Tundra fire increases the likelihood of methane hotspot formation in the Yukon–Kuskokwim Delta, Alaska, USA

Rapid warming in Arctic tundra may lead to drier soils in summer and greater lightning ignition rates, likely culminating in enhanced wildfire risk. Increased wildfire frequency and intensity leads to greater conversion of permafrost carbon to greenhouse gas emissions. Here, we quantify the effect of recent tundra fires on the creation of methane (CH4) emission hotspots, a fingerprint of the permafrost carbon feedback. We utilized high-resolution (∼25 m2 pixels) and broad coverage (1780 km2) airborne imaging spectroscopy and maps of historical wildfire-burned areas to determine whether CH4 hotspots were more likely in areas burned within the last 50 years in the Yukon–Kuskokwim Delta, Alaska, USA. Our observations provide a unique observational constraint on CH4 dynamics, allowing us to map CH4 hotspots in relation to individual burn events, burn scar perimeters, and proximity to water. We find that CH4 hotspots are roughly 29% more likely on average in tundra that burned within the last 50 years compared to unburned areas and that this effect is nearly tripled along burn scar perimeters that are delineated by surface water features. Our results indicate that the changes following tundra fire favor the complex environmental conditions needed to generate CH4 emission hotspots. We conclude that enhanced CH4 emissions following tundra fire represent a positive feedback that will accelerate climate warming, tundra fire occurrence, and future permafrost carbon loss to the atmosphere.


Introduction
The Arctic is warming four times faster than the global average (Rantanen et al 2022), accelerating regional ecological changes that impact the entire Earth system.Thaw of ice and carbon-rich permafrost is concerning due to its potential to reintroduce large quantities of previously sequestered ancient permafrost carbon to the atmosphere as greenhouse gasses, carbon dioxide (CO 2 ) and methane (CH 4 ) (Schuur and Mack 2018, Turetsky et al 2020, Miner et al 2022).The magnitude and timing of this positive feedback to further warming are still debated given the large uncertainties around quantifying current and future changes in Arctic vegetation, hydrology, and permafrost dynamics and how they relate to gas exchange (Moubarak et al 2022).Acting as an accelerant to ecosystem change, wildfires have increased in the high latitudes since the 1980s (Kasischke et al 2010), threatening further permafrost carbon losses and amplifying the complexity and uncertainty around forecasting impacts on future climate (Hu et al 2015, Young et al 2017, Foster et al 2022).
Fire occurrence in tundra permafrost ecosystems can result in rapid and deep talik formation (perennially thawed substrate) (Jones et al 2015).Though historically rare, tundra fire produces talik on timescales of a few decades that would otherwise take a century or more based on forecasted climate warming and gradual processes (Rey et al 2020).When ice-rich tundra with abundant near-surface groundice burns, vegetation and surface organic soil horizons are removed, deepening the seasonal active layer, leading to rapid thaw subsidence (thermokarst) and alteration of soil hydrology (Jones et al 2015, Michaelides et al 2019).In poorly draining areas, thermokarst promotes saturated, oxygen-poor conditions, favoring the conversion of freshly thawed permafrost carbon into greater proportions of CH 4 emissions, as opposed to CO 2 (James et al 2021, Walter Anthony et al 2021).Previous work linked this mechanism to extreme CH 4 emission hotspots in lakes and adjacent terrestrial environments across broad regions where thermokarst is most prevalent (Elder et al 2021, Walter Anthony et al 2021).Abrupt disturbances, like wildfire, mobilize permafrost carbon to the atmosphere as greenhouse gasses on magnitudes equal to the emissions predicted from gradual thaw alone in the coming centuries (Turetsky et al 2020).Despite the potential impact of disturbancedriven permafrost carbon emissions, they are not currently represented in Earth system models (Turetsky et al 2020); and few studies have quantified the effects of Arctic/boreal wildfire on gas exchange between the burned land and the atmosphere, especially for CH 4 emissions (Ribeiro-Kumara et al 2020, Moubarak et al 2022) and especially during post-fire succession.
Future wildfire occurrence is likely to increase as northern regions warm beyond critical temperature thresholds during the growing season (Higuera et al 2011, Sae-Lim et al 2019).Mechanistically, greater warming leads to more convective storms capable of generating cloud-to-ground lightning, the primary ignition source for tundra fire (Veraverbeke et al 2017).Of all circumpolar landscapes, the tundra biome will likely experience the greatest increase in lightning flash rate by the year 2100 (149 ± 72% increase relative to present-day rates) (Chen et al 2021).Greater lightning ignitions, coupled with the likely expansion of woody shrubs and lengthening fire seasons in tundra regions, point towards drastically different tundra fire behavior in the warmer future (Hu et al 2015).These anticipated changes will likely bear significant societal impacts to resident communities; namely those who rely on the tundra's ability to support lichen forage for reindeer populations (Jandt et al 2008).Tundra fire's combined societal and ecological consequences for the vast permafrost carbon reservoir necessitate a deeper understanding of the connections between fire, permafrost thaw, and greenhouse gas emissions across spatiotemporal scales.
Here, we used airborne imaging spectroscopy to investigate the influence of recent fires  on the generation of CH 4 emission hotspots in the Izaviknek-Kingaglia region of the Yukon-Kuskokwim (YK) Delta in western Alaska, USA (approximately centered on 61.335 • , −162.901 • ).Previous airborne in situ measurements (Chang et al 2014, Miller et al 2016) and modeling efforts (Zhu et al 2013, Peltola et al 2019) identified the YK Delta as a regional hotspot for CH 4 emissions; however, few studies have operated on spatial or temporal scales able to determine the influence of fire on process-level carbon dynamics and gas emissions (Ludwig et al 2022).In a scale-bridging approach, we utilized NASA's Next Generation Airborne Visible/Infrared Imaging Spectrometer (AVIRIS-NG) to map CH 4 hotspots at ∼5 m pixel resolution over a total imaged area of 1780 km 2 (∼70 million individual observations) in August 2018 (Miller et al 2022).Hotspot detections were overlaid on existing maps of historical fire burn scars and a high-resolution land classification map (Ludwig et al 2022), elucidating patterns of tundra fire influence on CH 4 hotspot generation.

