Modeling Dispersal and Predator-Prey Interactions of Invasive
Species in ArcGIS
Megan C. Fencil
Final project report for CE 394K: GIS in Water Resource
Management
Table
of Contents
Background
Species information and distributions
Data collection and processing
Background
Invasive species are
organisms which inhabit areas outside of their indigenous distribution(s). A very large number of species in
In the
Species information and distributions
Zebra mussels
The
invasive species under investigation in this project is the zebra mussel Dreissena polymorpha . This species is native to
A zebra mussel (www.sgnis.org)
These maps were created by
the USGS to illustrate the rapid spread of zebra mussels:
Round Gobies
Round gobies Neogobius
melanostomus are native to the same
area as zebra mussels and were also introduced in freighter ballast. This species is troublesome because it has a
wide variety of negative impacts on native fauna. It aggressively eats
the eggs of native fish and outcompetes
other fish for habitat because of its aggressive nature and excellent
sensory system that gives it a feeding advantage at night. This fish spawns over a long time period in
summer and is capable of rapid population growth. Each female can produce up to 5,000 eggs
which have a good chance of surviving because the male guards the nest. Round gobies are very robust and thrive in
small dark places, so ballast tanks are alluring to them.
Despite their numerous unpleasant
qualities, round gobies are the only known predator to target
feeding on zebra mussels. This is
probably a result of their similar native habitats because their ancestors
evolved in the same place. Round gobies
prefer to eat small mussels near the substrate.
The number of mussels that a goby can eat per day is a function of the
body sizes of both, but this will be explained in further detail later in the
predation model. Because zebra mussels
are filter feeders, their body tissues accumulate and
concentrate toxins in the water they have
filtered. When round gobies eat such
high numbers of mussels, they further concentrate the toxin. The effect would be even worse for sport fish
high on the food chain (bass, walleyes, perch, trout, etc) that eat the
gobies.
A
round goby (www.sgnis.org)
The following map was created by the USGS to raise public
awareness of the increasing threat of round goby invasion.
Gain skill building maps in ArcGIS that allow analysis of
interactions between biological and physical/hydrological/environmental
factors
Model a predator-prey interaction between
round gobies and zebra mussels to determine whether
gobies can efficiently decrease zebra mussel populations
Determine whether
transplanted individuals will be successful in a previously uncolonized site,
based on comparison of habitat requirements and observed colonizations
Verify freighter
ballast as the main avenue for invasive species to disperse to new sites
Use data on sightings of invasive species in
well-studied areas to project how the species might be
distributed in unstudied areas
Data collection and processing
1.
I downloaded a shapefile of zebra mussel distribution in the
U.S. from the National Atlas. This map layer displays data from a variety
of sources for zebra mussel sightings between 1988 and 1998, but unfortunately
does not include HUC as an attribute
2.
Zebra
mussels have spread very widely since their introduction to the
3.
There
is no shapefile of round goby distribution
available on the web, perhaps because it is not considered as severe a pest as
the zebra mussel (yet!). The only
available national information is qualitative, which means that there is no
differentiation between sighting that consists of a single individual or a
large number of individuals. I chose to
deal with this obstacle by using hypothesized population sizes (explained in
detail later) rather than abandon the project.
Lack of data for biological applications in GIS is common because
biological analysis was a relative latecomer to the GIS scene. This means that biological data have not
historically been collected in a systematic manner amenable to computerized
spatial analysis. For example, site
descriptions in journals are often qualitative and do not provide clear enough
location information to even allow georeferencing. The situation is improving as biologists
become aware of the benefits of GIS, but data suitable for GIS analysis are
still scarce on the web. Fortunately, a
well-developed model can be created and tested with “dummy” values. This is actually a bit of a misnomer because
the values should definitely be based on educated expectations to the extent
that it is possible. When field data become available, the model can be tested
with it and modified to more accurately reflect the reality and stochasticity of nature.
