The development of thermokarst in ice-rich permafrost regions is a natural hazard, causing irreversible geomorphic changes [1
]. Thermokarst is the process by which characteristic landforms result from either the thawing of ice-rich permafrost or the melting of massive ice [2
]. The formation of large depressions and lakes or swamps produced by thermokarst processes is observed in discontinuous and continuous permafrost zones, especially in Alaska and Northeastern Siberia. These regions are often underlain by highly ice-rich permafrost, known as Yedoma ice-complex (Yedoma). Yedoma is a unique Quaternary permafrost deposit, consisting of excessive amounts of ground ice (50%–90% in volume) and organic-rich sediments [3
]. Permafrost degradation caused by thermokarst implies the mobilization of not only huge amounts of currently confined organic carbon, but also a certain amount of water preserved as ground ice. Therefore, feedbacks from permafrost degradation to the surface ecology, landscape, and hydrological processes have been of great scientific and social concern, and the need for elucidation of their role in the ecosystem and global climate system has been emphasized by many authors (e.g., [4
The amount of organic carbon preserved in permafrost regions is estimated to be twice as large as the amount of carbon in the atmosphere, and 211 + 160/−153 Gt are stored in Yedoma regions [9
]. Zimov et al.
and Walter et al.
] reported that thawing of Yedoma may release a significant amount of methane (~3.8 Tg/year), which can lead to further climate warming. A warming climate can induce environmental changes including thermokarst, which accelerates climate change through the microbial breakdown of organic carbon and the further release of greenhouse gases [12
]. Despite the recognition of uncertainty about the fate of Arctic regions and global climate change due to permafrost degradation, information about the spatial extent and rates of thermokarst processes is still limited.
One promising technique for the quantification of thermokarst in remote wide areas underlain by ice-rich permafrost is the deployment of Interferometric Synthetic Aperture Radar (InSAR). InSAR allows the remote detection of deformation on the terrestrial surface associated with earthquakes, volcanic activities, or other natural and anthropogenic alteration underground (e.g., [13
]). There have been several studies regarding surface change related to frozen ground dynamics. The first successful attempt to use InSAR to detect thaw settlement was made by [15
]. Short et al.
] then compared TerraSAR-X, RADARSAT-2, and ALOS-PALSAR interferometry for monitoring permafrost environments, concluding that ALOS-PALSAR, in particular, was the most promising data source for identifying permafrost and landscape change. In recent years, both seasonal thaw settlement [17
] and an increase in surface subsidence caused by Arctic tundra fire [18
] in a permafrost region of north Alaska were detected by InSAR time series analysis using ALOS-PALSAR data. Liu et al.
] provided temporally averaged levels of seasonal thaw settlement in individual drained thermokarst lake basins, demonstrating the ability to detect surface movement associated with ground freeze-thaw cycle at tens of meters spatial resolution. Large-scale thermokarst subsidence in the Anaktuvuk River Fire (ARF) scar was detected using InSAR for the first time by [18
]. Although spatial variation in thermokarst subsidence at the regional scale was shown, field information and InSAR analysis served to limit understanding of local thermokarst processes, as these phenomena occur at much smaller scales (from several to one hundred meters). To date, InSAR techniques for quantifying thermokarst subsidence with enough spatial and temporal resolution for understanding local thermokarst processes in detail, including evidence from fieldwork, have not yet been reported for the ARF scar.
The objectives of this paper are to demonstrate the ability of L-band InSAR to quantify thermokarst subsidence at spatial resolutions of an order of tens of meters, including supporting evidence from field surveys and analysis of optical satellite images, and to discuss the efficiency and limitations of this method as a monitoring tool for thermokarst. To this end, we investigated the northern part of the same ARF scar in detail, using both optical and microwave remote sensing, as well as in situ fieldwork.