Study area description
Our study area is an 18 km × 100 km box in the tundra of the YK Delta ∼85 km northwest of Bethel, AK, covering a series of tundra fire burn areas (figure 1).The YK Delta is characterized as subarctic tundra with discontinuous, ice-rich permafrost that currently hovers just below 0 • C, making it particularly vulnerable to widespread degradation via both abrupt and gradual disturbances (Pastick et al 2015).Instrumental records in Bethel, AK indicate the region has a mean summer (June-August) air temperature of 13.0 • C, mean annual air temperature of 0.6 • C, and mean annual precipitation of 500 mm (1990-2020) climate normal period, (NOAA NCEI 2022).While brackish marshes and low-lying tundra ecosystems dominate the coastal stretches of the YK Delta, our specific study region is characterized by low-gradient streams and fen networks that slowly drain upland peat plateaus (Ludwig et al 2022).Vegetation is predominantly a mixture of lichens, Arctic grasses, Sphagnum, and dwarf shrubs (Frost et al 2020).Open water bodies are abundant, covering roughly a third of the study area and ranging in size from a few m 2 to several km 2 (Ludwig et al 2022).Soil organic carbon contents remain understudied in the YK Delta, however, recent soil core measurements from the uppermost 1 m from the region show means of 72 ± 16.5 and 80 ± 14.2 kg C m −2 in unburned terrestrial peat plateaus and 2015 burn scars respectively (Ludwig et al 2018), a comparable amount to North Slope soils (Hugelius et al 2013).
The YK Delta has both direct and indirect records of infrequent fire occurrence dating back around 1000 years (Sae-Lim et al 2019).Around 35% of the 1780 km 2 study area has burned since 1971 (figure 1).The Aropuk Lake, Izavlknek River, and Kuka Creek burn scars from 2015 represent roughly 84% of the total burned area within our AVIRIS-NG imagery.Although recorded fires occurred as early as the 1950s in our general study area, only fires occurring during the 1970s, 1980s, 2000s, and 2010s overlapped with AVIRIS-NG imagery (figure 1).
Time series of satellite-derived Normalized Difference Vegetation Index (NDVI) show the expected positive or 'greening' trends following fire occurrence in our study area; however, unburned areas also show positive NDVI trends over the same period (Frost et al 2021).Post-fire succession is characterized by a decadal-scale recovery of plant functional diversity, with bryophytes and evergreen and deciduous shrubs dominating early successional stages (Frost et al 2020).

