Snow avalanches: A hazard and driver of landscape change

Video Transcript
Download Video
Right-click and save to download

Detailed Description

Snow avalanches kill, on average, 27 people in the United States each year and impact infrastructure and commerce in mountainous areas. Erich will present an overview of his research group's ongoing projects on snow and avalanches and dive a little deeper into some recent results of using tree rings to reconstruct avalanche chronologies, derive return periods of large magnitude avalanches, and examine relationships with regional-scale climate drivers in the context of a changing climate.

Peitzsch E (2021). Snow avalanches: A hazard and driver of landscape change. USGS Landslide Hazards Program Seminar Series, 22 September 2021.


Date Taken:

Length: 00:47:16

Location Taken: US



Okay. Well, I can see we’re at
one minute after the hour,

so we’re just going to
go ahead and get started.

My name is Matt Thomas,
and thank you for tuning in

to the USGS Landslide Hazards
Program seminar series.

For those of you that are new to
this meeting, you have the ability

to submit questions via the chat window
or to use the raise-your-hand feature

in combination with your
microphone and video camera.

We typically wait until the end of
the presentation to take questions,

so in the meantime, please just do
your best to make sure your

microphone and video camera are off
when you aren’t intending to speak.

Erich Peitzsch is a research physical
scientist at the U.S. Geological Survey

Northern Rocky Mountains
Science Center.

He has worked for the USGS since 2007
studying avalanches, snow, and glaciers.

He earned in M.S. and Ph.D.
in Earth sciences from

Montana State University.
His current research focuses on

avalanche frequency and associated
climate processes, wet snow processes,

and remate sensing to examine
snow-depth change.

When not studying the cryosphere,
he’s usually chasing his two sons

around the – around the
mountain on bikes, skis, or feet.

And that sounds like
a job in itself.

So, Erich, thank you very
much for joining us today.

- Thanks, Matt.
I appreciate it.

And, yeah, thanks to
everyone for tuning in.

I’m going to share
my screen here.

And how’s that look?

- It looks like it’s just
about to pop up, Erich.

There we go. I can see it live now.
And just a reminder about your

video camera related to your …
- Yeah.

- You know, when you
do your video. Up to you.

- Okay. And I’m also going to
turn off the camera here.

Great. Yeah. Just let me know
if anything looks bad. [laughs]

But thanks again,
Matt, for the intro.

And thanks to all of you for, again,
tuning in on this Wednesday

afternoon or – probably afternoon
wherever most folks are.

All right. Let’s get started here.
So going to talk about snow avalanches

and how they’re a hazard and
a driver of landscape change.

[clears throat]
Excuse me.

So first I want to acknowledge
my research team at NOROCK –

the Climate Change in
Mountain Ecosystems team.

And all of these folks have contributed
to all of this work that I’ll present today

in major ways. And, without their
effort, none of this work is possible.

I’d also like to acknowledge my
co-authors on the avalanche climate

work from the Northern Rocky
Mountain region that I will present

today, as well as the USGS Ecosystems
Land Change Science Program.

All right. Let’s get into it.
So why do we study avalanches?

Well, as I mentioned, they are a hazard.
And they are also a disturbance.

So, in the western U.S., avalanches
are the most frequently occurring

lethal form of mass movement
on an annual basis.

They’re creating impacts to public
safety, commerce, and recreation.

Avalanches – they’re also
a driver of landscape change.

You can see on the right here just
an example of some disturbance

along a railway in southern Glacier
Park. And this landscape change,

it’s actually important to
flora and fauna and habitats.

And how this landscape interacts –
or, how avalanches interact with

other disturbances, such as wildfire,
to create new avalanche terrain

in areas that were once forested.
So avalanche ecology is sort of

the side of avalanches
that we know not as much about

as we do from a
hazard perspective.

So this is from the Colorado Avalanche
Information Center just documenting

avalanche fatalities per year from 1950
to two – or, the winter of 2019 and ’20.

And this just provides some context
that, over the last 10 winters,

an average of 27 people died in
avalanches each winter in the U.S.

So, last year, of course,
was very tragic with 37 fatalities.

And that was due to a number of
reasons, including a fairly widespread

unstable snowpack throughout
the western U.S. and potentially

even more users in the –
in avalanche terrain.

So regional U.S. Forest Service
Avalanche Centers and transportation

corridor operations, they’re doing
a fantastic job of forecasting

and educating the public,
especially considering the increase

in the number of people in
avalanche terrain over the years.

But avalanche research itself at
the federal level is fairly limited.

