The Arctic National Wildlife Refuge has heterogenous landscapes with diverse ecosystems that are subject to a wide variety of geomorphic and vegetation processes that drive change. During the ~50-year study period, 18% of the Refuge underwent some type of landscape change, primarily due to wildfire, postfire succession, changes in shrub and tree cover, river dynamics, and ice-wedge degradation. This is similar to the amount of change detected in the Arctic Network of National Parks, located to the west of the Refuge, where a similar methodology found 24% of 206 systematically distributed plots showed change between 1975–1985 and 2008–2010 [27
]. Our finding also compare to an Alaska-wide analysis, in which Pastick et al. [37
] found 14% of the landscape had undergone change from 1984 to 2015, based on an analysis of spectral trends in Landsat imagery.
Below, we discuss some of the dominant changes affecting the landscape, compare and contrast the major drivers of change across the Refuge, evaluate rates of change during early and recent time intervals, and discuss the limitations of our remote sensing approach.
4.1. Change Types
We documented 19 broad categories of change associated with geomorphic and ecological processes. The major change types that emerged as most prevalent on the Refuge landscape included fire, river channel dynamics, tree or shrub increase, ice-wedge degradation, changes to the coastline, and hydrologic changes that included lake expansion and drainage, discussed in more detail below.
Wildfire caused the most change in our study Refuge-wide (6%) and was an important driver of change in the Boreal Biome (18%). Wildfire is a natural part of the boreal forest ecosystem (Figure A6
). While fire frequency has increased in recent decades in Alaska [38
], in the Refuge a huge area that burned in 1950 has been unequaled by any subsequent year. Therefore, large areas have been recovering from that fire, and our study found more tree increase than decrease after wildfire. Fire occurred only at our forested grids. Fires are known to occur in the tundra of northern Alaska but are uncommon [39
]. The recent large fire near the Anaktuvuk River, visible on Figure A6
, indicates that fire activity within the Tundra Biome could increase with climate warming, which could exacerbate thermokarst [40
]. Severe fires accelerate thermokarst by removing the insulating soil organic layer, allowing summer heat to penetrate and thaw the permafrost [42
]. We found some evidence that this process has occurred within the Refuge, as we noted several active-layer detachment slides that began within five years after fires (Figure 6
and Figure A5
Channel dynamics of active river floodplains can result in rapid changes. Half of our grids included points on active floodplains and we found river erosion and deposition were important drivers of change, affecting 5% of the Refuge. Differences between erosion and deposition can be linked to the rapid melting of glaciers in the Brooks Range [43
]. The glacially fed river floodplains had twice the frequency of changed points as the nonglacial river floodplains (63% vs. 30% of active floodplain area), including more erosion, deposition, and shrub increase or decrease. The ratio of river erosion to deposition was skewed slightly towards erosion on the glacial rivers, and towards deposition on the nonglacial.
Shrub expansion onto tundra is widespread in the Arctic, with large ramifications for ecological processes and climate feedbacks [44
]. In the Refuge, shrub cover increase in the absence of fire occurred almost entirely in the Mountain and Boreal Biomes, on alluvial substrates (e.g., active or abandoned floodplains, banks along floodplains, or alluvial fans). This change type affected 4% of the Refuge. In comparison, Swanson [45
] photointerpreted tall shrub presence and density at 471 plots (mostly Mountain Biome) in the Arctic Network of National Parks and found that 8% had dense canopies of tall shrubs (often on floodplains), associated with higher summer temperatures, deep summer thaw, and well-drained soils. Tape et al. [46
] also observed widespread shrub expansion on floodplains or nearby slopes. The scarceness of shrub increase we detected in the Tundra Biome and on nonalluvial surfaces throughout the Refuge appeared to be linked to soil conditions, and perhaps also to seasonality. Peak solar radiation at these latitudes occurs in late June, at which time soils may be thawed in Brooks Range valleys, especially on well-drained alluvial substrates, allowing plants to begin growth. The Tundra Biome, narrower within the Arctic Refuge than elsewhere in Arctic Alaska, is more affected by colder temperatures nearer to the Arctic Ocean. This causes soils to remain frozen near the ground surface at summer solstice, retarding the ability of shrubs to take full advantage of that period of maximum solar radiation.
