Snow avalanches affect transportation corridors and settlements worldwide. In many mountainous regions, robust records of avalanche frequency and magnitude are sparse or non-existent. However, dendrochronological methods can be used to fill this gap and infer historical avalanche patterns. In this study, we developed a tree-ring-based avalanche chronology for large magnitude avalanche events (size
Snow avalanches are hazardous to human safety and infrastructure (Mock et al., 2016; Schweizer, 2003), as well as an important landscape disturbance affecting mountain ecosystems (Bebi et al., 2009). In the United States, an average of 27 people die in avalanche accidents each winter (CAIC, 2020). Avalanches, especially large magnitude events, also affect transportation corridors and settlements throughout the world. For example, avalanches impact numerous roadways and railroad corridors in the western United States (Armstrong, 1981; Hendrikx et al., 2014; Reardon et al., 2008). Consequently, understanding general avalanche processes and associated large magnitude avalanche return intervals (RIs) is critical for local and regional avalanche forecasters, transportation agencies, and land use planners.
Long-term, reliable, and consistent avalanche observation records are necessary for calculating avalanche return intervals which can be used in infrastructure planning and avalanche forecasting operations. However, such records are often sparse or non-existent in many mountainous regions, including areas with existing transportation corridors. Thus, inferring avalanche frequency requires the use of dendrochronological methods to document damaging events or geomorphic response within individual trees at individual path to regional scales. Even in regions with historical records, tree-ring dating methods can be used to extend or validate uncertain historical avalanche records, which has led to the broad implementation of these methods in mountainous regions throughout the world (e.g., Corona et al., 2012; Favillier et al., 2018; Schläppy et al., 2014).
Numerous studies reconstructed avalanche chronologies in the United States using tree-ring methods (Burrows and Burrows, 1976; Butler et al., 1987; Carrara, 1979; Hebertson and Jenkins, 2003; Potter, 1969; Rayback, 1998). Butler and Sawyer (2008) provided a review of current methodologies and types of tree-ring responses used in avalanche dendrochronological studies. Favillier et al. (2018) provided a more recent comprehensive graphical summary of dendrochronological avalanche studies throughout the world. Numerous studies used dendrochronological techniques to develop avalanche chronologies for remote regions without historical avalanche records or areas with inconsistent avalanche observations (Butler and Malanson, 1985a; Germain et al., 2009; Reardon et al., 2008; Šilhán and Tichavský, 2017; Voiculescu et al., 2016), and many studies used these techniques to examine avalanches across space and time (Table A1).
Tree-ring avalanche research is resource and time intensive. Like other scientific fields, it is not feasible to completely sample the variable of interest in infinite detail due to logistical and financial constraints (Skøien and Blöschl, 2006). Thus, a strategic spatial sampling method is necessary. Here, we strategically sampled 12 avalanche paths in four distinct subregions of the US northern Rocky Mountains of northwest Montana to examine spatial differences at a regional scale. The sampling strategy is based on the concept of scale triplet, which defines the spacing, extent, and support of our sampling scheme (Blöschl and Sivapalan, 1995). Incorporating the scale triplet concept helps us understand the nature of the problem, the scale at which measurements should be made, and how we can estimate the measurements across space. Often the scale at which samples are collected differs from the scale necessary for predictive purposes (Blöschl, 1999). For example, if we are interested in avalanche frequency relationships with regional climate patterns but tree-ring samples are collected at an avalanche path scale, then a network of sampled paths need to be spaced and aggregated across the core of the climatically similar region. In our study, the extent is the entire region and subregions, the spacing is the distance between avalanche paths and subregions, and the support is the size of the area being sampled. In addition, the process scale is the natural variability of avalanche frequency, the measurement scale is the tree-ring proxies used to represent avalanche occurrence on an annual temporal scale, and the model scale relates to aggregating all of the sample areas to derive a regional avalanche chronology.
