Rainfall-Induced Debris Flow Risk Analysis – Yellowstone

National Park
D.P. Hess

Abstract

Yellowstone National Park is an area of intense geological research. Hydrothermal features, evidence of large-scale volcanic
activity, and geomorphologic processes are studied in the relative absence of unnatural influence. Rainfall-induced slope failure
is a common occurrence within the park boundaries, with some events impacting the travel corridors that have been developed
for visitor usage. The application of Geographic Information Systems (GIS), in conjunction with accepted geomorphologic
theory, is an effective way to determine degree of risk from a spatial and temporal frame of reference. This study focuses upon a
model that utilizes dimensional analysis to simplify the understanding of the complex interplay of precipitation intensity and
potential, physical characteristics of the surface under investigation, and local vegetation dynamics as they relate to debris flow
mobilization. GIS is used to illustrate model returns based upon the variation of key parameters. Finally, risk to human life is
qualified by spatial analysis of slope failure risk with respect to park infrastructure.

1. Introduction

Yellowstone National Park is the oldest entity in the of the most remote regions within the lower forty eight
United States National Park System. The park covers states. A large portion of the area is covered in dense,
more than two million acres of wilderness and resides coniferous forest that tends to experience stand-
within the states of Wyoming, Idaho, and Montana replacing burn in a periodic fashion. The fires of 1988
[1]. Yellowstone has been accessible to visitors since devastated more than thirty percent of Yellowstone
the 1890’s through the use of a system of stagecoach with effects still visible today [2].
roads which, eventually, evolved into the park road Active volcanism is the park’s most significant feature
system in place today. Many of these roads pass from the standpoint of geologic research. More than
along steep hillsides, at the base of precipitous cliffs, ten thousand thermal features are scattered about the
and over ancient debris flows and landslides. Federal park and are subtle indicators of the destructive nature
landscape architects limit the use of retaining walls, of the area. Violent eruptions of the supervolcano
rip-rap areas, and other preemptive engineering upon which Yellowstone National Park resides have
solutions in an effort to maintain the beauty and occurred in the past leaving ash and pyroclastic flow
natural appearance of the park. In addition to the deposits throughout the area [3].
immediately surrounding areas, Yellowstone is one
 2

Precipitation varies widely in type and intensity

throughout the area of study. Localized convective
cells migrate through the park on a NE tract during
the summer and have the capability of yielding high
levels of rainfall in a short period of time. At the
other end of the spectrum, many areas of the park are
classified as semi-arid and see very little precipitation
throughout the year [4]. Wide variability in surface
material composition, lithology, vegetation cover,
slope, and precipitation, when coupled with distinct
independence from human interaction, creates an area Fig 1. Debris Flow on East Entrance Rd.
ideal for studying geomorphologic processes. The
impact of development within the park’s boundaries
Table 1
is clearly illustrated by hazardous debris flows (Fig.
Data Source
1).
10m DEM NPS.gov/gis
30m DEM NPS.gov/gis
2. Data
Yellowstone Surface Material NPS.gov/gis
A wide range of data types were required for
Pre-1988 Vegetation Cover NPS.gov/gis
performing a risk analysis study of this magnitude
Post-1988 Vegetation Cover NPS.gov/gis
(Table 1). The Digital Elevation Model (DEM) was
6hr-2yr Return Period NOAA.gov
a critical component and was be used to determine
24hr-2yr Return Period NOAA.gov
slope – one of the key factors in determining debris
Annual Precipitation NPS.gov/gis
flow potential. Two resolutions of DEM, 10m and
Park Infrastructure NPS.gov/gis
30m, were compared and contrasted.
Surficial material data provided spatial distribution of Debris Flow Polygons NPS

various regolith throughout the study area, with the

classes and respective physical characteristics Influence of vegetation on model output was

outlined by Table 2. Density and mean grain size investigated by contrasting scenarios comprised of zero

were acquired from a publication titled Soils of vegetation input to those of the pre-1988 and post-1988

Yellowstone National Park and were not originally cases.

provided with the surface material layer [5]. Precipitation characteristics are necessary for accurate

Vegetation cover is a key parameter in determining modeling of geomorphology within the area of study.

debris flow potential. Cover data for the park prior to Thirty year average precipitation data was used to

and after the fires of 1988 was obtained from the provide an overall understanding of precipitation

National Park Service and compared. A root index potential throughout the park, yet instantaneous data is

was generated for each vegetation class and was required for analyzing rain-induced debris flows.

normalized then joined to the respective layers.