Tree increase unrelated to fire was found almost entirely in the Boreal Biome (5%). One Boreal grid showed altitudinal tree line advance, where spruce trees had advanced out of gullies and onto high tundra between the earliest and latest time series images (Figure 8
). Overflights of this grid together with a hand-drawn map showing tree extent in 1911 [47
] indicate that the advance was not due to vegetation recovery following wildfire. At another grid on an inactive floodplain trees had become denser in wetlands, perhaps attributable to reduced flooding. The overall increase in tree cover was greater in the recent interval than in the early one. Little tree increase occurred at other grids (without wildfire). A review of worldwide tree line studies found that only 52% of sites showed tree line advance [48
Ice-wedge degradation (a thermokarst process) is affecting ecosystems throughout the Arctic [21
]. It affected 2% of the entire Refuge and was the dominant change detected in the Tundra Biome (12%). In comparison, ice-wedge degradation was observed at a few of 206 photointerpreted plots in the Arctic Network of National Parks [50
]. Within small, targeted areas in northern Alaska, the extent of ice-wedge degradation increased from 0.5% to 4.4% (1945–2001) near Fish Creek [21
] and increased from 0.9% to 7.5% (1949–2012) near Prudhoe Bay [30
]. Farquharson et al. [51
] found thermokarst troughs and pits covered 7% of 12 small mapped areas across northern Alaska. For central and northern Alaska, Jorgenson et al. [52
] found thermokarst features occurred on 8% of sample points on airphotos from 2005 and 2006, with the frequency of occurrence much higher in the continuous permafrost zone in arctic Alaska (13.5%) compared to the discontinuous zone in boreal Alaska (5%). Ice-wedge degradation usually causes radical redistribution of water, resulting in newly wetting or drying conditions [53
]. The only type of thermokarst recorded in our study was ice-wedge degradation, which overall caused much more wetting (1.5%) than drying (0.2%). Other types of thermokarst may be common in the Refuge, going undetected in our study due to small areal extent. For example, several small active-layer-detachment slides (ALDs) were incidentally observed to have occurred at one forested site after a wildfire, although not at a point (Figure 6
). Many slides up to 90-m long have also occurred in severely burned forest 2 km NW of another study site (Figure A5
). In comparison, 848 ALDs and 276 retrogressive thaw slumps were mapped within the 2.7 million hectare Noatak National Preserve [54
On the Beaufort Sea coast, there were changes associated with coastal erosion, deposition, and salt water intrusion during storm surges. Deposition occurred at the mouths of rivers, and elsewhere the shore eroded gradually, seldom more than 1 m/year. Jorgenson and Brown [55
] compiled mean annual erosion rates (1950s to 1980s) for sections of Refuge’s coastline, and found coastline changes that ranged from erosion at ~1 m/year to accretion at ~12 m/year. Rates depended on the coastline type and soil texture. A long-term monitoring site in the Refuge had a mean annual erosion rate of 0.5 m/year between 1949 and 2001 [56
]. In comparison, Jones et al. [57
] documented a maximum erosion rate of 18.3 m/year at a point north of Teshepuk Lake, in low-lying thaw lake terrain that is rare within the Refuge.
Hydrologic changes included changes in lake area, river channel migration, irregular surface water changes on vegetated ground, water redistribution associated with ice-wedge degradation, and surface drying on inactive floodplains. Changes in the Tundra Biome tended to be related to landscape wetting (mainly ice-wedge degradation and surface water increase), while changes in the Boreal Biome tended to involve landscape drying (including reduced area of lakes and recent wildfire). Lake area tended to increase in the Tundra Biome and decrease in the Boreal Biome, where we found the process of lake drying on inactive floodplains left concentric rings of shrub and graminoid vegetation in former lake beds. Our results are consistent with those of Riordan et al. [24
], who examined surface water, lakes, and ponds at 11 regions of Alaska, using images from the 1950s to 2002. They found a decrease in the area of closed-basin ponds in all locations except the Arctic coastal plain. In the adjacent Yukon Flats National Wildlife Refuge, historic aerial imagery indicates that lake drying and vegetation invasion have occurred in the Boreal Biome since about the 1980s [58
]. Necsoiu et al. [59
] mapped waterbodies on a time series of high-resolution imagery for the Kobuk Valley and found total surface area decreased by only 0.4% during 1951–1978, but then decreased by 5.5% during 1978–2005. In contrast, Plug et al. [60
] used a time series of Landsat imagery (1978–2001) to show that lakes mostly increased during 1978–1992 and decreased during 1992–2001. Our limited sampling did not detect the decline in river icings documented by Pavelskiy et al. [61
4.4. Limitations and Applications of Remotely Sensed Change