We adopt the definition of Martin and Germain (2016) that large magnitude avalanches are events characterized by low and variable frequency with a high capacity for destruction. This generally translates to a size 3 or greater on the destructive classification scale – i.e., ability to bury or destroy a car, damage a truck, destroy a wood frame house, or break a few trees (Greene et al., 2016).
Understanding the spatiotemporal behavior of large magnitude avalanches on
the regional scale will improve avalanche forecasting efforts, especially for operations involving avalanche terrain that impacts transportation corridors. Here, we aim to answer three specific questions.
What is the regional, subregional, and path-specific frequency of large magnitude avalanches in the US northern Rocky Mountains of northwest Montana? How does the spatial extent of the study region affect the resulting avalanche chronology? What is the probability of detecting regional avalanche activity by sampling different avalanche paths?
To our knowledge, this is the first study to look at how various spatial
scales compare when reconstructing a regional avalanche chronology from
dendrochronological data from a large dataset (
Our study site consists of 12 avalanche paths in the Rocky Mountains of
northwest Montana, United States (Fig. 1 and Table 1). We sampled sets of three avalanche paths in four distinct subregions within three mountain ranges:
the Whitefish Range (WF; Red Meadow Creek) and Swan Range (Swan; Lost Johnny
Creek) in the Flathead National Forest and two subregions within the Lewis
Range in Glacier National Park (GNP), Montana. The sites in GNP are along
two major transportation corridors through the park: the Going-to-the-Sun
Road (GTSR) and US Highway 2 in John F. Stevens (JFS) Canyon. These two areas were utilized for previous dendrochronological avalanche research (Butler and Malanson, 1985a, b; Butler and Sawyer, 2008; Reardon et al., 2008). A robust regional avalanche chronology reconstruction will help place the previous work in the context of the wider region. The other two sites, WF and Swan, are popular backcountry recreation areas with access via snow machine in the winter along a US Forest Service road. The avalanche paths in each subregion encompass a range of spatial extents from adjacent (i.e.,
Study site. The red rectangle in the state of Montana designates the general area of the four sampling sites. The sites are
Topographic characteristics of all avalanche paths. The
n/a stands for not applicable.
Northwest Montana's avalanche climate is classified as both a coastal transition and intermountain avalanche climate (Mock and Birkeland, 2000), but it can exhibit characteristics of both continental and coastal climates. The elevation of avalanche paths within the study sites range from approximately 1100 to 2700 m, and the starting zones of these paths are distributed among all aspects (Table 1).
We eliminated or minimized influence from exogenous disturbance factors such as logging and wildfire by referencing wildfire maps extending back to the mid-20th century. We selected sites undisturbed by wildfire since this time except for Lost Johnny Creek, which was purposeful as this area burned most recently in 2003. We also minimized the influence of logging by selecting sites not previously logged. Using historical logging parcel spatial data, we determined logging in some sites was limited to very small parcels adjacent to the farthest extent of the runout zones.
The historical observational record in this area is limited. In this study
region, the Flathead Avalanche Center (FAC), a regional US Forest Service
backcountry avalanche center, records all avalanches observed and reported
to the center. However, not all avalanches are observed or reported given
the approximately 3500 km
Our sampling strategy targeted an even number of samples collected from both
lateral trimlines at varying elevations and trees located in the main lower
track and runout zone of the selected avalanche paths. This adequately
captured trees that were destroyed and transported, as well as those that
remained in place. The definition of a large magnitude avalanche in this study
refers to avalanches of approximately size D3 or greater (Greene et al., 2016) that may not run the full length of the avalanche path. We sampled
spatial extents within each avalanche path that are representative of runout extents
The sample size for avalanche reconstruction using tree-ring data requires careful consideration. Butler and Sawyer (2008) suggested that a few damaged trees may be sufficient for avalanche chronologies, but larger target sample sizes increase the probability of detecting avalanche events (Corona et al., 2012). Germain et al. (2010) examined cumulative distribution functions of avalanche chronologies and reported only slight increases in the probability of extending chronologies with a sample size greater than 40. This also depends on the available length of record within a given avalanche path. Thus, given the large spatial footprint (
We collected three types of samples: (1) cross sections from dead trees, (2) cross sections from the dead leaders of avalanche-damaged but still living trees, and (3) cores from living trees. We predominantly used cross sections in this study for a more robust analysis as events can potentially be missed or incorrectly identified in cores. We emphasized the selection of trees with obvious external scars and considered location, size, and potential age of tree samples. A limitation of all avalanche dendrochronology studies is that large magnitude events cause extensive damage and high tree mortality, thereby reducing subsequent potential tree-ring records.
