The Yukon-Kuskokwim Delta in southwestern Alaska is located between the Yukon and Kuskokwim Rivers within a region characterized by ice-rich discontinuous permafrost . It is a subarctic lowland marked by numerous ponds and lakes, with deposits composed of silt, sand, gravel, and organic materials . The Yukon-Kuskokwim Delta is the warmest region in our study, with average ground temperatures at 2 m depth of approximately 1.3 °C in the areas encompassing the fire scars . The investigated fires lie within the Yukon-Kuskokwim uplands, which are part of the inland areas of the delta , an area with a long history of tundra fires and relatively warm permafrost . Here, wildfires are predominantly fueled by lichens, which dominate the gentle, rolling hills of this region alongside shrubs and sedges .

The Inuvik study region, located east of the Mackenzie Delta, lies in the zone of continuous permafrost and is dominated by glacial sediments . It is predominantly composed of a mixed cover of trees and other natural vegetation, interspersed with patches of shrubs and herbaceous vegetation . With average ground temperatures at 2 m depth of about -3.9 °C, this region represents the second coldest site in our study .

The Noatak study region in northwest Alaska comprises the Noatak River Basin, which lies within the Noatak National Preserve at the western edge of the Brooks Range. The valley of the Noatak River is situated between the DeLong Mountains to the north and the Baird Mountains to the south and is characterized by predominantly low shrub or tussock tundra and underlain by continuous permafrost . Mean annual ground temperatures are approximately -2.4 °C in the vicinity of the fire scars , which primarily occur in low-lying areas near the Noatak River , in more productive vegetation types .

The study region in the Central North Slope of Alaska includes the upland areas at the Brooks Foothills (Yedoma permafrost region), situated between the Brooks Range to the south and the Beaufort Coastal Plain to the north, and is underlain by continuous permafrost . This region is characterized by silt, rocky silt and peat lithology, and it is dominated by shrubby tussock tundra and low shrubs . Here the lowest mean annual ground temperatures at 2 m depth are found, with approximately -6.5 °C . The most prominent fire of this region is the Anaktuvuk River Fire, which burned an area of about 1000 km² in 2007 .

The Sentinel-1 satellites, operated by ESA, are equipped with C-band SAR sensors at a wavelength of 5.6 cm. The first Sentinel-1 satellite was launched in 2014, followed by the second in 2016, which ceased operations at the end of 2021. An additional Sentinel-1 satellite was launched in December 2024. While the dual-satellite constellation theoretically enabled a 6 d repeat cycle, the typical interval was 12 d due to the mission’s global observation plan. The data utilized in this study was obtained in VV polarization using the Interferometric Wide (IW) swath mode, offering a 2.3 m × 13.9 m (range × azimuth) ground sampling distance for 250 km-wide swaths.

For InSAR analysis, data from the thawing season were used, starting once interferogram coherence became satisfactory after snowmelt and ending when the freezing period began, as determined by ERA5 data. For backscatter analysis, all available data from the month of August were employed. To investigate winter backscatter, only data from November to January within the approximate temperature range of -1 to -15 °C were considered, due to the temperature sensitivity of the measurements . Additionally, dates impacted by rain-on-snow events were excluded .

PALSAR-2 aboard JAXA's ALOS-2 satellite is a L-Band SAR sensor (24 cm wavelength) and was launched in 2014. Data used in this study was acquired in ScanSAR mode in HH polarization, with single look pixel spacing of about 9 and 3 m in range and azimuth and an applied nominal resolution of about 23 m. ALOS-2 has a revisit time of 14 d; however, images are not acquired regularly, so the selection of available images is much more limited. Depending on the study region, data were used either from the beginning of the thawing season (Inuvik region and Noatak River Basin), from the end of the thawing season (Central North Slope) or August to September (Yukon-Kuskokwim Delta). For each region, the dataset covers a timespan of at least three years, with the dates of the observations listed in Table .