Airborne remote sensing of CH 4 hotspots
AVIRIS-NG imagery was acquired over the study region as part of the NASA Arctic Boreal Vulnerability Experiment (ABoVE) airborne campaign (Miller et al 2019) on 18 Aug 2018.The acquisitions consisted of six individual 3.0 km × 100 km flight lines which overlap to form a contiguous 18 km × 100 km map with a spatial resolution of ∼5 m × 5 m (figure 1).The flight lines were intentionally positioned to cover a chronosequence of tundra fire burn areas, including the large 2015 fires.The sampling strategy for ABoVE includes co-locating airborne measurements from multiple sensors, thus, the AVIRIS-NG data also overlap with the ABoVE airborne L-band synthetic aperture radar (SAR) acquisitions in the region (Miller et al 2019, Parsekian et al 2021); however, the SAR observations were not considered in this analysis.AVIRIS-NG radiance and reflectance data are available from the Oak Ridge Distributed Active Archive Center (Miller et al 2022).
Methane hotspots were retrieved from AVIRIS-NG imagery as previously described in detail (Thompson et al 2015, Elder et al 2020, 2021, Baskaran et al 2022).Briefly, AVIRIS-NG measured reflected solar radiation from 350-2500 nm with 5 nm spectral resolution and sampling (Hamlin et al 2011).The retrieval algorithm utilized in this study determines CH 4 enhancements as excess column CH 4 absorption (in integrated concentration path length units, ppm m) in the shortwave infrared (SWIR) wavelengths using a matched filter approach (Thompson et al 2015, Thorpe et al 2017).We define CH 4 hotspots as pixels with ppm m CH 4 enhancement greater than 3.5 standard deviations of the background spectrometer noise.For the purposes of this study, if a CH 4 hotspot is detected in a pixel, the entire 25 m 2 pixel is considered to have a methane enhancement, independent of surrounding pixels.The background noise level is a function of available photons, which relates to solar/observer geometry; for example, observing the shaded side of vegetation yields a darker background, weaker CH 4 signal, and noisier detections.To normalize any inconsistencies in the matched filter sensitivity caused by variations in illumination conditions across and within flight lines, hotspot detection thresholds were determined twice for each flight line.Specifically, individual flight lines were bifurcated lengthwise (split in the along-track direction) into subset swaths 'a' and 'b' and hotspot detection thresholds were determined on each bifurcation to remove bi-directional bias related to solar illumination and viewing angles.Methane hotspot threshold values were typically 2500-3000 ppm m, meaning that if the total column CH 4 enhancements in the air above hotspots were condensed to a unit pixel volume of 1 m height, the CH 4 concentration in that volume would be 2500-3000 ppm.In the absence of sun-glint, water has negligible reflectivity for the SWIR wavelengths used in our CH 4 retrieval algorithm; therefore, we did not analyze any CH 4 enhancements over open water surfaces.Since followup ground observations were outside the scope of this work, we consider all enhancements above the threshold as binary hotspot detections and do not attempt to further quantify hotspot magnitude or derive surface fluxes.
Elder et al (2021) demonstrated AVIRIS-NGbased detections of CH 4 fluxes in the 8000-20 000 mg CH 4 m −2 d −1 range from thermokarst wetlands in Alaska using ground-based, closed-chamber experiments.Since even strong CH 4 point sources quickly dilute to background levels over distances of 1-100 m (Thorpe et al 2016, Elder et al 2021), we attribute the CH 4 hotspots observed in our YK Delta study area to co-located emissions and not wind-transported CH 4 .
The spectral features of CH 4 in the SWIR wavelengths span a region that can also be affected by variability in surface reflectance.To mitigate this, we use several methods to prevent interference by surface reflectance features.First, the matched filter estimates the distribution of the background variability to exclude detections that lie within this distribution.Second, we estimate the surface reflectance using the method of Thompson et al (2018), and remove any CH 4 bias using a local linear model as in Elder et al (2021).Finally, we assess the burned area reflectances to see if there are measurable differences in the distribution of surface materials, such as bare soil, which could influence the detection rate.