We sampled stem cross sections at the location of an external scar or just above the root buttress from downed or standing and dead trees and from stems of trees topped by avalanche damage. We extracted tree-ring core samples from living trees with obvious scarring or flagging along the avalanche path margins and runout zone using a 5 mm diameter increment borer. We collected a minimum of two and up to four core samples per tree (two in the uphill-downhill direction and two perpendicular to the slope). We photographed each sample at each location and recorded species, Global Positioning System (GPS) coordinates (accuracy 1–3 m), amount of scarring on the cambium of the tree, relative location of the tree in the path, and upslope direction (Peitzsch et al., 2019). We also recorded location characteristics that identified the tree to be in place versus transported from its original growth position (i.e., presence or absence of roots attached to the ground or the distance from an obvious excavated area where the tree was uprooted).
To prevent radial cracking and further rot, we dried and stabilized the cross sections with a canvas backing. We sanded samples using a progressively finer grit of sandpaper to expose the anatomy of each growth ring and used the visual skeleton plot method to account for missing and false rings and for accurate calendar year dating (Stokes and Smiley, 1996). We assessed cross-dating calendar year accuracy of each sample and statistically verified dating against measured samples taken from trees within the gallery forest outside the avalanche path and from preexisting regional chronologies (Table A2) (NOAA, 2018) using the dating quality-control software COFECHA (Grissino-Mayer, 2001; Holmes, 1983). For further details on cross-dating methods and accuracy calculation for this dataset, see Peitzsch et al. (2019).
We analyzed samples for signs of traumatic impact events (hereafter “responses”) likely caused by snow avalanches. We adapted a classification
system from previous dendrogeomorphological studies to qualitatively rank
the trauma severity and tree growth response from avalanche impacts using
numerical scores ranked 1 through 5 (Reardon et al., 2008). This classification scheme identified more prominent avalanche damage responses
with higher-quality scores and allowed us to remain consistent with
previous work (Corona et al., 2012; Favillier et al., 2018) (Table 2). To
compare our ability to capture avalanche/trauma events using cores versus
those captured using cross sections, we sampled a subset (
Avalanche impact trauma classification ratings for which C1 represents the strongest and easily detectable trauma and C5 represents subtle and difficult-to-detect trauma.
To generate avalanche event chronologies and estimate return periods for
each path and for the entire study site, we utilized R statistical software
and the package
We used a multistep process to reconstruct avalanche chronologies on three
different spatial scales: individual paths, four subregions, and the entire
region. We also calculated a regional avalanche activity index (RAAI) (Fig. 2). The process involved first calculating the ratio of trees
exhibiting growth disturbances (GDs) over the number of samples alive in year
General workflow of analytical methods to reconstruct regional
avalanche chronology, regional avalanche activity index, and the probability
of detection.
We then used double thresholds to estimate the minimum absolute number of GD
and a minimum percentage of samples exhibiting GDs per year (
We then used the chronologies derived from this process to calculate a
weighted index factor (
Next, we classified
Next, we compared the number of avalanche years and return periods identified in the full regional chronology to subsets of the region to determine the number of paths required to replicate a full 12-path regional chronology. We assessed the full chronology against a subsampling of 11 paths in total by sequentially removing the 3 paths with the greatest sample size. We then randomly sampled two paths from each subregion for a total subsample of eight paths, followed by generating a subsample of four paths by choosing the path in each subregion with the greatest sample size. Finally, we selected a random sample of one path from each subregion to compare against a total of four single path subsamples.
