Evaluating rockfall frequency from natural slopes (multiple methods)

Video Transcript
Download Video
Right-click and save to download

Detailed Description

Understanding of rockfall frequency-magnitude relationships is important for managing rockfall hazards, but characterizing these relationships is a challenging problem due to limited data, limited access, and the difficulty of accurately dating historic rockfalls. Developing frequency-magnitude relationships can be particularly difficult for natural rock slopes, which can still present significant hazards, but where data is often sparser due to greater distance from roads and buildings. This talk will provide an overview of the topic of rockfall frequency measurement by briefly discussing the relevant literature and presenting examples from the application of two different methods to the same study sites. The literature review will present a summary of methods previously applied to measuring rockfall frequency, including advantages and disadvantages of various approaches. The two case studies will come from original research in Glenwood Canyon, CO, an area where natural slopes present significant rockfall hazard to Interstate 70, but where previous knowledge of rockfall behavior consists mainly of the anecdotal insights gained by highway management personnel, without the benefit of systematic study. The first case study was conducted using lichenometry, and the second is based on an ongoing drone-based monitoring campaign. Comparisons between the results for the two methods will be made, and implications of these results for rockfall behavior in Glenwood Canyon will be discussed.

Graber A (2021). Evaluating rockfall frequency from natural slopes at multiple timescales using multiple methods; Examples from Glenwood Canyon, CO. USGS Landslide Hazards Program Seminar Series, 20 October 2021.


Date Taken:

Length: 00:52:39

Location Taken: Glenwood Canyon, CO, US

Video Credits

Video thumbnail: Aerial view of Glenwood Canyon overlooking the Hanging Lake exit and the west portal of the Hanging Lake Tunnel. Photo taken July 14, 2021, Andrew Graber, Colorado School of Mines.



My name is Stephen Slaughter,
and I’m today’s temporary host

for the USGS Landslide Hazards
Program seminar series.

Your usually host, Matt Thomas, is in
the field and will return next week.

During today’s presentation,
please remember to keep your

microphones muted and video off.
We ask the speakers to reserve

10 minutes for the top of
the hour for questions.

So, following the presentation,
you can submit questions

via the chat function or, preferably,
use the raise-the-hand feature

to ask questions using your
microphone and camera.

Today’s speaker is Andrea Graber.
Andrew completed his B.S. in geology

at Wheaton College in Illinois and is
currently a geological engineering

Ph.D. candidate at the Colorado School
of Mines working with Paul Santi.

His research thesis,
which he’ll be discussing today,

focuses on understanding rockfall
frequency and depositional behavior

from natural rock slopes.
In addition to his thesis work,

Andrew has published papers on
design methods for post-wildfire

erosion mitigation and on modeling-
based investigations on critical

groundwater conditions for
large landslides in southern Peru.

Andrew received the 2019 AEG
Marliave scholarship and has received

research and field work grants
from AEG, GSA, the Wyoming

Geological Association, and the
Colorado Scientific Society.

So welcome, Andrew,
and the virtual stage is yours.

And I turned on
the transcript.

Let me turn the transcript off here.
There we go.

Okay, Andrew.
Sorry, go ahead.

- Sounds good. No worries.
Thanks, Stephen.

Just to check as I get started, are you
seeing my mouse pointer on the slides?

- I am not seeing it.
- Laser pointer should work then.

- You showing it?
- Yes.

- I don’t see it, unfortunately.
- Cursor.

Well, we can probably get away
without that. Thanks.

- Okay.
- Well, thanks, Stephen.

And thanks to the USGS Landslides
group for inviting me to talk here today.

So what I’m hoping to accomplish
in this talk is to talk a little bit about

my thesis research. But I also want to
talk about this subject of rockfall

frequency because it’s kind of
a niche topic, but it’s an interesting one

from a geomorphic perspective,
and it’s an important one

from a hazards perspective.
And so I hope to be able to give

an overview of a little bit of the theory,
some applications that other people

have done to try to assess
rockfall frequency in various ways,

and then to talk about the
particular case studies

in Glenwood Canyon
that I’ve been a part of.

So, to start off with some
acknowledgements, I wanted to

thank my adviser, Paul Santi,
whom I’ve been working with

for several years now,
and his help has been invaluable.

And then some folks at CDOT –
Matt Tello, Ty Ortiz, and Bob Group –

who have been helpful organizing
funding for me, but also making sure

I have access to facilities in the canyon.
Folks who have helped me in the field –

Claire Graber – my wife, Julia Payne,
Josh Clyne, and Kyle Radach,

who are all School of Mines alums
who helped me with field work.

And then, for funding, School
of Mines, Colorado Scientific –

or, sorry – Colorado Department of
Transportation, Colorado Scientific

Society, AEG Foundation, Wyoming
Geological Association, and GSA.