Geospatial analysis
To assess the occurrence of CH 4 hotspots in tundra regions, a geospatial analysis was conducted using fire history, the location of surface water, and the CH 4 hotspot dataset.We used areal CH 4 hotspot occurrence ratios in per mil notation: ( R = n hotspot pixels × pixel area total area of interest × 1000 ) .
These ratios served as the primary metric for understanding how general landscape segments (unburned, burned, proximity to water, proximity to water within burn scars, etc; see table 1 for a full description of all general landscape segments) and specific land surface classifications (i.e.fen, degrading permafrost, tundra, etc) either promote or abate hotspot occurrence.Hotspot occurrence ratios were determined for each of the 12 swaths individually such that each swath was treated as an independent observation.This created some replication in areas where flight lines overlap (∼17% of the total area studied), however, the percent of burned area was similar between the overlapping and nonoverlapping regions (37% and 31%, respectively), effectively providing increased sampling, and thus, strengthening our statistical analysis in those regions.

Fire datasets and landscape partitioning
We synthesized readily-available data from the Alaskan Large Fire Database and the Monitoring Trends in Burn Severity products to determine fire history across the landscape in our YK Delta study domain.Merging these two products allowed for the best available fire history data for each fire event in the study domain.The Alaskan Large Fire Database (Kasischke et al 2002, Hrobak et al 2022) provides current and historical fire perimeter information for Alaska beginning in the 1940s and is maintained by the Alaska Fire Service.Additionally, the Monitoring Trends in Burn Severity (Eidenshink et al 2007, MTBS Data Access 2023) product provides Landsat satellite-based estimates of fire perimeters from 1984 to the present.The record of fire events in the Alaskan Large Fire Database provides critical long-term information on past fire activity, however, the more detailed spatial information available through the Monitoring Trends in Burn Severity program provides more accurate delineations for more recent fires.Combining these two datasets resulted in a detailed fire history for the YK region from the 1940s to 2018 (the year of the airborne imagery used in this study) where earlier fire events are incorporated from the Alaska Large Fire Database, and beginning in 1984, fire events from the higher-resolution (30 m) satellite-derived data product are included.
Presence-absence maps of fire history were developed using the burned and unburned areas of the fire history product.These maps were analyzed across each of the six bifurcated AVIRIS-NG flight lines (12 swaths total) to isolate burned and unburned areas by fire year and decade of fire.To account for differences in spatial resolution between the AVIRIS-NG dataset (∼5 m pixels) and the fire history map (30 m pixels) a 50 m buffer on either side of the fire perimeter (total distance of 100 m) was created based on the fire history presence-absence maps.This 100 m total distance (herein called the fire buffer zone) is greater than a Landsat pixel length that spans the boundary of the fire perimeter, accounting for uncertainty in the delineation of fire perimeters between the two sensors.
Proximity to water is an important regulator of CH 4 emissions hotspots (Elder et al 2020); therefore we determined CH 4 hotspot occurrence ratios in segmented regions near surface water bodies.Since water features were masked in the AVIRIS-NG CH 4 retrieval process, this layer was used to develop a water buffer zone around each water feature within each swath.As Elder et al (2020) found that a 40 m region near water bodies correlated to higher methane emissions, this same distance (40 m) was applied as a buffer in this analysis (herein called the water buffer zone).It is important to note that water often delineated the fire boundaries in our study area, thus our segmentation protocol was designed to test for potential compounding effects created by the nexus of fire and water buffer zones.
To assess the role different general landscape segments play in the formation of CH 4 hotspots, we segmented our study region by burned and unburned areas, proximity to water, proximity to the burn scar perimeter, and proximity to water within burned areas (defined in table 1).
The resulting 12 general landscape segments were represented in each of the 12 swaths.This allowed us to use an ANOVA and Tukey's Honestly Significant Difference test (α = 0.05) to determine differences in CH 4 hotspot occurrence ratios across the coarsest subdivision of the general landscape segments (summarized in table 2).We recognize that flight-swath-based subsampling of landscape segments is somewhat arbitrary relative to the natural ecosystem divisions on the landscape.Thus, we did not further subdivide the landscape segments given its potential to artificially inflate sampling statistics.
Time since disturbance plays an important role in post-fire succession (Johnstone et al 2016, Frost et al 2020) and could affect CH 4 emission rates.To test this, burned areas were grouped by decade and ANOVA was used to test for differences in the mean decadal CH 4 hotspot occurrence ratios.