Aside from us at NOROCK,
the Forest Service National Avalanche

Center does some opportunistic
research with their adjunct duties,

but they’re not a research entity.
So we’ve partnered with the National

Avalanche Center, the U.S. Forest
Service Rocky Mountain Research Station,

for some of the work
that I’ll present today.

And we’ve also partnered
with folks in southeast Alaska,

especially with the Alaska Climate
Science Center and – to do some of

the similar work, again, that I’ll
present up in southeast Alaska.

So the results of this work – it’s used –
the results are used by avalanche

forecasters and many operation types.
Again, including regional backcountry

avalanche centers, avalanche forecasters
for transportation corridors like the

I-70 that goes through Colorado,
some of the transportation corridors

in California as well,
and then here in Montana.

It’s also used by infrastructure planners
and natural resource managers.

So the image on the right shows
an avalanche protective structure

above a neighborhood –
a small neighborhood in Colorado.

And this is just
outside of Aspen.

And the engineer that designed
this structure designed it to withstand

a 100-year avalanche event.
So, in March 2019, a widespread

avalanche cycle occurred
throughout Colorado.

And a very large avalanche
hit this particular structure.

Some of the debris spilled
over the dam, just breaking

a few windows on the house.
And it could have been much,

much worse. So we’re actually
researching the return period of

an avalanche of this size, and in this
location in Conundrum Creek here,

if you’re familiar with it,
in a recent project in Colorado.

And, again, I’ll explain a little
bit more about that in detail later.

So our group focuses on four major
areas of avalanche research at

NOROCK. We study how often
large-magnitude avalanches occur.

And we look at associated climate
and atmospheric drivers of those

large avalanche events.
We also use drones to look at

snow-depth change in complex
avalanche terrain and how that

affects both the avalanche hazard
and snow as a water resource.

We collaborate with the National Park
Service on an avalanche forecasting

program for the annual spring opening
of the Going-to-the-Sun-Road in

Glacier National Park in Montana.
And then, in concert with those

forecasting efforts, we also do wet-
snow avalanche research, particularly –

or, sort of motivated by the
potential increases of wet

avalanches due to a warming climate.
So today I’m going to discuss these

four areas of research and present
some recent results from each.

So let’s get started off – or, let’s start
off talking about avalanche frequency.

So our main objective of this
component of our research group

is to reconstruct avalanche
chronologies from the past

so that we can examine
climate drivers.

And this ultimately helps us
understand future avalanche behavior.

We published some work that we
did in the Northern Rockies region.

And today I’ll present the
results of that work.

But, as I mentioned earlier, we’re also
currently working in southeast Alaska

and Colorado so that we can compare
frequency as well as associated climate

drivers in different snow climates.
So, for example, you know, Colorado

has a continental snow climate sort of
characterized by shallow snowpack,

high elevation, and, you know,
predominantly cold and dry conditions.

While southeast Alaska is a maritime
climate located next to the ocean

with much wetter and
warmer conditions.

So, to achieve – excuse me –
to achieve our objectives in this study,

we used dendrochronology,
which is the study of tree rings,

to reconstruct a long-term
magnitude avalanche chronology.

And, in the northern Rockies,
we were able to reconstruct

a chronology from
1867 to 2017.

And we defined large-magnitude
as a size 3 or greater on the

avalanche destructive size scale.
And this scale ranges from 1,

which is relatively harmless
to humans, to 5, which is the ability

to destroy wide swaths
of forest and gouge –

basically transform
the landscape.

So a D3, right in the middle there, is
defined as having the ability to bury or

destroy a car, damage a truck, destroy a
wood-frame house, or break a few trees.

And you can see in these images – the
one on the right is from Colorado, and

you can see the destruction of the forest
there from this particular avalanche.

And this was from
the March 2019 cycle.

So, if anyone here is from Colorado
and may remember that cycle,

it was pretty widespread.
Caused a lot of disruption

to transportation
corridors as well.

The image on the left is from the
Going-to-the-Sun Road in Glacier Park.

And you can see the heavy machinery –
that’s a bulldozer and a excavator –

removing about 30 feet of combined
snow and woody debris from the

Going-to-the-Sun Road during
spring clearing operations.

So how do we do this?
So here’s a little video.

It should work. We tested it out.
And I’ll just kind of talk us through it.

But what we do is, we suspect
this is an avalanche scar.

We’ll cut into that tree, just like Adam
is doing there with the chainsaw.

We’ll take out a little
cross-section from each tree.

And then we will sand it so that it has
a pretty polished and smooth surface.

