Next Article in Journal
Identification and Functional Characterization of the Nonexpressor of Pathogenesis-Related Genes 1 (NPR1) Gene in the Tea Plant (Camellia sinensis)
Next Article in Special Issue
Technological Bases for Understanding Fires around the World
Previous Article in Journal
Detection of Pine Wilt Disease Using Time Series UAV Imagery and Deep Learning Semantic Segmentation
Previous Article in Special Issue
An Ensemble Model for Forest Fire Occurrence Mapping in China
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Forests
Volume 14
Issue 8
10.3390/f14081577
forests-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Predicting the Unpredictable: Predicting Landcover in Boreal Alaska and the Yukon Including Succession and Wildfire Potential

by
Monika P. Calef
1,*,
Jennifer I. Schmidt
2,
Anna Varvak
3,† and
Robert Ziel
4
1
Environmental Studies, Soka University of America, Aliso Viejo, CA 92656, USA
2
Institute of Social and Economic Research, University of Alaska Anchorage, Anchorage, AK 99508, USA
3
Math/Science, Soka University of America, Aliso Viejo, CA 92656, USA
4
Geophysical Institute, University of Alaska Fairbanks, Fairbanks, AK 99775, USA
*
Author to whom correspondence should be addressed.
†
Retired.
Forests 2023, 14(8), 1577; https://doi.org/10.3390/f14081577
Submission received: 23 May 2023 / Revised: 25 July 2023 / Accepted: 27 July 2023 / Published: 2 August 2023
(This article belongs to the Special Issue Forest Fires Prediction and Detection)
Browse Figures
Versions Notes

Abstract

:
The boreal forest of northwestern North America covers an extensive area, contains vast amounts of carbon in its vegetation and soil, and is characterized by extensive wildfires. Catastrophic crown fires in these forests are fueled predominantly by only two evergreen needle-leaf tree species, black spruce (Picea mariana (Mill.) B.S.P.) and lodgepole pine (Pinus contorta Dougl. ex Loud. var. latifolia Engelm.). Identifying where these flammable species grow through time in the landscape is critical for understanding wildfire risk, damages, and human exposure. Because medium resolution landcover data that include species detail are lacking, we developed a compound modeling approach that enabled us to refine the available evergreen forest category into highly flammable species and less flammable species. We then expanded our refined landcover at decadal time steps from 1984 to 2014. With the aid of an existing burn model, FlamMap, and simple succession rules, we were able to predict future landcover at decadal steps until 2054. Our resulting land covers provide important information to communities in our study area on current and future wildfire risk and vegetation changes and could be developed in a similar fashion for other areas.

1. Introduction

The boreal forest of Alaska and western Canada covers an extensive area, is partially underlain by permafrost, contains vast amounts of carbon in its soils and vegetation, and is home to only very few tree species. This area is also characterized by frequent and extensive wildfires, especially crown fires, which are integral to the ecology of this biome, since they rejuvenate cold, wet sites by thawing the underlying permafrost, provide nutrients, and reset succession [1,2,3]. Annual area burned varies widely from year to year and has been linked to large-scale climate teleconnections but is ultimately a result of available dry fuels, ignition by lightning or people, and weather conducive to fire spread [4,5,6,7]. While people ignite many fires near their towns and roads, these fires do not generally contribute much to overall area burned [8,9,10].
The population of Alaska has increased more than five times since 1950 (US Census) leading to extensive development into the highly flammable wildland–urban interface (WUI) thereby drastically increasing people’s exposure to risk of catastrophic losses. Communities are faced with burning structures, air pollution, and loss of subsistence foods and ecosystem services [11,12] as well as projected wildfire response costs of around USD 1.1–2.1B from 2006 through 2100 in Alaska [13]. The simplest prevention of future loss is to limit development in hazardous zones [14]. However, this requires communities to identify areas most prone to hazards such as wildfire, permafrost collapse, and flooding.
While the boreal forest of Interior Alaska has evolved with frequent wildfires for the past 4000 years [15], plant response to fire depends on the species’ flammability, ability to resprout or reseed quickly, and fire severity and return interval. Recovery after fire is usually lead by graminoids and herbaceous plants, followed by deciduous shrubs and trees such as quaking aspen (Populus tremuloides Michx.) and paper birch (Betula papyrifera Marsh.) and eventually by slower-growing evergreen trees [16]. Both more severe fires that remove much of the soil organic layer [17,18] and more frequent fires [19] seem to inhibit full evergreen recovery and lead to a greater post-fire abundance of deciduous shrubs and trees.
Temperatures in Alaska have been warming twice as fast as the global average since the mid-20th century [20]. A detailed analysis of Alaskan climate trends from 1920–2012 indicates underlying warming and drying which is confounded by large interannual variability, as well as larger climate patterns such as the Pacific Decadal Oscillation (PDO) and Pacific-North American (PNA) [4,21]. Projections indicate that Alaska is likely to warm by 4.4–5.6 °C in the Interior by 2050 [22], which is not accompanied by increases in precipitation. Several remote sensing-based analyses have tried to understand recent forest responses to this changing climate [23,24,25,26]. These studies found areas of greening mostly due to shrub expansion into tundra and browning in the forests of Interior Alaska. A decrease in NDVI occurs immediately after fire but can also be associated with insect defoliators, dry conditions, and mid- and late successional changes from deciduous to evergreen dominance [23,27]. There is concern that recent and predicted changes in climate as well as increase in fire extent, frequency, and severity will lead to a biome shift away from especially black spruce (Picea mariana (Mill.) B.S.P.)-dominated ecosystems [18,26,28,29,30].
Understanding these vegetation and wildfire changes and their impacts on human risk exposure requires detailed and frequent landcover datasets. Yet very few such datasets are available. The National Land Cover Datasets [31] mostly focus on the conterminous US but are available for Alaska for 2001, 2011, and 2016 [32]. Unfortunately, they do not extent to Canada. The North American Land Change Monitoring Systems [33] is only available for three years (2005, 2010, and 2015) and contains a unique ‘urban and built-up’ class [34]. The landcover with the largest spatial and temporal extent in the Western Arctic is based on NASA’s Arctic-Boreal Vulnerability Experiment (ABoVE) [35]. ABoVE included a major field campaign with the purpose to provide a “better understanding of the vulnerability and resilience of ecosystems and society to this changing environment” in the Western Arctic of North America [36]. The resulting 31-year landcover data, based on LANDSAT data, landcover training data, and very high resolution imagery, provide a 15-class land cover dataset at 30 m resolution from 1984 to 2014 [35]. However, just like NLCD and NALCMS, the ABoVE classification includes only a single evergreen forest category for the entire domain.
We wanted to understand forest changes at decadal increments from 1984 to 2054 near three major towns in the western North American boreal zone in order to aid these communities with wildfire exposure assessment and planning. Therefore, (1) we further refined the singular “evergreen forest” category in the existing ABoVE land cover dataset to distinguish between evergreen tree species with different flammability and then expanded this refined classification across four historic decadal instances from 1984 to 2014. (2) We predicted future vegetation and wildfire occurrence at decadal intervals to expand our landcover data until 2054 by including future fire modeling and succession.

2. Materials and Methods

2.1. Study Areas

We chose three major communities in the far north of the western boreal zone as study areas, Fairbanks and Anchorage, Alaska, USA, and Whitehorse, Yukon, Canada (Figure 1). The Fairbanks and Whitehorse study areas are much larger than the Anchorage area (Table 1). The Anchorage study area has the largest proportion of developed area: 9% of the study area according to the 2016 NLCD [34] and the largest population.
The Anchorage study area near the southern coast consists mostly of the Municipality of Anchorage (MOA), the most populous city in Alaska. Most of the human development in the MOA is located along the western and southern coastline, while the rest of the area is mountainous and inaccessible. Only 28% of the land is covered by forest consisting mostly of paper birch, quaking aspen, balsam poplar (Populus balsamifera L. subsp. Balsamifera), and white spruce (Picea glauca (Moench) Voss s.l.) [37]. These species prefer well-drained, permafrost-free sites [38]. White spruce is a fire avoider that regenerates from seeds after fire [29] while the deciduous tree species are much less flammable, regenerate via suckering and seeds, and are often found in early successional stages [39]. Black spruce and mountain hemlock (Tsuga mertensiana (Bong.) Carriere) are less common. Black spruce generally grows on cold soils shaded from the sun, which are also poorly drained, nutrient poor, and acidic [38,40]. Black spruce is a highly flammable fire embracer with semi-serotinous cones that fuels severe crown fires [40,41,42]. It generally self-replaces after fire. Mountain hemlock grows near the treeline, is fire-sensitive, and regenerates from seeds after fire [43]. Due to the proximity of the ocean, this study area receives plentiful rainfall, 417 mm a year, (NOAA 1991–2020 normals), and fire is infrequent.
The Fairbanks study area (Table 1) is located in Interior Alaska and includes the majority of the Fairbanks North Star Borough (FNSB) which is home to the town of Fairbanks. Study area boundaries are similar to the FNSB but were adjusted to include areas most relevant to the communities impacted by wildfire [44]. This area is covered by extensive boreal forest dominated by black spruce on poorly drained acidic soils (usually also containing permafrost) in valley bottoms and on northern slopes. White spruce, trembling aspen, and birch are found at warmer, well-drained sites on south-facing slopes [39,40,45,46]. This study area experiences extensive wildfires almost every year; 30% of its area has burned since 1984.
Whitehorse is the capital of the Yukon, Canada, and is located east of the Rocky Mountains. The study area boundaries were drawn to capture surrounding communities and wildfire activity that could encroach especially from the south. Dominant tree species include mostly white spruce and lodgepole pine (Pinus contorta Dougl. ex Loud. var. latifolia Engelm.) on well-drained sites, trembling aspen on warm sites, and subalpine fir (Abies lasiocarpa (Hook.) Boivin) and paper birch on finer textured soils [47,48]. Much like black spruce, lodgepole pine is a highly flammable fire embracer sustaining crown fires. The species is still expanding its range migrating northward in northwestern Canada but has not yet reached Alaska [49]. Subalpine fir is fire sensitive and regenerates from seeds after fire. Other less common tree species include balsam poplar and black spruce on some wetlands. This study area has hardly burned for several decades, and there is concern that flammability is becoming critically high.
Table 1. Population, size, and area burned from 1984 to 2021 for the three study areas. Note that the Alaska fire perimeter database [34] includes fires > 100 ha prior to 1988 and fires > 10 ha since; the Yukon fire perimeter database [35] includes fires > 200 ha until 1997 and smaller fires especially near communities since. Population data are given for 2021 [50,51].
Table 1. Population, size, and area burned from 1984 to 2021 for the three study areas. Note that the Alaska fire perimeter database [34] includes fires > 100 ha prior to 1988 and fires > 10 ha since; the Yukon fire perimeter database [35] includes fires > 200 ha until 1997 and smaller fires especially near communities since. Population data are given for 2021 [50,51].
Study AreaPopulationArea (km2)Area Burned (km2, %)
Anchorage, AK288,1212703.8418.43, 0.68
Fairbanks, AK95,59313,822.024190.75, 30.32
Whitehorse, YK31,91312,305.27102.85, 0.84

2.2. Landcover Modeling

We developed the following major landcover modeling approaches for each of the three study sites:
  • A refinement of ABoVE’s Evergreen Forest category into different types of needle leaf species relevant to wildfire potential using several predictors in a compound modeling approach for the year with the best cross-reference data. We then applied this refined model to existing ABoVE data for the remaining years of 1984, 1994, 2004, and 2014 and used a succession sequence based on analyses of existing ABoVE data to simulate recovery after fire disturbance.
  • A future wildfire and vegetation prediction approach that used the historic fire record as reference, Monte Carlo simulations to represent expected future area burned, FlamMap’s Burn Probability module to randomly identify future large fire locations and perimeters, and succession rules derived from ABoVE data to guide burned area revegetation.

