Informing Recovery Management of the Threatened
Blackside Dace, Chrosomus cumberlandensis, using a
Bayesian-Belief Network Model
Kevin T. McAbee, Nathan P. Nibbelink, Trisha D. Johnson, and Hayden T. Mattingly
Southeastern Naturalist, Volume 12, Special Issue 4 (2013):143–161
Full-text pdf (Accessible only to subscribers.To subscribe click here.)
143
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist Vol. 12, Special Issue 4
Informing Recovery Management of the Threatened
Blackside Dace, Chrosomus cumberlandensis, using a
Bayesian-Belief Network Model
Kevin T. McAbee1,2,*, Nathan P. Nibbelink1, Trisha D. Johnson3,4,
and Hayden T. Mattingly3
Abstract - Integrated modeling frameworks allow resource managers to incorporate
multiple sources of information (both data and expert judgment), acknowledge uncertainty,
and make quantitative predictions about resource outcomes. To demonstrate the
utility of an integrated-modeling approach for recovery planning of imperiled species,
we developed a comprehensive model in the form of a Bayesian-belief network to support
recovery of a federally listed stream fish, Chrosomus cumberlandensis (Blackside
Dace). Our model quantitatively combined expert judgment and data from empirical
studies to create a comprehensive model that is testable, transferable, and easily communicated.
Sensitivity- and scenario-building analyses demonstrated that mining
impacts such as elevated stream conductivity were the most influential variables affecting
predicted local Blackside Dace population persistence. Our results suggest
that mining impacts are a logical focal point for research and recovery actions for the
species, but additional review and revision of the model are recommended. Taken as a
whole, our effort enhances the current and future capacity for informed recovery-management
of Blackside Dace populations.
Introduction
Improving the condition of an imperiled species to a more stable, protected
status (i.e., recovery) fundamentally requires the explicit inclusion of science
into management decision-making (Boersma et al. 2001). As with most natural
resource management decisions, planning for the recovery of imperiled species
must consider four primary sources of difficulty: ecological complexity; multiple
sources of uncertainty; multiple, possibly competing objectives; and differing
stakeholder values (Clemen 1996). These difficulties are minimized when complex
ecological relationships (such as life history, demographics, environmental
inputs, and human disturbances) are incorporated into a comprehensive, integrated-
modeling framework (Clemen 1996, Peterson and Evans 2003). Furthermore,
such an integrated modeling-framework assists recovery planning by explicitly
linking research and monitoring to management (Peterson and Evans 2003).
In addition to the primary sources of difficulty in decision-making, recovery
planning for imperiled species, such as those protected under the US Endangered
1Warnell School of Forestry and Natural Resources, The University of Georgia, D.W.
Brooks Drive, Athens, GA 30602. 2Current Address - US Fish and Wildlife Service, Utah
Ecological Services Field Office, 2369 West Orton Circle, Suite 50, West Valley City, UT
84119. 3Department of Biology, Box 5063, Tennessee Technological University, Cookeville,
TN 38505. 4Current Address - The Nature Conservancy, 2021 21st Avenue, Nashville,
TN 37212. *Corresponding Author - Kevin_McAbee@fws.gov.
Ecology and Conservation of the Threatened Blackside Dace, Chrosomus cumberlandensis
2013 Southeastern Naturalist 12(Special Issue 4):143–161
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist
144
Vol. 12, Special Issue 4
Species Act of 1973 (ESA), presents unique management challenges (Marcot
and Molina 2007). Imperiled species are often highly specialized, narrowly dispersed,
newly classified, or very sensitive, attributes which make them difficult
to detect in the field or to use in experimental studies (Flather and Sieg 2007).
These obstacles are associated with data deficiencies and high uncertainty about
system dynamics (Marcot and Molina 2007). As a result, managers often rely on
expert judgment when initially creating recovery plans. Because these judgments
can become ingrained as fact rather than be treated as valuable yet untested hypotheses,
it is imperative to re-evaluate recovery goals over time. To this end,
we suggest that recovery management for many imperiled species would benefit
from the use of a predictive model that uses an integrated-modeling approach that
incorporates multiple sources of information.
Modeling frameworks such as visual diagrams and probabilistic models allow
managers to incorporate multiple sources of information (e.g., data, empirical
models, expert judgment), acknowledge uncertainty, and make quantitative predictions
of outcomes under various management alternatives (Blomquist et al.
2010, Clemen 1996, Peterson and Evans 2003). Here, we present an integratedmodeling
framework developed specifically to support recovery of the federally
listed stream fish, Chrosomus cumberlandensis (Starnes and Starnes) (Blackside
Dace). Our framework is formatted as a Bayesian-belief network (BBN), a
probabilistic graphical model that describes conditional dependencies between
variables. The BBN includes a description of system components (e.g., biological
and physical variables) thought to most influence recovery of Blackside
Dace through ecological relationships. One important aspect of our work was
the collaborative process used to develop the BBN, a process that mimicked
recovery-team planning commonly used under the ESA (the most common form
of imperiled species management in the United States).
Our study had two primary objectives: (1) to develop a functional model
(BBN) incorporating the best empirical data and expert knowledge available,
and (2) to evaluate the BBN’s utility through sensitivity analysis and scenario
building. An initial BBN of Blackside Dace system dynamics was created to
document our current understanding of system structure. A sensitivity analysis
was then used to assess the relative influence of system components on management
outcomes. Finally, several land-management scenarios were simulated to
predict Blackside Dace population response to various combinations of ecological
stressors.