The Alaska large fire database includes both current and historical wildfire data starting from 1940. The location and perimeter of fires are mapped using three techniques: ground surveys, airborne surveys, and interpreting aerial photography. For the Central North Slope, the oldest fire scar included in this study was the Shivugak Bluffs Burn described in , which was identified in a 1948 aerial photograph. Due to the extent of thermokarst development, the fire was suggested to have occurred some decades earlier. Therefore, an approximate fire year of 1930 was assumed here. For Inuvik, the Fire History dataset from the Government of the Northwest Territories (NWT Centre for Geomatics) was used, with recorded fires starting in 1950. Data sources include GPS, MODIS, Landsat, and Sentinel-2.

To allow for a comparison with pre-fire conditions each fire scar was compared to its unburned surrounding area by calculating the difference between the median values of the fire scar and the median of the adjacent area, which was defined as a 5 km buffer around the scar, excluding all recorded fire scars within the buffer. This method additionally helps to counteract spatially limited phenomena such as insufficiently corrected atmospheric or ionospheric effects, or poorly chosen reference points in the case of InSAR, as well as differences in vegetation cover and prevailing moisture conditions in the case of backscatter analysis. Areas with multiple fire events were excluded. Results were masked for terrain slopes less than 5° to exclude areas potentially affected by mass movements, such as solifluction . For this, the Copernicus DEM was used. Furthermore, for the Noatak River Basin and Central North Slope, areas along river courses with significantly different InSAR signals and γ0 values were masked. Land cover information from CALU was used for this purpose. The classes “partially barren” and “shallow water/abundant macrophytes” were masked with a 100m buffer for large polygons. In addition, CALU water classes were used to mask out open water areas, ensuring that no open water pixels remained in the datasets. InSAR and backscatter results were masked identically to ensure comparability. Fire scars were selected based on their masked 5 km buffer areas, which mostly displayed homogeneous results.

The results were analyzed with respect to the time since initial burn, with multiple results per year averaged per region, and functions were fitted to the data for each study site. For thaw season deformation rates (-αDDT), an exponential model was applied to account for the decrease in thaw season deformation rates over time. To evaluate the annual deformation rate, both exponential and damped oscillation models were tested, as proposed by and , respectively. In contrast, backscatter values exhibited a more linear trend and therefore a linear model was applied in this case. These comparisons were also conducted in relation to the corresponding ground temperature at 2 m depth to analyze potential influences.

Radiometrically Terrain Corrected (RTC) Sentinel-1 data generated by the Alaska Satellite Facility (ASF) was used for backscatter analyses. γ0 data with 10 m pixel spacing was utilized. Sentinel-1 γ0 data from unfrozen summer conditions in August and selected frozen winter season data (see Sect. ) were averaged for each year. The difference between the mean August backscatter and the mean backscatter of the following winter, which primarily reflects soil moisture information , was further analyzed. In the case of the backscatter data, each year was treated separately in the analysis instead of being averaged over multiple years. To visualize the overall patterns, Figs. , , and show the median across all investigated years.

For the evaluation of Sentinel-1 thaw season deformation rate results (-αDDT), they were compared to the ABoVE thaw season deformation products . Because the ABoVE thaw season deformation product is expressed in absolute values, while -αDDT represents the thaw season deformation rate in mm DDT⁻¹, these two observables are compared in a qualitative consistency assessment. This comparison highlights whether regions with higher absolute thaw season deformation in the ABoVE product correspond to higher thaw season subsidence rates in -αDDT. Spatial variations in -αDDT arising from local differences in DDT values are expected to be small when considering individual study areas of limited size, making the qualitative comparison between the two datasets justified.

Since the ABoVE products are from 2017, we compared them with our data from the same year for regions where measurements were available. Excluding pixels masked in this study, a common area is covered of about 460 km² for the Noatak River Basin, 2400 km² for the Central North Slope region and 1600 km² for the Yukon-Kuskokwim Delta.

In order to assess the reliability of annual deformation rates derived from PALSAR-2 InSAR, used in this study, a comparison to annual deformations rates delineated from LiDAR DTMs was conducted. An area of 50 km² within the Anaktuvuk River tundra fire scar (North Slope) was investigated. The fire event occurred in 2007, and DTMs are available for the years 2009, 2014, and 2021. The InSAR data covers a time period of 2015–2019. Although the time interval for the DTMs does not precisely align with that of the InSAR data, the annual deformation rate derived from the DTMs for the period 2014–2021 was still used to make an approximate comparison.