2.2.1. Refinement of the ABoVE Classification

Starting with one annual ABoVE dataset in each study area, we reclassified the evergreen forest category into more detailed vegetation classes based on the species’ wildfire response using a compound modeling approach that incorporates machine learning where better information was not available. The start year was chosen based on available cross-reference vegetation data such as Community Wildfire Protection Plans (CWPP) [44,52] or a vegetation inventory dataset [53].
We started with Fairbanks, splitting the 2004 evergreen forest category into black spruce and white spruce as identified on the 2005 CWPP by the Fairbanks North Star Borough within the boundaries of the CWPP extent (which was restricted to the borough boundary) [44]. Within this area, we developed and trained a gradient boosted decision tree machine learning model using the XGBoost package (version 1.0.1) in R (version 3.6.3, https://xgboost.ai/, accessed 1 June 2022) [54,55,56] on black spruce locations. We could then use this model to extrapolate black spruce locations to the Evergreen Forest category throughout the entire study area. All remaining evergreen pixels were classified as white spruce.
For Anchorage, we used the 2004 CWPP [52] data to identify hemlock locations in the entire study area and then reclassified the rest of the 2004 evergreen forest pixels as white and black spruce with a new gradient boosted decision tree machine learning model.
For Whitehorse, we first identified subalpine fir locations on a 5 k vegetation inventory dataset [53] from 2012 which covered approximately 2/3 of our study area. Outside the area of overlap, a minimum elevation of 1200 m was used based on a statistical analysis of mean elevation of subalpine fir pixels, which is also confirmed by species observations [53]. All evergreen pixels in the 2014 ABoVE landcover which identified as subalpine fir were reclassified accordingly. Then, we trained a new gradient boosted model on lodgepole pine locations in the same 5 k inventory dataset, which allowed us to extrapolate to the rest of the study area. All remaining other evergreen forest pixels were classified as white spruce.
Once the machine learning models were developed for the year with vegetation training data, they could be propagated to the evergreen forest category in ABoVE landcovers for all remaining decadal time steps (1984, 1994, 2004, and 2014) and domains following two rules:
  • If the pixel remains classified as evergreen forest, then the modeled classification remains the same.
  • If the pixel’s ABoVE classification changes to the evergreen forest category, then we use the trained machine learning model to refine that category.
We only applied minor additional corrections; e.g., the 2014 Fairbanks dataset did not include the impact of a very large fire the previous year; therefore, we reclassified evergreen forest pixels within the fire perimeter as sparsely vegetated. In addition, some pixels classified as herbaceous prior to 2014 were classified as deciduous forest in 2014 without any indication of an earlier disturbance or a successional response during the 30-year period. Inspection of current and historical satellite imagery by members of our group studying permafrost loss determined this earlier classification of areas as herbaceous was an error and the 2014 deciduous classification was more appropriate (Dmitry Nicolsky and Louise Farquharson, personal communication). We revised herbaceous pixels in earlier decades (29% in 1984, 25% in 1994, and 11% in 2004) to deciduous forest when they were classified as such in 2014 and lacked wildfire activity to better represent actual vegetation.

2.2.2. Model Predictors, Testing, and Validation

After substantial exploratory analysis, we settled on the following predictors that were available for our study areas. Additionally, four-fold cross-validation on training data identified the parameters, which boosted performance, as measured by area under the ROC curve (AUC-ROC):
  • (For all three study areas) Topology: elevation, aspect (split into north–south and east–west components), and slope;
  • (For all three study areas) Climate (averages for ten years prior to the year of interest from CRU 4.0 [57,58]: growing season (MJJA) temperature, growing season precipitation, summer temperature (JJA), and winter temperature (DJF);
  • (For Anchorage and Fairbanks) Wetland status using NWI GEN data [59];
  • (For Fairbanks) Time since last fire for the year of interest from the Alaska large perimeter database (https://fire.ak.blm.gov/ (accessed on 1 January 2023)) and Canadian National Fires Database [60];
  • (For Fairbanks) Soil information from the SSURGO database [59]: annual water storage, groundwater water pH, and topsoil pH.
To spatially de-correlate the data, the dataset was divided into multiple spatial rectangles (about a thousand per each geographical study area), and then 20%–25% of those rectangles were randomly assigned as test data. The remaining data were used for model training and were randomly assigned into several groups for cross-validation in tuning model parameters.
Like a random forest machine learning method, XGBoost is an ensemble learning algorithm that combines predictions from multiple decision trees. Whereas a random forest builds independent decision trees and uses bagging to aggregate the results from the decision trees, XGBoost uses a form of gradient boosting, where each additional decision tree is built to correct the prediction errors of the previous trees. This should result in a model with better performance [45]. To prevent overfitting, each decision tree is shallow and trained on a random subset of the training data. We have chosen to use the XGBoost model rather than a random forest because the R XGBoost package is significantly faster in training its models and can result in better model performance.
For each of the three locations, we thus arrived at a nonparametric model, which takes as inputs values corresponding to the independent variables described above and outputs a number between 0 and 1 which represents a weighted “vote” between the aggregate of the decision trees.
Since we wanted to use our models to give a binary result (e.g., whether or not the observation is “black spruce” or “other evergreen”), it is necessary to establish a particular cutoff for the resulting output. The cutoff for each model was determined to maximize the F1-score, which is the harmonic mean between precision and recall, which we chose to balance the tradeoff between the positive predictive value and the sensitivity of the model.
After the models were trained and the cutoffs were established, we measured each model’s performance by area under the receiver operating characteristic curve (AUC-ROC) by comparing true positive rate with false positive rate for different cutoffs. For Fairbanks, we additionally calculated the area under the curve for precision and recall (AUCPR). We then created the confusion matrices on test data and used them to determine precision and recall of each model.

2.2.3. Succession Model

To determine future changes to the landscape, we needed to develop general successional rules based on ecological knowledge of the boreal forest [1,61] and exploration of the 31-year ABoVE dataset. While not all sites follow the same pathway at the same speed, a detailed analysis of vegetation changes across the ABoVE data record combined with published information and expert knowledge allowed us to derive general transition rates and pathways for our study areas which are explained in more detail in the Appendix A (Figure A1, Figure A2 and Figure A3 and Table A1).
There were three main reasons to use the existing literature in addition to the ABoVE data. First, data become more limited as we go back in time, especially for vegetation types less prone to burning 10 years after a fire. Therefore, we combined data-driven results with current knowledge to develop informed succession rules. Second, during our data analysis we observed that fires in the early 1980s occurred in the ABoVE dataset prior to them actually occurring (for example, 1986 burns were recorded in 1984). Third, the fire perimeter boundaries for Alaska often include unburned areas (i.e., inclusions). This in addition to edge effects can affect the results from the ABoVE analysis. The resulting rules were reviewed by ecologists and fire experts in Alaska and Canada and further fine-tuned. The final succession rules were then applied to burned areas to determine landcover type after disturbance.
The Fairbanks study area experienced the most landcover change due to extensive wildfires. Since the Anchorage study area experienced essentially no disturbance, vegetation was assumed to be persistent. The succession pathway for Whitehorse is similar to the others, except for the addition of sub-alpine fir and lodgepole pine categories. The ABoVE data from the Fairbanks area were analyzed in three ways to understand succession and guide our model. We calculated the change in vegetation composition based on the time since the last fire irrespective of the vegetation type prior to burning (Figure A1). Then, we identified the vegetation type in the year prior to a wildfire (i.e., pre-fire) and tracked the composition of vegetation within each pre-fire vegetation type after a wildfire (Figure A3). Previous research indicates that pre-fire vegetation is important for post-fire succession [1,18,62]. Lastly, we created a chord graph using the statistics software R’s circlize package v. 4.15 [63] to visualize flows among vegetation types between 1984 and 2014. These results along with expert opinion were used to develop succession rules based on pre-fire vegetation type and time since last fire. When predicting future landcover, vegetation was assumed to remain the same between time steps unless there was enough time since the last fire that succession would be expected by the model or the pixels burned.