Methods
Blackside Dace
The Blackside Dace is a small-bodied, cyprinid fish species endemic to tributary
streams in the Upper Cumberland River drainage of southeastern Kentucky
and northeastern Tennessee (Black et al. 2013a [this issue]; Etnier and Starnes
2001; Starnes and Starnes 1978, Starnes and Starnes 1981). The species was federally
listed as threatened in 1987 by the US Fish and Wildlife Service (USFWS)
145
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist Vol. 12, Special Issue 4
because of limited distribution, extirpation of local populations (O’Bara 1985),
and threats to habitat integrity from forestry and mining activities (USFWS
1987). A number of Blackside Dace research studies and other efforts related
to recovery have been completed since the species was listed (McAbee 2008;
see Supplemental Appendix 1, available online at https://www.eaglehill.us/
SENAonline/suppl-files/s12-Sp4-1040f-McAbee-s1, and, for BioOne subscribers,
at http://dx.doi.org/10.1656/S1040f.s1). For example, the Tennessee Wildlife
Resources Agency (TWRA) and other interested parties began developing the
Northern Cumberlands Forest Resources Habitat Conservation Plan (NCFRHCP)
in 2006 to guide activities associated with timber harvest on wildlife management
areas in Tennessee (Blomquist et al. 2010, NCFRHCP 2012). The Blackside
Dace was one of several imperiled species selected for coverage by the NCFRHCP
(Blomquist et al. 2010, NCFRHCP 2012).
To assist in developing required habitat conservation plan (HCP) components
for Blackside Dace, we composed an up-to-date species account from the published
literature and other sources. Next, we developed a survey questionnaire to
administer to Blackside Dace experts, with several questions specifically related
to HCP components (e.g., monitoring). An “expert” was defined as a person
whose knowledge and experience with Blackside Dace clearly exceeded that of
the average fish biologist in Kentucky or Tennessee. Typically, such experience
was acquired through field research or population monitoring of Blackside Dace.
The main purpose of the survey questionnaire was to complement the literature
review with expert knowledge that might not be available in the literature. Fifteen
individuals completed the questionnaire for Blackside Dace. The species account
and survey results are presented in Supplemental Appendix 1.
Modeling approach
Modeling complex interactions and incorporating multiple sources of
information necessitates a logical framework in which to easily communicate
relationships and results (Clemen 1996). We chose to structure the Blackside
Dace model in a specialized influence diagram, a BBN. BBNs are directed, acyclic
graphs used to represent the relationship between variables in a probabilistic
manner (Pollino et al. 2007), and they are increasingly being used for natural
resources problems, such as water planning (Cain 2001), land management (Rieman
et al. 2001, Martin et al. 2005), habitat assessment (McNay et al. 2006,
Smith et al. 2007), population viability (Marcot et al. 2001, Steventon et al.
2006), and decision analysis (Conroy et al. 2008, Marcot et al. 2006a, Peterson et
al. 2013, Pollino et al. 2007). In wildlife management, BBNs are used to describe
the cascade of influences that shape management outcomes of interest (Marcot et
al. 2006b).
In order to incorporate the most comprehensive and current ecological knowledge,
a working group of 11 Blackside Dace experts was assembled to assist in
construction of the BBN. This diverse group of scientists represented federal and
state agencies, research universities, and non-profit organizations, and provided
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist
146
Vol. 12, Special Issue 4
similarly diverse skills, knowledge, and experience. More than half of the working
group members (7 of 11, or 64%) had previously completed the Blackside
Dace survey questionnaire described above. Involving experts offered many
benefits to the model-development process, including incorporating as much
knowledge as possible, tailoring the model to end users, and promoting a sense
of teamwork within the management process.
Defining and refining the elements of the model are important for model transparency,
utility and effectiveness (Clemen 1996). Elements within our Blackside
Dace model were first defined using scientific support and empirical data, with
the goal of including dynamic, explicit science within the plan (Boersma et al.
2001). Where empirical data were not available, elements were defined using the
expert judgment of working group members. Model refinement was performed
throughout the model development process. Our overall modeling approach is
illustrated in Figure 1.
Figure 1. Sequence of actions taken by the authors to develop and refine a Bayesian-belief
network to inform recovery management of Blackside Dace, Chrosomus cumberlandensis.
BBN = Bayesian-belief network.
147
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist Vol. 12, Special Issue 4
Development of initial BBN
We used the software Netica (Norsys 2007) to develop and refine our Blackside
Dace BBN. Following Marcot et al. (2006b), the first step of BBN development is
the creation of an influence diagram representing ecological linkages that affect
the species of interest. To derive these linkages, we compiled the species account
and analyzed survey results (Supplemental Appendix 1), thereby establishing a
scientific foundation for the model. As discussed above, survey questions were
designed primarily to guide development of NCRFHCP components, but survey
responses also provided BBN facilitators with sufficient information to draft an
influence diagram. In December 2007, the draft influence diagram was presented
to the Blackside Dace working group, where it was discussed and modified until
consensus was reached. The updated influence diagram was developed and ratified
through an open dialogue concerning future applications of the model that
included the exchange of knowledge (data, conditions, and trends) and respectful
debate of ecological hypotheses. Therefore, the resulting ecological causal web
was supported by the group as a whole, but may not represent the exact views of
any individual.
To simplify the model, modifications to the diagram were made by removing
variables that had little or no support in published literature, were not present
across the complete range of the species, or had high correlation to a retained
model variable within the same component class (detailed in McAbee 2008).
When finalized, the group structured the influence diagram with four types of
model components representing: (1) human and environmental inputs, (2) key
ecological correlates of (2a) associated upland habitat conditions and (2b) stream
habitat conditions, (3) biological (life-history) effects, and (4) population response
(Fig. 2). The influence diagram contained a set of model components and
the linkages between them, but in order to convert this diagram to a BBN, each
component was partitioned into a set of discrete states (Cain 2001, Marcot et al.
2006b, Pollino et al. 2007). To accomplish this task, each group member was
asked to suggest a method by which to measure each variable (with units), and
to define ecologically meaningful discrete states within the range of measurements.