USGS has published a map of round goby
distribution (shown in the previous section), but it is only available as a
picture file. To deal with this, I
created a map of round goby distributions based on the subregions of reported
sightings that were provided by HUC in text format. This information was downloaded from the Sea Grant Nonindigenous Species Site (SGNSS). The data did not specify where in the
subregion the gobies were found, so I decided to select the entire affected
subregion rather than arbitrarily choose a point to represent the sighting
location. This map looks different than
the map for zebra mussel distribution because there was no x,y data available for gobies, so I selected the
entire subregion and the presence of gobies therefore appears as an entire
selected subregion rather than a single point as for zebra mussels. The model is robust to this assumption of
generalized location because if a fish has been sighted in an area the size of
a subregion, there is a good chance that it has already spread throughout the
subregion by the time the data are published.
To create the map, I made a list of the affected subregions in MS
Excel. I then imported it as a table
into ArcGIS and selected from the attribute table
those subregions which contained gobies.
The map itself was much less visually impressive than the map created by
the USGS because it had “blobs” (based on subregions) instead of points for
sightings, but that was the best way to deal with the limitation of not having
the sighting data available in x, y format.
It suited its intended purpose, which was to provide a spatial framework
for the predation model.
4.
The
coverage files that I downloaded were in Arc/Info
interchange file format (.e00
file extension), so I converted to a form which would let me import them into
my basemap. I imported them to my geodatabase using the “Import from
Interchange file” function in ArcToolbox.
5.
After
creating a map of zebra mussel locations and building the basemap, I wanted to
determine possible natural routes of dispersal. Natural dispersal of fish and mussels is
dependent on movement through adjacent bodies of water to new uncolonized
sites. I downloaded the spatial data
file for EPA Reach File
1 from the USGS water resources site. This very large file contains over 62,000
features, so it required a long to download and processing. To incorporate data about lakes and water
bodies other than streams and rivers, I downloaded the appropriate shapefile
from the National Atlas and added it to my basemap. This was used in step 9 when I needed to
search for isolated lakes.
6.
There
is inherent location bias in detection of invasive
species. As I analyzed
distribution data, I soon found that areas of high population tend to have
intensive ecological monitoring programs, usually associated with industry or
government regulatory agencies. Also,
recreational anglers often find and report invasive species while they are
fishing because they have a personal interest in maintaining healthy
populations of native species.
Therefore, the reported data on distributions of invasive species favors
proportional over-reporting of high population and recreational areas while
presence of invasives may go undetected for decades in less populated or
isolated areas. To verify this
suspicion, I downloaded a shapefile of urban areas from the National Atlas. I added this as a layer to my basemap and
turned off the other layers to simplify the view. The features were difficult to see at the
scale of the entire country, so I focused in on southern
7.
After
confirming that reported sightings of zebra mussels are concentrated in urban areas,
I became concerned about the inverse of that problem: How often do invasives colonize a poorly studied or isolated waterbody and remain undetected?
This is a difficult problem, because it essentially requires one to make
predictions about data that do not exist.
For most of the country’s isolated water bodies, there is no certain way
to tell whether zebra mussels are present because no biological surveys have
been done in many such areas. It is
obviously impractical to sample every lake, pond, quarry, and stream for every
invasive species. To determine where
zebra mussels are most likely to exist but be undetected, I used the kriging
tool in ArcGIS’s spatial analyst. This
tool interpolated between the known locations and densities of zebra mussel
sightings to project the most probable distribution in underrepresented
areas. *It is important to note that the map created by kriging is sensitive to over-reporting in
urban and recreational aquatic areas.
I had to accept this assumption and consider it a source of error
because I had no way to determine what percentage of the higher number of
sightings near urban areas was due greater sampling effort, and how much was
due to environmental factors unique to urban areas (increased shipping,
eutrophication due to higher nutrient input, etc.)
8.
The
shapefile of zebra mussel locations that I downloaded
from the National Atlas did not contain a category for HUC in the attribute
table. To discover in which HUC zebra
mussels were found, I downloaded spreadsheet data from the USGS Nonindigenous
Aquatic Species site which allows a query the database for zebra mussel
sightings in each state. However, this data is only available in spreadsheet format, so
I had to import it into Excel. Excel did
not recognize the form of the imported data and I had to make several manual
adjustments to format the data. This was
quite time-consuming, and I quickly realized that I needed to reduce the
spatial scale of the predation model.
After formatting the data, I added it to the attribute table for the
shapefile of zebra mussels that I had previously downloaded from the National
Atlas.
Based on how long it took to format these data in
Excel, I chose to limit the spatial scale of my predation model to the 
9.