Next, we used the
We also calculated the probability of detecting an avalanche year identified
in the regional chronology as if any given individual path was sampled. The
probability of detection for a given year (POD
We also calculated the probability of detection for each path for the period
of record (POD
Using a 10 m digital elevation model (DEM), we calculated a number of
geomorphological characteristics for each path, including mean elevation (m;
full path and starting zone), elevation range (m), eastness (
Histograms of
We collected a total of 673 samples from 647 suitable trees impacted or
killed by avalanches (trees:
The avalanche response subset analysis that compared results as if samples
were from cores versus full cross sections showed that core samples alone
would have missed numerous avalanche events and generated a greater proportion of low-quality growth disturbance classifications (Fig. 4). For the subset of 40 samples analyzed as cores, we identified only 124 of 191
(65 %) GDs in total. Of the 67 GDs that we would have missed just by using
cores, 24 were classified as C
Example of cross section sample where four cores taken uphill, downhill, and perpendicular (2) would have missed at least one scar (1933) and potentially the pith of the tree. The black lines indicate the potential cores using a 5 mm width increment borer. Note the scale on lower right of sample.
Number of individual avalanche paths in which an avalanche event
occurred in any given year. Avalanche years with
There were 49 avalanche events identified from GD responses across all 12 individual paths in the study region. The avalanche years most common
throughout all of the individual path chronologies were 2014 (seven paths),
1982 and 1990 (five paths), and 1933, 1950, 1972, and 1974 (four paths) (Fig. 5 and Table 3). We identified the year with the greatest number of individual
GD responses (2002) in three paths – two from JFS subregion and one in the WF subregion. There was no clear pattern of similarly identified years from
paths physically closer in proximity to each other. However, paths within
the WF subregion produced the most similar number of large magnitude
avalanche years. When we applied the
Avalanche chronologies and return interval (RI) statistics of all
12 avalanche paths in the region. Avalanche years in bold indicate years
identified in at least two avalanche paths in the subregion. Underlined
avalanche years indicate years in common in at least one path from at least
three of the four subregions,
Across all individual paths, the median estimated return interval was 8 years with a range of 2 to 28.5 (Fig. 6). Hereafter return intervals indicate median return intervals unless specified. JGO, located in the GTSR subregion, exhibited the greatest spread in estimated return intervals,
followed by LJB. The avalanche paths within the WF subregion had the most
similar return intervals of any of the subregions. The return interval for
JGO differed significantly from several other paths: RMA, RMB, RMC, and Shed 10-7 (
Boxplot of return intervals for individual avalanche paths in each
subregion:
We subset the period of record for each path from 1967 to 2017 and compared RI values to the full record. In this subset, nine paths exhibit no change in RI values when compared to the full record. In one path, 54-3, RI values decreased from 14 to 10 years. We observed larger changes in the other two paths: the JGO path had only 1 avalanche year recorded (down from 5 years), and the median RI in LJC changed from 22.5 to 35 years. If we removed 54-3, JGO, and LJC for this comparison, the records from the subset period of record are similar to the complete records for the other paths in the study.
When the paths were aggregated into subregions (three paths per subregion) the median return periods for each subregion were similar and all less than 10 years (Fig. 6e and Table 4). The number of avalanche years for all of the subregions ranges from 12 to 18 with the greatest number of identified years in the JFS subregion and the fewest in the WF subregion. The JFS subregion has the shortest median return interval, followed by the Swan, WF, and GTSR subregions. The number of avalanche years for each aggregated subregion is greater than the number of avalanche years for any individual path within each subregion except for the JFS subregion where 18 avalanche years were identified but Shed 10-7 totaled 20 avalanche years (Table 5).
Avalanche chronologies and return interval (RI) statistics of all
four subregions;
Number of avalanche events for each subregion, the mean of three individual paths in each region, and the overall aggregated region.