Members were asked to provide documentation to support their choices.
Facilitators analyzed the responses provided and used the suggestions to create
mutually exclusive states (with units) for each variable. In February 2008, after
slight modifications by the working group, the mutually exclusive states (see
Fig. 3) and the units upon which they were defined were ratified (see Supplemental
Appendix 2, available online at https://www.eaglehill.us/SENAonline/
suppl-files/s12-Sp4-1040f-McAbee-s2, and, for BioOne subscribers, at http://
dx.doi.org/10.1656/S1040f.s2).
Using these unique states for each variable, initial BBN structure was parameterized
with conditional probabilities. The conditional probability is the
probability that a given state would be achieved, given the states of the contributing
variables. For example, the probability that non-native predators are
present is a function of the presence of Castor canadensis Kuhl (Beaver) (Fig. 3),
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist
148
Vol. 12, Special Issue 4
because Beaver create habitat suitable for non-native predators. Group members
were asked to work independently to assign conditional probabilities for all variables
and discrete states. Individual member responses were averaged to parameterize
probabilities for variables within the model. All members then accepted
the conditional probabilities, which resulted in the alpha model (sensu Marcot et
al. 2006b) of the Blackside Dace BBN shown in Figure 3.
Model context and objectives
Temporally, our BBN represents the influence of current conditions on
Blackside Dace populations over a 5-y period. Ecological conditions (including
a population estimate) are measured at year 0, while the population response
is predicted for year 5. The working group decided to measure population response
on a 5-y time step because the lifespan of Blackside Dace is estimated to
be 3–4 y (Etnier and Starnes 2001, USFWS 1988, Supplemental Appendix 1).
By using a time period slightly longer than the lifespan, users can predict local
extirpation if no reproduction or colonization occurs.
The spatial extent of the model was 200-m stream reaches known to harbor
Blackside Dace populations. This extent was chosen to duplicate the population-
Figure 2. Influence diagram demonstrating the cascade of ecological influences on
population response of Blackside Dace, Chrosomus cumberlandensis. The overall structure
demonstrates that human and environmental inputs (level 1) influence ecological
variables (levels 2a and 2b) and life history (level 3) that shape local Blackside Dace
population response (level 4).
149
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist Vol. 12, Special Issue 4
density and habitat-modeling studies conducted by Black et al. (2013a, b [this
issue]). Data for model variables describing fish densities and stream habitat
conditions were considered in the context of 200-m stream reaches. However,
human inputs and associated upland habitat variables were based on conditions
in the entire watershed upstream from these sites. We recognized that only one
200-m reach should be established within a 12-digit hydrologic unit boundary
(HUB) (US Geological Survey 2007).
The fundamental objective of the working group was to maintain Blackside
Dace populations in currently inhabited streams. Therefore, the model outcome
was designed to predict population trends at currently monitored sites. The
population response was represented by four states: local extirpation, population
decline, stable population, and population growth (Fig. 3). Population growth
and decline were defined as a population change >10% in the positive or the negative
direction, respectively. A stable population was defined as one with a change
of less than 10%. Model outcomes were broken into states that represent absolute failure
(extirpation), biological take (population decline), and success (stable and growing
populations). These outcomes may be interpreted differently based on the
state of the initial population, but nonetheless they provided a way to characterize
population response to inputs.
Figure 3. Division of influence diagram variables into mutually exclusive states, constituting
the initial BBN structure for Blackside Dace, Chrosomus cumberlandensis. Gray
boxes indicate model components that are not influenced by other model components.
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist
150
Vol. 12, Special Issue 4
It is important to note that we did not consider specific populations during
model creation and analysis, but rather a general or example population. Therefore,
the responses were not conditional to existing population sizes, but were
simply bounds for expected responses to guide the working group. When working
group members were developing the model, they simply needed to know
how population growth and decline were defined. If the model was to be used for
specific populations, a “current population size” input would li kely be needed.
The means objectives of the project included maintaining quality habitat for the
species. Stream habitat conditions have been the most heavily investigated aspect
of Blackside Dace ecology, and therefore, were the key proximal influences within
the BBN (Fig. 2: level 2b). The habitat modeling study by Black et al. (2013b [this
issue]) demonstrated that Blackside Dace presence/absence during summer is
strongly associated with conductivity, with water temperature also identified as an
important contributing variable. Furthermore, Blackside Dace reproductive activity
is negatively related to substrate embeddedness (Mattingly and Black 2013 [this
issue]). Therefore, conductivity, water temperature, and substrate embeddedness
were central to the model. Each of these variables was also divided into unique
states based on the results of their associated studies (Supplemental Appendix 2).
For example, habitat models show that Blackside Dace are much more likely to inhabit
streams with conductivity below 240 μS (Black et al. 2013b [this issue]). The
working group also chose to include a measure of invertebrate community health
to represent the Blackside Dace forage base, and to act as a surrogate for other water
quality parameters (Barbour et al. 1999).
Land-use inputs, represented by nodes at the top of the network (Fig. 2:
level 1), are those conditions most likely to occur in the region and subsequently
affect the ecology of the Blackside Dace, specifically the three local habitat
variables described above. Information provided in Supplemental Appendix 1
suggests that mining and forestry practices pose the greatest threat to the persistence
of Blackside Dace. Secondary threats throughout the species’ range include
human development (e.g., urbanization, construction), poor livestock practices,
road crossings that restrict fish movement, and sedimentation of streams from unpaved
roads (Supplemental Appendix 1). The presence of Beavers was selected
as a land-use input because Beavers alter stream conditions in ways believed to
degrade Blackside Dace habitat (Compton et al. 2013 [this issue]). Streamflow
was selected as an environmental input because water levels and flow regime are
key aspects of fish habitat (e.g., Freeman and Marcinek 2006).
Ecological model components (Fig. 2: level 2a) included upland habitat variables
linking land-use inputs to stream-habitat conditions and biological effects.
For example, forestry practices can increase substrate embeddedness through
sedimentation inputs, but riparian buffers can mitigate this effect (Barbour et
al. 1999). The watershed completeness variable represented percent forest cover
outside the riparian zone, but within the watershed. Two additional variables,
non-native predators and physical barriers to movement, were included as binary
states of present or absent.
151
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist Vol. 12, Special Issue 4
Human and environmental inputs and associated upland conditions influence
stream habitat conditions (conductivity, stream temperature, substrate embeddedness,
and the invertebrate community, Fig. 2: level 2b). These variables
subsequently influence Blackside Dace individuals at different stages of life
history. Therefore, three model components described the success of juveniles,
adults, and immigrants (Fig. 2: level 3). Finally, the three life-history population
parameters shape the management outcome of interest, Blackside Dace population
response (Fig. 2: level 4).
Sensitivity analysis
As part of model refinement, sensitivity analysis was performed using Netica
for the completed Blackside Dace BBN to identify variables having the greatest
influence on the outcome of interest (Clemen 1996, Rieman et al. 2001). BBN
sensitivity analysis determines the relative influence of an individual variable by
varying it across all possible states, while keeping all other variables constant
(Rieman et al. 2001). Resulting variation in the outcome probability can therefore
be attributed to changes in the variable being tested. Specifically, we calculated
the relative influence of network components on the population response outcome.
Although the outcome contained four states, for ease of communication we
reviewed only the sensitivity of the “local extirpation” state because minimizing
the probability of local extirpation was a fundamental objective of interest. Results
of the sensitivity analysis were presented to the working group before the
model was ratified to assess whether parameterization of model components met
the expectations of experts (Marcot et al. 2006b).
Scenario building
Changing one or more aspects of the model to test “what if ” scenarios is
especially important when the influencing variables in the model contain uncertainty
(Clemen 1996). In our model, uncertainty was high because a substantial
portion of the model was parameterized using expert opinion rather than data
from empirical studies. Using this type of analysis, users may want to refine the
model, carefully consider variables of importance when making decisions, or
design monitoring strategies to reduce key uncertainties. In our scenario analysis,
we focused on creating combinations of inputs that represented real-world management
conditions for the Blackside Dace by setting input variables to specific
states, but allowing intermediate variable states to vary based on probability
tables. Input variables set to known states for scenario building were mining
impacts, livestock practices, forestry practices, development, unpaved roads,
presence of Beavers, road crossings that restrict fish passage, streamflow, and
size of nearby and distant Dace populations.
Four scenarios were developed to explore effects of alternative input conditions
on model outcomes: (1) optimal input states (described below), (2) single
environmental stressors (drought and Beaver presence), (3) land-use change
caused by humans (logging and mining), and (4) multiple stressors. Optimal
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist
152
Vol. 12, Special Issue 4
input states (scenario 1) included all human land-use variables (mining, logging,
livestock, human development, and unpaved roads) set to “none”, and the road
crossing variable set to “best management practices” (for detailed description
of BMPs see Supplemental Appendix 2). Optimal input states also included
environmental inputs being set to represent Beaver absence, strong metapopulation
interaction (large Blackside Dace populations both nearby and distant), and
above-normal stream flow.
We tested the effect of single environmental stressors (scenario 2) by
changing the state of a single input variable to a less-than-optimal state, but
maintaining all other inputs at the optimal state. This test demonstrated how the
associated component influenced Blackside Dace population response when all
other conditions were optimal. For example, changing the stream-flow state to
“below normal” represented drought conditions. We assessed the response of
the model to human land-use components by changing the prior probabilities
of certain model components to known states (scenario 3). For example, when
the forestry practices node was set to “BMPs” and all other inputs were at optimal
values, the model represented a pristine watershed being logged. Finally,
given that optimal conditions are uncommon under field conditions and watersheds
are often exposed to multiple stressors acting together, in scenario 4 we
tested the response of our population outcome by varying the prior probabilities
of environmental (drought and Beaver) and/or anthropogenic (logging and livestock)
model components.
Results
Sensitivity analysis
The probability of local extirpation of Blackside Dace was strongly influenced
by the density of adults and juveniles (Fig. 4), an expected result because,
logically, future population status would be sensitive to current population size.
Among other model parameters, extirpation was most strongly influenced by
conductivity in the stream habitat-condition level, and mining practices in the
human land-use level of the BBN. Varying stream conductivity across the three
possible states caused the probability of local extirpation to vary between 26%
and 74%, illustrating that changes in conductivity greatly altered the predicted
Blackside Dace population response. The direct connection between mining
practices and elevated conductivity was a product of the working group’s parameterization
of the BBN, and likely stemmed from their awareness of the recent
study by Black et al. (2013b [this issue]), and from field experience and professional
judgment.
The likelihood of extirpation was moderately sensitive to a number of other
variables such as invertebrate density, substrate embeddedness, and riparian buffer
quality (Fig. 4). Extirpation probability showed almost no sensitivity to eight
variables, including the size of nearby or distant Blackside Dace populations,
unpaved roads, and road crossings (Fig. 4).
153
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist Vol. 12, Special Issue 4
Scenario-building results
Optimal conditions. Under optimal conditions (defined above), a Blackside
Dace population sampled in a 200-m reach within a 12-digit HUB was predicted
by the BBN to have a 10.3% probability of local extirpation and 9.3% probability
of declining after 5 y. In addition, the population was predicted to have a 45.7%
and 34.7% probability of remaining stable or growing, respectively (Table 1). By
summing the probabilities of the two undesirable states (extirpation and decline),
the model yielded a <20% chance of a population being negatively impacted under
optimal habitat conditions. The group agreed that a 10% probability of local
extirpation in 5 y seemed higher than expected, but chose to not to make changes
to the model parameterization because they believed the model represented the
current state of knowledge.
Single environmental stressors (drought and Beaver). The presence of drought
increased local extirpation (12.6%) and declining population (10.9%) probabilities
Figure 4. Sensitivity of the mean probability of local extirpation state at the population
response node for Blackside Dace, Chrosomus cumberlandensis. The probability of
local extirpation is most influenced by adult density, stream conductivity in the stream
habitat condition level (level 2b) and mining practices in the input level (level 1). The
bars show the change in local extirpation probability when the associated node (y-axis)
is changed across their possible states, while all other nodes remain constant. Human
land-use inputs are in dark gray, and biological inputs are in light gray.
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist
154
Vol. 12, Special Issue 4
only slightly from the optimal scenario, indicating that Blackside Dace populations
should be minimally affected by drought conditions over monthly time
periods when other inputs are optimal (Table 1). Similarly, the presence of Beavers
only slightly increased the probability of local extirpation (13.9%) and population
decline (13.1%) from optimal conditions (Table 1). This prediction was influenced
by the model time extent (5 y). Beavers are thought to have a legacy effect such
that they may not greatly influence habitat structure until years after colonization.
A model representing a longer time step (10–15 y) would likely indicate a stronger
effect of Beaver presence on Blackside Dace extirpation.
Changing land use. The predicted Blackside Dace population response to
a logging operation using BMPs was only slightly more negative (11.5% extirpation
and 10.3% decline probabilities) than optimal conditions (Table 1).
In contrast, mining in the watershed showed a much stronger effect. When the
mining practices node was set to “low”, extirpation had the highest predicted
probability (33.8%) of the four population response states in that scenario
(Table 1). Moreover, with mining being set to “high impact”, the probability of
extirpation (51.6%) was higher than the three other population response states
combined for that scenario (Table 1). These results followed the expectations
of the group (and results of the sensitivity analysis) indicating that of all the
input nodes, mining has the greatest potential to influence Blackside Dace
population response.
Multiple stressors. Combinations of the less-influential nodes from sensitivity
analysis (see Fig. 4) such as drought, Beaver activity, and logging with BMPs,
did not predict less desirable outcomes for Blackside Dace populations compared
to the mining influence discussed above (Table 2). In contrast, if drought
were to occur in a watershed containing low- or high-impact mining, Blackside
Dace populations were predicted to show negative trends with higher probability
than positive trends (Table 3). In fact, combining mining with just one other
Table 1. Predicted population response under optimal conditions and when changing the state
of a single node. Under optimal conditions, Blackside Dace have a relatively low probability of
negative population trends. Unless described, all other input nodes are set to levels optimal for
Blackside Dace. Mining showed the greatest potential for population effects, tripling the probability
of extirpation, even for low-impact mining, whereas other disturbances (e.g., drought, Beaver,
logging) increased the probability of local extirpation only slightly.
Predicted population response
(% probability of being in each state)
Local Population Stable Population
Inputs extirpation decline population growth
Optimal 10.3 9.3 45.7 34.7
Drought (below normal streamflow) 12.6 10.9 44.9 31.6
Beaver present 13.9 13.1 44.4 28.6
Logging with BMPs 11.5 10.3 45.1 33.1
Low-impact coal mining 33.8 14.9 31.7 19.6
High-impact coal mining 51.6 16.2 21.2 11.1
155
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist Vol. 12, Special Issue 4
sub-optimal node state predicts a more than one-third probability that Blackside
Dace will become locally extinct over 5 y in all instances (Table 3). The influence
of mining on the population response was most strikingly demonstrated by the
prediction that even low-impact mining (with all other nodes being set to optimal)
would have a more negative influence on Blackside dace populations than
if mining were absent and all other input nodes were set to their least favorable
states (Table 4).
Table 3. Predicted population response under multiple stressors that include mining. Highly negative
population response trends are predicted when less influential inputs occur in combination
with mining. Unless described, all other input nodes are set to levels optimal for Blackside Dace.
Predicted population response
Inputs
(% probability of being in each state)
Local Population Stable Population
Initial stressor Mining level extirpation decline population growth
Drought Low impact 37.9 14.8 29.6 17.6
High impact 57.1 14.9 18.6 9.4
Logging with BMPs Low impact 34.5 15.1 31.3 19.0
High impact 52.7 16.2 20.6 10.5
Beavers present Low impact 38.6 15.9 29.0 16.4
High impact 56.9 15.8 18.2 9.1
Table 2. Predicted population response under multiple stressors. Highly negative population responses
are not predicted for combinations of less influential inputs. Unless described, all other
input nodes are set to levels optimal for Blackside Dace.
Predicted population response
(% probability of being in each state)
Local Population Stable Population
Stressors extirpation decline population growth
Drought and Beaver presence 16.3 14.4 42.9 26.4
Drought and logging without BMPs 17.8 14.4 41.7 26.1
Logging without BMPs and livestock 16.3 13.7 42.6 27.4
accessing the stream
Table 4. Comparison of mining influence to the cumulative influence of all other input nodes. The
probability of negative population responses are more likely when mining occurs alone than when
all other stressors act cumulatively.
Inputs Predicted population response (% probability of being in each state)
Other input Local Population Stable Population
Mining nodes extirpation decline population growth
None Least optimal 25.3 18.2 36.8 19.7
Low impact Most optimal 33.8 14.9 31.7 19.6
High impact Most optimal 51.6 16.2 21.2 11.1
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist
156
Vol. 12, Special Issue 4
Discussion
The objective of this project was to improve recovery management of
Blackside Dace by creating a comprehensive, integrated model based on current
ecological knowledge of the species. We accomplished the objective by
constructing a BBN and providing a quantitative framework that can describe potential
changes to the system, both human and environmental. Because the model
is quantitative, managers can compare predictions with real-world outcomes and
update the model based on new information. Our model allows users to assess
the expert belief upon which the model is based, refine model structure to better
represent actual outcomes, and update ecological knowledge. The model is also
transferable because it was intentionally comprehensive, incorporating conditions
throughout the species’ range. Managers can therefore apply the model to
any Blackside Dace population across the range of the species, and in a range of
environmental conditions. Most importantly, the model is transparent. By explicitly
describing an influence diagram for Blackside Dace (and par ameterizing the
diagram in a BBN framework) important ecological relationships have been effectively
defined and communicated, thereby facilitating communication among
stakeholders and peer review of the model.
Because our model incorporated ecological complexity, multiple sources of
uncertainty, and differing stakeholder values, further development of this model
through the addition of management decision alternatives would allow it to be
used in a formal structured decision-making (SDM) context (Clemen 1996).
BBNs are well suited for incorporation into a SDM framework because they are
conceptual models that can convert to a decision-support model (Peterson et al.
2013). We believe adding key recovery elements for the species would provide
the necessary components of a SDM framework. Outcomes important to the
ESA recovery planning process (i.e., recovery goals) could serve as the decision
outcome; and action alternatives that management agencies may undertake could
serve as the decision inputs.
Decision-making for species recovery under the ESA presents unique management
challenges (Marcot and Molina 2007). Imperiled species are often highly
specialized, narrowly dispersed, newly classified, or very sensitive, making them
difficult to detect in the field or to use in experimental studies (Flather and Sieg
2007). These obstacles are associated with data deficiencies and high uncertainty
about system dynamics (Marcot and Molina 2007). As a result, managers often
rely initially on expert judgment when creating recovery plans. Because these
judgments can become ingrained as fact rather than be treated as valuable yet untested
hypotheses, it is imperative to re-evaluate recovery goals over time. To this
end, we suggest that recovery management for many listed species would benefit
from an integrated modeling approach like our BBN, ultimately incorporated into
SDM approach appropriate for adaptive resource management decision-making.
Limitations and future work
Before an alpha-level model is used to implement management actions, it is
recommended that the model undergo a formal peer review process to ensure
157
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist Vol. 12, Special Issue 4
credibility and effectiveness (Marcot et al. 2006b). Because peer review and
refinement are important aspects of the modeling process, the model should be
sent to both BBN and Blackside Dace experts in order to receive independent
agreement that the model is mathematically valid and contains appropriate
ecological relationships.
One structural component that is important to assess is the differing hypotheses
that were combined in order to reach model consensus (see Development of
initial BBN section). By striving for consensus, we may have masked structural
uncertainty of competing hypotheses. Assessing the variation or range in elicited
probabilities is a key sensitivity analysis that can test this structural uncertainty.
We acknowledge that structural uncertainty arising from alternative structures
can be important to assess (Regan et al. 2002), but at this stage, the group felt that
consensus was paramount. Therefore, testing of structural uncertainty was tabled
for future iterations.
The strong influence of conductivity on Blackside Dace persistence is supported
by habitat models (Black et al. 2013b [this issue]), but these authors
suggest that conductivity may only be an indicator or surrogate for another detrimental
phenomenon. For example, heavy metals are known to be associated with
acid mine drainage (Gray 1997, Petty and Barker 2004), and have both lethal and
sublethal effects on fish (e.g., Henry et al. 1999), but this relationship has not
been investigated for Blackside Dace. In addition, to our knowledge, a comprehensive
analysis of natural variations in the conductivity of streams in the Upper
Cumberland River drainage has not been completed. Because conductivity is one
of the most influential variables in our model, we advise that increased research
emphasis be placed on developing a mechanistic understanding of the relationship
between conductivity and Blackside Dace persistence.
To more effectively describe how the environmental variables are affecting
Blackside Dace, our model can be refined by improvements to environmental
input states. For example, Beavers are described in the model as being either
“present” or “absent”. However, this does not effectively convey the legacy
effect that Beaver colonization may have on stream habitat structure and Blackside
Dace population response (Compton et al. 2013 [this issue]). Beaver can
alter aquatic habitat by increasing channel width, opening riparian canopy, and
altering flow regimes (Naiman et al. 1988). Alterations by Beaver have temporal
effects on aquatic habitat, such that stream habitat may differ one, five, or fifteen
years after Beaver have colonized the area (Naiman et al. 1988). Rather than
classify Beaver as present or absent, the model might be improved by setting the
states of Beaver activity based on the number of years since Beaver colonized a
stream, with zero indicating Beaver absence. For our BBN, however, the working
group members did not approve any assumptions about the effects of Beaver
beyond simple presence/absence.
Changes to the definition of average streamflow would offer similar improvements.
Rather than considering streamflow at a single point in time, it would be
more ecologically sound to describe the compounding effects of multiple years of
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist
158
Vol. 12, Special Issue 4
drought. Consecutive years of drought have different impacts on fish communities
than seasonal droughts (Lake 2003), and may prompt Blackside Dace to alter
habitat usage as well.
Potential changes to the model highlight the ability for peer review and
criticism of the model to be quickly adopted. Revising the recovery plan for
Blackside Dace in which new goals are quantified by this model is certainly
possible. New recovery goals could be defined based on the probability of
populations becoming locally extirpated. For example, one such goal that
builds on the current recovery plan might read, “at least 3 streams in each of
8 sub-basins must always have <15% probability of local extirpation over the
next 20 years.” In this instance, a new model based on a 20-year time period
would need to be constructed.
Conclusions
In the recovery plan for Blackside Dace (USFWS 1988), as with many recovery
plans, the multiple aspects affecting species recovery are only weakly or
conceptually connected, and uncertainties are not addressed in any formal fashion.
In contrast, the BBN presented here provides a framework to quantitatively
link these aspects together in a flexible model to assist Blackside Dace recovery
management in the present, and it can be updated through time to reflect new
scientific knowledge. Potential conflicts between human land use and Blackside
Dace can now be evaluated with a contemporary understanding of the system.
It is noteworthy that stream conductivity and mining practices emerged from
our analysis as major drivers of Blackside Dace population sensitivity, which
suggests the need for additional research to confirm expert opinion and identify
causal mechanisms. In summary, our modeling framework enhances the capacity
for recovery management today while opening the door to an informed Blackside
Dace management system for the future.
Acknowledgments
Funding was provided by a Challenge Cost Share agreement between the National
Park Service and the University of Georgia, administered through the Piedmont-South
Atlantic Cooperative Ecosystem Studies Unit as Task Agreement J5028000705.
K.T. McAbee was supported by funding from the University of Georgia Graduate School
and the Warnell School of Forestry and Natural Resources. Additional support was
provided by US Fish and Wildlife Service (USFWS) HCP Planning Assistance grants
to TWRA administered through The Nature Conservancy to Tennessee Technological
University (TTU). Completion of the manuscript was facilitated by a 2011-2012 TTU
Faculty Non-Instructional Assignment to HTM. Amy Knox constructed Figure 1. We especially
thank Blackside Dace species experts who completed survey questionnaires and
served on the working group. Special appreciation is extended to Michael Floyd, USFWS
Kentucky Field Office, for coordinating field visits, and to the Big South Fork National
River and Recreation Area for providing field housing. Earlier drafts of the manuscript
were improved by comments from Shannon Albeke, C. Rhett Jackson, James Long,
James Peterson, Angela Romito, Julie Wilson Stahli, and two anonymous reviewers.
159
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist Vol. 12, Special Issue 4
Disclaimer
The findings and conclusions in this article are those of the authors and do not necessarily
represent the views of the US Fish and Wildlife Service.
Literature Cited
Barbour, M.T., J. Gerritsen, B.D. Snyder, and J.B. Stribling. 1999. Rapid bioassessment
protocols for use in streams and wadeable Rivers: Periphyton, benthic macroinvertebrates
and fish. US Environmental Protection Agency, Office of Water, Washington,
DC. 339 pp.
Black, T.R., J.E. Detar, and H.T. Mattingly. 2013a. Population densities of the threatened
Blackside Dace, Chrosomus cumberlandensis, in Kentucky and Tennessee. Southeastern
Naturalist 12(Special Issue 4):6–26.
Black, T.R., B.K. Jones, and H.T. Mattingly. 2013b. Development and validation of
habitat models for the threatened Blackside Dace, Chrosomus cumberlandensis, at
two spatial scales. Southeastern Naturalist 12(Special Issue 4) :27–48.
Blomquist, S.B., T.D. Johnson, D.R. Smith, G.P. Call, B.N. Miller, W.M. Thurman, J.E.
McFadden, M.J. Parkin, and G.S. Boomer. 2010. Structured decision-making and
rapid prototyping to plan a management response to an invasive species. Journal of
Fish and Wildlife Management 1(1):19–32.
Boersma, P.D., P. Kareiva, W.F. Fagan, J.A. Clark, and J.M. Hoekstra. 2001. How good
are endangered species recovery plans? Bioscience 51:643–649.
Cain, J. 2001. Planning improvements in natural resources management: Guidelines for
using Bayesian-belief networks to support the planning and management of developing
programmes in the water sector and beyond. Centre for Ecology and Hydrology,
Wallingford, Oxon, UK. 136 pp.
Clemen, R.T. 1996. Making Hard Decisions: An Introduction into Decision Analysis.
Duxbury Press, Belmont, CA. 664 pp.
Compton, M., M.A. Floyd, and D.E. Stephens. 2013. Changes in fish community
structure and effects on Blackside Dace (Chrosomus cumberlandensis) populations
following beaver colonization in Davis Branch, Cumberland Gap National Historic
Park, Bell County, Kentucky. Southeastern Naturalist 12(Special Issue 4):112–128.
Conroy, M.J., R.J. Barker, P.W. Dillingham, D. Fletcher, A.M. Gormley, and I.M. Westbrooke.
2008. Application of decision theory to conservation management: Recovery
of Hector’s Dolphin. Wildlife Research 35:93–102.
Etnier, D.A., and W.C. Starnes. 2001. The Fishes of Tennessee, 2nd Printing. The University
of Tennessee Press, Knoxville, TN. 689 pp.
Flather, C.H., and C.H. Sieg. 2007. Species rarity: Definition, causes, and classification.
Pp. 40–66, In M.G. Raphael and R. Molina (Eds.). Conservation of Rare or Little-
Known Species. Island Press, Washington, DC. 392 pp.
Freeman, M.C., and P.A. Marcinek. 2006. Fish assemblage responses to water withdrawals
and water supply reservoirs in Piedmont streams. Environmental Management
38:435–450.
Gray, N.F. 1997. Environmental impact and remediation of acid mine drainage: A management
problem. Environmental Geology 30:62–71.
Henry, T.B., E.R. Irwin, J.M. Grizzle, M.L. Wildhaber, and W.G. Brumbaugh. 1999.
Acute toxicity of an acid mine drainage mixing zone to juvenile Bluegill and Largemouth
Bass. Transactions of the American Fisheries Society 128:919–928.
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist
160
Vol. 12, Special Issue 4
Lake, P.S. 2003. Ecological effect of perturbation by drought in flowing waters. Freshwater
Biology 48:1161–1172.
Marcot, B.G., and R. Molina. 2007. Special considerations for the science, conservation,
and management of rare or little-known species. Pp. 93–125, In M.G. Raphael
and R. Molina (Eds.). Conservation of Rare or Little-Known Species. Island Press,
Washington, DC. 392 pp.
