Customizable Agent-based modeling framework (OvCWD)

• OvCWD has two component models: OvPOP (Odocoileus virginianus POPulation simulation model) and OvCWDdy (Odocoileus virginianus Chronic Wasting Disease dynamics model).
• The two models link host (white-tailed deer) demography and behavior with disease (CWD) transmission dynamics.
• Insights gained from model explorations can help improve our understanding of the complex CWD system and provide defensible recommendations for deer and disease management.
• Specifically, virtual experiments (or scenarios) can be implemented using OvCWD models to analyze the effects of alternative disease management strategies and provide policymakers with a defensible decision-making context.

Open access, peer-reviewed, certified, models.

Model programs are freely available in the digital repository
CoMSES Net Computational Model Library
(https://www.comses.net/codebases/)

• OvPOP
  

https://doi.org/10.25937/kv07-3e08

• OvCWDdy
  

https://doi.org/10.25937/6qeq-1c13

Also see the ‘From Practice’ article in Ecological Solutions and Evidence
(in collaboration with Michigan DNR management specialist Chad Stewart)
where we have described the new decision-making tool for managers
working to control chronic wasting disease in wild cervid populations.

Belsare, A, Stewart, CM.
Ov CWD: An agent‐based modeling framework for informing
chronic wasting disease management in white-tailed deer population
Ecol Solut Evidence. 2020; 1:e12017.
https://doi.org/10.1002/2688-8319.12017

• Both models are coded in NetLogo, an open source Java-based modeling environment.
• The ODD (Overview, Design concepts, Details) protocol for both the models will be downloaded alongwith the models (see ‘docs’ folder).

Model calibration for a mid-west County.

• The objective of the calibration exercise is to find a permutation of initial deer population parameters that generates a realistic in silico deer population for the region of interest.

• Ten replicates of MIOvPOP using a selected set of initial population parameter values are undertaken for each calibration attempt.

• For each replicate, the model deer population is projected over a period of 15 years.

Model generated population abundance compared to DNR estimates

• Pre-harvest abundance in the 15th year of the model run in each iteration is compared with the Montcalm County deer population estimate derived from field and harvest data (51,800 with an assumed ± 5% standard deviation).

Model-generated population’s age structure and sex ratio

• Age structure and female: male ratio in the 15th year, pre-harvest season model deer population is compared with Michigan DNR estimates for Montcalm County deer population.

Age-sex composition of the model-generated deer population

Age-sex composition is assessed throughout one model iteration, and we see a stable age distribution after 8 years.

Age-sex class wise harvest innthe 15th year of the model run

Age-sex classwise harvest numbers from the 15th year of the model run are compared with Michigan DNR numbers for Montcalm County.

Finite population growth rate of the model deer population

The model-generated deer populaion reaches a stationary growth rate (lambda ~ 1) by the 14th year of the model run.

Baseline Scenario: CWD trajectories assuming current harvest strategy in Montcalm County MI

• The baseline scenario is simulated to generate a CWD outbreak trajectory over a 25 year period in a fully susceptible, naïve Montcalm County deer population.
• CWD introduction is simulated in the first year of the model run via one CWD+ dispersing male yearling.
• Note that the user can specify the number and characteristics (age-sex class and group association) of deer that seed CWD outbreak in the model deer population (slider ‘seed-infection’ and chooser ‘CWD_introduced_by’ on the model interface).
• The annual harvest regime implemented in the baseline scenario reflects the current deer harvest levels in Montcalm County.

Plotting baseline CWD trajectories for better comparisons

• CWD outbreak trajectory in the model-generated deer population. Blue circles represent the mean CWD prevalence (± SE) for iterations where CWD outbreak persisted (from a total of 100 iterations). Notice the emergent pattern here: CWD prevalence remains low, below 1%, for 14 years post-introduction. This temporal pattern and the outbreak size pattern, both are in agreement with field and modeling data from other studies in North America. For instance, see the surveillance data from Colorado mule deer herds (Figure 4, page 34: Scientific opinion on chronic wasting disease (II). EFSA Journal 2018;16(1):5132, 59 pp. https://doi.org/10.2903/j.efsa.2018.5132)

• Similar CWD outbreak patterns have been documented in white-tailed deer populations from Wisconsin, Pennsylvania and West Virginia.