And we’ll take it to the lab.
And then, in the lab, we basically

take these cross-sections,
look at them under the microscope.

And that’s where we’re
able to date the tree.

So, when a tree is knocked over,
we don’t necessarily know

when it’s knocked over,
so we don’t know when it died.

So we don’t – we know – we can say
how old the tree was, but we don’t know

the exact dates of when the tree started
growing and then when it died.

So we’re able to cross-date
these trees with live trees.

So we take cores of live trees
outside of the avalanche path.

And that allows us to date those trees
so that we can the look at each ring,

and we know the exact year
for each of those rings.

And, from there, we are able to
look at any sort of avalanche signal

or disturbance in the tree ring.
And I’ll get to that in just a sec.

And that allows us to
date each tree ring.

All right.

So, as I mentioned, we use – you know,
we’re using dendrochronology here.

And this sort of method allows for
a pretty robust assessment of

avalanche frequency across
a really large spatial extent.

And we use dendrochronology
because observational records

of avalanches in many areas
really don’t date back that far.

Colorado, for example, has a pretty
good observational record, I’d say,

from, you know, roughly
the 1950s or ’60s to present.

And, of course, it gets better as we –
as we get closer to the present.

But – excuse me – prior to 1950,
it’s not really good.

And that’s actually a pretty
long record for the U.S.

[laughs] When I talk to my colleagues
in Europe, of course, they – you know,

they look at their observational
records and, because they’ve just sort of,

you know, lived in the mountains
for centuries, they have some

pretty darn good observational
records that extend into, you know,

the 1700s –
or, back to the 1700s.

So I’m pretty jealous of that.
But, regardless, one way of getting

around it is by using the study of tree
rings, and that’s what we’re doing.

So you can see the image
in the lower right here.

If we were to use only cores –
you know, using an increment borer

like the image on the bottom left – if we
were to just take our increment borer,

put it in the tree, and take out a small
core, it’s about 5 millimeters wide.

And, in that lower-right image,
you can see that, even if we took

a core in all four directions on the tree,
we could potentially miss some

really important information.

And you’ll see here that it’s
labeled as a scar from 1933.

So, if we took our cores and went in
in those directions, then we potentially

would have missed that scar from 1933,
and we would have missed the pit,

which is the center of the tree, from
1891. So it would have been really hard

for us to date – to actually get the
full information from this tree.

So that’s why cross-sections
are really important.

So, in one of our papers in Natural
Hazards and Earth System Science,

we looked at, had we used only cores
instead of cross-sections as well – or,

predominantly cross-sections,
then we would have missed about

30% of the years that we identified
as major avalanche years in our

data set from the northern Rockies.
So, yeah, it takes a lot of time – I mean,

sure, it’s fun to go out and, you know,
use the chain saw and collect these data.

But it takes a lot of time, especially
the processing and the lab work.

But it’s worth it because
otherwise we would have missed

a pretty substantial
amount of data.

All right. So our study site,
as I mentioned,

is located in the northern Rockies
of northwest Montana.

And we sampled in four distinct
mountain ranges you can see here

labeled by the red dots.
We call these four distinct ranges or –

we call them sub-regions.
And they all have similar

avalanche climate types. And so this
sampling strategy is based on

the concept of scale triplet,
if you’re familiar.

But it basically defines the spacing
and the extent of the support of

our sampling scheme.
So it allows us to understand

the nature of the problem,
or the process – in this case, avalanches.

It allows us to understand the scale
at which the measurements

should be made. And then it also
allows us to estimate the measurements

across space. So we can take these point
measurements and actually estimate

them for the general region when we
combine all four sub-regions together.

So just a little more detail about
some of the study sites here.

This is the [clears throat] – excuse me –
in the southern part of Glacier National

Park is John F. Stevens Canyon.
And it’s a major transportation corridor.

U.S. Highway 2 and the Burlington
Northern Santa Fe Railway

run through this canyon here.
And you can see the avalanche –

you can see the road in the
bottom of the canyon there.

And the railway sort of to the
looker’s right of the road.

And then all of the avalanche paths are,
of course, the treeless chutes that

come down and affect
the railway and the road.

Some of the railway –
some of it has sheds that protect it.

But, over time, the avalanches have
basically extended beyond the sort of

planned, or expected, boundary.
And so some of the sheds now

come up short, and debris
can still reach the railway.

On the right is the
Going-to-the-Sun Road.

And I’ll talk more in depth about the
road itself and the avalanche paths that

affect it, but that’s some of
the terrain that affects

the Going-to-the-Sun Road –
the very large avalanche paths there.