There were some limitations on our ability to detect change related to image quality and high spatial variability. We chose to manually interpret the changes we could see on images, using visual cues, such as pattern, texture, brightness, and juxtaposition, as well as ecological knowledge of the interpreters. When imagery was high quality, this worked very well and proceeded rapidly. The aerial photographs varied in quality and resolution, however. In the Mountain Biome, some of the oldest photographs in the southern Brooks Range were of poor quality. In addition, the aerial photographs were difficult to rectify on mountain slopes due to steepness and to lack of reliable ground reference points on slopes that had only scree and dwarf shrub vegetation. We believe we are correct in concluding that there was little change in the Mountain Biome other than on river floodplains, despite the image limitations. Images for the Boreal Biome were generally acceptable and changes in forested types were easy to detect. Imagery was excellent for the Tundra Biome but we likely could not detect subtle vegetation changes since most plants are <0.3 m tall and are hard to differentiate on imagery, partly due to lack of shadows. We could not reliably detect increases in dwarf shrubs, but if taller shrubs (e.g., alder) were to invade the tundra, they would be easy to detect. A vegetation type was assigned for each time period, but in practice, types could not be photointerpreted on the 1950s images without referring to the later images, so we did not analyze changes in vegetation type over the study period. Stereoscopy could have improved interpretation but was not used.
The combination of high variability in landscape characteristics (e.g., the diverse vegetation types and substrates) and high variability in drivers of change across the Refuge landscape, combined with a small sample size (35 sites spread across three biomes), limited our ability to detect significant differences in the data. In particular, large differences in the abundance of highly dynamic alluvial terrain, polygonized ground that is subject to ice-wedge degradation, and fires that are highly variable in space and time lead to high variability in both vegetation and change types. Due to high between-site variability and low number of sites, confidence intervals overlapped for most of the comparisons we made. As high-resolution satellite imagery becomes more available, cost decreases, and methods are developed to efficiently automate the image rectification process, larger sample sizes will be feasible. Similar studies to ours could have larger sample sizes with little increase in interpretation effort by having more sites and fewer points per site. Yet, the high variability among sites related to different vegetation types being affected by different change drivers in different regions at different times will remain a large obstacle in assessing whether changes are significant.
The 15 grids in the Tundra Biome had the best-quality aerial photographs, so we are most confident of our results for that biome. The photographs from the first time period (~1952) were of better resolution and higher quality than those available for the rest of the Refuge. For the second time period (~1982), most Tundra grids had excellent aerial photographs from four years: 1981, 1984, 1985, and 1988. We used the 1985 photograph to record data for that time period. However, we eventually georeferenced and examined the other photos at most grids to aid in interpretation. This was useful for detecting ice-wedge degradation because it gave us a range of different water levels to determine what normal seasonal variability was (Figure A2
). We found observed widths and extent of ice-wedge polygon troughs remained fairly constant for the 1981–1988 period despite expected rising and falling water levels over the summer season, allowing more confidence in our interpretation of ice-wedge degradation. We believe this is because actively subsiding troughs are steep sided, minimizing changes to the aerial extent of water as water rises and lowers in the troughs. We were conservative about assigning ice-wedge degradation change. For example, if a point had similar patterns of surface water in ~2004 to any one of the 1980s years, we did not interpret it as changed, even if the area of surface water was different than in 1985.
These empirical data can be applied to modelling efforts to improve prediction of future change by providing realistic input variables to models. The dataset developed by this study has been used for projecting future changes across a broader region of northern Alaska using state-transition modeling [62
] and for landscape change analysis using satellite remote sensing and decision-tree modeling [37
]. We found that environmental variables were very useful in explaining variations in change types across the region and can be incorporated into other studies. Sormunen et al. [63
] showed that including local environmental conditions, such as topography and soils information, in models of subarctic vegetation change greatly improved the predictive accuracy and changed the model outputs by constraining possible vegetation shifts using more realistic data. They found that climate-only models overestimated the amount of vegetation change. Results including environmental data fine-tuned the predictions and could also predict potential refugia in future climates. Our findings of the large differences between change types on alluvial vs. nonalluvial substrates, such as shrub changes on alluvial and ice-wedge degradation on nonalluvial terrain, can be used to improve modelling of future landscape changes with climate change.