Uncertainties of depicted data points for curve fitting are provided as error bars indicating the standard deviation across multiple results within each time-since-burn year. Additional uncertainty indicators are provided as error bars for the InSAR-derived estimates. For the thaw season deformation rate (-αDDT), the standard error of the linear regression between DDT and surface displacement was derived. For the annual deformation rates the standard deviations of the linear model with respect to the time-series input data are provided. For the Yukon-Kuskokwim region, where annual deformation rates were estimated using the DDT-based approach by , the standard deviations of displacement residuals were calculated. In order to illustrate the uncertainty of the fitted functions, the 95 % uncertainty envelopes are provided, similarly to , estimated as the predicted value ± 1.96 times the residual standard error (RSE = RSS/(n−p)), where RSS is the residual sum of squares, n is the number of observations, and p is the number of model parameters). For the regionally merged result an additional 95 % confidence interval is estimated based on a parametric bootstrap (2000 samples), defined as the 2.5th and 97.5th percentiles of the resulting predictions.

The temporal distribution of the fire events, which were selected based on homogeneity, varies across the different study regions (Fig. ). In the Inuvik region, 25 fire scars were applicable for the final analysis, with fire events dating back to 1968. The temporal distribution of fire events shows good coverage starting from the 1980s, with the longest interval between events being 14 years, from 1968 to 1982. The maximum number of fire events used in a single year for this region was eight, occurring in 2012. In the Noatak River Basin, many available fire scars were not selected due to too heterogeneous buffer reference areas. In total, 20 fire events were selected, with 6 occurring between 1971 and 1987 and 14 between 2005 and 2020. In the Central North Slope, not many fire scars were available, resulting in the smallest number used per region in this study. In total, 12 fire scars were used, mostly from between 2003 and 2019. An additional fire scar from 1983 was available, as well as the oldest fire scar in the study, estimated to date to approximately 1930. In the Yukon-Kuskokwim Delta, the highest number of fire scars was used, totaling 43, with multiple events occurring in the same years. The oldest event in this region is dating back to 1953. The maximum data gap between events is 18 years, occurring between 1953 and 1971.

Although the fire scar values exhibit considerable variability, no correlation between standard deviation and fire scar size was found (Table ).

The investigation of August backscatter revealed consistent, highly significant trends across the regions of Inuvik, Central North Slope, and Noatak River Basin (Figs. d and ). Younger fire scars displayed lower backscatter values, with reductions of, on average, 0.5 dB or less compared to their surroundings (Fig. d). Over time, the backscatter intensity increases, eventually exceeding the surrounding values after approximately 15 to 35 years. The Yukon-Kuskokwim Delta exhibits contrasting behavior compared to these regions (Fig. d), however the fitted linear function is not statistically significant (p-values > 0.05, Fig. ). Backscatter values of fire scars are here consistently higher than those of the surrounding reference areas, decreasing over time (Fig. d).

When considering the corresponding ground temperatures (within fire scars only, difference to surroundings on average 0.02 °C, with 0.33 °C SD), a clear difference between colder and warmer regions becomes evident (Fig. c). While the three colder regions, all below -2 °C, initially exhibit lower backscatter values for fire scars compared to the surrounding areas, followed by a gradual reversal, the warmer Yukon-Kuskokwim Delta region tends to show higher backscatter values within the fire scars, especially for younger fire scars (Fig. d). The prevailing backscatter level of the surrounding region does not appear to influence the direction of the trends, as Yukon-Kuskokwim Delta values differ only minimally from those of the colder Noatak River Basin and Central North Slope, whereas the Inuvik region, the second coldest study region, exhibits surrounding backscatter values more than 1 dB higher than the other regions (Fig. ).