2.2.4. Future Fire Disturbance and Landcover

We utilized the 72-year record (1950–2021) of wildfire history within the Fairbanks area to understand recent trends in area burned. While there is much variability in annual area burned and the record is less reliable further back in time, the area burned seems to be increasing [64]. We calculated the 10-year moving sum of annual area burned and number of wildfires (Figure 2). Figure 2 identifies two notable departures in decadal area burned, based on extreme fire seasons of 1957, 1958, and 2004. Otherwise, decadal totals were much less variable. Both the elevated decadal totals that included the 2004 season and the elevated norms after 2014 suggest increasing areas burned and thus the impact of climate change. We used this historic trend of decades with extraordinary area burned separated by three decades with less area burned to guide assertions of future area burned totals for Fairbanks.
This history of 10-year cumulative sums (63 in total) of annual area burned were used in a Monte Carlo simulation consisting of 100,000 iterations to randomly cumulate ten years of area burned and estimate future decadal projections. The decadal number of large fires has not varied nearly as much (n = 20–32) as the total area burned over the historic record in Fairbanks. As such, we assumed little increase in the number of ignitions (kept around 25 except for the last decade which was set to 30) and attributed most of the variability in area burned to increased fire sizes. Most of the increase in area burned was reserved for one extreme decade of area burned anticipated 2045–2054.
With these assertions for the number of large fires and total area burned for each succeeding decade, we used the fire simulation model FlamMap [65] to simulate individual fire perimeters that meet those expectations. FlamMap is a fire mapping and analysis system used in fire management that calculates fire behavior independently for each cell. It cumulates spread rates and fire line directions over specified time periods to produce burned areas and perimeters. Calculations are based on land cover and terrain characteristics converted to a FlamMap fuelscape, and we tuned the model by its ability to match historic fire perimeters. Like similar fire behavior modeling systems such as BehavePlus [66,67], FARSITE [68], and NEXUS [69,70], FlamMap uses the Rothermel surface fire spread model [71,72,73]. FlamMap combines this fuelscape with flammability parameters consisting of fuel moistures and wind speeds responsible for large fire growth. The Burn Probability module starts with a specified number of randomly ignited fires and produces burned area perimeters for each specified ignition. Though there is no way to predict the accurate location and size of individual future fires, we utilized FlamMap to elicit a representative distribution of simulated fire perimeters that provides a picture of future impact to the landscape based on our expectations for each of the specified future decadal totals for ignitions and total area burned.
Subsequent to those decadal disturbance depictions, the landscapes were modified using our succession rules before modeling the next decade’s disturbances and successional progressions.
Since wildfires played a minor role in the other two study areas with <1% burned from 1984 to 2021 in each (Table 1), we concluded that it was not useful to simulate significant reductions in wildfire hazard and exposure based on unreferenced increases in area burned in future decades. Instead, because large catastrophic impacts are plausible and need to be planned for, we used simulated wildfire perimeters from FlamMap simulations to suggest where and how wildfires in the Wildland Urban Interface (WUI) around Anchorage and Whitehorse may impact those communities. For those communities, landscape, and therefore wildfire hazard, changes were estimated primarily with succession rules applied to other noted disturbances, such as planned fuel treatments and major insect and disease impacts.

3. Results

3.1. Refined ABoVE Classification

Using our approach, we were able to develop consistent and comparable landcover datasets for our three study areas across four decades that provided the forest details necessary to distinguish evergreen forest types with significantly different wildfire behavior (Figure 3).
Nearly 70% of the Anchorage study area contains non-forest landcover (Figure 3a, Table A2a). The predominant forest types are mixed forest (13% of the study area in 2014), deciduous forest (8%), and white spruce (5%). The presence of highly flammable black spruce is low (0.52%) though concentrated adjacent to several developed areas. Overall landcover change is barely noticeable through the four decades despite some spruce beetle mortality in the area.
In contrast, the Fairbanks study area (Figure 3b, Table A2b) is dominated by woodland (20% in 2014) followed by black spruce (19%). This study area is predominantly covered by trees as is indicated by the distribution of mixed forest (9%), deciduous forest (9%), and white spruce forest (8%). The landscape changed through the four decades. For example, 14% of 1984 evergreen forests, 19% of woodlands, and 18% of deciduous forest were converted to other landcover types by 2014. The distribution of sparsely vegetated land and low and tall shrubs increased dramatically, indicating disturbance.
The Whitehorse study area (Figure 3c, Table A2c) is dominated by white spruce forest (29% in 2014) followed by lodgepole pine forest (13%). Subalpine fir accounts for only 6%. From 1984 to 2014, there seems to be general vegetation recovery as herbaceous, sparsely vegetated, and barren landcover types are decreasing, while all forest types, as well as tall and low shrubs, are increasing.

3.2. Model Validation

Overall, the models were sufficiently accurate. Table 2 shows the confusion matrices for the three study areas followed by statistics in Table 3. AUC-ROC was 0.88 for Anchorage, 0.77 for Fairbanks, and 0.86 for Whitehorse. AUCPR for Fairbanks was 0.89. The Fairbanks model had the highest true positive rate probably due to the high number of black spruce pixels on which the model could be trained (Table 3). For example, for observations that are actually black spruce, the model correctly identified them 84% of the time and misclassified only 16% of them. For observations that are not black spruce, the model correctly identified them 52% of the time and misclassified 48% of them. In contrast, Anchorage had a large proportion of true negatives because only 0.5% of the study areas included black spruce for model training and testing.

3.3. Succession Model

The analysis of ABoVE data yielded information on vegetation recovery based on pre-fire vegetation type (Figure 4). Pre-fire vegetation was important for determining post-fire succession with distinct vegetation compositions among the different vegetation types (Figure A3). Therefore, the derived post-fire succession rules were based on pre-fire vegetation at 5-year increments for each study area (Table 4). In general, the first 5–10 years are frequently sparsely vegetated followed by succession into shrubs (Figure A2). Deciduous forest returns 16 years after fire; mixed forest and woodland can take 20–30 years to recover; and evergreen forests do not return for >30 years with the exception of lodgepole pine, which could reestablish after 20 years ([1,58] and Jill Johnstone, personal communication). Our chord graph visualization supports the transition of flammable evergreen and woodland pixels into other landcover types over the 31 years within the Fairbanks area (Figure 5).

3.4. Future Fire Disturbance and Landcover

We were unable to draw relationships to climate trends through 2060 that could inform changes in flammability or fuel characterizations that would produce the anticipated increase in burned area using fire behavior modeling tools such as FlamMap. At the same time, there is indication in the historic fire record that suggests important trends are already manifest. The historic trends suggest that large departures in annual burned area totals for Fairbanks have occurred in both 1958/59 and in 2004, something less than 50 years apart. This is likely due to the maturation of boreal spruce forests in the intervening period [74]. The burned area in 2004 and its impact on decadal totals is much larger than for the earlier period. Also, burned area totals since 2014 have shown a marked increase from earlier periods as well.
Figure 6a shows the plotted distribution of those 100,000 instances. The simulation suggested a bimodal distribution with two distinctly different ranges analogous to the historic trend: a lower distribution with median decadal area burned of approximately 1000 km2 that occurred in 80% of the simulations and a second, higher, and distinct range of exceptional decadal burned area totals with a median of approximately 4000 km2 that occurred in the remaining 20% of simulations. Assuming the near future is not drastically different from the past, we let these two probable yet distinctly different scenarios serve as the basis for our four future simulated decadal area burned totals. The simulation runs were guided by the 1000 km2 estimate for the first three decades (2014–2023, 2024–2033, and 2034–2043). The 4000 km2 total was used to guide the simulated total for the final decade (2044–2053). These simulated totals are shown in Figure 6b alongside the historic decadal area burned total and demonstrate conformance to the historic trend, as well as the suggestion of climate-based increase. With now nine of the ten years of the 2014–2023 decade passed, the first projected decadal total is nearly complete and is very close to our Monte Carlo estimate.
We realize that our simulated fire perimeters are only one possible version of the future, but they give us a probable estimate of where the fires might be and how much area might burn. We know from historic fire analysis that many forest stands identified with a high wildfire exposure have actually burned recently. We can also see in Figure 7 that our model recognizes that some of the tall shrub area northeast of Fox, which burned in a massive wildfire in 2004, will start to be flammable again sometime within the next decade as indicated by the two smaller fire perimeters.
The landcover composition changed very little in the Anchorage study area over the years (Figure 8a). The Fairbanks study area will continue to experience the most change in landcover (Figure 8b, Table A3a). Evergreen and deciduous forests will decline in upcoming decades, which becomes more drastic during the high intensity fire decade preceding 2054. Mixed forest and woodland will expand slightly from 2024 to 2044 but suffer reductions by 2054. Many trees are replaced by low and tall shrubs by 2044. After the intense fire decade, tall shrubs will dominate 24% of the landscape, followed by woodland (15%) and sparsely vegetated (13%). In 2054, black spruce will cover 47% of its 2014 extent and 41% of its 1984 extent. White spruce will be reduced to 49% and 41% of its respective 2014 and 1984 extent. Forests will no longer dominate the Fairbanks landscape in 2054 when they will account for only 27% with an additional 15% of woodland (compared to 45% and 20% in 2014, respectively). The Whitehorse study area will continue its conversion of sparsely vegetated, tall, and open shrub to mostly woodland, white spruce, and mixed forest (Figure 8c, Table A3c). However, overall landscape change is minimal.

4. Discussion

In this study, we set out to create consistent landcover datasets for our three study areas at decadal increments. We then projected landcover into the future to better understand wildfire exposure for communities across eight decades.

4.1. Historic Land Cover and Succession

Ecosystems in boreal Alaska and western Canada are responding to recent changes in climate as well as the fire cycle. Several remote sensing analyses of unburned forest areas at various scales and resolutions ranging from 1980 to 2014 indicate that while the majority of the forest biome is stable or going through succession as expected [26,27,30,75], our analysis of ABoVE time series data confirms that the majority of the forest in our study still appears to be following historic succession pathways at the landscape scale.
However, there are indications of change along the northern and southern border, as well as in the eastern portion of Interior Alaska [26,27,30,75]. There, “browning” of forest pixels has been detected, indicating a decrease in NDVI or ecosystem productivity. Studies that include recent fires [24,27,29,76] describe an ecosystem shift from evergreen to deciduous forest which might have starting around 1990 [28].
The Fairbanks study area in Interior Alaska is dominated by forest and woodland (accounting for nearly 75% of the study area in 1984), contains the most highly flammable black spruce forest (21% in 1984), and experienced extensive wildfires (30% of the study area burned between 1984 and 2021). This led to a notable decrease in forest and woodland between 1984 and 2014 when they accounted for only 65% of the study area, while sparsely vegetated area increased by 84%. The chord diagram (Figure 5) visualizes this “flow” of evergreen and woodland pixels to other landcover types.
Several studies have discussed this shift to alternate succession trajectories as more frequent wildfires can prevent black spruce recovery when there is not enough recovery time [18,28,29]. White spruce regeneration might be impeded by moisture stress and higher temperatures at dry sites [77,78], while potentially encroaching on sites where permafrost is melting [78]. While the general assumption is a replacement of evergreen forest with deciduous, a tree-ring study from Interior Alaska found decreasing growth trends for aspen since the 1950s with sharper reductions during insect defoliation events, while recent birch growth also was slightly lower than the 20th century mean [77]. Our landcover analysis shows decreases in deciduous forest in the Fairbanks study area by 18% between 1984 to 2014.
Whitehorse is our second-most forested study area, where forests and woodlands accounted for nearly 55% of landcover in 1984 and increased to 59% in 2014, while herbaceous and sparsely vegetated landcover types decreased. This indicates general vegetation recovery and occurred in the near absence of fires, though some flammable vegetation was removed via fuel treatment. Anchorage contains the largest urban area and the highest population despite being the smallest study area. It had the least amount of forest and woodland cover in 1984 (32%) and experienced essentially no change in landcover. Wildfires were also essentially absent due to a moister climate and less flammable vegetation.
While needle-leaf trees are generally more flammable than deciduous species, being able to differentiate the major evergreen tree species based on their flammable vastly improves fire model predictions and thus risk assessment for communities. Since our study areas extended across two countries and three sites, a lack of widely available predictor data would have been problematic for a more static empirical model. Instead, our machine learning model was able to incorporate available data but also interpolate when data were insufficient. This was a critical ability for a model in a study such as ours where available data rarely covered the entire study area. Additionally, choosing a decision tree model over a random forest model, which would have been similar, vastly increased model processing speed. Overall, model accuracy could possibly have been improved by incorporating more predictors than the climate, soils, and topographic information we used, but after a year spent searching for and testing additional datasets, these datasets covered most of our study areas, were at an appropriate resolution, and seemed most robust.