Spatial bootstrapping of hotspot occurrence ratios
We also assessed the significance of CH 4 hotspot occurrence using a spatial bootstrapping approach.Since there was a large difference in spatial extent between landscape segmentations, e.g.much more area in unburned interior or fire scar than either of the water or fire buffers, we used bootstrapping to assess if this could lead to bias in occurrence ratios.For each of the six flight lines, we simulated hotspots for the entire region using the occurrence ratio from the unburned interior (without water) as a baseline.We used the stratifiedSample function in Google Earth Engine to randomly generate hotspots and repeated this 100 times using different random seeds.For each set of 100 simulated hotspots, we calculated the occurrence ratios in unburned interior, burned interior, water buffer areas, and burn scar perimeters, using the same approach by bifurcated swath on the simulated hotspots as was done with the observation dataset.We then compared the means, standard deviations, and 95% confidence intervals (CIs) of these bootstrapped hotspot occurrence ratios within each landscape category to the measured occurrence ratios.Hotspot occurrence ratios that overlap the bootstrapped 95% CIs indicate no significant difference between the landscape segment and the unburned interior, whereas landscape segments with hotspot occurrence ratios outside of the bootstrapped 95% CI are significantly different than the unburned baseline after accounting for their differences in area within each swath.

Landcover classification and relationships with hotspots
To evaluate CH 4 hotspot occurrence relative to more specific environmental settings, we also quantified hotspot occurrence ratios using a 5 m × 10 m resolution landcover classification (Ludwig et al 2023) independent of the general landscape segmentation described above.This landcover classification was originally developed to assess watershed-scale drivers of dissolved CH 4 concentrations and includes specific landcover types within unburned, recently burned, and older burned regions (Ludwig et al 2022).Individual resolution elements were classified using k-means and Sentinel-1 C band, Sentinel-2 multispectral, ArcticDEM, and derived layers, such as the Normalized Difference Wetness Index, as inputs.The landcover classifications include two types of unburned wetlands, recent and old burned wetlands, waterbody edges (shadowed banks, mud, ephemerally exposed), tundra vegetation types on unburned peat plateaus (including lichen, shrub, and sedge types), degrading permafrost on peat plateaus, tundra bordering degrading permafrost spots (often more saturated, moss dominated), old burned tundra on peat plateaus, and high and low severity recently burned tundra on peat plateaus.We clipped the six AVIRIS-NG flight lines to the area of overlap with the landcover classification and resampled the landcover classification at hotspot locations using the sampleRegions function in Google Earth Engine and the AVIRIS-NG scale (∼5 m).Hotspot occurrence ratios were calculated similarly to section 2.3 using the landcover classification areas within the flight lines and resampled classifications of hotspots.

Effects of fire on surface reflectance and CH 4 retrievals
Our assessment of reflectances in burned and unburned areas reveals no obvious artifact that could cause inadvertent CH 4 hotspot detections in burn scars.No single surface type or feature consistently differentiates the burned and unburned areas.However, one interesting trend is revealed by categorizing spectra by their NDVI into those that are thickly vegetated (NDVI >0.5), thinly vegetated or senescent (NDVI 0.35-0.5),and bare (NDVI >0.35).We calculated NDVI from AVIRIS-NG sensor data itself in order to characterize the observed land cover at the time of the overflight.The 'bare' category is typically water, mud, non-photosynthetic vegetation, partly-inundated pixels or shadow, and dark areas that might be noisier for CH 4 detection purposes.However, these surfaces are far more common in unburned areas, so they are unlikely to be responsible for the observed CH 4 hotspot occurrence ratios.