The road runs through it
through the valley there.

So, as I mentioned, we take these
samples to the lab, and we’re looking

at them under the microscope.
And what are we looking for?

So there can be numerous tree-ring
response types that result from

mechanical damage to the –
to the tree caused by avalanches.

So imagine that an avalanche hit it.
There’s an avalanche on a slope –

or, sorry [laughs] – a tree on a slope,
and an avalanche comes down

and hits that tree. Then, if it doesn’t
uproot the tree and totally, you know,

knock it – uproot it and then kill it,
then it can potentially leave a signal.

So it might destroy sort of the outside,
or the growing, part of the tree.

And then, in order to compensate for
that, the living tissue will sort of grow

around that now dead part –
those dead cells on the cambium there –

the outside of the
growing part of the tree.

So that can result in a scar.
A scar – so we use a scale, from 1 –

which is an obvious avalanche signal,
or scar, to a 5, where the growth

is disturbed – we know it’s a growth
disturbance, but we’re sort of uncertain

that it was an avalanche.
It could be, you know,

another exogenous factor, like,
you know, a rockfall or a landslide,

or even a large animal
coming by and, you know,

rubbing themselves [chuckles]
along the tree.

So it can range –
that’s why we have a scale.

But you can see here,
the C1 Impact Scar

and what that looks like
in the image on the screen.

And then another potential signal
is what we call reaction wood.

So imagine that that tree
is on the slope again.

And an avalanche comes down, hits the
uphill side of that tree, and it doesn’t –

you know, it doesn’t uproot it,
but it hits it, and maybe there’s a scar,

but maybe not.
Maybe it tilts the tree over.

And so what the tree is now going to do,
it’s going to put sort of extra growth

on the downhill side because the
tree wants to stand up straight.

So it’s going to put extra growth
on the downhill side to allow that tree

to stand up straight and to – we call that
reaction wood so that the downhill side

will now have sort of thicker – the ring
will look thicker on the downhill side.

And so, again,
that’s a pretty good signal.

We rated that as a C2.
So not quite as good as a C1,

but we have successive years of
reaction wood, so you can see,

from 1927, and 1933 as well,
you know, we have successive years

of reaction wood – so the
tree trying to right itself.

So, again, how do we know
it’s an avalanche here?

So how do we distinguish between
avalanche responses from noise in the

tree ring? So, again, first we sampled
and we used mostly cross-sections.

And second – again, don’t worry
about the equations here, but we

created an avalanche activity index.
We basically applied a series of

filters using thresholds.
These thresholds are based on

sample size and the number
of growth disturbances

per year within
each avalanche path.

And then finally, we placed more
weight on the higher-quality responses.

So those C1 responses that you saw
in that image prior, those are going to

be weighted more heavily than a C2,
and then a C3, and so on.

So what did we find?

We’ll get to this Venn diagram here
in a sec, but we ended up with 673 samples

from 647 trees that we – within
those 673 samples, we identified

over 2,100 growth disturbances.
The mean age of our trees was 73 years.

And we constructed an avalanche
chronology, as I mentioned before,

from 1867 to 2017.

And this resulted in us being able to
identify 30 large-magnitude avalanche

years across the region.
So, again, after we applied the

sort of filtering stages,
filtered out all the noise,

we came up with 30 major
large-magnitude avalanche

years throughout the region –
so those four sub-regions combined.

This little Venn diagram here – if we
take those four sub-regions separately –

and that’s what the initials up top
denote, the value in the center there,

1982, is the only year that was
common between all four sub-regions.

And then we have five other years
that were common between

at least three of the sub-regions.
So this – what this shows us is that

basically that scale is really important
and that sampling across a large area

requires pretty careful
and strategic sampling.

So, for statistical analysis, we took some
climate data – and this is looking at the

climate and avalanche relationships.
And we looked – we took our regional

reconstructed avalanche chronology.
So we [clears throat] – excuse me –

looked avalanche
versus non-avalanche year.

Took a univariate analysis.
And then we took those significant

variables and threw those
into a principal component analysis

to sort of
reduce collinearity.

We then took the principal components,
and we used those as explanatory variables

in a generalized linear autoregressive
moving average model.

This sort of accounts for the problem
that we – that happens when we’re

trying to collect tree rings is,
if you have an avalanche,

and it takes out a bunch of trees,
then you don’t have any trees there

to pick up an avalanche
that may occur next year.

And so what this does is it allows us
to actually sort of look back in time

and use a moving average for
avalanche activity through time.