 3

Table 2

Regolith Qualitative Description Mean Grain Size Density

(mm) (kg/m3)
Alluvial Fan Deposit Cobbly to Sandy, Local Loess in Surface Layers 4.12 2090
Stream Alluvium Coarse-Textured Stream Deposits, Stratified Deposits of 3.57 1910
Sand, Gravel, and Cobbles
Fine Textured Alluvium Relatively Fine Deposits of Silt, Sand, and Fine Gravel 0.19 1940
Deposited by Slow-Moving Streams
Colluvium-Talus Angular Gravel, Cobbles, and Boulders; Fragmented Soil 33.71 1635
Material with Few Fine Materials
Colluvium-Talus / Soil Angular Gravel, Cobbles, and Boulders, with Soil 9.41 1720
Material Filling Interstices in Rocks
Colluvium - Soil Mantle of Loose Material that is Primarily Soil, Some .79 1450
Rock Fragments
Glacial Till Unstratified Mixtures of Clay, Silt, Sand, Gravel, and 3.33 1940
Boulders
Flood Deposit Silty to Sandy with Few Coarse Intrusions 0.18 1880
Glaciofluvial Deposit Pebbly to Sandy with Fine-Grain Interstitial Material 2.87 1720
Lake Sediment Silty to Sandy Deposits Mixed with Sandy Particles 0.11 2200
Beach Sediment Sandy – Rounded and Angular Quartz Grains Intermixed 3.91 2010
With Heavies and Cobble to Pebble Sized-Particles.
Loess Silty Clay Sized Material Intermixed With Sand 0.47 1830
Landslide Debris Poorly Sorted Boulder to Sandy Particles with Significant 5.14 1990
Fine-Grain Soil Component
Residuum Fine-Grained Material Weathered in Place From Bedrock 0.97 1610

3. Analysis
Precipitation data capable of illustrating the
possibility of intense rainfall events, in the form of A debris flow is a mass of poorly sorted sediment,
return period maps, was obtained from the National agitated and saturated, surging down a slope in
Oceanic and Atmospheric Administration (NOAA). response to the attraction of gravity. The well-known
Two frequencies were compared to annual rainfall Bingham plastic model is sufficient for studying large-
data and contrasted with each other; two year- six scale homogeneous fields, however debris flows are
hour frequency and two year – twenty four hour very much dependent upon grain to grain interactions
frequency. [6]. Mobilization occurs when the cohesive strength of
Park infrastructure data, such as roads, buildings, an overall granular body is no longer sufficient to
trails, and campsites were used to determine areas of counteract the downslope component of gravitational
risk resulting from mobilization of debris flows. force. Degradation of cohesive force is caused by the
Ground-truth spatial representation of previous debris addition of liquid which acts as a lubricant between
flows were obtained from the National Park Service individual particles within the mass of study. In
and have been used to qualify model returns. addition, increased liquid content increases weight of
the volume of interest.
 4

Grain flow scenarios are one of the leading topics at Yellowstone to illustrate the spatial distribution of
the forefront of physical research. Mathematical debris flow risk.
emulation of such a complex problem is beyond the
scope of this project, however dimensional analysis 4. Processing
allows for a relative comparison while maintaining Please refer to Appendix A for a detailed step-by-step
sufficient accuracy [7]. The primary physical model description of the processing used for this analysis.
relating to a debris flow is the balance between the Determination of slope, in degrees, was the first
downslope component of gravitational force and the processing step. The slope was calculated from each
cohesive strength of the medium [8]. The non- DEM (10m and 30m resolutions) using the raster
dimensional representation of this model follows: calculator and a representative map was generated for
the area of study. Cos2θ, where θ is the slope
[cos 2 θ ][ ρ bulk ][ P]4 calculated in the previous step, was produced as a
ϕ=
[V ][ P][v g ][ ρ ] second map (Fig. 2). Relatively flat areas, as defined
by low slope, are defined by white areas. Steep areas
Downslope Component of Gravitational Force
= are shown as dark sections. Comparing the map shown
Cohesive Strength of Medium
in Fig 2 to the shaded relief map provided by the
where: National Park Service validates the normalized
θ = Slope [dimensionless] representation of steepness as generated by the cos2θ
ρbulk = Bulk Density of Medium [kg/m3] map.
P = Precipitation [m]
V = Vegetation Index [dimensionless]
νg = Mean Grain Volume [m3]
ρ = Mean Grain Density [kg/m3]

The dimensionless number φ yields a relative

understanding of the physical characteristics of a
particular volume of grains based upon the noted
input parameters. It will not predict a quantitative
slip condition. It will, however, describe a propensity
to mobilize or remain stationary based upon the ratio
of gravitational attraction to cohesive strength of the
medium of interest. The material will flow if the
numerator is larger than the denominator, however
will remain stationary if the denominator is larger
Fig 2. Map of Cos2θ Distribution
than the numerator. This relationship was mapped in
relative classes and projected upon the area of
 5

All possible combinations of DEM resolution,

precipitation input, and vegetation presence were
processed and stored as a layer.

5. Results
Fig. 4 shows a number of relevant cases illustrating a
variation of input parameters. Each map describes a
different scenario comprised of varying precipitation
data and vegetation coverage. It should be noted that
changing DEM resolution had little effect on model
results when compared to adjusting these parameters,
therefore DEM comparison maps are not included. All
scenarios shown in Fig 4 utilize the 10 meter DEM. It
is important to note that actual values cannot be
compared from scenario to scenario. Only the relative
Fig 3. 6hr-2yr Return Period Data
comparison of the spatial distribution of debris flow
risk may be compared and contrasted.
The first scenario was based upon annual rainfall data.
Bulk density for each surface material class was
Vegetation cover was not incorporated in this scenario.
extracted from the surface material data and
The annual precipitation data, even normalized,
multiplied by cos2θ, along with the precipitation input
overwhelmed the model, which can be verified by
(Fig. 3), to determine the gravitational force acting
observing the echo of the precipitation layer in the
upon the surface material subject to movement. The
model return. This was expected due to the intense
cell size resulting from the multiplication of the
variation in climate from one zone of the study area to
subject layers was limited to 480 as derived by the
another. The relationship between annual rainfall total
Feature to Raster conversion, therefore the resulting
and rainfall potential for a given area is quite weak due
data is relatively coarse as compared to the original
to convective cell storms that frequent otherwise arid
DEM. In addition, the return period cell size of 386
climates. The annual precipitation data simply
was compromised by the raster multiplication to the
describe areas that are wet or dry within the study area
density layer.
and, therefore, are not ideal for studying debris flow
The vegetation index was multiplied by the
mobilization.
precipitation data and grain properties to develop a
The second scenario utilized the 6 hour – 2 year return
representation of the cohesive bonding of the grains
period data. Vegetation cover was not used. This
contained within the volume of study. Again, the
model return was relatively homogeneous throughout
determining resolution was that of the Feature to
the study area and showed a strong dependence on
Raster conversion.
topography. The model was no longer dominated by
precipitation input. Moreover, flat areas within the
 6

1 2

3 4

5 6

Fig 4. Comparison of Results

 7

study area show little risk of mobilization, which was grain cohesion index. However, the comparison of
not illustrated by the first scenario. 24 hour return scenarios 3 and 4 highlights a problem. Matrix
period data, without vegetation cover, showed a reinforcement should increase with increased
similar response. The third and fourth scenarios vegetation coverage [9]. The model returns, however,
incorporate 6 hour return period data along with pre- show exactly the opposite. Scenario 3, which describes
1988 and post-1988 vegetation cover, respectively. the impact of pre-1988 forest fire vegetation coverage
These results are a perfect example of not being able on model results, should provide a more stable scenario
to compare scenarios on a quantitative basis. when compared to scenario 4 if all other parameters
Although the resulting maps appear to have areas of remain the same. It does not from a qualitative sense.
increased risk for mobilization when compared to This problem is discussed in more detail in a later
scenario one or two, they possess, as a whole, a more section.
reinforced structure as indicated by elevated values of

Fig 5. Overall Debris Flow Mobilization Hazard Map

 8

Scenarios 5 and 6 show the model returns from 24 to the close proximity to the Spatial Analysis Center.
hour – 2 day return period data coupled with pre- In any event, the model return matches up well in this
1988 and post-1988 vegetation coverage. The area. A high concentration of significant debris flow
distribution of risk is similar to that of scenarios 3 risk is shown near the southeast portion of the park.
and 4. The seemingly counterintuitive problem with This area is known as The Thoroughfare and is quite
vegetation cover remains. desolate. It is the place where problem grizzly bears
are relocated to keep them from interacting with park
6. Verification visitors and personnel. The area is also defined as the
Numerous locations within Yellowstone experience most remote wilderness in the lower 48 states. This
rainfall-induced debris flows on a regular basis. area has not been mapped completely by Yellowstone
These areas are, for the most part, located in the field technicians, thereby explaining the absence of
backcountry where they pose very little threat to debris flow polygons [10]. In addition, a number of
human life or developed infrastructure. Fig. 5 shows areas being shown as debris flow areas on the model
an overall hazard map for Yellowstone National Park. return reside upon mountains – more locations that
The model was based on post-1988 vegetation cover, most likely have not been mapped. These areas are
24 hour return period data, and a 10m DEM. In an most notably featured near the northwest portion of the
effort to validate this model, comparison was made to park. In summary, it should be noted that the debris
debris flow polygons obtained from the Yellowstone flow polygons used for model validation are not
National Park Spatial Analysis Center. These complete, however the available data corresponds well
polygons, shown in red on the map, represent with model returns.
ground-truth data consisting of various locations of
previous debris flows. 7. Limitations
Threat to park infrastructure, personnel, and visitors The proposed model has a number of limitations. The
is confined to small pockets widely scattered about area of study experiences seismic activity to some
park. The northeast corner of the park is known for degree on a daily basis, however this model is not
hazardous debris flows where road closures are quite capable of predicting seismically triggered
common during the summer thunderstorm season. mobilization events. In addition, runout distance is not
The debris flow polygons correspond well with the calculated. The areas specified on each risk analysis
model return for this area. Those areas not matching map are strictly points of slope failure initiation and not
up to the polygons may not have been mapped or, areas affected by overall movement.
quite possibly, just have not mobilized yet. The areas The model does not know, and therefore, cannot
along the North Entrance Road show a number of incorporate groundwater level in the relevant
high risk areas and red polygons are present in great calculations. It is possible that positive pore pressure
quantity. The Spatial Analysis Center is located 3 can be developed from elevated water table level,
miles south of this area in the village of Mammoth however this model cannot predict such an instance.
Hot Springs. It is a guess of the author that a focused Other models have been developed that incorporate
effort was made to map debris flows in this area due water table level, however it is the opinion of the
 9

