4.1 Spatial and Temporal Context of Fire on the North Slope of Alaska
 Between 1950 and 2012, 42 tundra fires are known to have occurred on the North Slope of Alaska, with the earliest reported burn in 1959 [AICC, 2012]. These 42 fires burned a combined area of 1635 km2. The majority (~83%) of these fires have been small, burning areas <10 km2. Four of the five largest fires observed on the North Slope over this period occurred in the past two decades, which define the pattern described by Rocha et al. . There has also been an increase in the total number of fires detected on the North Slope, particularly in the last decade, but many of these are very small (< 0.5 km2), and their detection likely a result of better observational coverage [Miller, 2010]. Spatially, tundra fires have primarily occurred in the Arctic Foothills between the Brooks Range and the Arctic Coastal Plain (Figure 1). Only three fires have been documented on the Arctic Coastal Plain and all three of these occurred between 2011 and 2012.
 Together, the Meade River (500 km2) and Ketik River (1200 km2) fires roughly equal the area burned on the North Slope of Alaska over the period of record keeping [AICC, 2012]. Addition of these sites to the fire history database would make the Ketik River fire potentially the largest documented event and the Meade River fire the third largest documented event in the region. However, these delineated areas should be viewed as approximations as it is even difficult to accurately delineate fire perimeters that occur during the remote sensing period, which began around the 1950s [Kasischke and Turetsky, 2006; Barrett et al., 2012]. These delineated areas should also be viewed as maximum burn extents since tundra fires exhibit a heterogeneous burn pattern with numerous unburned inclusions. A patchy burn pattern at the Ketik River site could explain the lack of charcoal encountered within the estimated burn perimeter. This could also result from the presence of a more uniformly graminoid-rich tundra at the time of the burn because this vegetation type often leaves no partially burned charred material [Cofer et al., 1990]. Alternatively, it could indicate that the fire severity varied across the burn extent or that we have over-estimated the size of this fire.
 Both the Meade River and Ketik River fires likely occurred as single events sometime during the last 300 years (Table 1). Due to limitations associated with radiocarbon dating during this period, we can only provide the possibility of the fires in several different age ranges. In addition, the charcoal material we dated was fixing atmospheric carbon before these fires occurred. We consider the period between C fixation and fire (in-built age) for the material we dated (burned twigs) to be a few years and at most a few decades [Gavin, 2006]. Nevertheless, we can further constrain the timing of the two fires by assessing the probability distribution of calendar year calibration ages, historic aerial photographs, and regional climate proxy records.
 For both the Meade and Ketik sites, the most recent calibrated age range occurred between AD 1952 and 1954. However, this age window accounts for a low probability of the sample (2–3%) making it very unlikely to be the calendar age of the charcoal. Furthermore, analysis of aerial photography acquired in the late-1940s revealed that the increased landscape texture and inferred fire perimeters at the Meade and Ketik sites were already present at this time (SOM Figure 1). The oldest possible age range for both fires occurred between AD 1695 and 1725. The probability that the Meade and Ketik fires occurred during this time period is 20% and 24%, respectively. This period is coincident with the Little Ice Age (LIA), and in particular the coldest and driest phase of the LIA on the North Slope [Bird et al., 2009]. While, presumably this was a period with little thunderstorm activity, large tundra fires could have occurred owing to the dry conditions. The final possible calibrated calendar year age ranges for these events occur between AD 1810 and 1920. The probability distribution for both the Meade and Ketik sites is highest during this time, 77% and 74%, respectively. Within this time period, results from the Meade River sample indicate four possible age ranges, with AD 1876 to 1918 yielding a 60% probability. Results from the Ketik sample only indicate the one age range (AD 1812 to 1919) during this period. Jorgenson et al.  noted the possibility of widespread thermokarst pit formation between AD 1850 and 1940 for a study site approximately 200–300 km to the northeast of these fires sites. Furthermore, Bird et al.  suggest that the period after AD 1880 marked a transition to regionally warm and dry conditions. Thus, given the probability distribution of ages from radiocarbon dating, the presence of fire altered tundra in the 1940s photography, and regional climate proxy data, both fires likely occurred between AD 1880 and 1920 or shortly thereafter when considering the in-built age factor [Gavin, 2006].
 The discovery of the Meade and Ketik Fires indicates that large tundra fires on the North Slope of Alaska have likely occurred as recently as the last ~100 to 130 years. It is likely that other fire events of variable size have also occurred in the region that remain to be discovered (SOM Figure 2) and that the role of fire as a disturbance agent in Arctic tundra in general has likely been underestimated. Efforts that seek to determine the historical context of large fires on the North Slope are still in their nascent stage. In addition, our cursory analysis of a subset of the remote sensing archive for the region identified eight small tundra fires not documented in the fire history database (Figure 1). While the focus of this paper is on the detection of the large prehistoric events, a recent study highlighted the important role of small fire events on fire regimes and that globally there was a 35% underestimate in the area burned as a result of omission of small fires [Randerson et al., 2012].
