Funding for this work was provided by the Eastern National Utley Grant Fund, Declining Amphibian Pop - PowerPoint PPT Presentation

1 / 4
About This Presentation
Title:

Funding for this work was provided by the Eastern National Utley Grant Fund, Declining Amphibian Pop

Description:

Epizootiology of sixty-four amphibian morbidity and mortality events in the. USA, 1996-2001. ... used in epidemiology to map sites of potential disease ... – PowerPoint PPT presentation

Number of Views:74
Avg rating:3.0/5.0
Slides: 5
Provided by: MKG34
Category:

less

Transcript and Presenter's Notes

Title: Funding for this work was provided by the Eastern National Utley Grant Fund, Declining Amphibian Pop


1
Can landscape features of amphibian breeding
pools predict mortality events?
Megan K. Gahl and Aram J. K. Calhoun
The University of Maine, Department of Plant,
Soil, and Environmental Sciences, Orono, Maine,
USA
Results
Introduction
Conclusions
High catchment position was in all of the 9
highest ranked logistic regression models from
the AIC comparisons (Table 2), demonstrating its
substantial influence on disease occurrence in
ANP. High catchment position and nearest disease
wetland were in the two best models, and both
ranked highest in relative importance comparisons
(Table 1). Contrary to our predictions, the
relationship of disease occurrence to nearest
disease wetland was positive, therefore, a
shorter distance between a site and the nearest
disease site did not necessarily mean a site was
more likely to endure a mortality event. This
negative spatial autocorrelation indicated that
disease events were not clustered over this
monitoring period. To test model predictive
power, we used the logistic regression
coefficients from the highest ranked model and
calculated a logit prediction for each wetland
site used in model development. For the highest
ranked spatial autocorrelation model with high
catchment position and nearest disease wetland as
predictors, the overall reclassification rate was
95 percent (Table 3). Field tests substantiated
model predictive power. We predicted disease
occurrence in 16 wetlands reserved from modeling
(Figure 1b.) and compared field health screening
(2003-2005) results to model predictions. Two
wetlands predicted to be disease sites, because
of high catchment position, both had mortality
events in 2003-2005, and 13 of 14 predicted
nondisease sites maintained healthy populations
(Figure 1c). In model accuracy tests, type II
errors in predicting disease occurrence were all
derived from wetlands affected by pathogens other
than Ranavirus. Therefore, our model was able to
predict Ranavirus susceptibility but not
necessarily predict mortality events caused by
other diseases.
Amphibian mortality events in relatively pristine
and protected settings cause particular concern
because they may be linked to larger scale,
off-site factors that interact with local
landscape features. While we have some
understanding of the landscape features of
amphibian declines caused by chytridiomycosis and
watermold infections, little is known about the
landscape features associated with Ranavirus
epizootics, even though Ranavirus mortality
events have been widely reported in the
literature (Bollinger et al. 1999, Green et al.
2002, Docherty et al. 2003). Predictive models
based on landscape features have been widely used
in epidemiology to map sites of potential disease
outbreak (Hess et al. 2002, Ostfeld et al. 2005).
Because Ranavirus mortality events are widespread
and affect multiple species in different biotic
landscapes, a predictive model based on abiotic
landscape features and landscape position to
identify wetlands at risk is an important first
step for conservation and disease control
efforts. In this study, we determined landscape
features that differentiated wetlands that
experience periodic mortality events from those
wetlands that have maintained healthy amphibian
breeding populations, and provide the first clear
connection between landscape features and
Ranavirus susceptibility.
Table 2. Highest-ranked logistic regression
models for predicting amphibian disease
occurrence in ANP breeding pools, incorporating
spatial autocorrelation (Nearest Disease
Wetland). The relationship of each variable to
disease occurrence is shown in parentheses. For
clarity, only models with evidence ratio shown. For comparison, the global model is shown.
High catchment position is the most important
landscape feature associated with amphibian
larval mortality events caused by Ranavirus in
Acadia NP wetlands. In addition, wetlands that
were farther away from known disease wetlands
were as likely to be affected by disease as those
nearby, suggesting that disease sites are not
clustered in the landscape or within a watershed.
This suggests that within-pond stressors,
resulting from a HCP landscape position, may play
a more important role in disease susceptibility
than the physical transportation of disease by
vectors to nearby sites in ANP.
Conservation Implications
Conservation and disease containment efforts in
Acadia National Park should be directed toward
higher catchment position and headwater wetlands.
We propose that larger scale factors, such as
climate and atmospheric deposition, can interact
with the landscape to create aquatic systems that
leave amphibians more vulnerable to disease,