In terms of commonality of years between the subregions, 1982 is the only year identified in all of the four subregions (Fig. 7). Avalanche years commonly identified in three subregions are 1933, 1950, 1954, 1974, and 2014. We identified the JFS subregion as having the greatest number of years exclusive to that subregion (10 years). The WF subregion shared the greatest number of years with other regions (11 years), followed by JFS (9 years), GTSR (8 years), and the Swan (7 years). In the only available comparison with an incomplete and limited historical record, the individual reconstructed avalanche chronologies of paths in the JFS subregion captured 10 %–50 % of the recorded large magnitude events over the years 1908 to 2017.
Venn diagram of avalanche years common between subregions. Overlapping areas of each ellipse indicate years in common with each subregion.
We identified 30 avalanche years in the overall region and a median return
interval of 3 years (Table 5). The number of samples increases through time
to a peak during 2005, and as expected, the number of GDs also increases through time (Fig. 8a). The
Comparison of the number of avalanche years and return intervals (RIs) when including all 12 paths in the region to using a combination of fewer paths to define the region. HLC signifies a high level of confidence and MLC a medium level of confidence as per Favillier et al. (2017, 2018). “Number not in regional” refers to avalanche years identified in that particular combination of paths but not identified in the regional record.
n/a stands for not applicable.
When we included all paths but S10.7 (one of two paths with the greatest sample size), we captured 80 % of all avalanche years and added 1 new
year to the chronology (Table 6). When we removed LGP (the other path with
the greatest size of sampled trees), we still captured all of the years in
the regional chronology but introduced 4 new years into the chronology
for a total of 34 years. A random sample of 8 (2 from each subregion)
of the 12 avalanche paths captured 83 % of the years in the chronology and
identified 2 new avalanche years. Finally, when using only one path from
each subregion with the largest sample size (Shed 10-7, 54-3, LJA, and
RMA), we captured 73 % of the avalanche years identified in the full
regional chronology. When using a random sample of one path from each subregion (1163, LGP, LJC, RMB), we captured only 43 % of the years
included in the regional chronology of all 12 paths. The RAAI is insensitive
(no significant difference,
The probability of detection for the avalanche years (POD
Probability of detection (POD
n/a stands for not applicable.
Finally, the probability of capturing all of the avalanche years identified
in the regional chronology by each individual path ranges from 7 % to
40 % (Table 8). The greatest POD
Probability of detection of each individual path (POD
The processing and analysis of 673 samples spanning a large spatial extent allowed us to create a robust regional large magnitude avalanche chronology reconstructed using dendrochronological methods. Cross sections provided a more robust and complete GD and avalanche chronology compared to a subsample generated from cores alone. Due to the reduced information value of working only with cores, Favillier et al. (2017) included a discriminatory step in their methods to distinguish avalanche signals in the tree-ring record from exogenous factors, such as abnormal climate signals or response to insect disturbance. By using cross sections to develop our avalanche chronologies, we were able to view the entire ring growth and potential disturbance around the circumference of the tree as opposed to the limited view provided by cores. This allowed us to place GD signals in the context of both climate and insect disturbances without the need for this processing step. We do not discount any studies that use cores for reconstructing avalanche chronologies and understand there are sampling limitations from environmental and policy perspectives in different regions, as well as financial and processing constraints. However, we are suggesting that if the ability to collect cross sections exists, then it is advantageous to collect them.
We targeted sample collection in the runout zones and along the trim line where large magnitude avalanches occurred in recent years. At several sites, we collected samples at the upper extent of the runout zones (S10.7, Shed 7, and 1163). Thus, some additional noise in the final chronology for those specific paths could be due to more frequent small magnitude avalanches. Though the oldest individual trees extended as far back as the mid-17th century, the application of the double thresholds processing steps restricted individual avalanche path chronology lengths since the minimum GD threshold requirements were not met. It is difficult to place confidence in these older recorded events due to the decreasing evidence back in time inherent in avalanche path tree-ring studies. Therefore, we chose to examine more recent time periods dictated by the avalanche years identified through the double threshold methods.