And sort of helps –
it doesn’t fully alleviate,

but it helps constrain
this limitation.

And then we were able to take that
and look at the GLARMA model

and use that to examine trends
in the probability of large-magnitude

avalanches through time.

One thing to note is that, while our
avalanche reconstruction was from

1867 to 2017, we only examined the
large-magnitude avalanches in the

context of the climate data from
1950 to 2017 because a lot of

those climate indices only
went back to about 1950.

So what did we find?
Well, we found a slight but significant

decrease in the probability of
large-magnitude avalanche years.

So the chance that a year
would be a large-magnitude

avalanche decreased
from 1950 to 2017.

This translates to roughly
a 2% reduction in avalanche

probability per decade.

And this coincides with the trend
of the Panel B there of maximum

snow depth decreasing over time.
And maximum snow depth was

the largest contributor variable
to our first principal component.

So basically, snow depth decreases
and the probability of avalanche

decreases as well. I’ll explain a little
bit more about this in a second,

but I do want to show that the
probability of avalanche year – so –

decreases, again, as the first principal
component becomes more positive.

So you can see that the probability
drops when we have less snow

and sort of high pressure –
high atmospheric pressure – you know,

nice, sunny weather. And this explains
about 40% of the variability.

And then finally, though, our second
principal component, which was

driven mostly by sort of warmer
winter temperatures, it was sort of

moderately negatively correlated.
But you can see that the probability

is still greater when we
have warmer temperatures.

So what we can infer here is that, yes,
the decrease in snowpack depth over

time is the most influential driver
in our avalanche probability.

But that decrease is actually buffered
by warming and spring precipitation.

We also found that March precipitation
was a significant predictor.

So it is buffered by this warming
and spring precipitation.

And the implications here –
our results suggest that large-magnitude

avalanches can occur during years
of below-average snowpack.

So it’s not just during big snow years.
But the probability a large-magnitude

avalanche year decreases through
time as snowpack decreases.

You know, one of the things – or,
one of the reasons we hypothesized

is that, you know, we’re still going to
get snow in the upper elevations.

So avalanches can still initiate at
these upper-elevation starting zones.

But, as an avalanche moves down
in elevation, and there’s – of course,

snowpack at lower elevations
is most susceptible to warming.

So, if we have less snow at lower
elevations, there’s going to be

a lot more surface roughness.
So more – basically,

more vegetation sticking out
and less material to entrain.

So the runout extent of these
large-magnitude avalanches

may be decreasing as well.
We didn’t – we didn’t look into that,

but that’s one of the reasons
we suggest why we might see

a decrease in probability as well
due to snowpack decreases.

So, again, will temperature play
an increasingly greater role?

We still need to do more
work on that in terms of,

will we see more wet-snow
avalanches, which I’ll discuss in a bit.

And – but we do know that studies
like this will help us understand

avalanche frequency
in a changing climate.

And this prepares us for
large avalanche cycles in general.

And then finally, dendrochronology
is useful for reconstructing these –

the large-magnitude avalanche
occurrence throughout a region

and can help forecasting efforts
and infrastructure planning.

So let’s move on to the remote sensing
portion of some of the work we do.

So we’ve been using drones to
map avalanches across the landscape

on mostly the slope or basin scale.
Because we’re more interested in

sort of a – both from an avalanche
as a hazard, but also from snow

as a water resource in specific basins
and how water – or, excuse me –

how snow lingers in certain terrain,
or not, into the spring and summer.

So we use drones and structure-from-
motion photogrammetry to generate

high-resolution digital
surface models, or DSMs.

And we’re talking about
the centimeter scale here,

so 2-centimeter resolution on some
of these models with very low error.

From here, we can determine the failure
layer or the bed surface of each of these

avalanches that we’re able to map.
Or, in the case of avalanches in

the upper left, these are –
the upper-left image here

with the boxes around them,
those are glide avalanches.

And then the – to the right there
is a digital surface model

of that same exact scene,
same exact day.

So we’re able to create
these digital surface models.

And the ones that fail to ground, we’re
able to look at the snow depth on those.

And determining the snow depth there,
or even just the failure layer of

avalanches that don’t fail to ground,
helps us with forecasting the timing

of other avalanches in other
locations that may occur later.

So, in those places that we can reach,
you know, on skis or foot or

snowmobile, we’re able to get there and
measure those depths for comparison.

So we created, again, a digital surface
model of snow depth change by

comparing the snow surface elevation
basically at different time steps.

Notice differencing.