The direction of the winter backscatter trends (Fig. e) is mostly consistent with the summer trends (Fig. d), although the trend lines for the Inuvik region, Noatak River Basin, and Central North Slope are less closely aligned compared to the August data. Initial differences are greater for Inuvik in winter than in summer, while the opposite is observed for the average values of the other two regions, where initial values inside and outside the burn area closely resemble each other on average. Nonetheless, backscatter values increase over time in all three areas, with values surpassing the reference values for the Inuvik region at an earlier time compared to the summer data. The trend observed in the Yukon-Kuskokwim Delta resembles that of the summer data, however, the values decrease more rapidly over time in winter than in summer.

The separation between warmer and colder regions appears less apparent in the winter data compared to the August data, with overlapping values (Fig. d). Nevertheless, it remains evident that initial lower winter backscatter relative to surrounding reference values is observed only in regions with temperatures below -2 °C for younger fire scars, whereas the warmer region exhibits higher winter backscatter than their surroundings throughout the entire investigated period. Similar to the summer backscatter data, the trend directions appear to be independent of the surrounding backscatter level, as the warmest region, the Yukon-Kuskokwim Delta shows winter backscatter values that more closely resemble those of the colder Noatak River Basin and Central North Slope, whereas the second-coldest region, Inuvik, exhibits notably higher winter backscatter values (Fig. ).

The investigation of the summer-minus-winter backscatter difference reveals distinct patterns across different regions (Fig. f). For the Central North Slope and the Noatak River Basin, the rates of change in summer and winter backscatter over time are approximately similar, with August backscatter initially showing larger deviations from the reference areas (Fig. d, e). No significant trend over time can be observed here (Fig. ).

For the Inuvik region, August backscatter initially exhibits smaller differences, but the rate of change in winter backscatter differences is faster than that of August (Fig. d, e). Similarly, in the Yukon-Kuskokwim Delta, winter backscatter changes faster over time, though it shows less deviation from the reference areas compared to summer backscatter.

When considering ground temperatures, particularly for older fire scars, a difference is observed between the warmest region, the Yukon-Kuskokwim Delta, and the coldest region, the Central North Slope (Figs. e and f). In the Yukon-Kuskokwim Delta, winter backscatter especially for slightly older fire scars is closer to the reference areas, whereas in the Central North Slope, the opposite pattern is evident (Fig. d, e).

The assessment of the comparison between Sentinel-1 derived thaw season deformation rates of the thawing season 2017 and the ABoVE thaw season subsidence products shows a linear trend visible in the value distributions for ranges with sufficient sample sizes (Fig. ). Although the ABoVE product represents absolute thaw season subsidence and -αDDT expresses the thaw season deformation rate in the DDT domain, the comparison shows that larger absolute ABoVE thaw season subsidence values correspond to more negative -αDDT values, indicating higher subsidence rates. This relationship is consistent within each study region where sufficient sample sizes are available and provides a qualitative indication of consistency between the two deformation metrics.

To assess the spatial consistency of the long-term annual deformation rate results from PALSAR-2 InSAR, we compared them with deformation estimates derived from LiDAR DTMs . Although the DTM data (2009, 2014, 2021) and InSAR time range (2015–2019) do not perfectly overlap, the annual deformation rate from 2014–2021 was used for an approximate comparison. An analysis of the histograms of the annual deformation rates (Fig. a) indicates an approximate alignment of median values (0.4 cm yr⁻¹ heave for InSAR results and 0.2 cm yr⁻¹ heave for DTM-derived results). The larger spread of deformation values derived from the LiDAR DTM may result from systematic differences between the InSAR and the ΔDTM approach, rather than from the DTM’s higher spatial resolution, as the data were downsampled to the same resolution for this comparison. On the other hand, the mean values of annual deformation rates for the time period closer to the fire event show subsidence rate values of 2.8 cm yr⁻¹ (Fig. a dashed line). Moreover, the spatial distribution of pronounced annual subsidence rates observed in the DTMs in parts aligns with annual subsidence rate patterns derived from PALSAR-2, as illustrated in Fig. . This similarity in spatial patterns is also reflected in a linear trend visible in the comparison of the value distributions (Fig. b). However, a quantitative comparison reveals systematic differences between the two datasets. An additive bias is indicated by the regression intercept of 0.35 cm yr⁻¹, while the regression slope of 0.10, which deviates substantially from the 1:1 line, indicates a conditional bias between the LiDAR-derived and InSAR-derived annual deformation rates. A Bland–Altman analysis (Fig. c) further assesses the agreement between the two datasets, revealing a mean bias of −0.25 cm yr⁻¹ and 95 % limits of agreement ranging from −2.07 to +1.57 cm yr⁻¹, with most pixels falling within these limits. however, a proportional bias is evident, with differences increasing systematically for pixels with larger deformation rates.

For the interpretation of the space-for-time analysis results, it is important to consider that the temporal coverage and the number of selected fires vary across the studied regions (Fig. ), which affects the quality of the derived trends. The Inuvik, Yukon-Kuskokwim Delta, and Noatak River Basin regions had an adequate temporal distribution and a sufficient number of usable fire scars, encompassing both younger and older scars. The Central North Slope, while containing the fewest investigated scars and being primarily characterized by younger ones, also included the oldest fire scar observed in this study.

Including both large fire scars, which may span different site-specific characteristics, as well as small scars, could introduce additional uncertainty to the results. The dependency of the standard deviation values of the investigated parameters on fire scar size was examined after binning the fire scar sizes into 50 km² increments and using median standard deviations to account for the bias toward smaller size ranges. While winter backscatter exhibited comparatively higher R2 values (0.33, Table ), the standard deviation values of the investigated parameters generally showed little dependency on fire scar size, although the standard deviations themselves remained non-negligible.

Furthermore, the reliability of our InSAR results must be considered. We assessed this by comparing thaw season deformation rate results (-αDDT) to ABoVE InSAR L-band thaw season subsidence products (Fig. ), providing a qualitative consistency check of deformation patterns, with the observed linear trends in the value distributions supporting the overall consistency of the results. While the values of the distributions for a given range vary across regions, potentially as a result of DDT normalization or spatial filtering, this does not affect our analysis, as we consider relative differences between fire scars and their surrounding areas. The qualitative consistency of annual deformation rates was assessed by comparing our InSAR results with LiDAR-derived ΔDTMs, demonstrating similar median values and broad correspondence in spatial deformation patterns (Figs.  and ). However, systematic biases between the datasets were identified, which may be attributed to the different observation periods covered and fundamental differences in measuring principles between the two approaches. In addition, the use of an unburned reference area helps to mitigate localized effects such as uncorrected atmospheric or ionospheric disturbances, as well as errors resulting from poorly selected reference points. Furthermore, the spatial filtering applied to the InSAR results reduces not only large-scale atmospheric and ionospheric artifacts but also removes deformation signals with spatial scales larger than the filter radius. This filtering centers the mean deformation around zero . In this context, the use of unburned reference areas allows the interpretation to focus on deformation within the fire scars relative to their undisturbed surroundings, rather than on potentially biased absolute values. This relative approach also helps to account for interannual fluctuations, since deformation trends in the burn scars are evaluated against prevailing conditions in the unburned surroundings, making the inferred trends more robust to annual regional variations. Additionally, for InSAR data, the seasonal median and the annual rate were computed over a maximum of three years to reduce atmospheric and ionospheric effects. Furthermore, although the coarse resolution of ERA5 air temperature data may introduce some uncertainties, its application for deriving -αDDT values has been demonstrated to be suitable in a previous study .

Furthermore, uncertainties of the fitted functions and observations need to be considered. The uncertainty band based on the residual standard error represents the scatter of the observed data around the fitted function and reflects the variability among individual observations. Consequently, the broad uncertainty bands indicate that recovery times of individual fire scars may deviate from the recovery time inferred from the fitted function. In contrast, the bootstrap confidence interval reflects the uncertainty of the fitted recovery function itself.

The error bars representing the standard deviation across all fire scars within each time-since-burn year (Fig. ) illustrate the variability in deformation rates among individual fire scars within the same region. The InSAR-specific error bars (Fig. ) represent the uncertainty of the individual InSAR-derived deformation estimates arising from the respective estimation procedures and illustrate the confidence associated with the estimated deformation rates.

The analysis of thaw season subsidence rates (Figs. a and a) revealed that, as expected, younger fire scars exhibit higher subsidence values. Thaw seasonal subsidence rates gradually decreased over time, with different regions displaying varying fire responses, some areas displaying a more linear trend and others a more exponential trajectory. Nevertheless, when averaged across all regions, thaw season subsidence rate values progressively converged toward those of the surrounding unburned reference areas, reaching alignment after approximately 50 years (Fig. a). The 95 % bootstrap confidence interval, incorporating all data points, suggests that alignment may occur as early as 43 years post-fire, whereas no alignment with the reference level is reached within the observed time period for the lower bound of the interval. The 95 % uncertainty band based on the residual standard error, representing the variability of the observed data around the fitted function, suggests that alignment with the reference level could occur as early as 12 years post-fire.

These findings differ from previous results on thaw season subsidence based on the bootstrap confidence interval. However, the residual standard error based uncertainty band encompasses the recovery times reported by previous studies. The larger number of fire scars investigated in our study across all regions may also contribute to the broader uncertainty range observed here. Although these studies investigated thaw season subsidence rather than the thaw season deformation rate (-αDDT) used in our study, comparing the recovery time remains valid because both quantities are expected to correspond to the same recovery timescale, as absolute thaw season deformation is inherently reflected in the thaw season deformation rates. In particular, reported in Fig. 7 that thaw season subsidence in the Yukon-Kuskokwim Delta returned to pre-fire conditions after approximately 15 ± 7 years (Fig. a). This delineated value greatly depends on the pre-fire thaw season subsidence assumed by . Contrary to , our study establishes the difference from undisturbed conditions by incorporating an unburned reference area, which allows for the comparison of reference regions under the same recording conditions, while also accounting for processing inaccuracies related to shifts in the value range. Furthermore unlike , our study did not exclude the contributions of annual effects resulting from “permafrost layer thinning” due to melting of excess ground ice from our thaw season -αDDT product, which may account for discrepancies, particularly if increased annual deformation rates persist over a longer time period and contribute to the observed -αDDT values. However, the overall annual deformation rates observed in our PALSAR data (Fig. b and c) suggest increased permafrost layer thinning over a period shorter than the observed thaw season results. In this case, equilibrium is reached on average after approximately 10 years.

The recovery time of thaw season subsidence was indicated at approximately 27 to 36 years in for an Arctic-boreal region in North Yukon. This time span is longer than suggested by , yet still shorter than what we observed in our study regions (Fig. a).

also investigated thaw season subsidence using a space-for-time approach in the Yukon-Kuskokwim Delta. Similar to our methodology, they employed buffer zones as unburned reference areas. Their results showed that fire scars reached equilibrium with reference areas after 29 years, which is similar to the results of in North Yukon. employed a linear fit, which our data likewise indicate to be appropriate for the same region. In our analysis, equilibrium is reached approximately 10 years later for this region (Fig. a). However, the coefficient of determination indicated by (R2=0.045) was substantially lower than the value we obtained for the same region (R2=0.62, Fig. ).

In , hardly any difference in ALT and thaw season subsidence was observed between burned and unburned areas across regions of the North American Arctic-Boreal zone. However, this study relied on a more simplified approach, not accounting for fire year or regional differences.

As reported in other studies , the increase in ALT typically peaks within 5 to 10 years. Nevertheless, the observed prolonged influence on -αDDT values (Fig. a) suggests that factors other than ALT may contribute to these thaw season results. In , a connection between -αDDT values and soil moisture was demonstrated. This relationship indicates that within fire scars, higher soil moisture levels within the active layer may persist for a longer period than what has been reported for ALT effects. reported increased soil moisture for burned areas in the Yukon-Kuskokwim Delta even after 46 years. Increased postfire soil moisture can be attributed to changes soil properties, thawing of the upper transition zone and a reduction in evapotranspiration . found that the greatest changes in soil moisture occurred near the surface. The influence of near surface soil moisture can also be inferred from the difference between summer and winter backscatter values (Fig. f). However, a direct relationship for the time period indicated by the -αDDT values (Figs. a and a) cannot be derived from the backscatter data.

