Basin Characterization Model (Flint et al. 2013). burn extent.
2. Changing climate affects fire.-Warming trends are likely leading to increased forest fire activity (Westerling et al. 2006), including in high-elevation forests, and this trend is predicted to continue increasing over the next several decades (Moritz et al. 2012). Climate during the fire season at higher elevations (approximately July through October) is also changing: earlier snowpack melt (Kapnick and Hall 2010) result ing in earlier dry down of soils and fuels, and longer, warmer and/or drier summers could all lead to an increase in fire, particularly at higher elevations. While high-elevation areas in the southern Sierra Nevada may resist snowpack declines (Mote et al. 2005), increasing night-time low temperature may be resulting in generally warmer and drier conditions during the fire season (Fig. 3).
We assessed changes in a climatic variable that may be associated with changing fire conditions, average August minimum daily temperature (as the most rapid increases in temperatures have been in daily minima), from 1900 to 2010 for two elevation zones in the study area (Fig. 3). The lower elevation zone, 2000-2250 m, was defined because it captures the mean upper elevation of fires prior to 1930, whereas the upper elevation zone, 2500-2750 m, represents the normal upper elevation boundaries of the highest fires in the 21st century (Fig. 2). The upward temperature trend in both curves (Fig. 3) suggests that the upper elevation band will soon experience minimum temperatures analogous to those in the lower elevation band in the first part of the twentieth century, suggesting an upward shift of the upper montane climate zone.
3. Increasing forest fuels.-Increasing fuels and connectivity between fuels at high elevations may also be driving increased incidence of fire. Several mechanisms could be responsible. Fire suppression throughout most of the 20th century has contributed to increasing stem density in forests of the Sierra Nevada, and allowed the build-up of both live and dead fuels (Vankat and Major 1978, Parsons and Debenedetti 1979, McKelvey et al. 1996, Ansley and Battles 1998. Combined with rising temperatures, this has contributed to increasing fire severity in low to mid elevation forests (Miller et al. 2009, Dillon et al. 2011, Miller and Safford 2012, Mallek at al. 2013. Historically, high-elevation forests of the Sierra Nevada, with their thin soils, sparse structure and cool climate, burned infrequently Fites-Kaufman 2006, Fites-Kaufman et al. 2007). However, while the significant increases in overall density experi enced in these forests over the last few decades (Dolanc et al. 2013) may be driven more by climate than fire suppression, the increased fuels at high elevation may connect previously discon tinuous forest stands and allow fires to carry farther upslope.
4. Increasing ignitions.-Fires on federal lands in the Sierra Nevada are caused both by lightning and by human ignition, at about an equal rate (Keeley 1982). There is also an elevation gradient effect, such that both lightning and human ignitions may be equally important at low to middle elevations. Records indicate that the density of lightning strikes are greatest between 1800 m and 3000 m, and reduce markedly above 3600 m (van Wagtendonk and Cayan 2008). With increasing human presence at higher elevations, however, there is the potential for increased ignitions, although data are lacking. High eleva tion fire has previously been limited by some combination of sparse fuels in some locations and burning conditions (weather and fuel mois ture) in others. Limiting ignition sources may not have been important. However, with stand density and potentially increasing fuel contiguity in subalpine forests, the likelihood of a lightning strike starting a fire at higher elevations may be rising.
5. Sampling bias.-Finally, sampling bias is also a possible driver of the recorded increase in maximum fire elevation. Record keeping has improved over time, and simply keeping better track of fire in remote locations may have contributed to our observation. Based on conver sations with knowledgeable FRAP experts, we consider the effect of fire sampling bias to have made only a negligible impact on our observa tion. Sampling bias would have resulted in an under-reporting of older fires. This would be less likely true for larger fires. We restricted our sample to fire perimeters .202 ha (500 acres), and found a similar increasing trend through time, reducing the likelihood that this pattern is due to under reporting of fires in older records.
Our hypotheses are not mutually exclusive, and a close examination of lightning ignitions, annual snowpack estimates, seasonal tempera tures increasing vegetation density, and fire may reveal linked drivers of increasing fire elevation.

MANAGEMENT IMPLICATIONS
Increased fire at high elevations could acceler ate vegetation shifts by increasing mortality rates of existing vegetation, and creating opportunities for upslope migration of species from lower elevations. Climate projection models predict a demise of Sierra Nevada subalpine forests as a direct or indirect consequence of changing climate (Lenihan et al. 2008). Empirical studies evaluating forest change in this region during the 20th century have recorded a different trend: increasing densities of subalpine forests, with relatively little upward distributional shifts of upper montane tree species such as red fir (Abies magnifica; Millar et al. 2004, Dolanc et al. 2013. The thin soils and harsh environment of the subalpine zone (Fites-Kaufman et al. 2007) may constrain species such as red fir, which grows on deeper soils (Potter 1998, Sawyer et al. 2009). Nevertheless, the pattern of increasing density of subalpine species suggests a shift toward conditions in which disturbance may accelerate climatically-driven transitions toward upper montane forest types, as also appears to be happening along the lower edge of conifer forests, where conifers are being replaced by hardwoods (Thorne et al. 2008).
Increased mortality rates among larger trees (Lutz et al. 2009, van Mantgem et al. 2009) could contribute to proportional shifts in tree density from larger to smaller trees in the upper montane and subalpine forests of the Sierra Nevada (Dolanc et al. 2013). The build-up of dead fuel biomass is also increasing with stand density in some parts of the high Sierras (Guarin andTaylor 2005, Smith et al. 2005), potentially enabling v www.esajournals.org more intense, and larger fires, which in turn could accelerate further changes in forest struc ture. Indeed, proportional increases in annual area burned were greatest in Sierra Nevada subalpine communities during the period 1984-2010 (Mallek et al. 2013).
A second issue relates to forest disease and pests. Currently, several high elevation white pine (Pinus spp.) species in the US and Canada are threatened by white pine blister rust (Cronar tium ribicola) (Tomback and Achuff 2010). How ever, so far, rates of mortality from the disease are lower in the Sierra Nevada than in other mountain ranges (Maloney 2011), which suggests the Sierra Nevada may provide refuge for some high elevation conifer populations. If this is the case, protection of such populations would have greater conservation importance and may be come a focal management objective. Increased fire, if it fosters succession toward replacement of subalpine tree species by upper montane species, could run counter to this objective. To date high elevation forests in the Sierra Nevada have not suffered the catastrophic losses to pine beetles (Dendroctonus spp.) seen in the central and northern Rocky Mountains over the last 5-10 years. However, recent (i.e., 2014) aerial surveys by the US Forest Service found notable increases in insect-driven mortality of high elevation pines (Z. Heath, personal communication) and continua tion of California's current drought may exacer bate this trend.
The demonstrated upward shift in high eleva tion fires, coupled with changes in tree mortality and stand density, and the potential of the Sierra Nevada as a refuge for disease-threatened tree species, all speak to the importance of carefully researching and monitoring the effects of increas ing fire elevation on ecosystem composition, structure and function.