4.1. Spatial Resolution and Variation of Captured Thermokarst Subsidence
Subsidence occurs preferentially along a network of ice wedges, establishing a connected trough. The obtained interferograms do not show sub-meter scale depressions along the trough, though they do show average overall subsidence. First-order estimation of subsidence within our ground truth plots ranged 12–61 cm (2–10 cm/year) during the six years after the fire (to late August of 2013). The total amount of subsidence observed in our interferograms (from 24 July 2008 to 14 September 2010) was 9.9 cm (3.3 cm/year). This subsidence rate for the first two-and-a-half years after the fire is comparable with the six-year average value for plots A and B (2.2 cm/year), though our InSAR observation was about 1.5 times higher for the first three thawing seasons after the fire. It is natural that rapid thermokarst for the first thawing season after fire stabilizes in following years, unless the disturbed surface has turned to water as thaw ponds or lakes, within which vegetation cannot recover. Induced subsidence gradually stabilized as surface vegetation recovered, acting as a modulator of surface energy exchange, which had been enhanced by the combustion of surface vegetation and organic mat. Regarding the following years after 2010, when ALOS observation was not obtained, Jones et al.
] reported that LiDAR-derived subsidence between 2009 and 2014 was about 6 cm/year as a spatial average for burned Yedoma upland areas. They also reported that visual analysis of high-resolution satellite imagery indicated marked ice wedge degradation between 2011 and 2014, while there were subtle differences in image texture between 2008 and 2011.
Comparing the range of temporally averaged subsidence amounts (2–8 cm/year) due to permafrost thaw estimated by [18
], our results showed a slightly lower subsidence rate (1–6 cm/year) for a post-fire three-year average, partly because our research area is a portion of the entire fire scar. On the other hand, it is unclear whether the subsidence measured by InSAR is reflected from spatial averages of surface subsidence (which is actually differential surface deformation in sub-pixel scale) or from the most active surface movements along the polygonal ice wedge troughs. The difference in surface height between trough bottoms and the central part of polygons was more than 50 cm. Having spatial resolution of 1 m, the LiDAR measurements of [29
] also clearly showed much larger subsidence (up to 25 cm/year) along troughs of degrading ice wedge polygons than that in the central part of polygons. It will be a future challenge to understand how the combination of mixed subsidence around the centers of polygons at a slow rate and along the troughs at a rapid rate will be averaged in InSAR signals.
Most dynamic subsidence observed in the Post-1 interval was 6.2 cm on average and 95% of measured values ranging 0.9–11.5 cm. Our results demonstrated that if we choose appropriate interferogram pairs and carefully treat seasonal surface movement, conventional two-pass differential InSAR analysis is capable of detecting the spatial variation of subsidence at a tens of meters scale (Figure 4
and Figure 5
), which clearly showed the difference between unburned stable areas and burned subsiding areas with various displacement rates. Although there was no observations of thaw slumps and active layer detachments in our study area, and we assumed only vertical displacement in our interpretation of observed InSAR results, it is likely that the larger subsidence rates on the steeper slopes were partly attributed to lateral displacement, about which we do not have evidence for discussion. Isotropic thaw subsidence, which is geographically uniform and is not apparent to observation at the surface [32
], may also have influenced surface movement in our study areas, and further discussion about spatial variation in subsidence caused by fire should be focused on its geographical heterogeneity. This kind of InSAR application for permafrost regions with accurate field surveys has the potential to reveal how ice-rich permafrost terrains deform upon their degradation.
It is also worth noting that large subsidence was calculated for fragmented unburned areas (for example, −4.4 cm on average for Post-1), as they were small patches (most of them smaller than 1 m2
) surrounded by burned surfaces in which thermokarst had been active (Table 2
). This fact seems to show that thermokarst areas tend to propagate into adjacent areas by the lateral influence of thermal and/or hydrological regime shifts in the ground surface.
4.2. Effects of Surface Changes due to Wildfire on InSAR Signal
It is important to consider change in soil moisture in active layer and vegetation recovery, when thermokarst in a wildfire scar is studied. The InSAR pairs in this study only contain time intervals either pre- or post-fire, which did not cross the fire moment. This selection avoided the inclusion of decorrelated interferograms from significant surface roughness changes due to the loss of surface vegetation and organic layer, localized thermokarst, and change in hydrology. Despite the advantage of using L-band SAR, which penetrates surface vegetation, we observed total decorrelation in the fire scar in the interferograms with time intervals across the fire. Although the degree of influence from vegetation recovery after the fire is unknown, there was little change in vegetation during the time intervals for the selected interferograms, except for the Pre-1 pair that bridged an entire growing season.