CH 4 hotspot data
AVIRIS-NG surveyed 1780 km 2 , but the total imaged area across six overlapping flight lines (12 swaths total) covered 2010 km 2 when including overlapping areas.The total imaged area consisted of ∼70 million pixels, where ∼1% were classified as CH 4 hotspots.Burned and unburned areas accounted for 36% (621 km 2 ) and 64% (1094 km 2 ) of the total terrestrial surveyed area (sans water surfaces), respectively.The fire buffer zone, a 100 m band straddling the fire perimeter, represents approximately 3.4% of the total surveyed area.Of the total CH 4 hotspots observed, 38% percent occurred in burned interior areas, 56% occurred in unburned exterior areas, and 6% were found in the fire buffer zone.
Mean decadal CH 4 hotspot occurrence ratios were not statistically different over time (p > 0.05) (figure S1).

Bootstrapping CH 4 hotspot ratios revealed no sampling bias
The mean CH 4 hotspot occurrence ratio from bootstrapping the baseline, unburned interior ratio was not different between landscape segmentation types within each flight line (as indicated by overlapping 95% CIs, figure S2), demonstrating that there is no sampling bias caused by the differences in the extent of landcover classifications within the flight lines.The variance in the bootstrapped ratio did increase in water buffer and fire perimeter areas, likely due to their relatively smaller area within the flight lines.As expected, the bootstrapped occurrence ratio for unburned interior areas was not different than the measured ratio (within the 95% CI e * Inclusive of all other landscape segments found within, including fire perimeter buffer zone areas.† Plotted in figure 2. NA = not applicable.The all-fire near water category includes shoreline buffers of waterbodies that were completely engulfed by fire. ‡ Letters in the bottom row represent the compact letter display output of the ANOVA post-hoc, pair-wise Tukey's Honestly Significant Difference test within the multicomp package in R (Hothorn et al 2008).If segments share the same letter(s), they are not significantly different from one another.
around the bootstrapped ratio).For burned water buffer segments, the measured hotspot occurrence ratio in all flight lines was significantly higher than the 95% CI around the bootstrapped baseline occurrence ratio from unburned interior regions (figure S2).Occurrence ratios in the fire perimeter were also significantly higher than the unburned interior ratio in five of the six flight lines.Occurrence ratios in fire interior segments were significantly higher than the bootstrapped unburned interior ratio in four flight lines.Occurrence ratios in unburned water buffer areas were also significantly higher than the bootstrapped baseline unburned interior ratio in three of the six flight lines, and significantly lower in the other three flight lines.

Hotspot occurrence in burned vs. unburned areas
We found that all swaths had significantly higher CH 4 hotspot occurrence in burned areas compared to unburned areas (p < 0.05) (figure 2).Specifically, we observed 17 CH 4 hotspots km −2 in burned areas on average, compared to 13 hotspots km −2 on average in unburned areas.Hotspots were 28.5% more likely (percent difference) on average to occur in burned tundra areas compared to unburned areas (table 2).Although hotspot occurrence was higher in burned areas overall, unburned areas were only statistically different from burned regions that fell within water buffer zones and the fire buffer zone more broadly (table 2).These findings provide strong evidence that burned areas play a significant role in promoting intense CH 4 hotspot emissions, especially in wet regions already prone to high emissions.

Fire and water buffer zones
Methane hotspots clustered along the fire perimeter.This effect was compounded where fire boundaries were delineated by water features.For example, all but two of the swaths had the greatest CH 4 hotspot occurrence ratio where the fire and water buffer zones overlapped.The two exceptions are apparent in lines 3b and 6b.Methane hotspots are approximately 62% more likely on average to occur within the fire buffer zone compared to unburned segments (table 2).And the greatest difference was observed between unburned segments and areas where the fire water buffer zones overlapped (87.1% difference).All segments where the effects of fire were combined with the presence of water were significantly distinct (ANOVA and Tukey's Honestly Significant Difference, α = 0.05) from all unburned segments and even burned areas without water (table 2).