while recognizing that management of these larger
scale factors is difficult, though important, for
any single park or preserve. Although our
predictive models were based solely on Acadia
National Park wetlands, the amphibian species
present in ANP and mortality events attributed to
Ranavirus are widespread and our model should be
instructive for monitoring and conservation
efforts elsewhere.
HCP refers to a high catchment position
wetlands. Catchment position (categorical
variable High, Mid, Low) was converted into two
dummy variables for logistic regression
analysis. Log-likelihood from the logistic
regression for each model. K is the number of
parameters in a model, including the y-intercept
(constant). AIC difference (?QAICc) is the
difference between the model and the
highest-ranked model (lowest AIC value HCP, NEAR
model). ?QAICc evidence to be included in the set of best
approximating models. Akaike weights (wi)
represent the likelihood of a model or model
probability. ? Evidence ratio is the relative
likelihood of a model as compared to the
highest-ranked model (HCP, NEAR). Evidence ratio
the set of best-approximating models.
Methods
We developed and evaluated models to identify
landscape features key to predicting disease
incidence using logistic regression and model
selection criterion (QAICc). We calculated
landscape attributes (Table 1) with ArcGIS for 60
small amphibian breeding wetlands monitored in
Acadia National Park (ANP), Maine, USA (Plate 1,
Figure 1a) from 1999-2001. Of these sites, 8 were
observed to experience amphibian larval mortality
events during that monitoring period (Figure 1a),
affecting four amphibian species (Plates 2-5). We
tested model accuracy with reclassification of
model building sites and field tests with 16
sites we monitored for disease occurrence in
2003-2005.
Table 1. Relative importance of individual
parameters in model building set.
Table 3. Classification matrix for HCP NEAR
Model (n60).
The importance weight (Swi) of a parameter is
the sum of Akaike weights from all candidate
models which contain that parameter.
Plate 1. View of example model building site on
Gorham Mountain with the Gulf of Maine behind.
Literature Cited
Acknowledgments
Bollinger, T. K., J. Mao, D. M. Schock, R. M.
Brigham, and V. G. Chinchar. 1999. Pathology,
isolation, and preliminary molecular
characterization of a novel iridovirus from
tiger salamanders in Saskatchewan. Journal of
Wildlife Diseases 35413- 429. Docherty, D. E.,
C. U. Meteyer, J. Wang, J. Mao, S. T. Case, and
V. G. Chinchar. 2003. Diagnostic and molecular
evaluation of three iridovirus-associated
salamander mortality events. Journal of Wildlife
Diseases 39556-566. Green, D. E., K. A.
Converse, and A. K. Schrader. 2002. Epizootiology
of sixty-four amphibian morbidity and mortality
events in the USA, 1996-2001. Annals of the New
York Academy of Sciences 969323-339. Hess, G.
R., S. E. Randolph, P. Arneberg, C. Chemini, C.
Furlanello, J. Harwood, M. G. Roberts, and J.
Swinton. 2002. Spatial aspects of disease
dynamics. Pages 197 in P. J. Hudson, A. Rizzoli,
B. T. Grenfell, H. Heesterbeek, and A. P. Dobson,
editors. The Ecology of Wildlife Diseases.
Oxford University Press, Oxford. Ostfeld, R. S.,
G. E. Glass, and F. Keesing. 2005. Spatial
epidemiology an emerging (or re-emerging)
discipline. Trends in Ecology and Evolution
20328-336.
Funding for this work was provided by the Eastern
National Utley Grant Fund, Declining Amphibian
Populations Task Force Seed Grant, Norcross
Wildlife Foundation AV Stout Fund, New England
Society of Wetland Scientists, Maine Association
of Wetland Scientists, UMaine Association of
Graduate Students, and the UMaine Graduate
School. Statistical advice from Bill Halteman.
Disease diagnosis by David E. Green. Historical
disease data was obtained from Mary Beth
Kolozsvary, Jesse M. Cunningham, and Bruce
Connery.
Plate 3. Wood frog (Rana sylvatica) adult.
Plate 5. Green frog (Rana clamitans) adult.
Plate 4. Spring Peeper (Pseudacris crucifer)
adult.
Plate 2. Bullfrog (Rana catesbeiana) adult.
2
Can landscape features of amphibian breeding
pools predict mortality events?
Megan K. Gahl and Aram J. K. Calhoun
The University of Maine, Department of Plant,
Soil, and Environmental Sciences, Orono, Maine,
USA
Introduction
Results
Conclusions
  • Ranavirus mortality events are widespread and
    can affect multiple amphibian species1,2,3, but
    little is known about their associated landscape
    features.
  • Epidemiologists use landscape models to predict
    disease outbreak sites4,5.
  • We investigated landscape features that
    differentiated wetlands that experience periodic
    mortality events from those wetlands that have
    maintained healthy amphibian breeding
    populations.
  • High catchment position is the most important
    landscape feature associated with larval
    mortality events caused by Ranavirus.
  • Our model was able to predict Ranavirus
    susceptibility but not necessarily predict
    mortality events caused by other diseases.
  • Wetlands farther away from known disease
    wetlands were as likely to be affected by disease
    as those nearby, suggesting that disease sites
    are not clustered in the landscape.
  • Within-pond stressors may play a more important
    role in disease susceptibility than the physical
    transportation of disease by vectors to nearby
    sites.

Table 2. Highest-ranked logistic regression
models for predicting amphibian disease
occurrence in ANP breeding pools, incorporating
spatial autocorrelation (Nearest Disease
Wetland). The relationship of each variable to
disease occurrence is shown in parentheses. For
clarity, only models with evidence ratio shown. For comparison, the global model is shown.
  • High catchment position was the most important
    landscape feature to predict disease (Tables 1,
    2).
  • High catchment position and nearest disease
    wetland were in the two best models (Table 1).
  • Sites nearer a disease wetland were no more
    likely to endure a mortality event than those
    farther away.
  • The model demonstrated predictive power. The
    overall reclassification rate was 95 percent
    (Table 3) and predicted disease sites in field
    tests did experience die-off events during
    2003-2005 (Figures 1b, 1c).
  • Type II errors (lack of disease prediction) in
    predicting disease occurrence were all derived
    from wetlands affected by pathogens other than
    Ranavirus.

Methods
HCP refers to a high catchment position
wetlands. Catchment position (categorical
variable High, Mid, Low) was converted into two
dummy variables for logistic regression
analysis. Log-likelihood from the logistic
regression for each model. K is the number of
parameters in a model, including the y-intercept
(constant). AIC difference (?QAICc) is the
difference between the model and the
highest-ranked model (lowest AIC value HCP, NEAR
model). ?QAICc evidence to be included in the set of best
approximating models. Akaike weights (wi)
represent the likelihood of a model or model
probability. ? Evidence ratio is the relative
likelihood of a model as compared to the
highest-ranked model (HCP, NEAR). Evidence ratio
the set of best-approximating models.
Conservation Implications
Table 1. Relative importance of individual
parameters in model building set.
  • Amphibian disease containment efforts in Acadia
    National Park should be directed toward higher
    catchment position and headwater wetlands.
  • The amphibian species present in ANP and
    mortality events attributed to Ranavirus are
    widespread and our model should be instructive
    for monitoring and conservation efforts
    elsewhere.

The importance weight (Swi) of a parameter is
the sum of Akaike weights from all candidate
models which contain that parameter.
Table 3. Classification matrix for HCP NEAR
Model (n60).
Plate 1. View of example model building site on
Gorham Mountain with the Gulf of Maine behind.
Literature Cited
Acknowledgments
1. Bollinger, T. K., J. Mao, D. M. Schock, R. M.
Brigham, and V. G. Chinchar. Pathology,
isolation, and preliminary molecular
characterization of a novel iridovirus from tiger
salamanders in Saskatchewan. Journal of Wildlife
Diseases 35413- 429 (1999). 2. Green, D.
E., K. A. Converse, and A. K. Schrader.
Epizootiology of sixty-four amphibian morbidity
and mortality events in the USA, 1996-2001.
Annals of the New York Academy of Sciences
969323-339 (2002). 3. Docherty, D. E., C. U.
Meteyer, J. Wang, J. Mao, S. T. Case, and V. G.
Chinchar. 2003. Diagnostic and molecular
evaluation of three iridovirus-associated
salamander mortality events. Journal of Wildlife
Diseases 39556-566 (2003). 4. Hess, G. R.
et al. Spatial aspects of disease dynamics. Pages
197-XXX in P. J. Hudson, A. Rizzoli, B. T.
Grenfell, H. Heesterbeek, and A. P. Dobson,
editors. The Ecology of Wildlife Diseases. Oxford
University Press, Oxford (2002). 5. Ostfeld, R.
S., G. E. Glass, and F. Keesing. Spatial
epidemiology an emerging (or re-emerging)
discipline. Trends in Ecology and Evolution
20328-336 (2005).
Funding for this work was provided by the Eastern
National Utley Grant Fund, Declining Amphibian
Populations Task Force Seed Grant, Norcross
Wildlife Foundation AV Stout Fund, New England
Society of Wetland Scientists, Maine Association
of Wetland Scientists, UMaine Association of
Graduate Students, and the UMaine Graduate
School. Statistical advice from B. Halteman.
Disease diagnosis by D. E. Green. Historical
disease data were obtained from M.B. Kolozsvary,
J. M. Cunningham, and B. Connery.
Plate 2. Bullfrog (Rana catesbeiana) adult.
Plate 3. Wood frog (Rana sylvatica) adult.
Plate 4. Spring Peeper (Pseudacris crucifer)
adult.
Plate 5. Green frog (Rana clamitans) adult.
3
Can landscape features of amphibian breeding
pools predict mortality events?
Megan K. Gahl and Aram J. K. Calhoun
The University of Maine, Department of Plant,
Soil, and Environmental Sciences, Orono, Maine,
USA
Results
Conclusions
Introduction
  • High catchment position is the most important
    landscape feature associated with larval
    mortality events caused by Ranavirus.
  • Our model was able to predict Ranavirus
    susceptibility but not necessarily predict
    mortality events caused by other diseases.
  • Wetlands farther away from known disease
    wetlands were as likely to be affected by disease
    as those nearby, suggesting that disease sites
    are not clustered in the landscape.
  • Within-pond stressors may play a more important
    role in disease susceptibility than the physical
    transportation of disease by vectors to nearby
    sites.

We investigated landscape features that
differentiated wetlands that experience periodic
Ranavirus mortality events from those wetlands
that have maintained healthy amphibian breeding
populations.
Methods
Conservation Implications
  • Amphibian disease containment efforts in Acadia
    National Park should be directed toward higher
    catchment position and headwater wetlands.
  • The amphibian species present in ANP and
    mortality events attributed to Ranavirus are
    widespread and our model should be instructive
    for monitoring and conservation efforts
    elsewhere.

Table 2. Highest-ranked logistic regression
models for predicting amphibian disease
occurrence in ANP breeding pools, incorporating
spatial autocorrelation (Nearest Disease
Wetland). The relationship of each variable to
disease occurrence is shown in parentheses. For
clarity, only models with evidence ratio shown. For comparison, the global model is shown.
Table 1. Relative importance of individual
parameters in model building set.
The importance weight (Swi) of a parameter is
the sum of Akaike weights from all candidate
models which contain that parameter.
HCP refers to a high catchment position
wetlands. Catchment position (categorical
variable High, Mid, Low) was converted into two
dummy variables for logistic regression
analysis. Log-likelihood from the logistic
regression for each model. K is the number of
parameters in a model, including the y-intercept
(constant). AIC difference (?QAICc) is the
difference between the model and the
highest-ranked model (lowest AIC value HCP, NEAR
model). ?QAICc evidence to be included in the set of best
approximating models. Akaike weights (wi)
represent the likelihood of a model or model
probability. ? Evidence ratio is the relative
likelihood of a model as compared to the
highest-ranked model (HCP, NEAR). Evidence ratio
the set of best-approximating models.
Plate 1. View of example model building site on
Gorham Mountain with the Gulf of Maine behind.
Table 3. Classification matrix for HCP NEAR
Model (n60).
Acknowledgments
Funding for this work was provided by the Eastern
National Utley Grant Fund, Declining Amphibian
Populations Task Force Seed Grant, Norcross
Wildlife Foundation AV Stout Fund, New England
Society of Wetland Scientists, Maine Association
of Wetland Scientists, UMaine Association of
Graduate Students, and the UMaine Graduate
School. Statistical advice from B. Halteman.
Disease diagnosis by D. E. Green. Historical
disease data were obtained from M.B. Kolozsvary,
J. M. Cunningham, and B. Connery.
Plate 2. Bullfrog (Rana catesbeiana) adult.
Plate 3. Wood frog (Rana sylvatica) adult.
Plate 4. Spring Peeper (Pseudacris crucifer)
adult.
Plate 5. Green frog (Rana clamitans) adult.
4
Word Count for 3 poster versions
Core Text excludes titles, acknowledgments,
literature cited, tables, legends
Write a Comment
User Comments (0)
About PowerShow.com