CWD outbreak phases: Context for scenario development

• If the objective is to design efficient, sustainable and locally relevant CWD management strategies, it is particulalry important to have a clear understanding of the phases of a CWD outbreak.
• The introduction event will be impossible to detect in the real world. In fact, based on our current understanding, the probability of a persistent outbreak resulting from an introduction event is relatively low. In other words, a persistent outbreak is preceded by multiple introductions that are abortive, and fail to persist in the deer population.
• If active surveillance fails to detect CWD in a population, it does not mean that CWD is not there. In the early phase of the outbreak (pre-establishment phase), prevalence is low and cases are clustered. Furthermore, harvest-based sampling is not random or probabilistic. We have shown (Belsare et al. 2020 https://pubmed.ncbi.nlm.nih.gov/32189826/) that these issues affect detection probabiliy - or the confidence in detecting CWD if present at low prevalence. Non-detection of CWD does not mean it is not there.
• Active surveillance is able to detect CWD when it is in the endemic phase. Cases will be readily discovered in the region (Arkansas CWD surveillance provides a great example: https://www.agfc.com/en/hunting/big-game/deer/cwd/cwd-arkansas/) Therefore, by the time CWD is detected in a population, it is already established and difficult to eliminate.

APR implemented when CWD outbreak is in the pre-establishment phase (year 13; CWD prevalence < 1%)

The blue trajectory is for the baseline scenario. Post-APR implementation, the mean CWD prevalence for APR13 scenario is significantly different compared to the baseline scenario (2 samples Wilcoxon test).

APR implemented when CWD outbreak is in the endemic phase (year 18; CWD prevalence ~ 3%)

The blue trajectory is for the baseline scenario. Post-APR implementation, the mean CWD prevalence for APR18 scenarios is significantly different compared to the baseline scenario (2 samples Wilcoxon test).

PREAPR: CWD introduced in a population already under APR (0.42 yearling male harvest)

Year 10 onwards, the mean CWD prevalence for PREAPR scenario (red circles) is significantly different compared to the baseline scenario (blue circles) (2 samples Wilcoxon test).

PREAPR: CWD introduced in a population already under APR (0.37 yearling male harvest)

Year 10 onwards, the mean CWD prevalence for PREAPR scenario (red circles) is significantly different compared to the baseline scenario (blue circles) (2 samples Wilcoxon test).

PREAPR: CWD introduced in a population already under APR (0.32 yearling male harvest)

Year 10 onwards, the mean CWD prevalence for PREAPR scenario (red circles) is significantly different compared to the baseline scenario (blue circles) (2 samples Wilcoxon test).

APR13C: APR implemented in the 13th year with compensatory increase in anterless harvest

APR implemented in the 13th year (pre-establishment phase of CWD; prevalence < 1%), along with an increase in antlerless harvest (so that adult female + yearling female harvest increases from the current level of ~3300 to ~3900). Red = APR13; Yellow = APR13 + Compensatory increase in anterless harvest; Blue = Baseline harvest scenario. CWD prevalence for the compensatory anterless harvest scenario is significantly different from the CWD prevalence for the baseline harvest scenario but not significantly different from the CWD prevalence for the APR13 scenario.

    Wilcoxon rank sum test

data:  bltpy and apr13comptpy
W = 1157, p-value = 0.06473
alternative hypothesis: true location shift is not equal to 0

Yearling male harvest rate increased (from 0.47 to 0.52) before CWD is introduced in the population

The blue trajectory is for the baseline scenario. The mean CWD prevalence for the increased yearling male harvest scenario (0.47 to 0.52) scenario is not significantly different compared to the baseline scenario (2 samples Wilcoxon test).

Yearling male harvest rate increased (from 0.47 to 0.57) before CWD is introduced in the population

The blue trajectory is for the baseline scenario. The mean CWD prevalence for the increased yearling male harvest scenario (0.47 to 0.57) scenario is not significantly different compared to the baseline scenario (2 samples Wilcoxon test).

    Wilcoxon rank sum test

data:  bltpy and incymhtpy
W = 618.5, p-value = 0.2224
alternative hypothesis: true location shift is not equal to 0

Yearling male harvest rate increased (from 0.47 to 0.62) before CWD is introduced in the population

The blue trajectory is for the baseline scenario. The mean CWD prevalence for the increased yearling male harvest (0.47 to 0.62) scenario is not significantly different compared to the baseline scenario (2 samples Wilcoxon test).

    Wilcoxon rank sum test

data:  bltpy and incymhtpy
W = 636.5, p-value = 0.2981
alternative hypothesis: true location shift is not equal to 0