4.2. Future Fire Disturbance and Landcover

Flammap [65] is a well-established fire behavior mapping and analysis program that calculates fire behavior cell-by-cell in a GIS environment. Since the model simulates fire spreading through surface fuels, it helps identify areas most at risk of burning and therefore at priority for fuel treatment or similar hazard reduction measure. For example, it was used to assess the impact of fuel loads and topography on fire in California montane forest [79], pre- and post-fire fuel treatment in Utah [72], and the burn potential of fire refugia in the Eastern Canadian mixed forest [80]. Using flammability settings based on our refined ABoVE landcover types, FlamMap could be used to predict areas likely to burn in the near future in our study areas. Our assumption of increased wildfire in the last decade of our prediction was based on historic incidence of high fire decades and matches predictions of climatic wildfire thresholds that will result in larger areas burned in the future [81,82,83].
Our model predicts a continued conversion of forests to non-forests in the highly flammable Fairbanks study area. This includes evergreen, deciduous, and mixed forest, while woodland remains essentially the same and results in a landscape where tall shrubs, woodland, and sparsely vegetated account for 52% of the area by 2054. While there are several predictions of conversions of evergreen forests to deciduous and some reduction in forest altogether in an impeding major ecological shift [28,30,84], concrete details on the speed of this conversion or the actual area are lacking.
We realize that our simulated future firescars are only one possible version of the future, but they provide a realistic estimate of where future fires are likely, based on flammability of the landscape and how much area might burn. We know from historic fire analysis that many forest stands identified with a high wildfire exposure have actually burned recently. We can also see in Figure 7 that our model recognizes that some of the tall shrub area northeast of Fox which burned in a massive wildfire in 2004 will start to be flammable again sometime within the next decade as indicated by the two smaller firescars. Otherwise, direct comparison of model results with other people’s simulations is nearly impossible due to vast differences in study areas and temporal and spatial resolution. Balshi et al. [81] predicted a doubling of area burned from 1991–2000 to 2041–2050 which is much lower than our numbers in Figure 6, which indicates a roughly fourfold increase in area burned. Simulations with ALFRESCO indicated that annual area burned would increase “markedly” after 2010 [28]. McCoy [83] predicted a doubling in area burned in the central Yukon by 2069 based on projected fire weather. But since our predictions are based on current vegetation, its flammability, and past fire history, not future climate, our estimates are not comparable.
By refining and expanding ABoVE landcover datasets beyond their temporal coverage, our modeling approach resulted in a consistent and sufficiently detailed land cover datasets for our three study areas from 1984 to 2054 that identified evergreen forest stands with different flammability. This provides critical information for communities and agencies preparing for climate-related hazards who will have to adapt to increased fire risk, exposure, and costs in the boreal zone [85].
This, in turn should eventually reduce available fuel and thus fires [86]. Additionally, defoliation-causing insects and disease seem to be spreading and intensifying in some areas [87]. These changes will be associated with carbon losses, but eventually, the lower flammability of deciduous ecosystems should stabilize the wildfire regime [86,88].

5. Conclusions

We developed a unique approach to landcover modeling in the western North American boreal forest by combining multiple models to fill the data gap at the appropriate level of detail and temporal and spatial extent. Resulting landcovers can be used to determine wildfire risks from 1984 to 2054 in our three study areas and will provide critical planning tools to affected communities. Our relatively simple modeling approach could serve as a blueprint for similar efforts elsewhere.

Author Contributions

Conceptualization, J.I.S.; methodology, M.P.C., J.I.S., A.V. and R.Z.; software, M.P.C., J.I.S., A.V. and R.Z.; formal analysis, J.I.S., A.V. and R.Z.; investigation, M.P.C., J.I.S., A.V. and R.Z.; resources, M.P.C., J.I.S., A.V. and R.Z.; data curation, M.P.C., J.I.S., A.V. and R.Z.; writing—original draft preparation, M.P.C.; writing—review and editing, M.P.C., J.I.S., A.V. and R.Z.; visualization, M.P.C., J.I.S., A.V. and R.Z.; supervision, J.I.S.; project administration, J.I.S.; funding acquisition, J.I.S. and M.P.C. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Science Foundation, grant number 1927563 and The Pacific Basin Research Center.

Data Availability Statement

The landcover layers, as well as the R scripts presented in this study, are openly available in https://osf.io/vr93u/ (accessed on 1 February 2023).

Acknowledgments

We would like to thank Jill Johnstone, Dmitry Nicolsky, and Louise Farquharson for feedback on the vegetation layers and succession sequences.

Conflicts of Interest

The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

Appendix A

Deriving general succession rules required a thorough analysis of AboVE data in multiple steps. We also used fire history between 1950 and 2020 while our analysis of pre-fire vegetation effects was limited to fires between 1985 and 2014. Fairbanks was the only study area with extensive wildfire activity with over 23% of the study area burnt in 2004 and subsequent impacts on the vegetation composition. To build a vegetation succession model, we used the ABoVE dataset to explore how the vegetation changed over time (Figure A1). We then reorganized this information into proportion of pixels for each year before and after fire by vegetation type (Figure A2). This then allowed us to show how vegetation changed after fire based on pre-fire vegetation (Figure A3). Pre-fire vegetation is the vegetation identified in the year prior to burning. Areas that burned were primarily woodland (35%) and evergreen (28%) followed by tall shrub (9%) and fen (6%). All other vegetation types accounted for less than 5% of the area burned.
In order to identify the dominant vegetation type at each 5-year increment following disturbance, we plotted distribution of vegetation types after fire (Table A1). Our analysis indicates that sparse vegetation dominates the landscape during the first 5 years after a wildfire. Even though woodland that burned tended to return to woodland immediately after a fire, we believe it would be more realistic to transition to sparsely vegetated, then open shrubs, and back to woodland. Information in this table was then simplified into Table 4. Lessons learned from the ABoVE data were also compared with the literature and vetted by plant ecologists and wildfire experts in Alaska and Canada to develop rules for succession.
Detailed landcover data resulting from our modeling efforts for each decadal timestep are given in Table A2 (1984 to 2014) and Table A3 (predicted future vegetation for 2024 to 2054).
Forests 14 01577 g0a1
Figure A1. Changes in land cover in the Fairbanks study area as identified in ABoVE by year.
Figure A1. Changes in land cover in the Fairbanks study area as identified in ABoVE by year.
Forests 14 01577 g0a1
Forests 14 01577 g0a2
Figure A2. Changes in land cover in the Fairbanks study area as identified in ABoVE pixels of time since the last fire.
Figure A2. Changes in land cover in the Fairbanks study area as identified in ABoVE pixels of time since the last fire.
Forests 14 01577 g0a2
Forests 14 01577 g0a3
Figure A3. Cell counts of the vegetation present after a fire based on pre-fire vegetation type. Only burns between 1985 and 2021 were used. Pre-fire vegetation for burns after 2014 was assumed to be the same as 2014 counts.
Figure A3. Cell counts of the vegetation present after a fire based on pre-fire vegetation type. Only burns between 1985 and 2021 were used. Pre-fire vegetation for burns after 2014 was assumed to be the same as 2014 counts.
Forests 14 01577 g0a3
Table A1. Landcover type by time since last fire in the Fairbanks study area using the 1984–2014 ABoVE vegetation dataset and the Alaska wildfire dataset from 1950 through 2021 (https://fire.ak.blm.gov/ (accessed on 1 January 2023)). Numbers are given as a percent of pixels at each time step. Bold text indicates the dominant vegetation. If vegetation types are within 2 percent of each other, they are considered co-dominant.
Table A1. Landcover type by time since last fire in the Fairbanks study area using the 1984–2014 ABoVE vegetation dataset and the Alaska wildfire dataset from 1950 through 2021 (https://fire.ak.blm.gov/ (accessed on 1 January 2023)). Numbers are given as a percent of pixels at each time step. Bold text indicates the dominant vegetation. If vegetation types are within 2 percent of each other, they are considered co-dominant.
Time Since Last Fire
Vegetation05101520253035404550556064
Barren33300000000011
Bog00000000000000
Deciduous2118121716171619161322
Evergreen15438881314151823242931
Fen5910107777657776
Herbaceous414161013997533322
Low shrub3101186554432222
Mixed3116810910121311121214
Open shrub00000011100010
Shallows/littoral00000000000000
Sparsely vegetated40272131110001122
Tall shrub613162422201717171515171212
Tundra01110000000000
Water00000000000000
Woodland1918172221222222242222223029
Table A2. Landcover change through time in the (a) Anchorage, (b) Fairbanks, and (c) Whitehorse study areas. Area units are given in km2 and percent and are based on 30 m pixels.
Table A2. Landcover change through time in the (a) Anchorage, (b) Fairbanks, and (c) Whitehorse study areas. Area units are given in km2 and percent and are based on 30 m pixels.
(a)1984199420042014
Vegetationkm2%km2%km2%km2%
Barren59221.8958921.7859121.8658921.78
Bog0-0-0-0-
Deciduous Forest1896.991987.322127.842057.58
Evergreen Forest: Black Spruce140.52140.52130.48130.48
Evergreen Forest: White Spruce1505.551515.581495.511455.36
Fen130.48140.52130.48140.52
Hemlock461.7461.7461.7461.7
Herbaceous45816.9445516.8344716.5345516.83
Low Shrub50.1850.1860.2260.22
Mixed Forest36413.4636613.5436513.535913.28
NA1585.841585.841585.841585.84
Open Shrubs1043.851013.74953.51943.48
Shallows/littoral20.0730.1120.0720.07
Sparsely Vegetated2198.12188.062268.362288.43
Tall Shrub2318.542278.392228.212328.58
Tussock Tundra10.0410.0420.0720.07
Water572.11562.07562.07562.07
Woodland993.661003.71003.71003.7
(b)1984199420042014
Vegetationkm2%km2%km2%km2%
Barren1961.421861.352631.92601.88
Bog70.0570.0570.0570.05
Deciduous Forest156111.29157011.36142610.3212759.23
Evergreen Forest: Black Spruce293921.26300821.76284220.56256318.54
Evergreen Forest: White Spruce12819.2713139.511218.1110717.75
Fen6414.646834.945834.227775.62
Herbaceous5223.784513.263862.796734.87
Low Shrub4233.064203.043532.554983.6
Mixed Forest11698.4612078.7312258.8613089.46
NA20.0120.0120.0120.01
Open Shrubs1080.78950.69380.27360.26
Shallows/littoral120.09130.09120.09120.09
Sparsely Vegetated2311.672131.54143410.3710247.41
Tall Shrub11558.3611017.971106813649.87
Tussock Tundra460.33450.33420.3520.38
Water1551.121591.151731.251711.24
Woodland337424.41334824.22281020.33272819.74
(c)1984199420042014
Vegetationkm2%km2%km2%km2%
Barren7075.756505.286345.156245.07
Bog00000010.01
Deciduous Forest540.44710.58960.781291.05
Evergreen Forest: Lodgepole Pine149212.13159312.95161813.15163413.28
Evergreen Forest: White Spruce339927.62352928.68356728.99357829.07
Fen2121.722922.372952.43142.55
Herbaceous126710.310328.399307.569267.52
Low Shrub2441.983242.633252.643262.65
Mixed Forest1841.52211.82612.122832.3
NA70.0670.0670.0670.06
Open Shrubs3072.492451.992391.942451.99
Shallows/littoral170.14190.15200.16200.16
Sparsely Vegetated123610.041107910448.489497.71
Subalpine Fir6415.216745.486915.626955.65
Tall Shrub9427.669417.659757.929837.99
Tussock Tundra400.33570.46520.42470.38
Water5834.745864.765884.785884.78
Woodland9737.919577.789647.839597.79
Table A3. Predicted future landcover in the (a) Anchorage, (b) Fairbanks, and (c) Whitehorse study areas. Area units are given in km2 and are based on 30 m pixels.
Table A3. Predicted future landcover in the (a) Anchorage, (b) Fairbanks, and (c) Whitehorse study areas. Area units are given in km2 and are based on 30 m pixels.
(a)2024203420442054
Vegetationkm2%km2%km2%km2%
Barren58821.7358821.7358821.7358821.73
Bog00000000
Deciduous Forest2047.552057.582057.582057.58
Evergreen Forest: Black Spruce130.47130.47130.47130.47
Evergreen Forest: White Spruce1445.321445.321445.321445.32
Fen140.52140.52140.52140.52
Hemlock461.68461.68461.68461.68
Herbaceous45416.7845416.7845416.7845416.78
Low Shrub80.370.2580.2980.29
Mixed Forest35713.235813.2535913.2735913.27
NA1575.81575.81575.81575.8
Open Shrubs943.47943.46943.46943.46
Shallows/littoral20.0820.0820.0820.08
Sparsely Vegetated2278.412278.42278.42278.4
Tall Shrub2358.672348.662338.612338.61
Tussock Tundra20.0620.0620.0620.06
Water562.07562.07562.07562.07
Woodland1003.681003.691003.691003.69
(b)2024203420442054
Vegetationkm2%km2%km2%km2%
Barren1931.391931.391931.391931.39
Bog70.0570.0570.0570.05
Deciduous Forest13229.5613349.6512909.3310817.82
Evergreen Forest: Black Spruce228216.51195314.13174212.6111928.62
Evergreen Forest: White Spruce9066.557965.767195.25273.81
Fen6284.546284.546284.556294.55
Herbaceous3902.823902.823902.823902.82
Low Shrub4873.525253.85153.7310587.65
Mixed Forest12228.8412549.0812819.279717.03
NA10.0110.0110.0110.01
Open Shrubs10197.382021.462721.963552.57
Shallows/littoral120.09120.08110.08110.08
Sparsely Vegetated9586.939196.658406.08176112.74
Tall Shrub207515.01253618.35300421.73329723.86
Tussock Tundra430.31430.31430.31430.31
Water1711.241711.241711.241711.24
Woodland210415.22285720.67271219.63213315.43
(c)2024203420442054
Vegetationkm2%km2%km2%km2%
Barren6235.066235.066235.066235.06
Bog10.0110.0110.0110.01
Deciduous Forest1291.051311.061301.061281.04
Evergreen Forest: Lodgepole Pine164113.33163913.32164413.36164413.36
Evergreen Forest: White Spruce355728.91356628.98357329.04359029.18
Fen3132.543132.543132.543132.54
Herbaceous9287.549287.549277.539267.53
Low Shrub3262.653272.663272.653252.64
Mixed Forest2832.302832.302832.302852.32
NA40.0340.0340.0340.03
Open Shrubs2492.032441.992451.992431.98
Shallows/littoral200.17200.17200.16200.16
Sparsely Vegetated9507.729277.549227.499227.49
Subalpine Fir6925.626925.626925.626945.64
Tall Shrub9988.1110088.1910028.149848.00
Tussock Tundra470.38470.38470.38470.38
Water5884.785884.785884.785884.78
Woodland9567.779627.829647.849667.85