 These data also add to the growing observations concerning the fire regime of the North Slope region. Based on recent (AD 1950 to present) [Rocha, et al., 2012] and paleo (last ~5000 years) [Hu et al., 2010; Higuera et al., 2011] estimates of fire rotation periods (the amount of time it would take for a specified region to burn) and fire return intervals (the number of years between successive fires at a given location) for the North Slope of Alaska converge on a time frame of 4000–5000 years, respectively. Addition of our sites to this estimate for the region using the highest probability calibrated age ranges (AD 1810 to 1920) and the definition of the North Slope as provided in Rocha et al.  indicate a fire rotation period that ranges from approximately 3000 to 7000 years. Up to this point, the North Slope has a scant and evolving data set of past fire frequency compared to other systems where fire rotation periods have been estimated based on decades of research [e.g., Sousa, 1984]. Estimates of a fire rotation period for the study area have proven to be highly variable when using different calibrated age ranges of the fires discovered here. For example, using the period AD 1810–2012 results in a fire rotation period of ~7000 years and using the period AD 1920–2012 results in a fire rotation period of ~3000 years. Given the uncertainties in a minimal number of radiocarbon samples, the in-built age factor, and the fact that future work is needed to better reconstruct the fire history in the region, we have included these calculations in this paper a bit reluctantly. In addition, a recent modeling effort focused on fire return intervals in Alaska found values that ranged from 100 to 200 years, 500 to 700 years, and greater than 2000 years for various locations on the North Slope of Alaska [Pfeiffer et al. 2013]. These wide ranging estimates in the repeated burning of Arctic tundra in this region highlight the need for more concentrated work with respect to potential past tundra burning.
4.2 Wildland Fires and Shrub Invasion of Tundra
 Direct measures of the ecological impacts of tundra fires on vegetation on the North Slope of Alaska over decadal and centennial time scales are sparse to nonexistent. Observations at the Anaktuvuk River Fire indicate that graminoid cover is substantially greater than shrub cover in the first four years postfire [Jandt et al., 2012; Bret-Harte et al., 2013]. Observations at the 1977 Kokolik River site spanning the first five years postfire also support this pattern [Racine et al., 1987]. At the other large historic tundra fire site, the 1993 DCKN190, graminoid cover remained greater than shrub cover 17 years postfire, but the proportion of grasses relative to sedges had increased [Barrett et al., 2012]. Remotely sensed observations of vegetation change in tundra sites that burned in the last several decades also follow this general pattern with an increase in both albedo and greenness that is interpreted as an increase in grass cover [Rocha et al., 2012]. Our observations from one site located in the northern portion of the Anaktuvuk River Fire also follow this short-term trajectory, but we note a substantial difference in grass cover (primarily Arctagrostis latifolia and Calamagrostis lapponica) relative to our unburned upland moist tussock tundra site in the first five years postfire (Figures 6b and 7b). Our results also indicate a greater abundance of grass cover at the 1993 fire site (Figures 6b and 7c) similar to that found by Barrett et al. , whereas vegetation present at the 1977 burned site resembled the younger fire sites but with the addition of lichens that were absent from the 2007 Anaktuvuk River and 1993 DCKN190 fires (Figure 6b and 7d).
 Addition of the Meade River and Ketik River sites to the chronosequence of burned upland tundra used to infer fire-related changes in vegetation indicates that the relative proportion of grasses is less than more recently burned sites, yet still far greater than unburned tundra, likely reflecting the legacy of past burning. Moreover, the proportion of shrubs is greatest at these two sites (Figures 6b, 7e, and 7f). We interpret the greater abundance of shrubs to reflect secondary succession from grass to shrub dominance postfire. On the more southerly, warmer Seward Peninsula and Mackenzie Delta, postfire secondary succession in tundra is known to be accompanied by a transition from grasses to shrubs over a time scale of several decades [Racine et al., 2006; Lantz et al. 2013]. However, on the North Slope of Alaska, this transition has not been documented previously, and our results provide the first compelling evidence of a shift toward shrubbier conditions postfire in the region. These findings have implications when interpreting the modern distribution of shrubs on the North Slope of Alaska [Beck et al., 2011] and indicate that fire could play a key role in the ongoing “shrubification” of the region [Sturm et al. 2001; Tape et al., 2006].
