Recent massive expansion of wildfire and its impact on active layer over pan-Arctic permafrost

Wildfire is recognized as an increasing threat to the southern boreal forests and the permafrost beneath them, with less occurring over the cold continuous permafrost than before. However, we show that continuous permafrost was a major contribution to wildfire expansion in the pan-Arctic over the last two decades. The expansion rate of burned area over continuous permafrost was 0.9 Mha decade−1, in contrast to a decreasing trend (−0.5 Mha decade−1) over the entire permafrost areas. Burned area has been rapidly growing in the north of the Arctic Circle in particular, where the total burned area in the major fire seasons during 2011–2020 nearly doubled that during 2001–2010. Wildfire expansion is closely linked to an increased soil moisture deficit, considering wildfires there combust more than 90% of belowground fuel. Continuous permafrost experiences more severe fire-induced degradation. Active layer thickening following wildfires over continuous permafrost lasts more than three decades to reach a maximum of more than triple the pre-fire thickness. These new findings highlight the massive expansion of wildfires over continuous permafrost, which can dramatically modify ecological processes, disturb organic carbon stock, and thus accelerate the positive feedback between permafrost degradation and climate warming.


Introduction
Permafrost is essential in maintaining the stability of Arctic ecosystems, covering about 24% of the Northern Hemisphere land area, in which continuous permafrost (90%-100% permafrost coverage) occupies about half of the area [1]. The permafrost soils store ∼1700 Pg organic carbon [2], twice more than the carbon in atmosphere [3]. Permafrost is sensitive to the changing climate, especially the rising air temperature because the atmosphere in high latitudes has warmed faster than elsewhere on Earth since the 1980s [4]. Among them, warming over continuous permafrost is faster than in the other permafrost zones [5]. As permafrost regions experience widespread and persistent warming, soil carbon is vulnerable to rapid permafrost thaw [6]. Carbon released from thawing permafrost exacerbates a positive feedback, thus further enhancing the temperature rise [7]. Furthermore, permafrost degradation can have serious consequences for hydrological cycle, land cover and infrastructure integrity [8]. Therefore, thawing of permafrost is considered one of the tipping points in the changing climate, and likely to trigger irreversible changes of frozen soil and vegetation, and dramatic release of carbon dioxide and methane [9].
Wildfire is a major driver of permafrost thaw [10]. Wildfires remove the insulating organic layer consisting of vegetation and surface organic material [11]. When vegetation is replaced by black char, ground heat flux increases due to increased net radiation at the surface following the reduced albedo and loss of shading from trees and shrubs [10,12]. As a result, the active layer is deepened, and taliks can be formed in the thawed soil layers [13].
Wildfire-induced thaw settlement on ice-rich permafrost may result in surface ponding, which facilitates development of thermokarst [14]. Because of the superficial root system of Arctic vegetation, even a light-intensity fire can cause serious damage to vegetation in permafrost regions [15]. After a severe burn, it can take hundreds of years for coniferous forests to recover, or they can never be recovered due to expansion of broadleaf trees driven by wildfires and warming [16,17]. Therefore, the resultant permafrost thaw depends on not only burn severity but subsequent changes in thermal properties of land surface due to loss or degradation of vegetation and organic layers [13,18].
Soil carbon is vulnerable to wildfire disturbance in carbon-rich permafrost landscapes [19]. Considering the massive carbon stored in permafrost zones, wildfires, especially with underground combustion, can result in high carbon emissions [20]. In addition to direct carbon emissions from organic matter combustion, wildfires over the permafrost may alter soil carbon fluxes for decades because wildfires modify soil physical and chemical properties, and decomposition processes [21]. As wildfires become more frequent and severe in boreal regions [22,23], the released carbon induced by wildfires from frozen soil may alter the global carbon budget and accelerate global and Arctic warming, thereby forming a positive feedback between permafrost degradation and climate warming [24,25]. Under a warming climate, wildfires can have a compounding effect with an accelerated temperature rise and permafrost thaw [26,27].
It has been reported that wildfires in the subarctic and boreal regions, such as the North American boreal regions [28] and Siberia [29], are increasing in both frequency and intensity due to climate warming, and the growing wildfire risk may continue by the end of this century [30]. These regions in the southern permafrost zones are naturally fire-prone and receive more attention [14], while knowledge gaps of wildfires and their impacts over different permafrost zones remain. In this study, we aim to explore the difference between wildfire variation and wildfireinduced active layer thickness (ALT) change, as well as wildfire drivers in different Northern Hemisphere permafrost zonations over the past two decades. With multiple remote-sensing datasets combined with field measurements, we show that recent fire expansion mainly occurs in the continuous permafrost region, which is also experiencing a more serious permafrost degradation process.