References

  1. Johnstone, J.F.; Chapin, F.S., III; Foote, J.; Kemmett, S.; Price, K.; Viereck, L.A. Decadal observations of tree regeneration following fire in boreal forests. Can. J. For. Res. 2004, 34, 267–273. [Google Scholar] [CrossRef]
  2. Yarie, J. Forest fire cycles and life tables: A case study from interior Alaska. Can. J. For. Res. 1981, 11, 554–562. [Google Scholar] [CrossRef]
  3. Van Cleve, K.; Chapin, F.S., III; Dyrness, C.T.; Viereck, L.A. Element cycling in taiga forests: State-factor control. Bioscience 1991, 41, 78–88. [Google Scholar] [CrossRef]
  4. Justino, F.; Bromwich, D.H.; Schumacher, V.; Dasilva, A.; Wang, S.-H. Arctic Oscillation and Pacific-North American pattern dominated-modulation of fire danger and wildfire occurrence. NPJ Clim. Atmos. Sci. 2022, 5, 52. [Google Scholar] [CrossRef]
  5. Duffy, P.A.; Walsh, J.E.; Graham, J.M.; Mann, D.H.; Rupp, T.S. Impacts of large-scale atmospheric–ocean variability on alaskan fire season severity. Ecol. Appl. 2005, 15, 1317–1330. [Google Scholar] [CrossRef] [Green Version]
  6. Kasischke, E.S.; Turetsky, M.R. Recent changes in the fire regime across the North American boreal region—Spatial and temporal patterns of burning across Canada and Alaska. Geophys. Res. Lett. 2006, 33, 1–5. [Google Scholar] [CrossRef] [Green Version]
  7. Krawchuk, M.A.; Cumming, S.G.; Flannigan, M.D.; Wein, R.W. Biotic and Abiotic Regulation of Lightning Fire Initiation in the Mixedwood Boreal Forest. Ecology 2006, 87, 458–468. [Google Scholar] [CrossRef]
  8. Calef, M.P.; McGuire, A.D.; Chapin, F.S., III. Human influences on wildfire in Alaska from 1988 through 2005: An analysis of the spatial patterns of human impacts. Earth Interact 2008, 12, 1–17. [Google Scholar] [CrossRef] [Green Version]
  9. Kasischke, E.S.; Williams, D.; Barry, D. Analysis of the patterns of large fires in the boreal forest region of Alaska. Int. J. Wildland Fire 2002, 11, 131–144. [Google Scholar] [CrossRef] [Green Version]
  10. Veraverbeke, S.; Rogers, B.M.; Goulden, M.L.; Jandt, R.R.; Miller, C.E.; Wiggins, E.B.; Randerson, J.T. Lightning as a major driver of recent large fire years in North American boreal forests. Nat. Clim. Chang. 2017, 7, 529–534. [Google Scholar] [CrossRef]
  11. Trainor, S.F.; Calef, M.; Natcher, D.; Chapin, F.S., III; McGuire, A.D.; Huntington, O.; Duffy, P.; Rupp, T.S.; DeWilde, L.O.; Kwart, M.; et al. Vulnerability and adaptation to climate-related fire impacts in rural and urban interior Alaska. Polar Res. 2009, 28, 100–118. [Google Scholar] [CrossRef]
  12. Nelson, J.L.; Zavaleta, E.S.; Chapin, F.S.I. Boreal Fire Effects on Subsistence Resources in Alaska and Adjacent Canada. Ecosystems 2008, 11, 156–171. [Google Scholar] [CrossRef]
  13. Melvin, A.M.; Murray, J.; Boehlert, B.; Martinich, J.A.; Rennels, L.; Rupp, T.S. Estimating wildfire response costs in Alaska’s changing climate. Clim. Chang. 2017, 141, 783–795. [Google Scholar] [CrossRef] [Green Version]
  14. Iglesias, V.; Braswell, A.E.; Rossi, M.W.; Joseph, M.B.; McShane, C.; Cattau, M.; Koontz, M.J.; McGlinchy, J.; Nagy, R.C.; Balch, J.; et al. Risky Development: Increasing Exposure to Natural Hazards in the United States. Earth’s Futur. 2021, 9, e2020EF001795. [Google Scholar] [CrossRef]
  15. Hu, F.S.; Brubaker, L.B.; Gavin, D.G.; Higuera, P.E.; Lynch, J.A.; Rupp, T.S.; Tinner, W. How Climate and Vegetation Influence the fire Regime of the Alaskan Boreal Biome: The Holocene Perspective. Mitig. Adapt. Strat. Glob. Chang. 2006, 11, 829–846. [Google Scholar] [CrossRef]
  16. Chapin, F.S.; Viereck, L.A.; Adams, P.C.; Van Cleve, K.; Fastie, C.L.; Ott, R.A.; Mann, D.; Johnstone, J.F. Successional processes in the Alaskan boreal forest. In Alaska’s Changing Boreal Forest; Chapin, F.S.I., Oswood, M.W., Van Cleve, K., Viereck, L.A., Verbyla, D.L., Eds.; Oxford University Press: New York, NY, USA, 2006; pp. 100–120. [Google Scholar]
  17. Shenoy, A.; Johnstone, J.F.; Kasischke, E.S.; Kielland, K. Persistent effects of fire severity on early successional forests in interior Alaska. For. Ecol. Manag. 2010, 261, 381–390. [Google Scholar] [CrossRef]
  18. Johnstone, J.F.; Hollingsworth, T.N.; Chapin, F.S., III; Mack, M.C. Changes in fire regime break the legacy lock on successional trajectories in Alaskan boreal forest. Glob. Chang. Biol. 2010, 16, 1281–1295. [Google Scholar] [CrossRef]
  19. Johnstone, J.F. Response of boreal plant communities to variations in previous fire-free interval. Int. J. Wildland Fire 2006, 15, 497–508. [Google Scholar] [CrossRef]
  20. Markon, C.; Gray, S.; Berman, M.; Eerkes-Medrano, L.; Hennessy, T.; Huntington, H.; Littell, J.; McCammon, M.; Thoman, R.; Trainor, S. Alaska. In Impacts, Risks, and Adaptation in the United States: Fourth National Climate Assessment; Reidmiller, D.R., Avery, C.W., Easterling, D.R., Kunkel, K.E., Lewis, K.L.M., Maycock, T.K., Stewart, B.C., Eds.; U.S. Global Change Research Program: Washington, DC, USA, 2018; Volume II, pp. 1185–1241. [Google Scholar]
  21. Bieniek, P.A.; Walsh, J.E.; Thoman, R.L.; Bhatt, U.S. Using Climate Divisions to Analyze Variations and Trends in Alaska Temperature and Precipitation. J. Clim. 2014, 27, 2800–2818. [Google Scholar] [CrossRef] [Green Version]
  22. Chapin, F.S.I.; Trainor, S.F.; Cochran, P.; Huntington, H.; Markon, C.; McCammon, M.; McGuire, A.D.; Serreze, M. Alaska. In Climate Change Impacts in the United States: The Third National Climate Assessment; Melillo, J.M., Richmond, T.C., Yohe, G.W., Eds.; U.S. Global Change Research Program: Washington, DC, USA, 2014; pp. 514–536. [Google Scholar]
  23. Fiore, N.M.; Goulden, M.L.; Czimczik, C.I.; Pedron, S.A.; Tayo, M.A. Do recent NDVI trends demonstrate boreal forest decline in Alaska? Environ. Res. Lett. 2020, 15, 095007. [Google Scholar] [CrossRef]
  24. Goetz, S.J.; Bunn, A.G.; Fiske, G.J.; Houghton, R.A. Satellite-observed photosynthetic trends across boreal North America associated with climate and fire disturbance. Proc. Natl. Acad. Sci. USA 2005, 102, 13521–13525. [Google Scholar] [CrossRef]
  25. Verbyla, D. The greening and browning of Alaska based on 1982–2003 satellite data. Glob. Ecol. Biogeogr. 2008, 17, 547–555. [Google Scholar] [CrossRef]
  26. Ju, J.C.; Masek, J.G. The vegetation greenness trend in Canada and US Alaska from 1984–2012 Landsat data. Remote Sens. Environ. 2016, 176, 1–16. [Google Scholar] [CrossRef]
  27. Wylie, B.; Rigge, M.; Brisco, B.; Murnaghan, K.; Rover, J.; Long, J. Effects of Disturbance and Climate Change on Ecosystem Performance in the Yukon River Basin Boreal Forest. Remote Sens. 2014, 6, 9145–9169. [Google Scholar] [CrossRef] [Green Version]
  28. Mann, D.H.; Rupp, T.S.; Olson, M.A.; Duffy, P.A. Is Alaska’s boreal forest now crossing a major ecological threshold? Arct. Antarct. Alp. Res. 2012, 44, 319–331. [Google Scholar] [CrossRef] [Green Version]
  29. Baltzer, J.L.; Day, N.J.; Walker, X.J.; Greene, D.; Mack, M.C.; Alexander, H.D.; Arseneault, D.; Barnes, J.; Bergeron, Y.; Boucher, Y.; et al. Increasing fire and the decline of fire adapted black spruce in the boreal forest. Proc. Natl. Acad. Sci. USA 2021, 118, e2024872118. [Google Scholar] [CrossRef]
  30. Beck, P.S.A.; Goetz, S.J.; Mack, M.C.; Alexander, H.D.; Jin, Y.; Randerson, J.T.; Loranty, M.M. The impacts and implications of an intensifying fire regime on Alaskan boreal forest composition and albedo. Glob. Chang. Biol. 2011, 17, 2853–2866. [Google Scholar] [CrossRef]
  31. Xian, G.; Homer, C.; Yang, L. Development of the USGS National Land Cover Database over two decades. In Advances in Environmental Remote Sensing: Sensors, Algorithms, and Applications; Weng, Q., Ed.; CRC Press: Boca Raton, FL, USA, 2011; pp. 529–544. [Google Scholar]