author that groundwater typically follows the path of A similar problem arises when utilizing the available
least resistance in any medium. Fractures and joints vegetation data. Each area of lodgepole pine is not
are commonly the favored path, therefore such completely covered by that species, therefore the
models can be highly inaccurate. Rapid snowmelt homogeneity of each vegetation cluster is problematic.
can initiate a debris flow, however the mechanics This is a possible cause of the seemingly erroneous
behind such an event are extremely complex and not results observed in the pre-1988 and post-1988 model
fully understood. Therefore, snowmelt initiation is returns. Old-growth forests, as represented by a
not included in this study. Furthermore, debris flow significant portion of the pre-1988 vegetation layer,
initiation commonly occurs at depth and not tend to eliminate growth of spreading, young plants by
necessarily at the surface under saturated conditions. providing a thick canopy that blocks the majority of
The mechanism necessary for prediction of such an sunlight. Younger forests have a wide variation in
event is not incorporated. The focus of this study is plant type and growth rate as each species is competing
centered upon debris flow mobilization at the surface for necessary sunlight in the multi-generational process
under saturated conditions. of natural selection. It is the opinion of the author that
It is accepted that triggering mechanisms, such as the post-1988 vegetation data, with its wide variety and
wildlife traversing a slope, steam explosions, elevated density of new plant growth, is capable of
hydrothermal venting, or human disturbance are not achieving a higher value of surface material
predictable and have not been considered. reinforcement than that of the pre-1988 data.
A number of problems are related to the data used for Significant investigation of the source data would be
the project. The surface material data is sufficient for required to validate such a claim. The data was
calculating the necessary parameters for this study, deemed acceptable for this study.
however there is significant room for improvement. The addition of a statistical validation scheme would
Per conversation with a soil scientist at Yellowstone, allow for an enhanced understanding of scenario
the values of mean grain density and mean grain comparison. This is an improvement that may be
diameter used in this experiment are based upon possible in the future once a greater amount of previous
laboratory study performed upon random samples of debris flow activity is mapped by Yellowstone
each respective population. Using a mean grain size personnel.
of 4.12mm for each alluvial fan deposit implies a
homogeneous field of 4.12mm spheres wherever an
alluvial fan is represented in the landform layer.
Obviously, this is not the case and the grain size for
such an area can range from sand to boulder. This
misrepresentation can have a tremendous impact on
the local hydraulics at the surface for each area,
however it was acceptable for our study of the entire
park. It would not be acceptable for a case requiring
a finer resolution, such as a single hillside.
 10

References

1. http://www.nps.gov/yell/technical/yellfact.htm
2. Meyer, G.A., Wells, S.G., 1995. Fire and alluvial chronology in Yellowstone National Park: Climatic and
intrinsic controls on Holocene geomorphic processes. GSA Bulletin. 107(10)1211-1230.
3. Allen, E.T., Day A.L., 1935. Hot springs of The Yellowstone National Park. Carnegie Institution of
Washington.
4. Meyer, G.A., 2001. Recent large-magnitude floods and their impact on valley floor environments of
northeastern Yellowstone. Geomorphology (40) 271-290.
5. Rodman, A., Shovic, H.F., Thoma, D. 1996. Soils of Yellowstone National Park. Yellowstone Center for
Resources, Yellowstone National Park, Wyoming, YCR-NRSR-96-2.
6. Iverson, R.M., 1997. The physics of debris flows. Reviews of Geophysics. 35(3) 245-296.
7. Allen, P.A., 1997. Earth surface processes. Blackwell Science Ltd.
8. Iverson, R.M., Reid, M.E., LaHusen, R.G., 1997. Debris flow mobilization from landslides. Annual of
Earth Planetary Science. 25:85-138.
9. Diodata, N., 2004. Local models for rainstorm-induced hazard analysis on Mediterranean river-torrential
geomorphological systems. Natural Hazards and Earth Systems Science. 4, 389-397.
10. Personal communication – Rodman, A. November 16, 2004.
 11

Appendix A – Procedure

The following text outlines the procedure used to generate the data and visualizations necessary to perform the
subject analysis. Each step of the project was carried out on the ArcGIS 9 platform unless stated otherwise.