Increase in surface soil moisture will cause an apparent phase shift in InSAR signal toward upheaval. For example, [33
] estimated that an increase in surface soil moisture (5%–30%) could decrease penetration depth of SAR microwaves from 100 to 2 mm. In permafrost regions, surface disturbance induces a relative increase in soil moisture at disturbed sites, as loss of vegetation causes a predominant effect on the decrease in evapotranspiration at the surface. For example, relatively higher soil moisture in the active layer was observed at an early stage after clear-cutting of a larch forest stand in a continuous permafrost zone in Eastern Siberia [34
], and at a burned area from tundra fire in a discontinuous permafrost zone on the Seward Peninsula of Alaska [35
]. In the case of ARF, soil moisture increase within the fire scar was probable and a certain magnitude of subsidence signal could have been offset. The increase in soil moisture at the burned area, however, enhanced ice segregation in the active layer, resulting in greater frost heave.
4.3. Uncertainly in Our InSAR Subsidence Detection due to Active Layer Change
Although we have offset surface movement due to seasonal thaw settlement by subtracting interferometric phase changes for the reference unburned areas, spatial variation in the obtained subsidence contains errors associated with spatial variation in seasonal thaw settlement—caused by differences in soil moisture content, soil texture, and active layer thickness. Additionally, uncertainty remains regarding the increase in active layer thickness and frost heave due to increased soil moisture after fire.
Frost heaving occurs in frost-susceptible soil when the amount of soil moisture is sufficient for forming segregated ice lenses in the active layer (e.g., [36
]). Ground heave occurs throughout the freezing period, whenever conditions for ice segregation are fulfilled. The degree of frost heave equals approximately the total thickness of segregated ice lenses in the active layer [37
]. Thaw settlement, on the other hand, occurs along with the melting of ice lenses as the thaw front progresses, from the top down within the active layer. Assuming similar seasonal weather conditions, as well as negligible changes in the moisture condition and structure of soil particles, the ground surface will return to its initial elevation after one year. Ice lenses tend to form in the upper active layer and near the permafrost table in the case of two-directional freezing in permafrost regions (e.g., [38
]), and we expect that total ice lens thickness is greater in the upper active layer because top-down freezing predominates freeze-up from the permafrost table. Therefore, the rate of thaw settlement is small in the later thawing seasons, as the progress rate of the thawing front is low in the late season. For instance, thaw front progress in Arctic tundra was dulled in early August, and then gradually reached maximum thaw depth in mid-September (e.g., [34
]). Thaw settlement is therefore active in the early thawing season, as the thaw front rapidly progresses, and becomes significantly slower later in the season when there is only a small change in thaw depth.
In this study, interferograms were created using SAR images obtained only during thawing seasons. While Pre-1 nearly spanned an entire thawing season, Pre-2, Post-2, and Post-3 time intervals contain later portions of the thawing season. We can infer the extent of spatial variation in seasonal thaw settlement from Pre-1 and Pre-2. The difference in calibrated subsidence between burned and unburned areas inside of the future fire scar was within 0.7 cm (Table 2
)—much smaller than the values of subsidence we observed in interferograms of post-fire periods. Additionally, Pre-2, Post-2, and Post-3 InSAR pairs covered only later thawing seasons (when thaw settlement is relatively inactive), while the time interval for Post-1 is the whole year (representing one total seasonal freeze-thaw cycle). Therefore, these interferograms showed small errors associated with spatial variation in seasonal thaw settlement.