  32. NLCD (National Land Cover Datasets). 2016. Available online: https://www.mrlc.gov/data/nlcd-2016-land-cover-alaska (accessed on 1 January 2023).
  33. NALCMS (North American Land Change Monitoring Systems). Available online: http://www.cec.org/north-american-land-change-monitoring-system/ (accessed on 1 January 2023).
  34. Jin, S.; Homer, C.; Yang, L.; Danielson, P.; Dewitz, J.; Li, C.; Zhu, Z.; Xian, G.; Howard, D. Overall Methodology Design for the United States National Land Cover Database 2016 Products. Remote Sens. 2019, 11, 2971. [Google Scholar] [CrossRef] [Green Version]
  35. Wang, J.A.; Sulla-Menashe, D.; Woodcock, C.E.; Sonnentag, O.; Keeling, R.F.; Friedl, M.A. ABoVE: Landsat-Derived Annual Dominant Land Cover across ABoVE Core Domain, 1984–2014; ORNL DAAC: Oak Ridge, TN, USA, 2019. [Google Scholar]
  36. ABoVE. About ABoVE. NASA. 2022. Available online: https://above.nasa.gov/about.html (accessed on 1 January 2023).
  37. Cheyette, D.; Rupp, T.S.; Rodman, S. Developing Fire Behavior Fuel Models for the Wildland–Urban Interface in Anchorage, Alaska. West. J. Appl. For. 2008, 23, 149–155. [Google Scholar] [CrossRef] [Green Version]
  38. Viereck, L.A.; Dyrness, C.T.; Batten, A.R.; Wenzlick, K.J. The Alaska Vegetation Classification; USDA Forest Service: Portland, OR, USA, 1992; 278p. [Google Scholar]
  39. Van Cleve, K.; Viereck, L.A. Forest succession in relation to nutrient cycling in the boreal forest of Alaska. In Forest Succession, Concepts and Application; West, D.C., Shugart, H.H., Botkin, D.B., Eds.; Springer: New York, NY, USA, 1981; pp. 185–210. [Google Scholar]
  40. Van Cleve, K.; Dyrness, C.T.; Viereck, L.A.; Fox, J.; Chapin, F.S., III; Oechel, W.C. Taiga ecosystems in interior Alaska. Bioscience 1983, 33, 39–44. [Google Scholar] [CrossRef]
  41. Rogers, B.M.; Soja, A.J.; Goulden, M.L.; Randerson, J.T. Influence of tree species on continental differences in boreal fires and climate feedbacks. Nat. Geosci. 2015, 8, 228–234. [Google Scholar] [CrossRef] [Green Version]
  42. Johnstone, J.F.; Boby, L.; Tissier, E.; Mack, M.C.; Verbyla, D.; Walker, X. Postfire seed rain of black spruce, a semiserotinous conifer, in forests of interior Alaska. Can. J. For. Res. 2002, 39, 1575–1588. [Google Scholar] [CrossRef] [Green Version]
  43. Tesky, J.L. Tsuga mertensiana. U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station, Fire Sciences Laboratory 1992. Available online: https://www.fs.usda.gov/database/feis/plants/tree/tsumer/all.html (accessed on 1 January 2023).
  44. State of Alaska-Division of Forestry, Fairbanks Area Office. Community Wildfire Protection Plan for at-Risk Communities in the Fairbanks North Star Borough, Alaska Phase I. 2006. Available online: http://forestry.alaska.gov/Assets/pdfs/fire/cwpp/misc/fnsb_cwpp_v12.pdf (accessed on 1 June 2022).
  45. Viereck, L.A.; Dyrness, C.T.; Van Cleve, K.; Foote, M.J. Vegetation, soils, and forest productivity in selected forest types in interior Alaska. Can. J. For. Res. 1983, 13, 703–720. [Google Scholar] [CrossRef]
  46. Yarie, J.; Van Cleve, K. Controls over forest production in Interior Alaska. In Alaska’s Changing Boreal Forest; Chapin, F.S., III, Oswood, M.W., Van Cleve, K., Viereck, L.A., Verbyla, D.L., Eds.; LTER Oxford University Press: Cary, NC, USA, 2006; pp. 171–188. [Google Scholar]
  47. Oswald, E.T.; Brown, B.N. Forest Communities in Lake Laberge Ecoregion, Yukon Territory; Information Report BC-X.; Canadian Forestry Service, Pacific Forestry Centre: Victoria, BC, Canada, 1986. [Google Scholar]
  48. Stanek, W.; Orlóci, L. Vegetation survey of the Alaska Highway, Yukon Territory: Types and gradients. Plant Ecol. 1980, 41, 1–56. [Google Scholar] [CrossRef]
  49. Johnstone, J.F.; Chapin, F.S., III. Non-equilibrium succession dynamics indicate continued northern migration of lodgepole pine. Glob. Chang. Biol. 2003, 9, 1401–1409. [Google Scholar] [CrossRef]
  50. American Community Survey 1-Year Estimates Subject Tables. Available online: https://data.census.gov/table?t=Populations+and+People&g=0500000US02020&y=2021&tid=ACSST1Y2021.S0101 (accessed on 1 January 2023).
  51. Census of Population. Available online: https://www12.statcan.gc.ca/census-recensement/2021/dp-pd/prof/index.cfm?Lang=E (accessed on 1 January 2023).
  52. Goodrich, C.P.; Rodman, S.U.; Stam, J. Municipality of Anchorage Community Wildfire Protection Plan; Municipality of Anchorage: Anchorage, AK, USA, 2008. [Google Scholar]
  53. Vegetation Inventory-5K. Forest Management Branch. Available online: http://yukon.maps.arcgis.com/home (accessed on 1 January 2023).
  54. Chen, T. XGBoost: A Scalable Tree Boosting System. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, CA, USA, 13–17 August 2016; Krishnapuram, B., Shah, M., Smola, A.J., Aggarwal, C.C., Shen, D., Rastogi, R., Eds.; ACM: New York, NY, USA, 2016; pp. 785–794. [Google Scholar] [CrossRef]
  55. xgboost: eXtreme Gradient Boosting; R Package Version 0.4–2.1. 2015. Available online: https://github.com/dmlc/xgboost (accessed on 1 June 2022).
  56. Friedman, J.H. Greedy function approximation: A gradient boosting machine. Ann. Stat. 2001, 29, 1189–1232. [Google Scholar] [CrossRef]
  57. High-Resolution Gridded Datasets (and Derived Products). Available online: https://crudata.uea.ac.uk/cru/data/hrg/ (accessed on 1 June 2022).
  58. Harris, I.; Osborn, T.J.; Jones, P.; Lister, D. Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset. Sci. Data 2020, 7, 109. [Google Scholar] [CrossRef] [Green Version]
  59. Web Soil Survey. United States Department of Agriculture. Available online: https://websoilsurvey.nrcs.usda.gov (accessed on 1 June 2022).
  60. Natural Resources Canada, Government of Canada. Canadian National Fire Database. Available online: https://cwfis.cfs.nrcan.gc.ca/ha/nfdb (accessed on 1 June 2022).
  61. Chapin, F.S., III; Hollingsworth, T.; Murray, D.F.; Viereck, L.A.; Walker, M.D. Floristic diversity and vegetation distribution in the Alaskan boreal forest. In Alaska’s Changing Boreal Forest, 1st ed.; Chapin, F.S., III, Oswood, M.W., Van Cleve, K., Viereck, L.A., Verbyla, D.L., Eds.; Long-term ecological network series. Oxford University Press: New York, NY, USA, 2006; Volume 6, pp. 81–99. [Google Scholar]
  62. Cater, T.C.; Chapin, F.S. Differential effects of competition or microenvironment on boreal tree seedling establishment after fire. Ecology 2000, 81, 1086–1099. [Google Scholar] [CrossRef]
  63. Gu, Z.; Gu, L.; Eils, R.; Schlesner, M.; Brors, B. circlize implements and enhances circular visualization in R. Bioinformatics 2014, 30, 2811–2812. [Google Scholar] [CrossRef] [Green Version]
  64. Calef, M.P.; Varvak, A.; McGuire, A.D.; Chapin, F.S.; Reinhold, K.B. Recent Changes in Annual Area Burned in Interior Alaska: The Impact of Fire Management. Earth Interact. 2015, 19, 1–17. [Google Scholar] [CrossRef] [Green Version]
  65. Finney, M.A. An Overview of FlamMap Fire Modeling Capabilities. In Fuels Management-How to Measure Success, 2006; Andrews, P.L.B., Bret, W., Eds.; USDA Forest Service: Madison, WI, USA, 2006; pp. 28–30. [Google Scholar]
  66. Andrews, P.L.; Bevins, C.D.; Seli, R.C. BehavePlus Fire Modeling System Version 4.0: User’s Guide; RMRS-GTW-106WWW; USDA Forest Service, Rocky Mountain Research Station: Fort Collins, CO, USA, 2008. [Google Scholar]
  67. Andrews, P.L. Current status and future needs of the BehavePlus Fire Modeling System. Int. J. Wildland Fire 2014, 23, 21–33. [Google Scholar] [CrossRef] [Green Version]
  68. Finney, M.A. FARSITE: Fire Area Simulator—Model Development and Evaluation; USDA Forest Service, Rocky Mountain Research Station: RMRS-RP-4: Ogden, UT, USA, 1998. [Google Scholar]