So the data collected in this way,
in these short intervals – you know,

so just before and after a storm or before
and after a wind event, this allows us –

or, gives us detailed snow depth
information that’s really critical

for avalanche forecasting.

And so, in this work, in complex terrain,
we collected data over the course

of a winter on a weekly time step.
And fortunately [chuckles],

we were able to capture a large
avalanche you can see in that

top image there on the right.
And that was about a half-day after

it occurred, so pretty
recent after occurrence.

And we were able to identify the
weak layer in the slab depth from

that digital surface model instead of
getting in there and having to expose

ourself to the objective hazard on
that slope, even after the avalanche.

Overall, we found that a single storm
or melt event can dominate

accumulation patterns for the season,
depending on the magnitude of

that storm event or melt event.
And that the spatial variability

of snow depth is greater in complex
terrain than when compared to

more uniform and
sheltered meadows.

We are currently working right now
on processing these data on Yeti –

the USGS HPC – and looking at the –
basically optimal spatial variability

of snow depth in complex
terrain at these time steps.

And so it’s a huge amount of data that
Zach Miller in our group is working on.

And it’ll be pretty cool because the
resolution is so incredibly high

that it’ll be really cool to see what
we get out of this for optimal spatial

variability in this type of terrain.
And this – you know, this type of

work informs where and when
we should collect data for avalanche

forecasting. And, again, for water
resources into the summer.

So we also are looking at avalanche
frequency and how that affects

forests and vegetation. We implemented
a novel approach using Lidar and

aerial imagery at a random forest model
to classify imagery-based observation –

or, excuse me – imagery-observed
vegetation within avalanche paths.

And, again, this – the avalanche path
we see here on this image with the

red outline is in John F. Stevens
Canyon in southern Glacier Park.

And you can see the railway there
and how the bottom – the runout zone

of the avalanche path – or, the railway
runs through that, and then that runout

zone also extends over the Highway 2 –
U.S. Highway 2 as well.

So we used tree ring chronologies and
observational records to basically come

up with a spatially explicit avalanche –
or, spatially explicit avalanche

return intervals for this path –
well, two paths in this area.

Return intervals are, of course,
just the number of years

an avalanche reaches any
given point on the landscape.

And we did this, again,
for two paths to test its efficacy.

So let’s see.
There we go.

We can see that, in Panel A here,
the colors, of course, correspond to

return periods. And so this is –
this is the one we were just looking at

in that previous
satellite image.

But, in this case, we have
mapped the return periods.

You can see that it has sort of two
feeder zones, or starting zones,

and they, of course,
converge into one runout zone.

You can see the railway there.
And the red denotes roughly

a 2-1/2-year return period.
So that sort of looker’s right starting

zone is certainly a more frequent –
or, the avalanches run more frequently

in that part of the avalanche path.
And then you can see that the middle

of the path – because it’s an incised
channel, the avalanche debris typically

flows down the middle
of the avalanche path.

And then the sort of lighter blue
on the very outside of that runout zone

at the bottom by the north arrow
is a 22-year return period.

And so that’s a – you know,
a much lower frequency event –

or, sorry, much lower
frequency return interval.

And then the pink dots, of course,
represent our tree rings in place

so you can see that, in this case,
we didn’t just sample near the bottom

of the avalanche path.
We sampled up into sort of

what we call the track or the area
where the avalanche runs down,

not quite into the starting zones,
but basically we wanted to capture –

or, we wanted to try and look at
the tree rings where avalanches

weren’t necessarily large-magnitude.
So try and capture some of

these smaller, or more
frequent, avalanches.

And then the panels on the right
are just the vegetation that

we were able to classify.
So we used aerial imagery

and Lidar to classify this vegetation
in these avalanche paths.

And then, within the return
period zones within each path.

And so this – we used an automated –
based on the random forest model,

and it differed slightly between the
avalanche paths, but the combination

of both Lidar and the spectral signature
from the aerial imagery provided

the best accuracy,
which is about 88 to 92%,

in classifying the vegetation within
the complex avalanche terrain.

And so we also just looked at
Lidar to basically see how well –

or, how useful Lidar is.
Because you can imagine that,

you know, instead of going through
the work of doing the dendrochronology

and the amount of time it takes,
can we – can we look at, you know,

sort of low-frequency return periods –
let’s say, like, 25 years or 50 years

or 100 years using Lidar.
And Lidar is definitely useful

for that lower-frequency event when
we’re looking at canopy height.

But, when we’re looking at higher
frequency return periods – you know,

basically anything sort of below
20 years, then Lidar just doesn’t really

prove useful for discriminating
the return periods there.