Step 1 - Critical data for Yellowstone National Park was established as the primary assignment for this project.
The following data were acquired from the National Park Service Data Store
(http://science.nature.nps.gov/nrdata/index.cfm): Roads, Buildings, Trails, Campsites, Patrol Cabins,
Lake Bathymetry, Park Boundary, Landforms and Surficial Materials, and 30yr Average Precipitation.
The National Park Service Technical Information Center in Denver, CO was contacted and provided the
following data on CD: 10m DEM, 30m DEM, Geology, Vegetation (Post and Prior 1988), and Shaded
Relief. The Spatial Analysis Center at Yellowstone National Park provided mapped areas of previous
debris flows and a copy of a report titled Soils of Yellowstone National Park.

Step 2 - The 10m and 30m digital elevation models were retrieved from the CD. Import71 from ArcView 3.2 was
used to decompress the e00 files. The mosaic command was used to join the north and south sections of
the 10m DEM as it was provided in a divided format. A slope map was created from the 10m DEM using
the Slope Function under Surface Analysis in the Spatial Analyst tool. The resolution was maintained at
10m. The process was repeated for the 30m DEM, although the primary data set was not provided in a
divided format.

Step 3 - Each DEM covers an area known as the GYA, or Greater Yellowstone Area. The focus of the study is
Yellowstone National Park, which is within and smaller than the GYA, therefore the data was cropped to
suit project requirements.
The boundary polygon was converted to raster format using the Convert Features to Raster utility found in
Spatial Analyst. The cell size was set at 10 to match the raster with the finest resolution. This approach
was acceptable because the layer being developed was only to be used for trimming and would have a
single value of 1 throughout the layer. The following statement was used to convert the boundary
polygon raster into a layer to be used for cropping other layers:

[trimmer] = con([bdryrast] > 0, 1)

Trimmer was used to crop the slope maps to Yellowstone National Park by entering the following
statement in the Raster Calculator:

[slopecropped] = [slopedeg] * [trimmer]

Step 4 - Cosine squared was calculated for each DEM resolution by using the following statement:

[cossq] = Sqr(Cos([slopecropped] div deg))

Step 5 - Return period data was downloaded from NOAA.gov. Two sets of data were gathered to be compared
and contrasted as a scenario: the maximum rainfall expected during a six hour period over a two year
span and the maximum rainfall expected during a twenty four hour period over a two year span. Data for
three states (Montana, Idaho, and Wyoming) were downloaded for each frequency set for a total of six
files. The files were unzipped and converted to .txt format. The text files were converted to raster format
by using the ASCII to Grid utility in ArcToolbox. Care was taken to ensure that the cell size of the
output raster was defined as the maximum of the input resolution. The output raster layers for each
frequency class were joined using the following script:

[return] = mosaic([[mt_2yr24hr],[id_2yr24hr],[wy_2yr24hr])

Two raster files – one for each frequency period and consisting of three states, were cropped to the
Yellowstone Park area by the following script:
 12

[returncropped] = [trimmer] * [asciito_na2_1]

The units of each raster were presented as inches *100000, therefore the conversion was performed by the
following script:

[returninch] = [returncropped] / 100000

Each layer was converted to meters by the following script:

[return24m] = [returninch] * .0254

The process was repeated for the 6hr data. Contours were applied to the resulting layers at an appropriate
interval.

Step 6 - General thirty year average precipitation data was used as a comparison to return period data. The
shapefile was converted to raster format by using the Convert Feature to Raster utility found in Spatial
Analyst. The default cell size of 480 was kept and the resulting layer was cropped to size using the
following script:

[precip_raster] = [precip - precip] * [trimmer]

The layer was contoured and appropriate labels were applied.

Step 7 - Surface material properties were extrapolated from “Soils of Yellowstone National Park” and entered in a
table using Microsoft Access. Gen_Land was defined as the primary key and columns for average grain
size and density were created. The table was joined with the landform table using the join command in
ArcGIS and converted to raster format using the Convert Feature to Raster utility. Individual density and
grain size raster layers were created during this process, each maintaining a cell size of 480.

Step 8 - Each scenario (such as no vegetation cover, annual precipitation, 10m DEM) was calculated using the
Raster Calculator and the appropriate script. The resulting layer for each scheme was saved for
comparison.
Appendix B - Flowchart