  69. Scott, J.H. NEXUS: A system for assessing crown fire hazard. Fire Manag. Notes 1999, 59, 20–24. [Google Scholar]
  70. Scott, J.H.; Reinhardt, E.D. Assessing Crown Fire Potential by Linking Models of Surface and Crown Fire Behavior; USDA Forest Service, Rocky Mountain Research Station: Fort Collins, CO, USA, 2001; RMRS Rp-29. [Google Scholar]
  71. Rothermel, R.C. A Mathematical Model for Predicting Fire Spread in Wildland Fuels; USDA Intermountain Forest and Range Experiment Station: Ogden, UT, USA, 1972. [Google Scholar]
  72. Frost, S.M.; Alexander, M.E.; Jenkins, M.J. The Application of Fire Behavior Modeling to Fuel Treatment Assessments at Army Garrison Camp Williams, Utah. Fire 2022, 5, 78. [Google Scholar] [CrossRef]
  73. Andrews, P.L. The Rothermel Surface Fire Spread Model and Associated Developments; USDA Forest Service, Rocky Mountain Research Station: Forest Collins, CO, USA, 2018. [Google Scholar] [CrossRef]
  74. Gaglioti, B.V.; Mann, D.H.; Jones, B.M.; Wooller, M.J.; Finney, B.P. High-resolution records detect human-caused changes to the boreal forest wildfire regime in interior Alaska. Holocene 2016, 26, 1064–1074. [Google Scholar] [CrossRef]
  75. Parent, M.B.; Verbyla, D. The Browning of Alaska’s Boreal Forest. Remote Sens. 2010, 2, 2729–2747. [Google Scholar] [CrossRef] [Green Version]
  76. Wang, J.A.; Sulla-Menashe, D.; Woodcock, C.E.; Sonnentag, O.; Keeling, R.F.; Friedl, M.A. Extensive land cover change across Arctic–Boreal Northwestern North America from disturbance and climate forcing. Glob. Chang. Biol. 2020, 26, 807–822. [Google Scholar] [CrossRef]
  77. Cahoon, S.M.P.; Sullivan, P.F.; Brownlee, A.H.; Pattison, R.R.; Andersen, H.E.; Legner, K.; Hollingsworth, T.N. Contrasting drivers and trends of coniferous and deciduous tree growth in interior Alaska. Ecology 2018, 99, 1284–1295. [Google Scholar] [CrossRef]
  78. Nicklen, E.F.; Roland, C.A.; Ruess, R.W.; Scharnweber, T.; Wilmking, M. Divergent responses to permafrost and precipitation reveal mechanisms for the spatial variation of two sympatric spruce. Ecosphere 2021, 12, e03622. [Google Scholar] [CrossRef]
  79. Airey-Lauvaux, C.; Pierce, A.D.; Skinner, C.N.; Taylor, A.H. Changes in fire behavior caused by fire exclusion and fuel build-up vary with topography in California montane forests, USA. J. Environ. Manag. 2022, 304, 114255. [Google Scholar] [CrossRef]
  80. Ouarmim, S.; Paradis, L.; Asselin, H.; Bergeron, Y.; Ali, A.A.; Hély, C. Burning Potential of Fire Refuges in the Boreal Mixedwood Forest. Forests 2016, 7, 246. [Google Scholar] [CrossRef] [Green Version]
  81. Balshi, M.S.; McGUIRE, A.D.; Duffy, P.; Flannigan, M.D.; Walsh, J.; Melillo, J.M. Assessing the response of area burned to changing climate in western boreal North America using a Multivariate Adaptive Regression Splines (MARS) approach. Glob. Chang. Biol. 2009, 15, 578–600. [Google Scholar] [CrossRef]
  82. Young, A.M.; Higuera, P.E.; Duffy, P.A.; Hu, F.S. Climatic thresholds shape northern high-latitude fire regimes and imply vulnerability to future climate change. Ecography 2017, 40, 606–617. [Google Scholar] [CrossRef]
  83. McCoy, V.M.; Burn, C.R. Potential Alteration by Climate Change of the Forest-Fire Regime in the Boreal Forest of Central Yukon Territory. Arctic 2005, 58, 276–285. [Google Scholar] [CrossRef] [Green Version]
  84. Chapin, F.S., III; Callaghan, T.V.; Bergeron, Y.; Fukuda, M.; Johnstone, J.F.; Juday, G.P.; Zimov, S.A. Global change and the boreal forest: Thresholds, shifting states or gradual change? Ambio 2004, 33, 361–365. [Google Scholar] [CrossRef] [PubMed]
  85. Berman, M.; Schmidt, J.I. Economic Effects of Climate Change in Alaska. Weather. Clim. Soc. 2019, 11, 245–258. [Google Scholar] [CrossRef]
  86. Foster, A.C.; Shuman, J.K.; Rogers, B.M.; Walker, X.J.; Mack, M.C.; Bourgeau-Chavez, L.L.; Veraverbeke, S.; Goetz, S.J. Bottom-up drivers of future fire regimes in western boreal North America. Environ. Res. Lett. 2022, 17, 025006. [Google Scholar] [CrossRef]
  87. Fettig, C.J.; Asaro, C.; Nowak, J.T.; Dodds, K.J.; Gandhi, K.J.K.; Moan, J.E.; Robert, J. Trends in Bark Beetle Impacts in North America During a Period (2000–2020) of Rapid Environmental Change. J. For. 2022, 120, 693–713. [Google Scholar] [CrossRef]
  88. Kelly, R.; Chipman, M.L.; Higuera, P.E.; Stefanova, I.; Brubaker, L.B.; Hu, F.S. Recent burning of boreal forests exceeds fire regime limits of the past 10,000 years. Proc. Natl. Acad. Sci. USA 2013, 110, 13055–13060. [Google Scholar] [CrossRef] [PubMed]
Forests 14 01577 g001
Figure 1. The three study areas in the western boreal zone. Vegetation is shown for 2014. Evergreen Forest: Black Spruce is only found in Fairbanks and Anchorage; Hemlock is only found in Anchorage; and SubAlpine Fir and Evergreen Forest: Lodgepole Pine are only found in Whitehorse. Thin gray outlines delineate fire perimeters for 1980 to 2019 in Alaska and from 1980 to 2022 in the Yukon.
Figure 1. The three study areas in the western boreal zone. Vegetation is shown for 2014. Evergreen Forest: Black Spruce is only found in Fairbanks and Anchorage; Hemlock is only found in Anchorage; and SubAlpine Fir and Evergreen Forest: Lodgepole Pine are only found in Whitehorse. Thin gray outlines delineate fire perimeters for 1980 to 2019 in Alaska and from 1980 to 2022 in the Yukon.
Forests 14 01577 g001
Forests 14 01577 g002
Figure 2. Historic area burned in the Fairbanks study area. Area burned (in km2) is shown annually (blue bars) and as a 10-year moving sum (orange line).
Figure 2. Historic area burned in the Fairbanks study area. Area burned (in km2) is shown annually (blue bars) and as a 10-year moving sum (orange line).
Forests 14 01577 g002
Forests 14 01577 g003aForests 14 01577 g003b
Figure 3. Landcover from 1984 to 2014 in the (a) Anchorage, (b) Fairbanks, and (c) Whitehorse study areas. Units are given as the percent of study area. Here Bla. Spr. = Black Spruce, Whi. Spr. = White Spruce, Mixed For. = Mixed Forest, Dec. For. = Deciduous Forest, Open Shr. = Open Shrubs, Spa. Veg. = sparsely vegetated, Lod. Pine = Lodgepole Pine, and Sub. Fir = Subalpine Fir. Other includes barren, bog, fen, NA, shallows/littoral, tussock tundra, and water.
Figure 3. Landcover from 1984 to 2014 in the (a) Anchorage, (b) Fairbanks, and (c) Whitehorse study areas. Units are given as the percent of study area. Here Bla. Spr. = Black Spruce, Whi. Spr. = White Spruce, Mixed For. = Mixed Forest, Dec. For. = Deciduous Forest, Open Shr. = Open Shrubs, Spa. Veg. = sparsely vegetated, Lod. Pine = Lodgepole Pine, and Sub. Fir = Subalpine Fir. Other includes barren, bog, fen, NA, shallows/littoral, tussock tundra, and water.
Forests 14 01577 g003aForests 14 01577 g003b
Forests 14 01577 g004
Figure 4. Vegetation changes after fire (year 0) in the Fairbanks study area based on ABoVE data and organized by pre-fire vegetation type. Only burns between 1985 and 2021 were used. Pre-fire vegetation for burns after 2014 was assumed to be the same as 2014 counts. Actual pixel counts decrease with time since fire and are provided in Figure A3.
Figure 4. Vegetation changes after fire (year 0) in the Fairbanks study area based on ABoVE data and organized by pre-fire vegetation type. Only burns between 1985 and 2021 were used. Pre-fire vegetation for burns after 2014 was assumed to be the same as 2014 counts. Actual pixel counts decrease with time since fire and are provided in Figure A3.
Forests 14 01577 g004
Forests 14 01577 g005
Figure 5. Landcover change from 1984 to 2014 in the Fairbanks study area. In this chord diagram, the size of each wedge indicates the proportion of the landscape and lines across the middle indicate transitions; 1984 landcover is set back from the edge, and arrows indicate change through time.