Because it’s based on, again,
canopy height on that vegetation.

All right. And finally,
in the remote sensing portion,

we just began a project evaluating
the efficacy of using satellite imagery

to create a regional chronology
of avalanche disturbance

across the landscape.
So we’re doing this for the entirety

of Glacier National Park, to start off,
and we’re using pattern recognition

technique. This looks at changes
in the vegetation within and around

avalanche paths. So this provides
some measure of avalanche frequency,

and then the associated landscape
change at the decadal scale.

So, using long-term Landsat and other
satellite imagery data sets, like Landsat,

we can analyze the spectral
representations of previously mapped

avalanche paths across the large region
to quantify landscape change

due to avalanche disturbance.
The bottom there is using a tool called

LandTrendr to basically segment
based on the spectral signature.

So stayed tuned
for more on this.

We’re just starting this
portion of the project.

So dive into the avalanche
forecasting component and then

wrap up with some of the
wet-snow avalanche work

that we do as part of this
avalanche forecasting work.

So we – as I mentioned before,
we have a collaboration with the

National Park Service in Glacier
National Park in avalanche forecasting.

And we work with them every
spring during the opening of the

Going-to-the-Sun Road.
The Sun Road is 80 kilometers,

of which 56 are closed every year,
mostly in the alpine section.

And 11 kilometers of road – actual
roadway are impacted by avalanches.

There are 37 total paths that you
can see on the map on the right.

And the colors denote basically
just the length of roadway that

the avalanches impact –
so red being a greater amount,

and then yellow being a – or, sorry –
orange being the smallest.

And, for each day it’s open,
the Sun Road adds roughly,

you know, 300 to almost
650,000 to the state economy.

This is actually an old figure.
So I imagine it’s much more now.

Particularly since visitation
in the park has gone up dramatically

in the past few years.

The median opening
of the road is June 8th.

So avalanches are a common
threat to both people and

infrastructure in Glacier Park.
So this is an example in 1953.

A wet-snow avalanche
killed two workers

during spring
opening operations.

In 1964, a bulldozer operator
triggered a wet-slab avalanche that –

while clearing the road.
It was knocked off the road.

Fortunately, he survived,
but with pretty substantial injuries.

and you can see the fracture
line up there by the road.

That fracture line sort of extends
sort of in a stepped fashion

to the left of the image.
And that, for a little scale,

is about – it’s about 10 to 15 feet in
height from the snow surface to the –

to the bed surface there.

So the public are allowed to bike
and walk and recreate along the

Going-to-the-Sun Road
after work hours.

So, once the road crew and associated
personnel are off the road for the

day or the weekend, they’re
allowed to recreate on the road.

So, in 2003, a cyclist was moving
along the road and stopped

for a little snack here at this spot
and heard a really large rumble.

She turned around
and saw this. [laughs]

So an incredibly close call.
She was, of course, just fine.

But this is a large wet-slab avalanche.
You can see that it’s pretty dirty,

which means that
it went to the ground,

or near the ground, and entrained
a lot of dirt and woody debris.

And that was another close call.
There have been more recent

close calls with recreationalists
along the Sun Road in both in 2018

and again in 2021,
just this past spring.

There were cyclists that had traveled up
the road, and an avalanche came down,

blocked the road,
like you can see in this image.

And basically blocked
them from getting back down.

The debris was, you know,
too steep for them to walk back across,

particularly with bikes.
And so they were stranded up top.

So we helped assist with sort of
forecasting all the road crew.

Road crew cleared the
debris off of the road there.

So the Park Service is re-evaluating
their interface with the public

regarding avalanche hazard,
and we’re involved in those

discussions with them to help –
to help consult with them.

But, for an internal forecast,
which is what we provide,

this is sort of
what it looks like.

And any employee of the park who is
working on the road can access this.

The map there on the left page
denotes the avalanche rating

on specific parts of the road.
So some parts are actually –

you know, would be rated differently,
depending on the amount of snow left,

the potential for any sort of avalanche
hazard, steepness – those sorts of things.

And so we have a sort of
an internal rating system that we use –

or four-scale system.

And then we provide a bit of
a discussion so that, you know,

they can understand why things
are rated the way they are.

And then we, of course,
provide the probability –

the expected size of avalanches and
then the trend over 12 to 24 to 48 hours.

And so this is an internal forecast.
Again, we work with Park Service

forecasters to do this work, but this
is put out as a daily product.

So part of the avalanche forecasting
partnership, as I mentioned,

allows us to collect data
related to wet-snow avalanches.