Hotspot correlation to landcover surface classification
Despite wide variance between flight lines, there were distinct differences in hotspot occurrence ratios between landcover classification types (figure 3).Methane hotspot ratios were highest in recently burned wetlands (wetlands that burned in the 2015 wildfires) compared to all other landcover classifications.Ratios were also high in water edges in older burned areas (inclusive of all older fire scars), though this is a very small component of the landscape and absent from some flight lines, leading to large variance.Other landcover classifications with higher ratios included the edges of peat plateaus in recent and old burned areas, degrading permafrost in unburned areas, and wetlands in both unburned and old burned areas.

Discussion
In this analysis, the wetlands in recently burned areas showed the highest CH 4 hotspot occurrence (figures 2 and 3).This is consistent with previous research in our study area which showed that fire occurrence strengthens the influence of landscape and watershed variables (i.e.abundance of degraded peat plateaus and/or percent surface water) on higher concentrations of dissolved CH 4 in burn-adjacent water bodies (Ludwig et al 2022).Recent work has also revealed significantly higher dissolved NO 3 concentrations in burned areas delineated by streams (Zolkos et al 2022).This is consistent with the concept that where fire occurs in permafrost environments, the resulting changes to permafrost thaw depths, soil saturation, geomorphology, and carbon mobilization likely become the predominant regulators of CH 4 dynamics.

Tundra fire promotes CH 4 hotspots
The CH 4 hotspot ratios showed agreement across the two independent landscape segmentation schemes (general landscape segments (figure 2) vs. landcover classifications (figure 3)).For example, the general water buffer zone in burned areas coincides with the burned wetland landcover classification types, both of which had the highest ratios of hotspot occurrence.Wetlands are well-documented sources of high CH 4 emissions due to near-surface saturation and the likely presence of small, sub-pixel surface waters, both of which could contribute to hotspot occurrence.It is important to note that our analysis is limited by the accuracy of our water mask, which likely underrepresented small waterbodies and effectively under sampled water-affected boundaries.While quantifying this uncertainty is outside the scope of this work, we expect that a more accurate accounting of surface water in our analysis would only strengthen our findings.
The other burned classifications with high hotspot ratios, such as peat plateau edges, which often have a higher abundance of shrubs and burned more intensely, could explain the higher ratio counts in interior burn scars observed in some swaths.These plateau edges are also more likely to be included in the general water buffer segmentation, especially if other terrestrial-aquatic transition zones such as wetlands are thin.Plateau edges in the recently burned areas were highly disturbed, often slumping into lakes,

CH 4 hotspots imply greater total emissions
AVIRIS-NG observes CH 4 hotspots in areas of intense CH 4 flux, and it is improbable that lower-magnitude fluxes more commonly associated with the average behavior of northern wetlands are observable (Elder et al 2021).It is possible that areas that have fewer AVIRIS-NG-observed CH 4 hotspots can have higher total emissions if they are more spatially distributed and fall below AVIRIS-NG detectability.This is however unlikely since multiple studies suggest that where hotspot behavior is observed, those locations disproportionally dominate total emissions from the local area (Walter Anthony et al 2016, Serikova et al 2019, Thornton et al 2020, Elder et al 2021).Work by Elder et al (2020) also showed that AVIRIS-NGobserved CH 4 hotspots spatially scale in a similar way to diffuse (i.e.non-hotspot) landscape fluxes, implying a congruency in the environmental factors that regulate both hotspot and non-hotspot emissions.Furthermore, Baskaran et al (2022) showed that AVIRIS-NG CH 4 hotspot occurrence followed expected patterns with respect to water table distribution in lowlands vs. uplands across a broad area survey of the Mackenzie Delta, NWT, CA.All in all, greater CH 4 hotspot occurrence likely translates to greater total emissions.However, given our lack of groundbased measurements of CH 4 and wind conditions, and the uncertainties involved in translating AVIRIS-NG hotspots to fluxes, we did not attempt to quantify CH 4 emissions from our study area.Future studies on the effects of fire on tundra emissions would benefit from site-scale emissions observations like eddy covariance flux towers stationed in both burned and unburned tundra.