Figure 5. Landcover change from 1984 to 2014 in the Fairbanks study area. In this chord diagram, the size of each wedge indicates the proportion of the landscape and lines across the middle indicate transitions; 1984 landcover is set back from the edge, and arrows indicate change through time.
Forests 14 01577 g005
Forests 14 01577 g006
Figure 6. Area burned trends for the Fairbanks study area. (a) Results of 100,000 Monte Carlo simulations of decadal area burned (in km2) based on random combination of historic annual area burned from the period 1950–2021. Here, frequency represents the number of simulations. (b) Historic (1974 to 2022) and simulated (2023–2053) decadal area burned (in km2), years on the x-axis indicate when the decade ended, so 1983 includes area burned from 1974 to 1983.
Figure 6. Area burned trends for the Fairbanks study area. (a) Results of 100,000 Monte Carlo simulations of decadal area burned (in km2) based on random combination of historic annual area burned from the period 1950–2021. Here, frequency represents the number of simulations. (b) Historic (1974 to 2022) and simulated (2023–2053) decadal area burned (in km2), years on the x-axis indicate when the decade ended, so 1983 includes area burned from 1974 to 1983.
Forests 14 01577 g006
Forests 14 01577 g007
Figure 7. 2034 landcover for Fairbanks with 2024–2033 modeled fire perimeters.
Figure 7. 2034 landcover for Fairbanks with 2024–2033 modeled fire perimeters.
Forests 14 01577 g007
Forests 14 01577 g008aForests 14 01577 g008b
Figure 8. Predicted future landcover in the (a) Anchorage, (b) Fairbanks, and (c) Whitehorse study areas. Units are given as percent of study area. Here, Bla. Spr. = Black Spruce, Whi. Spr. = White Spruce, Mixed For. = Mixed Forest, Dec. For. = Deciduous Forest, Open Shr. = Open Shrubs, Spa. Veg. = sparsely vegetated, Lod. Pine = Lodgepole Pine, and Sub. Fir = Subalpine Fir. Other includes barren, bog, fen, NA, shallows/littoral, tussock tundra, and water.
Figure 8. Predicted future landcover in the (a) Anchorage, (b) Fairbanks, and (c) Whitehorse study areas. Units are given as percent of study area. Here, Bla. Spr. = Black Spruce, Whi. Spr. = White Spruce, Mixed For. = Mixed Forest, Dec. For. = Deciduous Forest, Open Shr. = Open Shrubs, Spa. Veg. = sparsely vegetated, Lod. Pine = Lodgepole Pine, and Sub. Fir = Subalpine Fir. Other includes barren, bog, fen, NA, shallows/littoral, tussock tundra, and water.
Forests 14 01577 g008aForests 14 01577 g008b
Table 2. Confusion matrix. Here, 1 = black spruce for Anchorage and Fairbanks, and 1 = lodgepole pine in Whitehorse. A value of 0 represents other evergreens which are mostly white spruce.
Table 2. Confusion matrix. Here, 1 = black spruce for Anchorage and Fairbanks, and 1 = lodgepole pine in Whitehorse. A value of 0 represents other evergreens which are mostly white spruce.
AnchorageFairbanksWhitehorse
Actual
Predicted010101
038,6651635130,139108,045571,058111,373
12356815118,907572,70075,964177,110
Table 3. Statistics based on the Confusion Matrix.
Table 3. Statistics based on the Confusion Matrix.
StatisticAnchorageFairbanksWhitehorse
Accuracy90.8%75.6%80.0%
True positive33.3%84.1%61.4%
False positive66.7%15.9%38.6%
True negative94.3%52.3%88.3%
False negative6.1%91.4%13.3%
Precision25.7%82.8%70.0%
Recall33.3%84.1%61.4%
Table 4. Vegetation succession rules after fire based on analysis of ABoVE data, literature review, and expert feedback for (a) Anchorage and Fairbanks and (b) Whitehorse. For example, if a black spruce pixel burns, it is classified at sparsely vegetated for the following 10 years, then becomes tall shrub, and after 30 years is again recognizable as a black spruce pixel. Mixed forest and Lodgepole ine (only in Whitehorse) recover after 20 years, and deciduous forest recovers after 15. Woodland recovery can take 20 or 30 years depending on pre-fire vegetation and location (Alaska versus Yukon). Hemlock (only in Anchorage) recovers after 30 years. Only landcover types that are undergoing succession are shown here. Abbreviations used are Bla. Spr. = Black Spruce Forest, Spr. For. = Other Spruce Forest, Mix. For. = Mixed Forest, Dec. For. = Deciduous Forest, Open Shr. = Open Shrubs, Sha./lit. = shallows/littoral, and Spa. Veg. = Sparsely Vegetated. For more detail on how this table was derived, see Appendix A.
Table 4. Vegetation succession rules after fire based on analysis of ABoVE data, literature review, and expert feedback for (a) Anchorage and Fairbanks and (b) Whitehorse. For example, if a black spruce pixel burns, it is classified at sparsely vegetated for the following 10 years, then becomes tall shrub, and after 30 years is again recognizable as a black spruce pixel. Mixed forest and Lodgepole ine (only in Whitehorse) recover after 20 years, and deciduous forest recovers after 15. Woodland recovery can take 20 or 30 years depending on pre-fire vegetation and location (Alaska versus Yukon). Hemlock (only in Anchorage) recovers after 30 years. Only landcover types that are undergoing succession are shown here. Abbreviations used are Bla. Spr. = Black Spruce Forest, Spr. For. = Other Spruce Forest, Mix. For. = Mixed Forest, Dec. For. = Deciduous Forest, Open Shr. = Open Shrubs, Sha./lit. = shallows/littoral, and Spa. Veg. = Sparsely Vegetated. For more detail on how this table was derived, see Appendix A.
(a)
Pre-Fire VegetationPostfire Vegetation Recovery in 5-Year Increments
0–5 Years6–10 Years11–15 Years16–20 Years21–30 Years>31 Years
HemlockSpa. Veg.Low shrubTall shrubTall shrubTall shrubHemlock
Bla. Spr.Spa. Veg.Spa. Veg.Tall shrubTall shrubTall shrubBla. Spr.
Spr. For.Spa. Veg.Low shrubTall shrubTall shrubTall shrubSpr. For.
Mixed forestLow shrubTall shrubTall shrubTall shrubMix. For. Mix. For.
WoodlandSpa. Veg.Spa. Veg.Open Shr.Open Shr.WoodlandWoodland
Dec. For.Low shrubTall shrubTall shrubDec. For. Dec. For. Dec. For.
Open shrubsSpa. Veg.Open Shr. Open Shr. Dec. For. Dec. For. Dec. For.
Sha./lit.Sha./lit.Sha./lit.Sha./lit.Sha./lit.FenFen
Spa. Veg.Spa. Veg.Spa. Veg.Spa. Veg.Spa. Veg.Tall shrubTall shrub
(b)
Pre-Fire VegetationPostfire Vegetation Recovery in 5-Year Increments
0–5 Years6–10 Years11–15 Years16–20 Years21–30 Years>31 Years
Lodgepole pineSpa. Veg.Low shrubTall shrubTall shrubLodgepole PineLodgepole Pine
Sub-alpine firSpa. Veg.Spa. Veg.Spa. Veg.Low shrubLow shrubSub-alpine fir
Spr. For.Spa. Veg.Low shrubTall shrubTall shrubTall shrubSpr. For.
Mixed forestLow shrubTall shrubTall shrubTall shrubMix. For.Mix. For.
WoodlandSpa. Veg.HerbaceousHerbaceousOpen Shr.Open Shr.Woodland
Dec. For.Low shrubTall shrubTall shrubDec. For. Dec. For.Dec. For.
Open shrubsSpa. Veg.Open Shr. Open Shr.Open Shr.Open Shr.Woodland
Sha./lit.Sha./lit.Sha./lit.Sha./lit.Sha./lit.FenFen
Spa. Veg.Spa. Veg.Spa. Veg.Spa. Veg.Spa. Veg.WoodlandWoodland
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Calef, M.P.; Schmidt, J.I.; Varvak, A.; Ziel, R. Predicting the Unpredictable: Predicting Landcover in Boreal Alaska and the Yukon Including Succession and Wildfire Potential. Forests 2023, 14, 1577. https://doi.org/10.3390/f14081577

AMA Style

Calef MP, Schmidt JI, Varvak A, Ziel R. Predicting the Unpredictable: Predicting Landcover in Boreal Alaska and the Yukon Including Succession and Wildfire Potential. Forests. 2023; 14(8):1577. https://doi.org/10.3390/f14081577

Chicago/Turabian Style

Calef, Monika P., Jennifer I. Schmidt, Anna Varvak, and Robert Ziel. 2023. "Predicting the Unpredictable: Predicting Landcover in Boreal Alaska and the Yukon Including Succession and Wildfire Potential" Forests 14, no. 8: 1577. https://doi.org/10.3390/f14081577

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop