Crafting Better Decisions
Creating a link between belief networks and GIS
By Jeff Hicks and Todd Pierce, University of North Carolina,
Asheville’s National Environmental Modeling and Analysis Center
Considerable research has led to an increased
understanding of how human activity influ- Is the bulb blown? Is the lamp plugged in?
ences the landscape and has provided more
True 100 True 100
options for managing forests in an ecologi-
cally sound manner. With advances in GIS False 0 False 0
technology, decision-making techniques, and
environmental protection policies, more effec- 1 1
tive and integrated management approaches
The Comparative Risk Assessment Frame-
work and Tools (CRAFT), one such approach,
has been developed by the U.S. Forest Ser-
vice’s Eastern Forest Environmental Threat Does the light work?
Assessment Center (EFETAC) to improve the
quality of decisions for forest and natural re-
source managers. CRAFT is designed to help False 100
planning teams focus on the most important
issues, organize their analyses, and use the 0+0
right tools and data in a facilitated environ-
ment. CRAFT has four phases:
Figure 1: A simple network that predicts the outcome of a light working based on real-world observations
n Specifying objectives: What’s the
n Designing alternatives: What to do?
n Modeling effects: What could happen?
n Synthesis: What to communicate?
To better model the effects of different al-
ternatives, CRAFT uses belief networks [also
known as Bayesian networks, Bayes networks,
or causal probabilistic networks] and influ-
ence diagrams to model uncertainty about the
world by combining both common sense and
observational evidence based on the theory of
Essentially, a belief network includes a se-
ries of variables that represents real-world at-
tributes and each variable has several states.
For example, a variable could be whether
a lamp shines and its states could be true or
false. An expert on those attributes connects
the variables in a graphic network that shows
how one or more variables cause a change in
another variable (Figure 1).
The primary feature of a belief network is
its ability to “learn” and continually refine the Figure 2: Location map for the study
extent of a relationship between two variables
by using conditional probabilities. Instead of
making educated guesses between two factors,
a user (or in the case of CRAFT, a group of us-
20 ArcUser Fall 2009 www.esri.com
How Much? How Much?
Canopy Cover Distance to
How Much? How Much?
Figure 3: A basic conceptual model showing factors that
Invasives lead to the occurrence of invasive species. The yellow
boxes ask, What is the extent to which each of these factors
contributes to suitable locations for invasive species?
ers) can create a network, make observations cies to wildfires to landslides. Consequently, is incredibly complex, NEMAC simply sought
on those variables, and compile the findings NEMAC decided to write its own tool using to test a method for putting geographic infor-
as cases. It is from these cases that the belief the ArcGIS Desktop application ArcMap and mation into a Bayesian statistical context and
network software determines the conditional incorporating Python scripts and Netica, a returning the results to geographic space. As a
probabilities between two variables. program for working with Baysian belief net- result, the location variables used were based
While the theory underlying Bayesian sta- works from Norsys Software Corp. on a trusted data source, The National Map
tistics is complex, a software package com- As a test case to develop the method, Seamless Server, a data resource provided by
monly used for belief network modeling— NEMAC investigated the risk that an inva- the U.S. Geological Survey that is publicly
Norsys Netica—is approachable, graphic, sive species known as Japanese stilt grass, available and easily accessed.
and intuitive. In addition, outputs are not as or Microstegium vimineum (MIVI), would The process for preparing data in ArcMap,
intimidating as the results generated by many encroach on an area near Hot Springs in the exporting data to Netica, performing analysis
statistical packages. Pisgah National Forest in North Carolina. In- in Netica, and importing the results back into
Belief networks are useful for CRAFT and vasive species data was collected by Equinox ArcMap is summarized in the following five
other risk assessment tools but have not been Environmental (equinoxenvironmental.com), stages.
linked to GIS so variables can be placed in a a consulting and design firm.
spatial context. Although some networks, such Within the study area (see Figure 2), Equi- Stage 1: Location Data Preparation
as ones used to determine a likely disease di- nox Environmental collected GPS survey 1. Obtain elevation, streamline, and canopy
agnosis for a given set of symptoms, do not paths and marked every MIVI occurrence cover data from The National Map.
have an appropriate spatial context, for other as a point feature. The paths were locations 2. Derive aspect from elevation.
networks, such as models used to determine where MIVI was known to be absent and the 3. Create a multiple ring buffer around streams
likely forest health given a set of threats, spa- points were locations where MIVI was known and convert the vector layer to raster.
tial context is critical. This information can an- to be present. With the proposed process, this 4. Prepare location data so all rasters have
swer questions like, What areas of forests are information could then be used to assess the the same projection and resolution and that
most at risk? and Where can mitigation efforts risk of MIVI occurring in the rest of the study each raster cell snaps to the same grid.
be prioritized to leverage limited resources? area that had not been surveyed. Simply put, 5. Reclassify all data to appropriate classes.
Researchers at the University of North the absence of evidence was not evidence of Reclassification was an iterative process.
Carolina at Asheville’s National Environmen- absence. (Initially, aspect data was classified equally
tal Modeling and Analysis Center (NEMAC) First, a conceptual model (Figure 3) was based on the four cardinal directions. How-
looked for current solutions to tie belief net- created in consultation with scientists from ever, an EFETAC scientist pointed out that
work models to a GIS that would support the EFETAC. [EFETAC, established by the U.S. one class for north (270°–90°) and three
use of CRAFT but couldn’t find anything that Forest Service, uses an interdisciplinary ap- equal classes for southeast, south, and
allowed for in-depth risk analysis or had a proach in developing new technology and southwest, respectively, were more appro-
suitably generic process. It was critical that tools that anticipate and respond to threats to priate classifications.)
the process be general enough to apply to eastern forests.] While tracking the factors as- 6. Clip all data to study area boundaries.
any spatial risk assessment from invasive spe- sociated with the location of invasive species
Continued on page 22
www.esri.com ArcUser Fall 2009 21
Crafting Better Decisions
Continued from page 21
Stage 2: Survey Data Preparation
1. Convert vector survey data to raster with
the same snap grid as the location data.
2. Reclassify survey data into three classes:
known MIVI absence, known MIVI pres-
ence, and unknown MIVI presence or ab- Elevation
sence (i.e., areas not surveyed). Lowest 0
3. Clip the MIVI presence/absence raster to Moderate 0 North 0
the study area boundaries. Canopy Cover High 0 Southeast 0
Low 0 Highest 100 South 100
Moderate 100 5 Southwest 0
Stage 3: Data Combination and Export High 0 3 Distance to Stream
1. Use the Combine tool (Spatial Analyst 2
More than 60m 100
Tools > Local > Combine) to create an ag- Btwn 30 & 60m 0
Less than 30m 0
gregate raster. (The Combine tool visits 0
every cell in the study area. For each cell,
it records the value for the presence or ab-
sence raster and the values for all location
rasters.) Presence/Absence Figure 4: Representation of the
2. Use a Python script written by NEMAC Absence 83.3 data inputs overlaid with a screen
to export this dataset as a simple Presence 16.7 capture from Netica. In this
comma-delimited text data table where ev- 1.17 + 0.37
example, the user has selected
ery cell in the study area is a single row hypothetical states for each
and each variable—all location rasters and variable and Netica has updated
the presence/absence raster—has a unique the probabilities for the presence/
Stage 4: Export Data from ArcMap elevation state, had a south aspect, and was
and Import It into Netica more than 60 meters from a stream. Netica
1. Import the comma-delimited text file into moves on to the next row where the conditions
Netica. Netica automatically creates one might have been different. After Netica goes
node for every column. Each node repre- through the entire study area, it calculates the
sents a single variable for all location ras- probability of each state occurring in the pres-
ters and the presence/absence raster. ence/absence node given every possible com-
2. Configure each node so its states corre- bination of states in the location nodes. Net-
300,000 ESRI Customers spond to the classification applied to the ica can assert that for the stated combination
are within your reach. map in ArcMap. Note that the presence/ described previously, there is a 16.7 percent
absence raster has a value for unknown. A chance that MIVI will occur in places with
Advertise today! state should not be configured as unknown those conditions.
because this represents areas that were not In Netica, users can interact with the net-
Maximum Exposure. work to model what-if scenarios. As a user
surveyed. Statistics should be generated
Minimum Investment. solely on areas that were surveyed. By not clicks on different states in each node and sets
creating a state for unknown values in the them to 100 percent certainty, the probabilities
presence/absence node, Netica skips all represented in each other node (based on what
cases that represent areas that were not is known so far) are updated and displayed.
surveyed. These hypothetical situations do not alter the
3. Arrange and connect the nodes to represent probability tables. Rather, they show how oth-
the conceptual model. er variables respond when one or more vari-
4. In Netica, select Incorporate Case File to ables are set to certain states.
go through the entire data file and record
the observations for each row (i.e., every Stage 5: Export Data from Netica and
cell in the study area). Import It Back into ArcMap
5. The case file populates a table in every Netica stores the conditional probability tables
node, including the presence/absence as a Netica network file that shows the proba-
node, and determines probabilities based bility of each state in the response node for ev-
on these tables. ery combination of node state combinations.
For Rates and Media Kit, visit
www.esri.com/arcuser For example, the first row might repre- 1. Parse the network file to a text table using
or e-mail us at firstname.lastname@example.org. sent a location where MIVI was present, had another Python script written by NEMAC.
a moderate canopy cover, was at the highest Each presence/absence variable and each
22 ArcUser Fall 2009 www.esri.com
Figure 5: The first few lines of the
presence/absence node table in the
Netica network file. On the right, each
node feeding into the presence/absence
node has its own column. Each of its
possible states is combined so that each
row represents a unique combination
of the variable states. On the left, the
probabilities are represented for each
state of the presence/absence node.
state of the location variables has its own
column. Each row represents every combi-
nation of the states of the location variables
and the corresponding probabilities for
each state of the presence/absence variable.
This script also adds a new column to the
aggregate raster, created by the Combine
tool, for the MIVI presence probability
2. The script then goes through each cell in
the aggregate raster, matches the combi-
nation of states for each of its constituent
variables to that same combination in the
Netica output, and inserts the correspond-
ing MIVI presence probability value into
the column created in the previous step.
At this point, every cell of the survey raster
has a probability for MIVI presence. When this
field is symbolized and displayed, the result is Figure 6: The output of the Python script in ArcMap: the risk map for MIVI presence
a risk map for MIVI presence. Given the sim-
plicity of the variables investigated, this risk Acknowledgment research on creative ways of integrating geo-
map is probably not the most accurate assess- The authors thank Dr. Danny Lee and graphic data in visualization environments.
ment of where one might find MIVI. However, Dr. Steve Norman at EFETAC and Karin Dr. Todd Pierce has worked in GIS for
this method allowed NEMAC to successfully Lichtenstein, Jim Fox, and Alex Krebs at more than 18 years and has specialized in
take geographic information, use Bayesian NEMAC for their assistance and advice. GIS and Web programming for 12 of those
statistical analysis, and present the results in a years. He holds a B.S.E. degree in electrical
geographic context. About the Authors engineering from Tulane University and a
NEMAC is working with EFETAC to re- Jeff Hicks is a recent graduate of the Universi- doctorate in geography from Oxford Univer-
fine the belief network-GIS link and use it in ty of North Carolina Asheville Environmental sity in the United Kingdom. He is responsible
other studies and upcoming CRAFT projects. Studies program. With a varied background in for linking GIS and databases to the Web at
Most significantly, this process is not limited multimedia and graphic design, he was drawn NEMAC, where he leads the development of
to invasive species risk. NEMAC is investigat- to GIS because it combined his interests in an online multihazard risk tool for mitigation
ing other potential uses for the process to en- technology and the environment. He began planning. He also assists with development of
sure its generality and is also working to sim- his work on the belief network-GIS link as a geographic decision facilitation processes that
plify and automate the process, more tightly student intern for NEMAC and has gone on to use Web applications and data visualization
integrating ArcMap and Netica. become geospatial analyst at NEMAC. Hicks techniques to support public policy decision
For more information, visit the NEMAC currently is a key contributor to a collaborative makers with land-use planning; flood mitiga-
Web site (nemac.org) or contact the authors, effort with the U.S. Forest Service in the pro- tion and response; forest preservation; and
Jeff Hicks at email@example.com or Todd Pierce duction of the Western North Carolina Report other community issues.
at firstname.lastname@example.org. Card on Sustainability. He also assists with
www.esri.com ArcUser Fall 2009 23