So we’ll talk about
rockfall frequency as a topic.

I’m going to talk about some established
methods that people have already

published to evaluate rockfall
frequency – to quantify it.

And then I’ll give some background
on my study area, and I’ll talk about

these two case studies from my thesis
research, both in Glenwood Canyon,

looking at rockfall frequency
at the same sites but using

two different methods.

So rockfall is this erosional geomorphic
process, and it’s also a hazard.

And so, when we talk about rockfall,
there are a couple of ways

to think about characterizing it.
So one is this parameter of frequency.

And, as you’d expect, that refers to
the frequency of how many rockfalls

are happening per year.
But, like with other geologic events,

like earthquakes and floods,
rockfall – and landslides – rockfall

frequency is tied to rockfall volume.
So the way that we quantify it –

the best way to quantify it is to
quantify it in that context in a way

that includes volume.
So the way that people have usually –

or, have often done this –
not every study, but a common way

to do this is to use
a frequency-magnitude curve

like you might see for
floods or earthquakes.

So what that plots is magnitude –
in this case, volume – on the X axis

versus frequency on the Y axis.
And you can see a bunch of different

kinds of parameters on the Y axis
here – cumulative frequency,

frequency density – which is just
frequency normalized by the

bin width of your histogram,
probability of exceedance.

A number of different parameters,
but this relationship still holds up.

So what people have observed when
they plot up rockfall data this way is

that there – in log-log space,
there’s this linear relationship,

which is quantified by a power law.
It’s not a linear regression.

It’s a power law regression.

And what that does is, you can use the
coefficients of the power law fit to that

data to describe the trend of rockfall
between frequency and magnitude.

And this plot also shows something
that people have often observed,

which is that there’s frequently sort of
a drop-off of data towards the

lower end of the volume spectrum.
And that’s related to – what people

usually attributed that to is,
in their rockfall databases,

the small events get under-sampled.
And so that’s why they fall away

from this linear trend, where,
in theory, the linear trend

should continue all the
way down to zero.

So other options for quantifying
rockfall frequency – people would

use a recurrence interval,
like with floods or earthquakes,

but those also have to be tied with
magnitude. And, as you probably

all know, recurrence interval is
basically just the inverse of frequency.

And so then the other side of this is,
where is rockfall happening –

on the talus deposit? Is it running
out beyond the talus deposit?

And so that’s another piece of this topic
that’s important to consider and is

usually a part of what people are
looking at in these frequency studies.

So I’m going to talk about a little bit
of literature review of studies where

people have looked at rockfall
frequency and depositional behavior

in order to get timing information,
magnitude information, to better

understand these processes.
And note that a lot of studies use

a combination of methods,
but I’ve broken the big methods

into a couple of categories.
So one is to look back at some kind of

a historical record of rockfall
and to be able to make calculations

based on a database that way.
So using actual historical records

where people have written down
the times and locations and volumes

of rockfall, dendrochronology using
tree rings, cosmogenic isotopes,

lichenometry – all to date old events.
And your other main option is to

directly measure rockfall and
create an inventory that way.

So change detection methods are
becoming more and more popular.

Seismic monitoring is
another one that’s been used

in at least some fields,
and we’ll talk about that.

And then there are a few other options
that are out there that are published.

I’m not going to talk about them
in any detail in this presentation,

but I can send you some
references if you’re interested.

Feel free to contact me – terrestrial
SAR, rockfall collectors, using

radiocarbon under select conditions –
so those are some of the other options.

So historical records are great
for computing rockfall frequency,

but they’re hard to get. So they’re easier
to find along transportation corridors

or if there’s a specific site of interest.
Yosemite Valley is a good example

of a place where there’s a pretty
extensive rockfall record.

But the detail and availability of
these records can be highly variable,

and sometimes a record will have –
for some events, it’ll have all the

information you want, and it
won’t have that for other events.

And these are often biased towards
larger events simply because those

are the ones that are more likely to
make it all the way down the talus

and impact a road or a facility
that somebody cares about.

So this plot on the right is from
an example in British Columbia where

they had two transportation corridors,
and along each of them, there was –

there were rockfall records
from both a road and a railroad.

And so they could – they could plot up
frequency-magnitude data and be able

to compare them along two corridors
and have two data sets to look at

each corridor. And so they got the
sort of plot that we’re talking about –

this linear relationship in log-log space.
There’s a little bit of this tailing off

towards the smaller-magnitude events.
But they found some pretty nice

agreement between their first corridor
of the highway and railroad and

the second corridor of
their highway and railroad.

And so, if we look at these
B parameters, to put these in context,

if the – if the B parameter is greater,
that means your curve is steeper,

and that means the relative
contribution of small events is greater.

And so that’s how you can start
to compare these curves

between databases. And this is a case
where they had some independent dates

from some rock avalanches in the area,
and so they were actually able to

extend up to some very large
volumes in their plots that way.

So dendrochronology – well, I guess
I’ll start off by saying, what if you don’t

have historical records of rockfall?
How do you build a database using

some other kind of record?
So one example is using

dendrochronology. That’s basically
counting ages with tree rings.

And this is a pretty
well-established age relationship.

And, when you take cores of trees,
if your – if your core intersects any

sort of disturbance that might have
been caused by rockfall, like a scar

in the bark or a year where the growth
ring is thicker because the tree was

reacting to the trauma, you can use
that to build up a rockfall database.

So you need to have relatively
dense long-lived trees.

And it takes a lot of effort,
but it gives you this opportunity to

look at spatial distribution of rockfall.
So this example is from Switzerland.

And they had a database of trees –
those are these little dots in

the upper-left plot in
kind of the colored area.

There’s some black and gray dots.
So those are marking locations of trees.

And the – so the color ramp in
this central part of the map

is showing the recurrence interval
between growth disturbances.

And so that’s what they’re using
as the proxy for rockfall events.

And it means that they’re able to
develop recurrence intervals for growth

disturbances at each tree and then
contour those values so that they have

a plot of recurrence interval as it
varies across the talus deposit.

So what they showed – and if you
look down at the plot below that,

they were able to identify the same
damaging event in a whole bunch of

trees all the way down this
north end of the talus deposit.

So they said, while most of our events
are relatively small incremental

rockfalls, or fragmental rockfalls,
we have record of one large event

that made it all the way down
this talus deposit and damaged

a lot of trees
on the way down.

So this is a method that’s not
precise enough to capture

year-to-year variations in rockfall,
but these more decadal-type variations,

you can detect with

So lichenometry is another option for
trying to build a database of old events.

This is a Quaternary dating
method from glaciology.

And it uses calibrated lichen
growth rates to infer

minimum surface exposure ages.
So what that means is, you develop

a calibration curve that characterizes
the relationship between lichen size

and lichen age. And people do that
by dating surfaces where they

know how old the surface is,
and they can measure a lichen on it.

And then you go to places where you
don’t know the age of the surface,

and you measure the lichen size
and compare it to that curve

to get your age estimate.

The accuracy can be a bit limited – and
the age ranges as well – to a little bit

longer than dendrochronology, but still
in the last 1,000- to 2,000-year range.

The upside is, this is a method that lets
you collect a lot of data very quickly

because you’re just collecting
lichen measurements.

It assumes that the rock surface
is clean prior to growth.

And there have been some
concerns raised by lichenometry.

You know, these are such long-lived
organisms that people are concerned,

how well do we actually
understand the lichen growth?

How linear is it –
or, how consistent is it, I guess.

And so that’s worth
acknowledging here.

But there is a fairly well-established
relationship between age and size.

And so you can still say some things
about, well, larger lichens are older.

So this is a case study from the
Austrian Alps where they had

a really large lichenometry data set
for a whole bunch of talus deposits.

And the map on the right is
showing those deposits there.

And I think someone is unmuted.
I’m going to mute – there we go.

We’ve still got someone unmuted here.
I’m getting some …

- Yeah. Could someone – everyone
please check and make sure

you’re muted?
- I’m just getting some feedback

on my speakers.
- Yeah.


- Okay. That’s
better now. Thanks.

So, in this case study, they had
lichenometry measurements from

a whole bunch of talus deposits.
And what they’re able to show is

that the – they were looking at
lichen coverage to say, well,

at the tops of these talus deposits,
there’s a lot more rocks that are free,

or almost totally free, of lichen.
So we’re using that as a proxy

for greater activity
at the tops.

But they also found an association
between these more lichen-free fans

and permafrost. So they were
able to associate, well,

these permafrost-affected slopes
are causing more rockfall in certain

places around our study area.
And they also looked at

the calibrated lichen ages.
And, looking back, they were able to

find this peak – if you look at the
left-hand histogram down at the

bottom of the slide, so that peak
around 1890 or so, they associate that

with increased rockfall activity
at the end of the little ice age,

giving you more lichens that were
from that time period. And so they

were able to show a little bit of
temporal variation that way as well.

So another way you can get the ages
of previous rockfall is with

cosmogenic isotopes.
This gives you a good age range.

It also makes that assumption that
you have a boulder that was –

it was buried so deep, it was not having
any cosmogenic ray interactions.

And then the rockfall happened,
and it was exposed.

This is a method that’s more expensive,
and it takes a lot more effort to get

your samples and sample –
and do your sample processing.

And so that’s why it can be
difficult to get a big enough data set

to actually build a
frequency-magnitude curve.

But there’s still some
things you can learn.

So this is a case study from
Greg Stock’s work in Yosemite.

This is from a paper that’s
kind of a park-wide hazard

assessment for Yosemite.
And so they collected beryllium-10

cosmogenic ages for outlying boulders.
And so what they meant were boulders

that were beyond the edge of the
talus deposit, but were still obviously

rockfall-related and were also –
tended to be in and among

some of the campgrounds and,
like, facilities that already existed

in Yosemite Valley. And some of
these are really quite large.

There’s a picture of
one on the upper right.

So they collected these beryllium-10
ages and plotted them up.

And what they found –
they found a couple of things.

So one is that all of the ages fell below
their 15,000-year expected cutoff

because that’s the inferred age
for the deglaciation of the valley.

So all the rockfalls came out as
younger than that, which makes sense.

And that allowed them to use that
15,000-year upper age estimate to

get a recurrence interval for their
total data set of outlying boulders.

And so, if you take 15,000
and divide it by 258,

you get 50 to 60 years,
on average.


So, if we switch over to the idea of
directly measuring rockfall with

change detection – or, rather,
change detection is a method

that’s becoming a lot more popular
these days as technology becomes

more available and more usable.
So what I’m referring to with change

detection is taking two point clouds –
if you look at the graphic on the lower

right, taking two point clouds where, in
the left panel, it’s showing computing

a normal to the first point cloud using
some-size neighborhood of points.

And then, in the right panel,
it’s showing how you then align

a cylinder along that normal
and average the points in the

two clouds so that you can
actually calculate the difference

between those two
clouds along that line.

So, when you repeat this kind of
calculation thousands or

tens of thousands or – of times for
an entire point cloud, it means you can

map change across a large surface.
And so you – I’m sure many of you

are already aware of how
powerful of a technique this is.

So you can do change detection on
point clouds developed using Lidar or

structure-from-motion photogrammetry.
For those of you that don’t know,

photogrammetry is this technique
visualized in the upper right,

where you take repeated overlapping
photos of a subject with a moving

camera, and then a computer
algorithm reconstructs the shape

of your subject based on
the motion of those cameras

and identifying the features
in multiple frames.

So photogrammetry is a cheaper
version of preparing point clouds,

but you’ve got issues with moving
vegetation, and you’ve also got to

be very careful about how you
scale the point cloud – if you use

scale bars or GPS ground control –
to tell the model what the

dimensions should be
in real life.

Lidar gives you very
precise measurements,

but it’s more expensive, and so it
can be more difficult to apply.

But both of these give you point clouds
that let you detect changes down to the

centimeter scale, which can be really
powerful for detecting rockfall,

since all the methods we talked about
before pretty much leave out all

the small – the things at the small
end of the volume spectrum.

So this is a photogrammetry case study
from a pit mine in Australia where they

had several monitoring periods where –
so the top image is showing the

pit slope they
were monitoring.

And the colors are representing
different periods of rockfall monitoring.

So you can see all the rockfalls that
happened in periods A1, A2, etc.

So they detected a total of
645 events on the small end

of the volume spectrum from
February to March 2018.

And so then they were able to
prepare frequency-magnitude plots.

And this is a point where you can –
you can start to see a few

of the differences here.
So this red line is showing a period of

lower activity while the blue
and orange lines are showing periods

of higher activity. And they associated
the higher activity with wet periods and

then also with extremely dry periods
when they had these weak siltstone and

claystones drying out and contracting
and fracturing more because of that.

These figures are from a Lidar
change detection study

where they had several monitoring
periods, but I think about a year of it,

they took hourly Lidar scans of
this coastal bluff in England.

And what that allowed
them to do was compute

their frequency-magnitude plot –
that’s the left of the upper graphs –

at several different
time intervals.

So, and by the way, on the Y axis,
you’re seeing the probability of

exceedance instead of what I put on –
or, what was on the previous plots

with frequency, but the power law
relationship still holds.

So this is showing how, if you
monitor at a shorter time interval,

your frequency-magnitude curve
gets steeper because your relative

contribution of small events is
greater and the large events is less.

And that’s because, when you have
frequent rockfalls at a slope, often

rockfalls will happen from the same
location on the slope repeatedly.

And so, if you scan those at – a year
apart, it might look like just one event.

But if you were to scan it every month
of the year, you might see it was,

in fact, five different events that
all just occurred in the same place.

So you see – with more frequent
scanning, you see more smaller events.

And then the plot on the right is
showing the change in the B parameter

for the power fit with the
change in the monitoring interval.

So I wanted to say just a quick word on
seismic sensing for rockfall monitoring.

A lot of the case studies that
I’ve looked at for this apply

seismic monitoring of rockfall
in volcanic settings, where they

were looking at a lava dome and
monitoring rockfalls from the lava

dome to use as a measure of volcanic
activity, whether that’s the extrusion

rate of the magma or if it’s related to
seismicity in the magma chamber.

So this plot in the lower left is showing
how rockfall activity was increasing,

and they were able to correlate
that with some other signs of

volcanic activity. I’m not aware of any
examples that took a seismic plot –

or, a seismic record of rockfall
and created a frequency-magnitude

plot from it. And I think that’s because
of some concern over estimating the

volume from the seismic amplitude.
Because that’s – the amplitude is

dependent on how close you are to the
seismic event and some issues like that.

So now that we talked about the
literature review, that’s kind of

a quick view of a whole variety of
ways to look at rockfall frequency.

And so, with those things in mind,
I want to switch gears and talk about

these case studies that I worked on
in Glenwood Canyon.

So what I’ll start with is to talk about
background on the study area itself.

So any of you that are already
in Colorado or have explored

the west may have some familiarity
with Glenwood Canyon.

So this is a transportation corridor
for Interstate 70 as well as the

Denver & Rio Grande Railroad
in western Colorado.

It’s a 13-mile-long stretch
of the interstate.

And Glenwood Canyon suffers from
really frequent rockfall problems.

And I’m sure you all heard, as well,
about the debris flows that have

happened in Glenwood Canyon
this past summer.

And rockfall has been a problem
in Glenwood Canyon ever since people

started putting roads through there.
So the geology of this area – we’re just

going to talk about the geology of
the lower part of the canyon

because that’s what’s relevant to
the rockfall studies I’ve been doing.

So, at the – at the base, there’s
the Precambrian basement.

These are metagranitoids,
sometimes intruded with some

other younger granite dikes.
And then, above that,

there’s two Cambrian formations –
the Sawatch orthoquartzite

and the
Dotsero dolomites.

And these form the base of
the White River Uplift.

So that’s what the Colorado River cuts
through to form Glenwood Canyon.

So last year, the Grizzly Creek Fire
burned part of the canyon itself

as well as Grizzly Creek Canyon
and much of the White River Uplift

in the vicinity of Glenwood Canyon.
And we’ll talk a little bit more about

that later because it’s more relevant
to the drone scanning study.

But that’s another piece
of our context here.

So I like to include this picture
because it shows some of the

motivation for studying rockfall
in Glenwood Canyon at all.

So there are plenty of cut slopes
in Glenwood Canyon and,

as you all know, cut slopes are
a frequent source of rockfall.

But this is a place where
we actually have a lot of rockfall

from natural slopes. And you can
see the proximity of these –

so the big slope above the
tunnels to the right,

that’s all the Precambrian
granites and metagranites.

You can see the proximity
of these kinds of slopes

to our transportation facilities
here in the canyon.

So, for the first case study,
we studied six cliff and talus deposit

systems in the Glenwood Canyon area.
Those locations are marked by

the orange points in this map.
And then the background shading

is just a slope map, and it’s really
just to show the relief of the canyon.

So you can get an appreciation for it.
So these slopes – four of them

are granite rock masses.
There’s one orthoquartzite

rock mass and then another one
that has – another talus deposit

that has sources that are both
granitic and orthoquartzite.

And we choose the locations primarily
for access because they were

near to places where it was possible
to park and get out and walk to –

walk to these slopes
to collect data.

So we chose to use lichenometry
to quantify rockfall frequency

because we don’t have much
of a historical database.

Even with all the focus on Glenwood
Canyon to clean up rockfall hazards,

the database is really limited.
I have a copy of part of it,

and most of the events, there’s
no volume recorded for it.

So we have timing
information but no volume.

So that doesn’t help us
for frequency-magnitude.

We don’t have enough large long-lived
trees for dendrochronology.

We’re in a deep canyon, so cosmic
ray shielding is problematic

for cosmogenic dating.
And the other upside of lichenometry

is it’s just a lot cheaper than any
sort of direct methods that

require advanced equipment.
And, in addition, this was –

so I did the field work for this in 2018
and, even in the few years since then,

photogrammetry technology, especially,
has been getting a lot cheaper.

So, to collect the data for this,
I measured lichens on boulders

at each slope and sampling
along the contours.

And we’ll look a little bit
more at that in a moment.

So these images are showing
the study sites from this project.

The red dashed lines are
outlining the source areas.

And then the talus is
below in each photo.

and then the letters are
indicating lithology –

so G for granite,
Q for quartzite.

So, as I alluded to earlier,
with lichenometric studies,

to get your estimated ages,
you have to have some understanding

of what the local growth rate is
for your lichen species.

Because, since lichens are
living organisms, they’re affected

by climatic factors that can
change their growth rate.

So theoretically, you can –
you can get calibration data from

any surface that you know the age of,
which could be a foundation,

or a monument, or a quarry wall, or –
there are a whole bunch of possibilities.

But most frequently, the known
age surface information

comes from tombstones
in local cemeteries,

where you have a date right
there on the gravestone.

There’s some uncertainty with
gravestone ages just because they might

have been put up a few years later after
someone died, or a few years early.

But it gives an okay
approximation of age.

And so then you can start to develop
these curves, which I’ve put in the

lower right, that are plotting the
relationship between surface age

of these known-age surfaces
and then our lichen long axis.

So, to get our calibration curve,
we plot a linear fit to the

fastest-growing lichens. So those
are marked by the little crosses.

And you can – you can see that
this data is a little bit limited.

And so that’s worth bearing
in mind that our calibration curves

are not very well-constrained.
And so it’s just – so that’s

an uncertainty in
our calculated ages.

So, for the data set, there’s –
I have 1,200 lichen measurements

from these deposits.
And so these histograms on the

right are summarizing –
oh, and I forgot to mention,

the reason I have two columns
over here on this slide

is because I have two species.
So the idea behind choosing

two species was that we could collect
two semi-independent lichenometry

data sets and then be able to compare,
do they show the same picture

about rockfall or do they not.
And we’ll have sort of

a semi-independent
check on our results.

So that’s why these and some
other subsequent plots are

color-coded where the brown is
for one species, Lecidea atrobrunnea,

and the green ones are
for Lecanora novomexicana.

So the histograms are showing our
lichen data after we’ve converted

the size measurements to ages.
The data set represents

over 1,000 boulders with
a few that are – have volumes

up to 24 cubic meters.
That was the largest one.

And then a note about
an assumption in this method.

This method is assuming that
every block in the talus is

its own rockfall event.
And, as you can imagine,

that is an assumption that’s more likely
to be true for small blocks, which

theoretically are less likely to shatter,
and less true for large blocks.

So that’s another assumption.
We have to make it with the

precision of lichenometric dating.
It’s not really possible to date

five adjacent boulders and tell
for sure whether they all came from

the same rockfall event or not.
So that’s an assumption to list here.

So this gives us frequency-magnitude
curves where we’ve plotted

our rockfall volumes. Those are
from the boulder size measurements.

And then, on the Y axis
is frequency density.

So that’s our frequency where
we’ve computed – we binned the

lichen sizes based on volume,
divided the oldest age in that bin

by the number of measurements
in the bin, and then further normalized

that by the width of the bin.
So that’s why it’s frequency

density instead of
just frequency.

So that gives us our
frequency-magnitude curves.

With some of these, there’s pretty good
agreement between the two species at

a given slope – Slope 1 and Slope 2.
With others, there’s a little bit

more divergence. And another thing
that you can observe about these plots

is that the data start to fall away from
the curve towards larger magnitudes.

And we attribute that to the under-
sampling of the large volume rockfalls.

Because we’re assuming every boulder
is its own rockfall and larger volumes

are just going to be more likely
to disaggregate – you know,

a 24-cubic-meter boulder
is pretty large for this area,

given typical rock
mass characteristics.

So, from the frequency-magnitude
curve, we can calculate recurrence

interval, which is just
1 over our frequency.

And that lets us get recurrence intervals
for different volumes of rockfalls.

So these are color-coded –
so the blue is for a 0.1-cubic-meter

rockfall – 1, 10, 100 cubic meters.
And then, each of these is grouped

because we’ve got our
two species’ estimates again.

So the picture is pretty consistent
at the smaller volumes.

Our recurrence intervals in years
are in the 1- to 10-year range.

But, once we get into larger volumes,
there starts to be a lot more scatter.

There starts to be less agreement
between the species.

And that goes back to the mismatch
in our frequency-magnitude curves.

One thing that’s curious about this
is that the – so Slope 1 is one of

our quartzite slopes. And that
looks very similar to Slope 2,

which is a granitic rock mass.
And just from looking at them,

you might – it would be reasonable
to conclude, well, the quartzite

one should have more frequent
rockfall, right, because it’s

much more frequently jointed.
And so that’s one of the things that

was a little bit curious and that
I’ll come back to with the drone study

because that’s one of the things
we’re looking to check is,

does this result
actually make sense.

If we plot up the lichen ages spatially,
it gives an opportunity to try to look

for patterns in deposition to see,
are we seeing older ages at the

top of the deposit and young –
or, sorry – younger ages

at the top and
older at the bottom?

And the patterns are not
very clear in this data.

So we conclude that to indicate that
our rockfall activity over the time

span that we can look at here, which is,
you know, going back – in these plots,

the oldest age is about 500 years.
In that time span, rockfall at these

slopes is dominated by these smaller
incremental rockfalls instead of

a large event that might bury
a whole side of a talus deposit.

So then, to put these talus deposits
into a longer-term, like, geomorphic

context, we wanted to know, how does
the volume of the deposit relate to the –

relate to the frequency-magnitude
information that we have?