Future fires may accelerate the permafrost carbon feedback
While the interplay between fire frequency and intensity and their relationships to climate warming remains uncertain in the YK Delta, studies suggest a link between warming temperatures and the increasing frequency of fire in the region (Chipman et al 2015, Sae-Lim et al 2019).Arctic warming may result in a quadrupling of wildfire frequency in the YK Delta by the end of the century (Young et al 2017).This is concerning given that our work implies that the better-known positive warming feedback caused by concurrent tundra fire emissions (Moubarak et al 2022) is enhanced by successional, fire-promoted emissions which occur for years following the event.These notions, combined with the fact that the YK Delta experienced its two largest fires ever recorded during the writing of this manuscript in the summer of 2022 (Rosen 2022), motivate the need for further emissions monitoring both during and after tundra fire.

Conclusions
We used airborne hyperspectral imagery to survey 1780 km 2 of tundra in the YK Delta which has experienced repeated wildfires since the 1940s.We mapped CH 4 hotspots with ∼5 m spatial resolution across this domain and correlated them to individual burn events, burned scar perimeters, proximity to water, and other landscape categories.We found that CH 4 hotspots are roughly 29% more likely in recently burned tundra compared to unburned areas and that this effect is roughly tripled near burn scar perimeters delineated by surface water features.In fact, four of the top five CH 4 hotspot occurrence ratios were found in landcover classifications related to previously burned landscapes (figure 3).Our results imply that the successional changes following tundra fire favor the environmental conditions needed to generate CH 4 emissions hotspots, a positive feedback to the climate system, accelerating future warming, tundra fire occurrence, and permafrost carbon loss to the atmosphere.
The ABoVE airborne campaigns have collected airborne hyperspectral imagery over other tundra fire burn perimeters: near Kougarok on the Seward Peninsula (Liljedahl et al 2007, Iwahana et al 2016), in the Noatak River Valley (Loboda et al 2013, French et al 2015, He et al 2021, Masrur et al 2022), over the Anaktuvuk River burn scar, (Rocha and Shaver 2011a, 2011bJones et al 2009, 2015) and the 2022 Contact Creek fire near King Salmon, AK.We plan to extend the CH 4 hotspot analyses to these disturbances.The Kougarok area, in particular, offers extensive ground sampling data for comparison and validation (Hollingsworth et al 2021).Validation of the CH 4 fluxes from detected hotspots continues to be an area of active research (Elder et al 2021).

Figure 1 .
Figure 1.Top: Yukon-Kuskokwim Delta study region with six flight lines of AVIRIS-NG RGB imagery overlay.The red box in the inset map shows the location of the study region within Alaska.Bottom: a detailed view of YK Delta study area showing CH4 hotspots (yellow dots) amongst the complex interface between lakes (blue), fire buffer zones (hashing), combined fire/water buffer zones (purple), a 2015 burned region (deep red), a burn from the 1970s (orange), and unburned areas (light green) all within a few hundred meters.

Figure 2 .
Figure 2. CH4 hotspot occurrence ratios grouped by unburned (inclusive of all categories within, green bars), unburned within 40 m of water (water buffer zone, blue bars), burned (inclusive of all categories within, red bars), and fire buffer zones both overlapping water buffers (purple bars) and not proximal to water (grey bars with hashing) in each bifurcated AVIRIS-NG swath.The means of each category are plotted as horizontal lines of corresponding color.CH4 hotspot occurrence is enhanced in fire and water buffer zones across all swaths.

Figure 3 .
Figure 3. Mean and standard deviation of the flight line CH4 hotspot occurrence ratios by landcover classification type.Landcover classification types are unburned, unless stated otherwise.

Table 1 .
Terms and corresponding descriptions of general landscape segments.Terrestrial surface area inside the fire history map where it does not overlap with water buffer zone Unburned interior without water Terrestrial surface area outside the fire history map where it does not overlap with water buffer zone