Regarding the uncertainty that arises from increases in active layer thickness and frost heave due to increased soil moisture after fire, we concluded that this was insignificant for our conclusion as follows. We observed averaged thaw depths of 62 and 72 cm in the unburned and burned areas, respectively. Since these measurements were conducted at the end of August, they can also be treated as maximum thaw depth for 2013. Assuming a linear increase in active layer thickness in the burned area over the six years after the fire, the annual increase in the active layer was estimated as 1.7 cm/year. We also assumed progress of the thawing front during our InSAR observation, at the end of thawing, was about one fourth of the entire active layer for these years. For this increase in active layer thickness for burned areas, we can estimate an increase in thaw settlement by increasing the frost-heave strain (the ratio of total heave to active layer thickness minus the total heave), simulating the increase in soil moisture. Examples of frost-heave strain from field measurements are available in [40
], in which values of frost-heave strain vary from 0.03 to 0.40, corresponding to 4 and 20 cm of maximum frost-heave for organic-rich tundra and highly frost-susceptible ground, respectively, while active layer thickness is 70 cm on the North Slope, Alaska. In our case, hypothetical increases in frost-heave strains from 0.03 to 0.06 (1.8–4.1 cm maximum frost heave) and from 0.3 to 0.4 (14.4–20.7 cm maximum frost heave) during these six years result in 0.1 and 0.3 cm seasonal thaw settlement on average, which may influence InSAR signals. As the soil in our targeted area consists mainly of peat, the error associated with active layer thickness and moisture should be close to 0.1 cm (or less, because the increase in frost-heave strain due to moisture increase in the active layer was also smaller if we consider only a single thawing season).
4.4. Limitations and Implications of this Study
From the above-mentioned considerations and calibration, we concluded that errors due to changes in active layer conditions and vegetation recovery were typically less than 0.1 cm, insufficient for masking thermokarst signals due to fire. However, in an extreme case, such that an increase in active layer thickness and moisture at burned areas occurs during a single thawing season, we note that the fraction of the error can be larger, especially for Post-2 and Post-3, with relatively small subsidence signals. On the other hand, the vertical displacement errors associated with the inaccurate DEM used in the InSAR analysis and caused by the height ambiguity (e.g., [41
]) depend on the baseline lengths for interferogram pairs, shown in Table 1
. Though errors ranged from 0.3 to 1.0 cm for individual pixel values of subsidence (depending on baseline lengths), as far as we consider average values in certain areas (including numerous pixels), errors in measured average subsidence values should be much smaller than those listed in Table 1
. Considering every possible source of error affecting the subsidence signal, we can regard the overall uncertainty level for subsidence values reported in this study as less than the errors for individual interferograms, listed in Table 1
InSAR capability for capturing spatial variation in thermokarst subsidence can be utilized for the quantification of permafrost degradation of large spatial extent, and will hereafter lead to a more accurate estimation of ground-ice loss upon permafrost thaw, in the past or near future. The quantitative assessment of permafrost degradation will provide fundamental information for estimating carbon release due to permafrost thaw, together with increasing knowledge about carbon contents in permafrost as greenhouse gases and organic matter. Archived ALOS-PALSAR data can be utilized to assess permafrost degradation in the past in wide permafrost regions, with the current ALOS2 mission expected to provide further information regarding the surface of changing permafrost with improved data quality.
The two-pass differential InSAR technique using ALOS-PALSAR (L-band microwave) has been shown capable of capturing thermokarst subsidence at a spatial resolution of tens of meters, with supporting evidence from field data and optical satellite images. Significantly large amounts of subsidence (up to 6.2 cm/year spatial average) were measured within burned areas relative to unburned nearby by three independent InSAR pairs after a tundra fire, while relatively small spatial variation (less than 0.5 cm in spatial average) was observed from two independent InSAR pairs during the pre-fire period. The obtained interferograms did not show sub-meter scale depressions along the troughs of the depression network developed by thermokarst, though they could distinguish small land areas with stable and subsiding land surface at smaller than tens of meters scale and could also display detailed spatial variation of thermokarst subsidence. Post-fire interferograms were decorrelated along fire boundaries, where rapid surface changes due to lateral erosion can be expected, and clearly separated subsiding burned areas from stable areas of intact environment. The mean rate of subsidence observed in our interferograms (from 24 July 2008 until 14 September 2010) was 3.3 cm/year, with this value comparable to the rate estimated from field surveys at two plots on average (2.2 cm/year) for the six years after the fire. We introduced a calibration procedure comparing burned and unburned areas for InSAR subsidence signals to remove the noise from seasonal surface movement. Changes in active layer thickness and soil moisture, and recovering vegetation after the fire were estimated as factors insignificant enough to not mask thermokarst signals due to the fire. InSAR’s capability of capturing detailed spatial variation in thermokarst subsidence can be utilized for further understanding of thermokarst processes and quantification of permafrost degradation at a large spatial extent, hereafter leading to more accurate estimation of ground ice loss upon permafrost thaw in the past or near future.