All of the paths in the study are capable of producing large magnitude avalanches with path lengths greater than 100 m (typical length for avalanche destructive size 2, D2), and all but RMC have a typical path length of close to or greater than 1000 m (for avalanche destructive size 3, D3) (Greene et al., 2016). As Corona et al. (2012) noted, the avalanche event must be large enough to create an impact on the tree, and size D2 or greater will be evident from the tree-ring record (Reardon et al., 2008). However, the successive damage and removal of trees from events sized D2 or greater also impacts the future potential to record subsequent events of similar magnitude. In other words, if a large magnitude avalanche removes a large swath of trees in 1 year, then there are fewer trees available to record a slightly smaller magnitude avalanche in subsequent years. Therefore, dendrochronology methods inherently underestimate avalanche events by up to 60 % (Corona et al., 2012).
By examining three different spatial scales (individual path, subregion, and region), we produced a large magnitude avalanche chronology for the region captured in a small subset of the total number of paths across the large region. Accordingly, this sampling strategy may also alleviate the issue of recording large magnitude avalanches within a region in the successive years following a major destructive avalanche event that removed a large number of trees within specific paths but not others. Overall, a regional sampling strategy enables us to capture large magnitude avalanche events over a broad spatial extent, which is useful for regional avalanche forecasting operations and future climate association analysis. This strategy also allows us to understand large magnitude avalanche activity at scales smaller than the regional scale.
We applied the
We developed avalanche chronologies for 12 individual avalanche paths. The
path with the greatest number of identified avalanche years, S10.7, contains
two major starting zones that are both steeper (35 and 39
The range of return intervals across all paths (2–28.5 years) is similar to those reported for 12 avalanche paths across a smaller spatial extent in the Chic-Choc Mountains of Québec, Canada (2–22.8 years) (Germain et al., 2009). Although the authors in that study used a different avalanche response index, their study still suggests considerable variation in avalanche frequency across avalanche paths within a region.
The results from examining return intervals during a truncated period from 1967 to 2017 across all paths illustrate that several of the individual path return interval results should be treated with caution (e.g., JGO, LJC, and 54-3). The difference in minimum and maximum return interval values is a function of a decreasing sample size back in time. The minimum return interval values in many of the paths are concentrated during recent periods. This is a limitation of using dendrochronology to estimate return intervals. Comparing avalanche return intervals across individual paths should also be treated with caution given the variable nature of sample availability across paths. This variability across individual paths further provides reasons to evaluate the number of paths necessary to create a regional avalanche chronology from tree rings. Most of the paths have a reasonable record over this truncated period and also highlight the importance of strategic sampling in numerous avalanche paths. While dendrochronology underestimates avalanche activity, we show that sampling enough paths across a region provides a reasonable estimate of avalanche activity at this scale.
JGO contains the maximum return interval for any path in the study, and the return intervals are significantly different from numerous other paths. A lack of recording data after one large avalanche event could easily skew this value. To understand if this value is accurate, we would have to sample adjacent tracks to determine if the return intervals are similar or not. An appropriate sample base without large temporal gaps is necessary to fully provide an accurate estimate of return intervals within a single avalanche path. While the sample size is sufficient for this individual path, the results should be treated with caution due to the temporal gaps. In other words, the large return interval values may reflect the irregular preservation of evidence for large avalanches as opposed to an accurate estimate of return intervals. Therefore, we cannot fully explain the large maximum return interval for this path.
The return intervals for LJC in the Swan subregion were the greatest in this subregion, and this is likely due to wildfire activity in this path in 2003. LJC was heavily burned, and this created a steep slope with few trees that was once moderately to heavily forested. Substantial anchoring and snowfall interception likely created an avalanche path that did not have many large magnitude avalanches for decades since slope forestation plays a substantial role in runout distance and avalanche frequency in forested areas (Teich et al., 2012). In addition, wildfires in 1910 burned a majority of the JFS subregion as well, and the higher frequency of avalanche years recorded between 1910 and 1940 in S10.7 suggests wildfire impacts may also be a contributor to the high frequency of avalanche events in that location (Reardon et al., 2008). Additionally, the fire in LJC may also have removed evidence of previous avalanche activity.
Our results also suggest that return interval increases as path length
increases, though the sample size for this correlation analysis on individual paths is small (
The greatest number of identified avalanche years is in the JFS subregion. The avalanche paths in this subregion are all southerly or southeasterly facing, whereas the other subregions span a greater range of aspects. This narrow range of aspect may cause a bias toward overrepresentation of those aspects compared to the inclusion of other aspects in other subregions.
The differences between individual avalanche paths, as well as subregions, are likely due to localized terrain and weather/climate factors and the interaction of the two (Chesley-Preston, 2010), as well as local avalanche, forest stand, and fire history. For example, Birkeland (2001) demonstrated significant variability in slope stability across a small mountain range dependent upon terrain and weather. Slope stability and subsequent large magnitude avalanching are likely to be highly heterogeneous not only across the subregion but across a large region. This is also consistent with findings by Schweizer et al. (2003) that suggest substantial differences in stability between subregions despite the presence of widespread weak layers. Finally, climate drives weather but is not a first-order effect on avalanche occurrence in any one given avalanche path. In this study, we derived a regional avalanche chronology to provide a spatial scale that aligns more with the spatial scale of climate drivers than any one individual path. These are relationships that should be examined in future work.
The regional chronology we developed through the use of tree-ring analysis on collections made across 12 avalanche paths suggests, unsurprisingly, that the inclusion of more avalanche paths across a large spatial extent produces a more robust identification of major avalanche winters. When we aggregate all 12 paths together and apply thresholds to discriminate the signal from the noise, we identified 30 avalanche years throughout the region. This type of analysis allows us to place each avalanche year in the context of the region or the full extent of the scale triplet rather than simply collating all major avalanche winters identified in each individual path or subregion. However, we also account for the support and spacing by including adjacent avalanche paths within a subregion and multiple subregions throughout the region. This sampling strategy combined with the large sample size collected throughout the region allowed for a robust assessment of a regional avalanche chronology derived from tree-ring records.
We tested the sensitivity of the term regional by removing specific and random paths. Our results suggest that removing paths from this structure, and subsequently compromising the sampling strategy, introduces noise. By reducing the sample size, we reduce the ability of the thresholds to filter out noise, thereby increasing the actual number of avalanche years in the region. However, the sample size reduction also reduces the number of identified avalanche winters common to the full 12 path regional record (Table 6). Our results emphasize the importance of sampling more paths spread throughout the region of interest, as well as a large dataset across the spatial extent.
Avalanche path selection is clearly important when trying to assess avalanche frequency (de Bouchard d'Aubeterre et al., 2019). The importance of path selection is supported by our results suggesting that S10.7 is more influential than any other path in our study (Table 6), which is also illustrated by the large number of avalanche years detected in S10.7 due to increased sampling in the track. However, by selecting multiple paths representative of the range of geomorphic and potentially influential local weather controls throughout the region, we are able to provide a reasonable assessment of regional avalanche activity in areas without historical records. By quantifying the sensitivity of the number of avalanche paths within a given region, we illustrate that sampling a greater number of avalanche paths dramatically increases the probability of identifying more avalanche years and increases the ability to reconstruct major widespread avalanche events. However, as previously noted, dendrochronological techniques tend to underestimate avalanche frequency, which implies that caution should be used when interpreting a regional avalanche signal using this technique, particularly as sample numbers and qualities (e.g., cores versus cross sections) decline.
Interestingly, the difference in median return interval throughout the “region” using 12 paths compared to using 4 or 8 paths changes only slightly, suggesting that fewer paths are still able to represent the major avalanche return intervals across a region. However, choosing fewer paths appears to introduce more noise and therefore fewer years identified than a regional chronology with more avalanche paths.