Study region
We analyzed wildfires in the permafrost zones in the Northern Hemisphere excluding the Tibetan Plateau permafrost region where the occurrence of wildfire is negligible compared to that in the high latitudes. Continuous, discontinuous, sporadic and isolated permafrost zones are defined with 90%-100%, 50%-90%, 10%-50%, and 0%-10% permafrost coverage [1], respectively (figure S1). In the study region, the area of continuous, discontinuous, sporadic and isolated permafrost is 964 Mha, 287 Mha, 335 Mha and 493 Mha, respectively. We resampled the permafrost map from 1 km to 0.25 • spatial resolution using Nearest-Neighbour Interpolation.

Analysis of wildfire trends in permafrost zones
We used three wildfire datasets to analyze temporal changes of burned area in permafrost zones. The latest burned area product developed by the European Space Agency Climate Change Initiative (FireCCI51) is based on the combination of Moderate Resolution Imaging Spectroradiometer (MODIS) highest resolution (250 m) near-infrared band and active fire information from thermal channels over 2001-2020 [31]. The monthly total burned area is then aggregated into a regular grid at 0.25 • resolution. The AVHRR-LTDR Burned Area Grid product is derived from Advanced Very High Resolution Radiometer (AVHRR) Land Long Term Data Record (LTDR) dataset v5 [32]. It provides a monthly burned area at 0.25 • resolution from 1982 to 2018, in which the year 1994 is omitted due to lack of input data. The AVHRR time series have been used in boreal regional studies to derive the burned area time series, e.g. Canada [33] and Russia [34]. It can well capture the burned area trend according to the Canada national fire perimeters [32]. The latest MODIS Global Burned Area Product (MCD64) is generated from 500 m MODIS imagery coupled with 1 km MODIS active fire observations [35]. We used MCD64CMQ monthly grid burned area based on MCD64A1 algorithm with a 0.25 • resolution. MCD64A1 algorithm is more sensitive to small and moderate burns with less uncertainties in burned area [36].
The annual total burned area was calculated in each 0.25 • for FireCCI51, AVHRR and MCD64. We used these three datasets to analyze the trends of burned area after 2001. The annual burned area of FireCCI51 (2001-2020) was further used to analyze the temporal variation of burned area and its drivers in the continuous, discontinuous, sporadic and isolated permafrost subregions. We used Theil-Sen slope analysis [37] and the Mann-Kendall test [38,39] to investigate the temporal trends of annual burned area. These non-parametric trend estimators can reduce the adverse effects of anomalous values and have been widely used in geographical and climatic time-series data [40].
Given that remote sensing datasets tend to underestimate wildfire activity especially in high latitudes [41][42][43], we evaluated the datasets used in this study with other studies on wildfires in high latitude regions. The burned area in 2020 Siberian Arctic (latitudes > 66.5 • N) was estimated between 1.71 and 2.62 Mha [23]. Burned area from FireCCI51 in this study was 2.39 Mha over Siberian Arctic in 2020 (figure S4(a)), within the burned area range retrieved from Sentinel-3 (C3SBA10) which is identified as one of the most accurate products in representing burned areas of Siberian wildfire in 2019 and 2020 [23]. The mean burned areas of FireCCI51 during the growing season (May-September) of 2001-2014 in continuous and discontinuous permafrost were 2.70 and 1.58 Mha, respectively ( figure S4(b)), which about doubles the burned area derived from an early version MODIS burned area product (MCD45A1) [44]. This is because FireCCI51 has an advantage to detect small fires leading to higher global burned area than other products [31]. Moreover, the annual burned area of Eastern Siberia derived from FireCCI51 is highly correlated (r = 0.92, p < 0.001) with the 30 m resolution Landsat fire data during 2001-2020 [42] demonstrating that FireCCI51 can capture interannual variations in burned area (figure S4(c)).