4.3 Fire History and Permafrost Degradation
 Tundra fires can trigger thermokarst by removing insulating layers of vegetation and peat and decreasing albedo at the ground surface. In the southern portion of the Anaktuvuk River Fire, Mann et al.  noted the widespread occurrence of active layer detachment slides and several thaw slumps forming over the first three years following the burn. Rocha et al.  synthesized thaw depth measurements for all known tundra burn sites in Alaska and found that there was a sustained increase in the thaw depth within burned sites even three decades postfire. While we lack robust measures of active layer thickness variation and did not see widespread evidence of stabilized detachment slides and thaw slumps, we did notice widespread formation of thermokarst pits at our sites and conducted microtopographic surveys to capture the relief associated with ice-wedge troughs, as the formation of thermokarst pits is also an indicator of permafrost degradation [Jorgenson et al., 2006].
 The microtography of ice-wedge polygons (raised centers to troughs) in undisturbed upland tussock tundra settings was 0.2 to 0.5 m (Figure 7a). When viewing these sites in high-resolution remotely sensed imagery, it is difficult to distinguish individual polygons (Figure 3a). However, in images from the Anaktuvuk River fire four years after the burn, the increase in texture is evident (Figure 3b), and the ice-wedge trough microtopography show initial signs of thermokarst (Figure 7b). For the sites that had burned in 1977 and 1993, the relief associated with ice-wedge troughs ranges from 0.5 to 1.3 m indicating that the fires disrupted the soil thermal regime leading to permafrost degradation (Figures 7c and 7d). This shows up in the high-resolution imagery as an increase in texture (Figures 3c and 3d). DGPS surveys conducted at the Meade River and Ketik River sites were more similar to the burn sites than the unburned sites, with microrelief associated with the ice-wedge troughs on the order of 0.5 to 1.1 m (Figures 7e and 7f). However, one notable difference between these sites and the sites that burned in 1977 and 1993 (Figures 3c and 3d) was the lack of ponded water in the ice-wedge troughs (Figures 3e and 3f). We interpreted that this could be the result of a more integrated drainage network in these older burn sites or paludification of the troughs and infilling with peat and vegetation.
 The presence of near-surface, ice-rich permafrost terrain with massive ice in the form of wedges was essential for the identification of the previously unrecognized tundra fire events. The degradation of ice-rich permafrost terrain likely plays an important role in postfire vegetation succession trajectories as noted by Bret-Harte et al. . Thus, the increase in shrubbiness at the century-old fire sites is likely due in part to ice-rich permafrost degradation and it was the combination of vegetative and geomorphic change that aided in our ability to identify these sites. While it appears that large fires on the North Slope tend to occur in the eolian silt band of the northern Foothills where thick, Pleistocene-aged, ice-rich permafrost (Yedoma) is found, several small historic fires have occurred in areas not indicative of ice-rich permafrost terrain (Figure 1). Detection of fires that predate the period of record keeping in these ground-ice poor areas would be difficult to detect by using the techniques applied in this study. The distribution of Yedoma permafrost may also have implications for potential future successional trajectories associated with the Anaktuvuk River fire. Figure 1 shows that Yedoma is likely present below the northern two thirds of the area impacted by the Anaktuvuk River fire. This may result in more widespread ground subsidence as a result of postfire permafrost degradation relative to the southern one third of the burned area, where surface permafrost deposits tend to be less ice rich. Thus, spatially variable responses in vegetation succession should be expected in response to this fire event.
4.4 Broader Implications
 Identification of the Meade River and Ketik River fires may have Arctic-wide implications. Mack et al.  found that the 2007 Anaktuvuk River fire released 2.1 Tg of C to the atmosphere, which is equivalent to the annual net C sink for the entire Arctic biome. If the Meade River and Ketik River fires released similar amounts of C during combustion, these events may have had a similar immediate impact on atmospheric carbon. There may also be an important long-term impact of these disturbances because the fires here may promote the degradation of the Yedoma permafrost deposits (areas with Pleistocene-aged permafrost deposits that can be high in ice and carbon content) that occur in this region (Figure 1) [Kanevskiy et al., 2011]. The apparent shift from sedge-dominated tundra to shrub-dominated tundra that we documented in the older burn sites could also have a positive feedback on climate warming due to the combination of a decrease in albedo and an increase in evapotranspiration associated with this transition [Euskirchen et al., 2009; Bonfils et al., 2012]. Thus, during a period of already rapid shrub expansion in the Arctic [Myers-Smith et al., 2011] in which tundra fires are becoming larger and more frequent [Rocha et al., 2012], this process may be exacerbated [Lantz et al., 2013]. Further, an increase in shrub abundance may also lead to an increase in the frequency, severity, and areal extent of tundra fires [Higuera et al., 2008], which may lead to an increase in C release [Mack et al., 2011]. However, very little is known about the balance of C storage following tundra fires. The complexities associated with the long-term C budget at burned tundra sites underscore the importance of better understanding the spatial and temporal context of tundra fires in the Arctic.