A possible reason for prolonged increase in soil moisture might be an enhancement of roughness which leads to the trapping of blowing snow, which has been reported to linger 20 d longer in areas with degraded ice wedges on the Seward Peninsula, Alaska . As a result, wetter conditions develop in depressions, which subsequently also affect -αDDT. Whether these wetter conditions persist throughout the thawing season or are confined to the immediate post-snowmelt period remains to be confirmed by in situ measurements.

Ground temperatures within fire scars and their unburned surroundings showed no substantial differences (mean difference: 0.02 °C), indicating that the CCI ground temperature values used in our comparisons are unaffected by fire. Furthermore, comparisons with ground temperatures revealed no clear influence on the thaw season deformation rate results (Fig. a). However, when analyzing the trends of thaw season deformation rates per study region (Fig. a) in relation to their prevailing ground temperatures, it can be noted that fire scars in colder regions require somewhat more time to return to deformation values of the undisturbed surrounding.

In , the peak of “permafrost layer thinning” is reported to occur 25 years after fire events in the Yukon-Kuskokwim Delta, which is much later than what is observed from the fitted function of the annual deformation rate values of our study, where annual deformation rates of fire scars match those of surroundings after about 15 years for this region (Fig. b). This discrepancy is further amplified by the fitting function used, which allows for stronger reductions in values during the initial years compared to later years. However, the estimate reported by falls within the RSE-based uncertainty band of our fitted function for this region and is also consistent with the uncertainty of the individual InSAR-derived deformation rate estimates. This suggests that, although our fitted function indicates a faster recovery, the associated uncertainties overlap with previous observations (Fig. b). Compared to , our study incorporated a larger amount of fire scars and covered a longer temporal span of fire event years, which enabled a more comprehensive reconstruction of the system’s temporal evolution and may contribute to differences observed between the studies. The larger number of investigated fire scars across most study regions compared to previous studies may also contribute to the broader RSE-based uncertainty envelope observed in our study.

The timespan of the initial subsidence following fire events in our results, approximately 7–13 years across all regions’ data points based on the 95 % bootstrap confidence interval (Fig. b and c), however much closer resembles that reported by (Fig. c), who analyzed fire scars in the Arctic-boreal North Yukon region and also acquired more fire samples covering the post-fire progression. Consistent with , we found that, on average, this timespan is confined to the first decade after the fire event. For the Central North Slope, we also identified the reported uplift relative to the surrounding areas during the subsequent 30 years, which gradually attenuates to values comparable to the undisturbed surroundings (Fig. c). In contrast, other regions, especially the Inuvik region and Yukon-Kuskokwim Delta exhibited no discernible oscillatory behavior, resulting in an average trend that more closely aligns with an exponential fit, showing heave relative to the surrounding undisturbed areas.

A potential connection with ground temperatures could be suggested for annual deformation rates (Fig. b), where the warmest region, the Yukon-Kuskokwim Delta exhibited most pronounced relative heave for old fire scars compared to their surroundings (Fig. b, c). This may indicate that regions with higher ground temperatures may be more susceptible to permafrost thaw following fire events. However, this interpretation would need to be validated using data from additional warmer regions, as factors other than ground temperature may influence these outcomes, such as burn severity, ground ice content, vegetation cover, and soil moisture conditions.

Incorporating SAR backscatter data could enhance our understanding of how vegetation structure and soil moisture are affected in these regions.