Examine key drivers of wildfire change in permafrost zones
Soil water deficit can reflect the aridity of belowground fuels and is closely related to the occurrence and spread of wildfire. We used the Keetch-Byram drought index (KBDI) to evaluate the soil water deficit [45]. The KBDI index, representing a moisture regime from 0 to 20 cm of the soil layer, infers meteorological conditions favorable for the start and spread of fires. It has been demonstrated that KBDI is linked to the moisture content of fuel, i.e. higher values of KBDI (lower soil moisture) corresponds to lower fuel moisture and increasing flammability [46]. KBDI is calculated as [45]: where Q and Q 0 are the current and previous day KBDI, respectively, dQ is the KBDI incremental rate, T is the daily maximum temperature at 2 m above the ground (degrees Fahrenheit), dP is daily precipitation, R is the mean annual rainfall (inches) and dt is a time increment set to one day. Build-up index (BUI)is a component of Fire Weather Index (FWI) system [47]. The BUI accounts for moisture levels in the fuels tracked by two components from FWI, i.e. Duff Moisture Code and Drought Code. BUI is also used by forest fire management agencies as an indicator of the potential difficulty in extinguishing smouldering fires, or the tendency of a fire to remain smouldering deep in the ground or in large woody materials [48]. Higher BUI indicates a higher potential for smouldering fires. BUI is calculated as [47]: We used Initial Spread Index (ISI) to represent surface fuel availability. The ISI integrates the moisture content of surface fuels defined by the Fine Fuel Moisture Code (a component from FWI) and wind speed to be an indicator of the potential rate of spread of fires [48]. ISI is calculated as [47]: (4) where W is 10 m open-wind speed, m is fine fuel moisture content in percent. We obtained the monthly KBDI, BUI and ISI over 2001-2020 from the fire danger indices historical data of Copernicus Emergency Management Service [49]. The KBDI, BUI and ISI were calculated using the data generated from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis at 0.25 • [50].
Vapor pressure deficit (VPD) was calculated to infer atmospheric aridity as follows [51]: where SVP is the saturation vapor pressure (pa), T is 2 m air temperature ( • C), and RH is the relative humidity (%). The monthly 2 m (m) air temperature and surface air relative humidity at 0.25 • during 2001-2020 were obtained from ERA5 global reanalysis dataset. To reveal the relationship between wildfire and climate conditions, we performed linear regression between annual burned area and annual mean KBDI (and VPD) over 2001-2020 in different permafrost zones. The significance level of the correlations between annual burned area and aridity indices in this study were evaluated with the standard twotailed Student's t-test. We collected ground observational data from the Arctic-Boreal Vulnerability Experiment (ABoVE) to analyze the fuel combustion in permafrost zones. ABoVE is for field-based process-level studies and provides a foundation to improve understanding of ecosystem responses to climate change in the Arctic and Boreal region [52]. We collected the proportion of the below-ground component to the total carbon combustion of 701 site-years from ABoVE, in which 234 is for continuous, 343 for discontinuous, 102 for sporadic and 22 for isolated permafrost [53][54][55]. Surface biomass combustion mainly occurred in the top 20 cm of these sites.

Analysis of permafrost degradation after fire
The ALT of 71 site-years was obtained from 14 studies to analyze permafrost degradation after wildfire (table S1). These studies provide pre-fire and postfire ALT, or ALT change between post-fire and prefire, and the number of years from the year of fire to the year when the ALT was obtained. We first identified their location in different permafrost zones, and then calculated the relative change of ALT using prefire ALT and ALT change. We also calculated the average of ALT relative change over 0-5 years, 5-10 years, 10-20 years, 20-30 years and 30-40 years since the fire. A multiple regression was applied to the ALT relative change over the five periods to analyze the temporal change of ALT after fires in different permafrost zones.
The Circumpolar Active Layer Monitoring (CALM) observational network [56] is one of the global networks to monitor permafrost from the Global Terrestrial Network for Permafrost. We screened two cases with recorded historical wildfire and continuous pre-and post-fire ALT measurements from CALM program to show the ALT change after fire (figures 4(b), (c) and S1

Wildfire expansion over continuous permafrost
In the last two decades, the mean annual burned area in permafrost zones was about 13.65 ± 4.45 Mha. The burned area was 2.92 ± 1.69 Mha over continuous permafrost, 1.44 ± 0.93 Mha over discontinuous permafrost, 3.88 ± 1.76 Mha over sporadic permafrost and 5.41 ± 1.99 Mha in isolated permafrost ( figure 1(a)). The mean annual burned fraction over continuous permafrost is about 0.3%, lower than that of discontinuous permafrost (0.5%), sporadic permafrost (1.2%) and isolated permafrost (1.1%) ( figure 1(b)). The low burned fraction is consistent with low atmospheric and soil aridity ( figure 2) implying a naturally low wildfire risk over continuous permafrost relative to the other permafrost zones.
The increase in wildfires over continuous permafrost was a major contribution to recent wildfire expansion in the pan-Arctic region. Although the mean annual burned area over the entire permafrost zones decreased at a rate of 0.5 Mha per decade, the burned area over continuous permafrost increased at a rate of 0.9 Mha per decade during 2001-2020, and nearly 59.7% area of continuous permafrost with wildfire change experienced wildfire expansion (figure 1(c)). The greatest increase in burned area occurred over the Eastern Siberian continuous permafrost, where the growing rate of burned area reached over 20% per decade ( figure 1(c)). This increasing trend of wildfires over the continuous permafrost is captured by the three datasets, while the burned areas in the other three permafrost zones almost all decreased ( figure 1(d)). The burned area has been rapidly growing in the past few years in the north of 70 • N, where the total burned area in the major fire season during 2011-2020 nearly doubled the burned area during 2001-2010. In 2020 alone, the burned area in the north of 70 • N was about 0.13 Mha accounting for about 40% of the total burned area from 2001 to 2020 in this region, which implies that more areas of the cold northern range of permafrost are affected by wildfires lately.

Growing aridity over continuous permafrost
Changes in permafrost wildfires are related to the belowground and aboveground fuel availability and aridity, which can be reflected by belowground fuel availability (BUI) and aridity (KBDI), and aboveground fuel aridity (VPD) and fire spread rate (ISI) (figure 3). The continuous permafrost experienced significant KBDI growth (0.12 decade −1 , p < 0.05) and BUI growth (0.97 decade −1 , p < 0.01) over 2001-2020, while KBDI and BUI in other permafrost subregions did not show significant increase (figures 3(a) and S3(c)). Compared to KBDI and BUI which tended to significantly increase only in continuous permafrost, VPD and ISI significantly increased in all the permafrost zones. Among them, continuous permafrost experienced the most VPD increase (8.8 pa decade −1 , p < 0.01) and the least ISI increase (0.06 decade −1 , p < 0.01) (figures 3(b) and S3(d)).
The interannual variation of burned area over continuous permafrost is mainly linked to KBDI (r = 0.75, p < 0.01) and BUI (r = 0.46, p < 0.05) rather than VPD (r = 0.33, p = 0.15) and ISI (r = 0.28, p = 0.23) ( figure 3(d)). These results indicate interannual variations of burned area are more closely related to belowground fuel aridity because wildfires in permafrost zones mainly combust belowground carbon, thus rely strongly on moisture conditions of belowground fuel. In the continuous permafrost zone, belowground carbon combustion accounts for up to 91 ± 8% of the total carbon combustion (figure 3(c)). In other permafrost zones, the ratio of belowground carbon combustion is usually below 90% (figure 3(c)), and neither the atmospheric aridity nor soil moisture deficit is significantly correlated with interannual variation of burned area (figure S2).