Basically, to back out, if this is our
frequency-magnitude curve,

how long would it take us to get
all of the volume of talus that

we’re observing at these deposits?
So we were looking at

Glenwood Canyon here –
this is an oblique image –

to illustrate how talus tends to
fill these chutes and gullies

in these granitic bedrock slopes
rather than forming cones

or big aprons of material.
There’s some of those kind of cases,

but we didn’t really find
a volume estimation

in the literature that
matched the slopes that we had.

And so we came up with some
new ones to try to model,

if we have this gully, and we
fill it with talus, and it’s basically

bounded by bedrock, what kind of
volume does that give us?

And so, once we get the volume
estimate, then we can – we can

back it out to find out what kind of time
span of deposition are we looking at.

So, to illustrate how we get the volume,
this plot on the right is showing –

here’s the outline of our deposit.
Here’s the DEM.

In Part B, here’s the DEM
of the deposit surface.

And so, if we – if we estimate what this
channel steepness is from the bedrock

slopes around the deposit, and we get –
we use the interquartile range of these

bedrock slope measurements to get
a steeper channel – or, sorry,

a steeper chute and a shallower
chute so that we have

an upper- and a
lower-bound volume.

And then, in the lower group, it’s just
showing the DEM put together with

the lower surface, and that’s what
we use as our deposit volume model.

So this upper-right plot is showing
the volume ranges that we estimated

for each deposit. So those are varying
between about 10,000 cubic meters

all the way up to between
400,000 and 700,000 cubic meters.

That’s a much larger, more
extensive deposit – Slope 6.

And so we found an equation that
can use the frequency-magnitude

parameters, A and B, an estimate
of your largest and smallest possible

rockfall sizes, to calculate a volumetric
flux, which we then – we took that,

and we divided our total
volume estimate by that flux

to get our accumulation time.
And we constrained this max volume

using some measured events.
So there’s a rockfall that hit

the highway in 2004 from the quartzite
that was about 1,100 cubic meters and

then a 2010 rockfall of about
450 cubic meters from the granitoids.

And so these are our estimated
accumulation times from those.

And we have some good overlap
with some of these and some

not-so-great overlap [chuckles], or lack
thereof, for some of the other slopes.

And so that goes back to the shape
of the frequency-magnitude curve.

And where we have mismatch between
the two species in those curves,

we get a bigger possible range
for our accumulation time.

So we have these
recurrence interval estimates.

Some of them seem to
make possible sense.

A 1-cubic-meter rockfall
every few years seems reasonable

given the amount of lichen coverage
on these deposits, the amount of,

like, scars on boulders from,
you know, recent rockfalls.

So we have those estimates.

We have this picture that
the smaller rockfalls

are dominating over
our evaluated time scale.

And that it’s a fairly random
depositional pattern at that,

you know,
five-century-type time scale.

And we have these times that
we estimated to accumulate

the talus volumes.
But these left us with some

more questions about, well,
how do we know for sure that

these results actually make sense?
How do we validate this with

an independent method that’s
not dependent on lichen growth?

And so that’s how we got to doing a
drone study in Glenwood Canyon.

And I’ll talk about that in a minute.
I just wanted to acknowledge

some of these uncertainties about,
when you measure a lichen date,

you’re assuming that it grew since
the surface was exposed, but maybe

there was a lag in that time.
And we also have – we know

there’s some uncertainty in
our calibrated growth rates.

And our larger events are
probably under-sampled.

And so we have
some issues there.

So, for the second case study,
the Grizzly Creek Fire happened

in 2020 before we started
a drone monitoring campaign.

So that’s – the colors on this map
are showing the burn intensity.

And so we ended up trying to design
this study to be a hybrid, where,

one, we look at, how does drone –
what does drone scanning tell us

about the slopes that we have
lichenometric data for, but also, what

happened after the Grizzly Creek Fire?
Do we have more rockfall from

these granitic slopes because
of the wildfire damage?

So we reused three sites from the
lichenometry study, and then we have

one new one that’s our burn
slope, which I’ll show here.

So, on the left side,
this is our burn slope.

You can see the blackened trees.
And we chose it because it

was relatively close to a slope
where we had lichenometric data.

That’s the one
on the right.

And these two have similar aspects,
similar rock masses, and so we felt

they would be a reasonable
choice for comparing them to see,

do we see more rockfall
at one than the other.

Or, do we see – do we have
a similar picture from those two?

So, for the second case study,
it’s in progress.

These are a lot of goals that we
haven’t gotten to completing yet.

But we’re wanting to
look at the comparison

of burned and
unburned slopes.

We’re going back to that quartzite
and the granite slope –

the Slopes 1 and 2 in along
the highway that were –

that gave such similar rockfall
recurrence estimates and see,

does that picture actually make
sense based on drone scanning.

So that’ll help us with our soft
validation of the lichenometric study.