Marcot, B., R. Holthausen, M. Raphael, M. Rowland, and M. Wisdom. 2001. Using
Bayesian-belief networks to evaluate fish and wildlife population viability under land
management alternatives from an environmental impact statement. Forest Ecology
and Management 153:29–42.
Marcot, B.G., P.A. Hohenlohe, S. Morey, R. Holmes, R. Molina, M.C. Turley, M.H. Huff,
and J.A. Laurence. 2006a. Characterizing species at risk II: Using Bayesian-belief
networks as decision support tools to determine species conservation categories under
the Northwest Forest Plan. Ecology and Society 1 1(2):12.
Marcot, B.G., J.D. Steventon, G.D. Sutherland, and R.K. McCann. 2006b. Guidelines
for developing and updating Bayesian-belief networks applied to ecological modeling
and conservation. Canadian Journal of Forest Research 36:3063–3 074.
Martin, T.G., P.M. Kuhnert, K. Mengersen, and H.P. Possingham. 2005. The power of
expert opinion in ecological models using Bayesian methods: Impacts of grazing on
birds. Ecological Applications 15:266–280.
Mattingly, H.T, and T.R. Black. 2013. Nest association and reproductive microhabitat of
the threatened Blackside Dace, Chrosomous cumberlandensis. Southeastern Naturalist
12(Special Issue 4):49–63.
McAbee, K.T. 2008. Structured decision making for recovery management of a threatened
species, Phoxinus cumberlandensis (Blackside Dace). M.Sc. Thesis. University
of Georgia. Athens, GA. 93 pp.
McNay, R.S., B.G. Marcot, V. Brumovsky, and R. Ellis. 2006. A Bayesian approach to
evaluating habitat for Woodland Caribou in north-central British Columbia. Canadian
Journal of Forest Research 36:3117–3133.
Naiman, R.J., C.A. Johnston, and J.C. Kelley. 1988. Alteration of North American
streams by Beaver. Bioscience 38:753–762.
Norsys. 2007. Netica. Norsys, Inc., Vancouver, BC.
Northern Cumberlands Forest Resources Habitat Conservation Plan (NCFRHCP). 2012.
Cumberland Habitat Conservation Plans. Available online at http://www.cumberlandhcp.
org. Accessed 17 August 2012.
O’Bara, C.J. 1985. Status survey of the Blackside Dace (Phoxinus cumberlandensis).
Submitted to the US Department of the Interior, US Fish and Wildlife Service, Asheville,
NC.
Peterson, D.P., S.J. Wegner, B.E. Rieman, and D.J. Isaak. 2013. Linking climate change
and fish conservation using spatially explicit decision support tools. Fisheries
38(3):112–127.
Peterson, J.T., and J.W. Evans. 2003. Quantitative decision analysis for sport fisheries
management. Fisheries 28:10–21.
Petty, J.T., and J. Barker. 2004. Water quality variability in tributaries of the Cheat River,
a mined Appalachian watershed. Proceedings of the American Society of Mining and
Reclamation 15:1–21.
Pollino, C.A., A.K. White, and B.T. Hart. 2007. Examination of conflicts and improved
strategies for the management of an endangered Eucalypt species using Bayesian
networks. Ecological Modelling 201:37–59.
161
K.T. McAbee, N.P. Nibbelink, T.D. Johnson, and H.T. Mattingly
2013 Southeastern Naturalist Vol. 12, Special Issue 4
Regan, H.M., M. Colyvan, and M.A. Burgman. 2002. A taxonomy and treatment of uncertainty
for ecology and conservation biology. Ecological Applications 12(2):618–628
Rieman, B., J. Peterson, J. Clayton, P. Howell, R. Thurow, W. Thompson, and D. Lee.
2001. Evaluation of potential effects of federal land management alternatives on
trends of salmonids and their habitats in the interior Columbia River basin. Forest
Ecology and Management 153:43–62.
Smith, C.S., A.L. Howes, B. Price, and C.A. McAlpine. 2007. Using a Bayesian-belief
network to predict suitable habitat of an endangered mammal: The Julia Creek Dunnart
(Sminthopsis douglasi). Biological Conservation 139:333–347.
Starnes, L.B., and W.C. Starnes. 1981. Biology of the Blackside Dace, Phoxinus cumberlandensis.
American Midland Naturalist 106:360–371.
Starnes, W.C., and L.B. Starnes. 1978. A new cyprinid of the genus Phoxinus endemic to
the Upper Cumberland River drainage. Copeia 1978:508–516.
Steventon, J.D., G.D. Sutherland, and P. Arcese. 2006. A population-viability-based risk
assessment of Marbled Murrelet nesting habitat policy in British Columbia. Canadian
Journal of Forest Research 36:3075–3086.
US Fish and Wildlife Service (USFWS). 1987. Determination of threatened species status
for the Blackside Dace. Federal Register:22,580-22,585. USFWS, Asheville, NC.
USFWS. 1988. Blackside Dace recovery plan. Atlanta, GA. 23 pp.
US Geological Survey. 2007. National hydrography dataset (NHD). US Department of
the Interior, US Geological Survey, Washington, DC.
Supplemental Appendix 1 is Blackside Dace, Phoxinus cumberlandensis,
species account and Cumberland Habitat Conservation Plan (HCP) survey
results. (Available online at https://www.eaglehill.us/SENAonline/suppl-files/
s12-Sp4-1040f-McAbee-s1, and, for BioOne subscribers, at http://dx.doi.
org/10.1656/S1040f.s1).
Supplemental Appendix 2 is BBN user’s guide. (Available online at http://
www.eaglehill.us/SENAonline/suppl-files/s12-Sp4-1040f-McAbee-s2, and, for
BioOne subscribers, at http://dx.doi.org/10.1656/S1040f.s2).