And this is becoming increasingly
more important in the context

of a warming climate.
So one of the things we’re looking at

is characterizing the evolution of
snowpack properties as the snowpack

transitions from a cold, dry snowpack
to a wet snowpack in the spring.

The timing and during of this
transition – well, hopefully,

understanding that will help us
understand the timing of these

destructive wet slabs and glide
avalanches. So this schematic shows

how water moves through a cohesive
slab here, through flow fingers.

And it can potentially pool at
what we call a capillary barrier

between two different layers.
This happens when we have

fine grains – what we’re
calling rounds in this case –

over more coarse-grain snow –
more coarse-grain snow, like depth hoar,

which is a pretty common weak
layer in continental snow climates.

So, if we get water pooling at
that interface between the two,

and it erodes the bond between
the cohesive slab above and the

weak layer below, then we could
potentially get an avalanche.

As I mentioned, they’re destructive.
You saw the pictures.

They have a large impact
on the spring opening operations

on the Sun Road
each year.

Another type of avalanche
that impacts the road

that we have spent some time
studying are glide avalanches.

And glide avalanches – so snow cover
on a slope is always gliding downhill.

So gravity is just going to
slowly pull it downhill.

So it slides downslope,
but it slides at variable rates.

Of course, some parts move
downslope faster than others

due to a variety of factors, including
what’s underneath the snow.

So, if it’s a smooth rock slab,
then things are going to move and glide

downhill faster throughout the season.
And the glide, remember,

is a very slow process.
It’s not sort of this quick, spontaneous –

or, not spontaneous, but this quick
release of an avalanche that –

when you think of a snow avalanche.
This is a very slow movement.

But you get these variable glide rates,
and then you can have a tensile crack

that forms. We call it a glide crack.
So that’s sort of the frown face

that you can see there.

And the sort of prerequisites for a glide
avalanche are, we need, basically,

water moving through the snowpack
and moving along, or pooling,

at the bottom of the snowpack
and at the ground interface,

which is typically at the zero degrees,
once the snowpack has become

isothermal at zero degrees –
so same temperature throughout

the whole snowpack.
And, again, minimal surface roughness.

So smooth rock slabs are
a pretty common area

where we see these glide avalanches.
Here’s an example of a glide avalanche.

They involve the entire snowpack,
so they’re often very destructive.

So, for a little reference here,
the fracture line height in this image,

where the avalanche released
from the snowpack above this,

it averages about 15 feet,
and it tops out at about 50 feet.

Kind of can’t see that,
but it’s up in the upper-left corner

of that fracture
line there.

So the time before – or, the time
between crack formation, as you saw

in the previous image, and full snow
depth release can be highly variable.

It can be, in fact,
sometimes hours,

not even just days,
but hours, days, to weeks.

Often triggered by rain, snowmelt,
and even sometimes we’ve seen

these avalanches release
during cooling periods

after an intense
warming or rain event.

Glide avalanches also tend to occur
in the same location each year.

We call these
repeat offenders.

So we map the ones that are
repeat offenders that affect

infrastructure and roadways.
We developed an automated

GIS model that identifies potential
locations – or, allows us to identify

potential locations for other glide
avalanche release in other areas.

We use time-lapse photography to
pinpoint the timing of glide avalanches.

And the timing – or, pinpointing
the timing, of course, is the real crux.

As you all know.

So looking at these images,
we can ostensibly count the number

of dark pixels versus the
number of light pixels

in the image where –
that red box is an example.

And so we track that – the number of
light versus dark pixels in a sort of

area of interest over time.
This gives us a rate of glide

to see if the glide rates are
increasing during specific weather

events or, you know,
just prior to full depth release,

like on hot, sunny days,
for example.

So the camera we have
takes an image every 15 minutes.

Allows us to identify avalanche
release and glide rates on

a really short time step.
Of course, you know, the limitation

here is that, when it’s cloudy or
dark [laughs], then we can’t see it.

But fortunately, in the spring,
we have a lot of – we have a lot of

daylight up here in northwest Montana.
So that helps, but we have seen

avalanches release – they weren’t
there in the last picture in the evening,

and then they show up – the bed
surface is visible the next morning.

So that’s one of the techniques that
we use for glide avalanches as well.

So that’s all I’ve got.
I want to thank Matt and everyone

in the Landslide Hazards Program
for the opportunity to present our work

at NOROCK with a bit of
a focus on our avalanche science.

So thanks to all of you,
again, for tuning in,

and I’ll take any questions
that there might be.