The RAAI provides a measure of avalanche activity scaled to the number of active avalanche paths across the region through time, but RAAI is limited by the decreasing sample size back in time. The years with the greatest RAAI value coincide with substantial activity provided in the historical record, as well as previous dendrochronological studies from the JFS subregion (Butler and Malanson, 1985a, b; Reardon et al., 2008). The winter of 1932–33 was characterized by heavy snowfall and persistent cold temperatures leading to extensive avalanche activity that destroyed roadway infrastructure in the JFS subregion, 1950 saw a nearly month-long closure of US Highway 2 due to avalanche activity, and in 2002, an avalanche caused a train derailment. While these are all confined to the JFS subregion, with the exception of 2002, they are also years shared by at least two other subregions.
We examined the probability of detecting an avalanche year throughout the
region by sampling any one given path. In 7 of 30 years, the
POD
Our results also suggest that our sampling design using scale triplet increases the probability of detecting avalanche activity across an entire
region. We note that we are only able to scale our probability calculations
to our dataset with a limited historical observational record. However, our
results illustrate the importance of sampling more paths if the goal is to
reconstruct a regional chronology. In our dataset, the greatest value of
POD
Overall, our results suggest that sampling one path, or multiple paths in one subregion, is insufficient to extrapolate avalanche activity beyond those paths. Multiple paths nested within subregions are necessary to glean information regarding avalanche activity throughout those subregions, as well as the overall region. Our study is still limited by the underrepresentation inherent in dendrochronological techniques for identifying all avalanche events. While we analyzed 673 samples over the extent of the region, some of the paths in our study had relatively small sample sizes per individual path as compared to recent suggestions (Corona et al., 2012). This may have influenced the number of avalanche years identified and subsequent return intervals per individual path. However, we attempted to limit the influence of sample size by using full cross sections from trees, robust and critical identification of responses in the tree rings, and appropriate established threshold techniques.
We also recognize that sampling more avalanche paths in our region would certainly provide a more robust regional avalanche chronology, but time, cost, and resource constraints required an optimized strategy. Finally, our study would undoubtedly have benefited from a longer and more accurate historical record for comparisons and for the verification of the tree-ring record in all of the subregions. Overall, our study illustrates the importance of considering spatial scale and extent when designing, and making inferences from, regional avalanche studies using tree-ring records.
We developed a large magnitude avalanche chronology using dendrochronological techniques for a region in the northern US Rocky Mountains. Implementing a strategic sampling design allowed us to examine avalanche activity through time in single avalanche paths, four subregions, and throughout the region. By analyzing 673 samples from 12 avalanche paths, we identified 30 years with large magnitude events across the region and a median return interval of
List of previous avalanche dendrochronological studies
Continued.
Regional chronologies from the International Tree-Ring Database (ITRDB) used for cross dating.
Proportion of input types (tree-ring signals) to each growth disturbance (GD) class. Note that there could be multiple input types for each class. Termination of growth indicates that the tree was killed in that year and that it coincides with the historical avalanche record. Some of the termination of growth samples have earlywood if the avalanche occurred in the late winter or early spring. Refer to Table 3 for definitions of classes.
Summary data of subset of samples (
Correlation matrix (Pearson correlation coefficients) of the number of avalanche years, return interval (RI), starting zone slope angle (Slope), and path length (Length). Statistical significance is
Avalanche years identified in the regional analysis (Region;
Data for this work can be found in the ScienceBase repository:
EP was responsible for the study conception and design, data collection, analysis, and writing. JH contributed to the development of the study design, methods, editing, and writing. DS was responsible for data collection, tree-ring processing and analysis, and writing. GP, KB, and DF contributed to the study design, editing, and writing.
The authors declare that they have no conflict of interest.
Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the US government.
We extend gratitude to Adam Clark for his substantial data collection efforts and Zach Miller for his assistance processing samples. This work is part of the US Geological Survey Land Resources Western Mountain Initiative project.
This paper was edited by Pascal Haegeli and reviewed by Adrien Favillier and Brian Luckman.