Wildfire-induced permafrost degradation
Wildfire-induced permafrost degradation is marked by increased ALT. The ALT increases for years to decades after wildfires (table S1). For discontinuous permafrost, the ALT increases immediately after wildfire, which lasts for about a decade to reach its maximum thickness, being about 2.25 times thicker than its prefire status. It then takes another three to four decades for the ALT to gradually recover to the pre-fire status. For the continuous permafrost, the ALT gradually increases after fire. The increase of ALT can last for several decades. The post-fire maximum ALT in continuous permafrost can be 3-4 times thicker than its pre-fire status and it may take much longer time for ALT to recover to pre-fire status according to the very limited pre-and post-fire ALT measurements in continuous permafrost ( figure 4(a)).
The two sites with recorded historical wildfires showed remarkable differences in ALT changes between pre-and post-fire (figures 4(b) and (c)). At the Bonanza Creek LTER site, a burn event occurred in 2010. Before the fire, ALT increased at a rate of 0.64 cm yr −1 (R 2 = 0.60). After the fire, ALT increased at 17.66 cm yr −1 (R 2 = 0.97), about 28 times faster than the pre-fire increase rate ( figure 4(b)). ALT increased from an average pre-fire value of 55 cm to its recorded maximum value of 244 cm in 2018. At the Cosmostation site in the northern Tian Shan, within the southern edge of permafrost zones, the average pre-fire ALT was about 491 cm, much deeper than that at the Bonanza Creek LTER site. Two years after the fire, ALT reaches its maximum of 688 cm ( figure 4(c)). Moreover, the changes of land cover before and after the fire event surrounding the Bonanza Creek LTER site can be visualized from the satellite images. The vegetation was destroyed and bare land was exposed after wildfires (figure S5).