By analyzing changes in backscatter, we can gain insights into the post-fire dynamics of vegetation recovery and soil moisture variations. The comparison of summer backscatter values between fire scars and their surroundings revealed that fire scars in warmer ground temperature areas exhibit higher backscatter than unburned areas especially for younger fires, reducing over time, whereas colder regions initially tend to show slightly lower backscatter values, which increase over time (Figs. d and c). These patterns appear to be independent of the prevailing backscatter level of the surrounding areas, and thus of vegetation type and surface conditions (Fig. ). Our findings, however, are not consistent with research by and , who observed higher backscatter values up to 4–5 years after the Anaktuvuk River fire. In our results, this specific fire shows minimal deviations from surrounding backscatter values 10–15 years after the event (not explicitly shown). This finding aligns with 's observations, which also noted only limited contrast in C-band data after 10 years.

The deviations observed in previous studies of the Anaktuvuk River fire suggest that our findings regarding the relationship between ground temperatures and fire scar backscatter data may not be the sole contributing factor. For instance, reported that post-fire ecosystem changes contributed to a cessation in permafrost thaw, and recovery of the permafrost table within approximately 10 years after the Anaktuvuk River fire, with about 15 cm of aggraded permafrost observed on average. However, the increase in microtopography that developed due to thermokarst following the tundra fire remains evident. A similar recovery time span can also be inferred for summer backscatter data of this region.

Although there may be other factors that influence backscatter trends, many of the fire scars investigated in our study do indicate a connection between fire scar backscatter and ground temperatures (Fig. c), independent of the initial backscatter levels (Fig. ). Higher backscatter in fire scars has previously been attributed to increased surface roughness. This alone would however not explain why summer values of the warmer region are decreasing over time, while they seem to increase in colder regions (Fig. d), presuming the surface roughness would increase with the progression of thermokarst development. Another cause for enhanced summer backscatter values is increased soil moisture. Higher soil moisture in burned than in unburned areas has been reported for the Yukon-Kuskokwim Delta even after 46 years , which may as well be similarly deduced from backscatter data throughout the investigated period (Fig. f). Here, summer backscatter in fire scars is increasingly higher than for winter backscatter (Fig. d, e). As both winter and summer backscatter values gradually approach those of the unburned surrounding areas, the summer backscatter reduction in the Yukon-Kuskokwim Delta might also stem from differing causes, than from a reduction of soil moisture. It also has to be considered that higher winter backscatter in this region has previously been attributed to the presence of depth hoar in the snowpack, attributed to a freeze-up that occurred later than under pre-fire conditions .

Whereas this warmer region experiences a presumed initial rise in near surface soil moisture, some colder regions like the Noatak River Basin and Central North Slope seem to exhibit lower soil moisture than their surroundings (Fig. f). If it is correct to assume that the resulting differences between summer and winter backscatter are primarily due to variations in soil moisture, then these effects are not clearly linked to ground temperatures, since the Inuvik region also exhibits initially elevated deduced soil moisture levels, which decrease over time.

While the initially higher summer backscatter values in burned areas of the warmer Yukon-Kuskokwim Delta (Fig. d) might in some part be explainable due to enhanced soil moisture, the observed initial lower backscatter in colder ground temperature areas might additionally be attributed to reduced volume scattering resulting from vegetation loss. This may account for the greater reduction in backscatter values within the Inuvik region, where taller vegetation heights are to be found . The backscatter increasing over time might be explained by shrubification .

A clear increase of backscatter values could for instance also be observed for the oldest fire scar investigated within this study, the Shivugak Bluffs Burn of the Central North Slope region (Fig. d and e at about 90 years since initial burn). examined the long-term biophysical impacts of this tundra fire. The fire led to profound and persistent changes in vegetation structure and composition. Compared to adjacent unburned tundra, the burned area is now dominated by tall willows and grasses, while sedges, mosses, and other cold-tolerant species have significantly declined. This shift reflects a transition from a stress-tolerant plant community to one shaped by post-fire competition and warming. Thinning of the organic soil layer after fire has amplified permafrost thaw and thermokarst development, contributing to sustained ground warming. While the burned tundra has dried over time as ice-wedge troughs connected and drained, the unburned tundra is currently undergoing a parallel but slower process of wetting and thaw. The findings suggest that fire can act as a catalyst for vegetation and permafrost regime shifts, with consequences that persist for centuries . The reported conspicuously tall willows and terrain rugosity of this fire scar further explain the observed high backscatter values associated with it.