I
was originally interested in the habitat requirements that invasive species
need to establish themselves in a new habitat when they are introduced. Aquatic species have requirements for depth,
pH, and temperature, and I expected to be able to predict the success of new
colonizers based on these factors. However,
I quickly discovered that the habitat “requirements” stated for zebra mussels
were in most cases actually habitat preferences. Many of the sites that are currently
colonized with great success by zebra mussels have one or more habitat
parameters that is/are notably different from levels that zebra mussels are
supposedly able to tolerate. Because
there was such disparity between the published habitat requirements of zebra
mussels and their actual distribution in sub-par habitat, there would be little
predictive power in a model that tried to predict successful colonization based
on those factors. It is possible that a
better model could be created by greatly broadening the acceptable range of
habitat conditions, though I do not know enough about zebra mussels’ tolerance
ranges to estimate reasonable boundaries.
10.
The
literature on zebra mussels states that ballast water has been the greatest
contributor to spread of zebra mussels, but I
wanted to test the role of other
factors, particularly how the mussels are transported overland. Invasive aquatic species are frequently
transported over hundreds of miles in the tanks of trailered
recreational boats. This is the most
valid explanation for how invasives arrive in isolated lakes or quarries that
are not closely connected to colonizing sources. To determine possible overland routes for
dispersal, I downloaded a shapefile of the nation’s major roads and overlaid it
with the hydrology layer. Because major
roads often run along lake coastlines, I wanted to avoid confounding the
effects of those 2 variables on dispersal when determining the relationship
between roads and spread of invasives.
This required searching around on the map to find an area containing
zebra mussels in isolated locations, rather than along a shoreline or
river. I found several examples of this,
but the clearest is from rural
As you can see from the
map below (overlay of zebra mussel distribution (green triangles), major roads
(red lines) and stream/rivers blue), it is obvious that ballast water from
freighters traveling along major rivers are the primary avenues for dispersal. If trailered boats
were responsible for a large amount of dispersal, then there should be much
more zebra mussel colonization in the western part of the map where there are
major roads but no major freighter shipping routes.
11.
After
determining that trailered boats are a legitimate but
not very powerful dispersal method, I turned toward validation of the ballast
water dispersal hypothesis. The first
step was to turned on only the zebra mussel
distribution, HUC outline, and hydrolines. This showed the location of zebra mussels in
each HUC, and emphasized that their distribution is highest along the shores of
the lake. After noticing that inland
zebra mussels were concentrated along one linear water
body, I queried that feature and found that it was the Chicago Sanitary and
Ship Canal. It is obvious from the
overlay of these layers that this was the route for zebra mussels to disperse
inland from
.
Predation
model in GIS
One of my goals was to use
ArcGIS to model the outcome of a predator-prey
interaction in which zebra mussels are consumed by round gobies.
Predation
of zebra mussels by round gobies was modeled by the Lotka-Volterra
predation model:
dH/dt = r* H – b1*H*P , where
H =
number of prey
P =
number of predators
r =
rate of growth for prey population
b1
= predation rate (a coefficient expressing the efficiency of predation)
I applied this model in HUC
subregions where round gobies and zebra mussels overlapped. If both species were present in the same HUC
subregion, I assumed that they were interacting in a predator-prey
relationship. I also assumed that the
population size structure of zebra mussels in the 4 Chicago-area HUC regions
was the same as their naturally occurring size structure elsewhere. I mentioned earlier that my model was based
on hypothesized population densities. I
estimated densities of predators (round goby) and prey (zebra mussels) by
averaging densities I encountered in a literature search.
In laboratory experiments,
an adult round goby can consume over 100 zebra mussels per day. However, gobies prefer zebra mussels less
than 4 mm in length. For mussels 4 – 13
mm in length, average consumption by gobies was only 36 to 47 per day (Ghedotti et al.
1995). My estimates for b1 (predation
rate) were relatively conservative based on these findings; gobies probably
feed less efficiently in their stochastic natural habitat than in a lab where
all of their habitat requirements are carefully met. Despite the high predation rates, the model
predicted that goby predation caused little decline (less than 6%) in zebra
mussel population size after 5 years.