Discussion
In this study, we found areas of continuous permafrost have experienced rapid wildfire increases compared to the decreased trends in other permafrost subregions during the last two decades. The increased wildfire risk is expected to continue in continuous permafrost by the end of the 21st century with warming of 2 • C-5 • C in the boreal zone [30,58]. In permafrost regions, wildfires are usually initialized by lighting [29]. In continuous permafrost particularly, 90% of the burned area in the Interior Alaska and over 80% of burned area in northwestern Canada were initialized by lighting over 1975-2015 [59]. In this study, however, we found the burned areas within the Arctic Circle mainly increased during April-June in the last decade. This increase is more likely due to the growing overwinter fires in boreal regions [60,61]. The overwinter fires mainly emerge in early spring, earlier than occurrence of lightning-induced fires in the summer [62].
We found wildfires significantly increased across Siberian continuous permafrost. Siberia has been reported to be undergoing expansion of wildfire since the 21st century, and occurrence of fires has migrated northward to the Siberian Arctic in the last decade [29]. The years 2019 and 2020 experienced anomalously high burned area in northeastern Siberia [63], which accounted for about 44% of the total burned area in Siberian Arctic from 1982 and 2020 [23]. The belowground combustion is more important in northern boreal forests compared with southern stands [20]. According to the studies in Siberia, the fraction of belowground to the total combustion was about 65% in Siberian pine forests and 75% in larchdominated forests [19]. The fraction was about 70% in larch forests of northeastern Siberia underlain by continuous permafrost [64], and about 51% in darkconiferous forests of central Siberia, underlain by sporadic and isolated permafrost [65], demonstrating higher fraction of belowground combustion over continuous permafrost as measured at ABoVE sites ( figure 3(c)). It has been reported that the number of available field measurements on carbon combustion from wildfires in Arctic-boreal Eurasia is one order of magnitude smaller than that in North America [19]. Therefore, we considered the ABoVE data as a reliable proxy to represent the combustion pattern in the entire permafrost region.
Northeastern Eurasia, underlain by continuous permafrost, is dominated by deciduous larch (light taiga) accompanied with continental climate, sparser forests and relatively frequent burns with lessintensity, especially in the Far East [66]. Compared to the North American boreal forests (e.g. black spruce and jack pine) which have evolved to spread and be consumed by severe and intense fires as part of their life cycle, the Eurasian boreal forests have evolved to resist and suppress severe fires and turn them into low-intensity fires [67]. Therefore, due to the distinct adaptation strategies of tree species to fires [68], the Siberian belowground combustion ratio is lower than that of North America (89.9% combustion from belowground in black spruce forests [19]) due to less intense fires. However, the deciduous needle-leaf forests, which dominate in Siberian continuous permafrost, provide thick organic layer and much fuel loading due to annual litter fall and low decomposition [58]. Wildfires are more likely to increase when meteorological and fuel moisture conditions in fire season are worse in Eurasia [67].
We found that burned area changes over continuous permafrost are closely linked to belowground aridity inferred by KBDI, because wildfires over continuous permafrost are mainly attributed to belowground combustion ( figure 3(c)). Higher values of KBDI correspond to lower fuel moisture and increased belowground flammability [69]. The decrease in soil moisture reflected by KBDI ( figure 3(a)) has a strong influence on the combustibility of fuels, resulting in an increase in fire risk [70]. As the aridity of the fuel increases, so does the availability of belowground fuel inferred by fuel BUI. However, BUI cannot infer the interannual variation of burned area as good as KBDI. This is because BUI is a relative indicator of fuel available for combustion by a moving flame front rather than the actual fuel load on the ground [48].
Increasing Siberian wildfires are closely related to the continued warming, drying and lengthening of growing seasons [42], and the growing water deficit [23]. When fuel loading is accumulated through the lengthened growing season, the warmer and drier conditions facilitate wildfire development [71]. When the Arctic warms, the enhanced evapotranspiration tends to significantly decrease surface soil moisture [72], thereby leading to increased belowground fuel aridity. Considering the original low flammability of continuous permafrost (figure 2), the rapidly growing aridity of surface and atmospheric conditions facilitates fire burn and spread. Under a relatively dry background condition in the south, e.g. sporadic and isolated permafrost zones, the burned area is naturally high. However, the interannual variations in burned area are not well correlated with aridity ( figure S2). This may be because fire ignition and spread there are more regulated by human activities [62,73]. In southeastern Siberia, human activities cause more than half of the fires [74].
Wildfires can bring about serious consequences to the permafrost, i.e. rapid increase of post-fire ALT as we found, and these consequences can be amplified when wildfires increase under climate warming. Wildfires can result in the development of thermokarst in ice-rich permafrost settings [14]. The occurrence of thermokarst leads to melting of massive ice, causing ground thaw and surface subsidence [15]. When wildfires damage vegetation, surface organic and moss layers, soil thermal properties and heat transfer are modified, exerting substantial impact on permafrost temperature and increasing ALT [13] (figure 4). As shown in figure S5, permafrost landscape cannot recover to its pre-fire status eight years after the fire. For a severe burn, it even takes 50-200 years for black spruce to recover to pre-fire conditions [15]. Model simulations revealed that the permafrost extent in the Canadian subarctic region will be reduced from 67% at present to 2% by the end of this century due to climate warming, and wildfires will contribute to additional reduction of permafrost extent and accelerate disappearance of permafrost [27].
We found continuous permafrost undergoes the greatest and longest change of ALT after fire ( figure 4(a)). As overwinter fires are increasing in the boreal regions [60], especially in the eastern Siberia [62] covered by a large area of continuous permafrost, these smoldering fires cause more organic layer combustion and affect the permafrost thermal regime [75]. Fire effects on the active layer increased significantly after 1990 due to climate warming [27]. The synergistic effects of climate warming and changing fire regime will lead to long-term impact on ecological successions, and the post-fire secondary plant succession may take more than 100 years [15]. Permafrost shows resilience to wildfires, i.e. permafrost can recover to pre-fire conditions after decades, which sustains the northern ecosystems [16]. However, resilience can be compromised by the synergistic effects of climate warming and wildfires. Less resiliency near the southern limit of permafrost is due to the ongoing climate warming, increasing wildfires and common vegetation succession [14].
In conclusion, we found that increased wildfires over continuous permafrost regions was a major contribution to wildfire expansion in the Arctic in the last two decades. The increase in wildfires over continuous permafrost was linked to a growing soil moisture deficit and belowground fuel availability because over 90% of the combustion occurs belowground. Wildfires cause severer and longer-lasting post-fire degradation of continuous permafrost than the other permafrost zones. Under a warming climate with increased wildfires, permafrost degradation will be undoubtedly accelerated. The warming and permafrost thaw will continue in response to climate warming and wildfire expansion, especially in the cold continuous permafrost region, where temperature rise is particularly amplified.

Data availability statement
All satellite, reanalysis and site data used in this study are publicly available under the following URLs: The www2.gwu.edu/∼calm All data that support the findings of this study are included within the article (and any supplementary files).