And then we’ll also be able to look at
change detection results from our

source zone versus the talus.
Do we get the same number of

rockfalls if we do change detection
on the talus as we do on the source?

Because, in the lichenometry study,
we’re using the talus as a proxy

for our rockfall activity.

So, like I said, this is a year-long
scanning campaign.

We’re aiming for monthly scans.
It’s been interrupted by the summer

debris flows that you’ve
probably all heard about.

But I’ll be continuing to collect
data at least through next spring.

I’m using a DJI Mavic 2 Pro.
And I had a ground control network

that I set up at each slope for the
first scan, and then I’ll be referencing

every subsequent scan
to that first point cloud.

And I put these figures in to illustrate
the flight automation that we’re using.

So, to try to make our data collection
as consistent as possible, we’re trying

to fly the same flight paths.
So automation helps with that.

These are just – I wanted to show
a couple photos of the debris flows

because they’re quite dramatic.
But we’ve been able to get back

in the canyon. And so,
once we process these scans,

we’ll be looking at, did that rainfall
have any effect on rockfall.

For the sake of time, I’m going to
skip beyond the workflow slide

and just talk a little bit more
about our preliminary results.

So we – so far, I have
24 scans of our four study sites.

So, Slopes 1, 2, and 3, those are
reused from the lichenometry study,

and Slope 4 is our new one.
I’ve got a table to summarize

all the many photographs taken so far,
and there will be many more.

So this is showing a couple of the
change detection calculations.

So the output you get is
a point cloud where every point

has a distance value
associated with it.

It’s the distance between
the two clouds at that point.

So blue is into the page as we’re
looking at it, and red is out of the page.

So what this is illustrating is that,
when you get these results,

you’ve got to look through
and so a little interpretation

to see what do these
results actually represent.

So, in this case, there’s some
vegetation that’s showing up

as a change out of the page.
So that’s not what we’re interested in.

But we do have a rockfall
over here that I’ll zoom in on.

So, if we zoom in, we’re seeing,
all right, here’s our blue marking

that there’s a change
into the slope here.

And so, if we look at the photographs,
what we see is this sort of

crescent-shaped block marked
by the pink arrow in the upper right

is gone in the second image in July.
And a quick volume estimate of that

is about half a cubic meter.
So that’s the one rockfall I’ve seen

so far in the data processing.
Like I said, there’s – I have a lot

of scans that I haven’t
even touched yet.

So there’s plenty more
to do to look at the data.

So, in addition to that rockfall
at Slope 2, when I was setting up

the ground control points,
I saw there was a rockfall that

happened while I was standing there
watching it, which is pretty interesting.

And it was also in that
about half-meter,

maybe a little bit bigger –
half-cubic-meter size range.

We haven’t seen any rockfall
at the other slopes yet.

There’s a lot of data to process,
but so far, we’re not seeing a lot of

rockfall in the data that I’ve looked at.
And this is – even though these change

detection intervals are spanning at least
part of our thawing season – so what

that might imply is that thawing is not –
it might say that, at these slopes,

under their current geotechnical
conditions, thawing is not as

important for initiating rockfall.
Or it might just mean that we’re at

a low background rate, and we’re
not seeing a lot because it’s

just happening at a –
maybe a few-year recurrence interval

at these particular slopes.
And so we might see some more after

those September and October scans
related to rainfall rather than thawing.

So the rockfalls we’ve seen at this
Slope 2 so far are actually fairly

consistent with the results that
we got from the lichenometry.

So, if we go back to that recurrence
interval plot, the half-meter size

would put us at a recurrence interval
around three to four years.

And so, if we’re only seeing one or two
of those in that kind of size range

in a single spring season,
that fits in fairly well with the idea

that those would be on
a couple-year recurrence interval.

If we were seeing, for example,
10 half-cubic-meter rockfalls

in a single spring season,
then we might be more concerned

that these lichenometry results
are not so accurate.

But, so far, what we’re seeing
is somewhat consistent.

But, like I said, there’s
a lot of data left to look at.

So just, on future work, there’s going
to be a lot more scanning to do,

at least through this next spring,
and a lot more data analysis.

And so what I’m hopeful we’ll
see is that we’ll be able to fill out

the picture of rockfall.
We’ll be able to – we’ll be able to

have a really good sense of,
what did the lichenometry tell us

that we believe and we can definitely
support it with other independent

evidence, and where does it fall short?
And be able to – be able to sort of

put those thoughts together
as we keep moving forward.

So thank you all for listening.
That was a lot really fast –

a lot to go through,
but I appreciate your attention.

And I’d be happy to hear your
feedback and answer your questions.

Oh, and by the way, I have
a reference slide at the end here,

but any of those literature methods
that you’re interested in, feel free

to reach out to me, and I’d be happy
to provide more papers if you –

if you want to look at anything
like that. So thank you.