This indicates that round gobies alone will be
ineffective at eradicating zebra mussels, a finding which is supported
by ecologists. If the size structure of
the goby population changes, then the size structure of the zebra mussels will
probably change because the body size of gobies is proportional to the size of
zebra mussels it can consume.
After running the model on
the hypothesized population near
Problems and
Solutions
1.
The
HUC coverage map for the entire
2. Published data
for invasive species tends to be qualitative
(presence/absence) rather than quantitative
(number of individuals). Of course it is
sensible to focus on qualitative data initially, so that presence and spread of
invasives can be determined. However,
once an invasive has been documented qualitatively, then a more detailed survey
of its density must be done if any predictive analyses are to be done. Without measurements of field density, models
of dispersal can only be built on hypothesized population sizes that may not be
realistic. In my model based on estimated
predator and prey densities rather than observed data for the entire site, the
effect on prey population size was so small that it would not have conveyed
meaningful information visually. The
small decrease in prey population (6%) was interesting numerically, but such a
small decrease does not translate well to a map. However, if the model had been based on
actual observed numbers, then the map would have been applicable to the real
population, if not exciting.
3. The USGS website offers
plentiful spreadsheet data on spatial distribution of invasive species. It also displays completed maps of the
distributions (I assume their primary purpose is for public awareness), but
these have been converted to picture files so that the
attributes are inaccessible.
Excel has difficulty handling the spreadsheet format containing the
data, and I spent a lot of time manipulating the format to make it
workable. It would be helpful for USGS
to offer several data formats, such as Excel, comma-delimited, etc.
ConclusionS
The
most important concepts illustrated by this project are:
Freighters are
probably the main cause for the spread of invasive species. Many invasive aquatic species are well-suited
to surviving in the ballast tanks of ships, so they can be transported great
distances. Trailered recreational
boats are a lesser source of colonization.
Published data on habitat
requirements of invasive species may be based on the species’ preferences or
range in its native habitat. In reality,
invasives become successful as invaders precisely because they can survive in a
wide range of conditions. This makes it difficult to predict whether transplanted species will
successfully colonize, even with knowledge of the habitat conditions at
the site.
GIS was useful
for developing the predation model
but not for displaying its results because the population
decline due to predation was so small.
It provided useful numeric results, but did not lend itself well to
visual analysis other than clarifying original data entry. Visual representation of data would be more
impressive for an interaction in which the predator did significantly decrease
prey density.
Data
collection efforts for invasive species are heavily skewed
towards representation of urban and recreational areas. To project the presence of the species in unsurveyed areas, kriging in
ArcGIS is quite informative as long as you are able
to deal with the assumption that the resultant distribution is limited by the
fact that the interpolation is based on uneven data collection.
Future
Work
I learned a great deal
about creating predictive ecological models by combining data sources and
spatially analyzing them with GIS. I
would like to georeference some of the distribution
maps that are published on the USGS website, so that the labor-intensive method
of manually converting data to the appropriate format in Excel and re-importing
it to map format in ArcGIS can be avoided. This would allow me to analyze the entire
country because computing power would be greatly improved. I intend to apply this methodology to my
graduate work on larval red drum in the
The USGS is currently developed a set of online analytical
tools called the NBII (National Biological
Information Infrastructure) that will allow biologists to
collaboratively monitor, interpret, model, and manage natural resources. These tools will be a great asset to
conservation biology by furthering the shift of GIS from simple spatial mapping
towards more complex and powerful modeling.
references
The Round Goby Neogobius melanostomus (Pallas): A Review of
European and North American Literature. Patrice M. Charlebois, et al.
The Round Goby (Neogobius melanostomus): Another Unwelcome
Invader in the
Electric Fish Fence Aug. 2002. Earthwatch Radio, University of Wisconsin-Madison
Exotic-nonindigenous Creatures Invading the
Great Lakes Great Lakes Sport Fishing Council
Possible Impact of Gobies and Other Introduced
Species on Habitat Restoration Efforts. David J. Jude
and Scott F. DeBoe. Center for
Round Gobies: Cyberfish of the Third
Millennium. David J. Jude. [
Zebra Mussel Predation by Round Gobies in the
Laboratory. Michael J. Ghedotti,
et al. J. Great Lakes Res. 21(4) pp. 665-669 Internat. Assoc.
Questions or comments?
E-mail me at mcfencil@hotmail.com