Snapshot Wisconsin Scientific Publications
The millions of photos collected by Snapshot Wisconsin volunteers become valuable data used by DNR scientists and university collaborators for wildlife research.
When researchers finish a research project and would like to share their findings, they submit a scholarly publication for peer review by the scientific community. Publications are how the scientific world shares its findings and ensures integrity. Peer review is a long process that can take 3-18 months to complete.
Scientific publications using Snapshot data have been published on a variety of topics, from deer behavior to predator-prey interactions. These important research findings have used the power of this large (and growing) dataset to improve our understanding of wildlife populations and behavior. The information from these papers can help inform wildlife management in Wisconsin and help other scientists, researchers, and management officials worldwide further wildlife research and make decisions.
Below is a list of known scientific publications that have used Snapshot Wisconsin data in their work. Current and former DNR staff are indicated in bold. An asterisk indicates graduate and post-doctoral students who have worked with Snapshot Wisconsin.
Mammal responses to global changes in human activity vary by trophic group and landscape (2024)
Burton, A.C., Beirne, C., Gaynor, K.M., Sun, C., Grandados, A., …, Anhalt-Depies, C., …, Stenglein, J.L., …, Whipple, L.S., …, and R. Kays
Key Findings:
- In less developed landscapes, mammals were less active when human activity increased.
- In more developed landscapes, mammals were more active when human activity increased, but shifted to become more nocturnal.
- Large carnivores were the most sensitive to changes in human activity.
Gilbert, N.A.*, Stenglein, J.L., Van Deelen, T.R., Townsend, P.A., and B. Zuckerberg
Key Findings:
- Snapshot Wisconsin data were used over two winters to determine how deer changed their behavior during extreme cold and warm events.
- During extreme cold temperatures, deer became more diurnal and used conifer-dominated landscapes, while on abnormally warm winter days, deer were more nocturnal and used more deciduous landscapes.
- Deer demonstrated stronger responses to abnormal temperatures later in winter.
Clare, J.D.J.*, Zuckerberg, B., Liu, N., Stenglein, J.L., Van Deelen, T.R., Pauli, J.N., and P.A. Townsend
Key Findings:
- An entire year of Snapshot Wisconsin data was used to understand fine-scale spatial and temporal changes in deer and wolf use of the landscape.
- Deer responses to the presence of wolves varied depending on snow, vegetation growth and time of year.
- Deer were more likely to avoid areas recently used by wolves in spring and fall and less likely to avoid these areas in winter.
- Deer spent time in rich foraging areas recently used by wolves in summer, mitigating predation risk by increasing vigilance.
Human Disturbance Compresses the Spatiotemporal Niche (2022)
Gilbert, N.A.*, Stenglein, J.L., Pauli, J.N., and B. Zuckerberg
Key Findings:
- Snapshot Wisconsin trail camera data were used to assess how the time between detections for pairs of species changed on a gradient of increasing human disturbance.
- Pairs averaged fewer days between detections and higher connection (i.e., more co-occurrences between possible species pairs) in high human disturbance landscapes.
- Human-caused landscape changes may alter how species interact, decreasing available space and time and leading to increased species interactions.
Tabak, M.A., Norouzzadeh, M.S., Wolfson, D. W., Newton, E. J., Stenglein, J.L., et al.
Key Findings:
- Machine learning algorithms tend to require advanced skills to utilize and need improvement in order to better recognize species in different locations.
- Three million camera trap images from 18 studies, including Snapshot Wisconsin in 10 U.S. states, were used to train two deep neural networks to create improved models for classifying camera trap images.
- The resulting software (MLWIC2: Machine Learning for Wildlife Image Classification in R) addresses the limitations of using machine learning to classify images and does not require high-level programming skills to be used.
Shahinfar, S., Meek, P. and G. Falzon
Key Findings:
- Deep learning algorithms are useful for classifying a large volume of wildlife camera trap images, but not enough is known about the ideal sample size to train models to classify at certain accuracy levels.
- Using three datasets, including Snapshot Wisconsin, a formula and guidelines were developed from this experiment to help ecologists add deep learning to their toolbox.
Identifying Animal Species in Camera Trap Images Using Deep Learning and Citizen Science (2018)
Willi, M., Pitman, R.T., Cardoso, A.W., Locke, C., Swanson, A., Boyer, A., Veldthuis, M. and L. Fortson
Key Findings:
- 497,204 Snapshot Wisconsin images and corresponding metadata were used in this dataset to evaluate the usefulness of deep learning and human effort in classifying camera trap images.
- Combining a trained model with classifications from citizen scientists decreased human effort by 43% while maintaining overall accuracy.
- Using deep learning networks and citizen scientists will allow faster processing of large volumes of camera trap data.
Estimating eastern wild turkey productivity using trail camera images (2025)
Butkiewicz, H.E.*, Stenglein, J.L., Riddle, J.D., Truckenbrod, S.A., C.D. Pollentier
Key Findings:
- Used Snapshot Wisconsin camera data to calculate turkey productivity metrics in a first-ever attempt to use camera trap data to calculate these metrics.
- Snapshot data provided reasonable estimates for reproductivity metrics across spatial and temporal scales that were otherwise impossible.
- Turkey productivity metrics were higher in the northern and eastern management zones than in the southern and eastern management zones.
Fractional Richness: An index for camera trap networks (2024)
Berman, L.M.*, Schneider, F.D., Pavlick, R.P., Stenglein, J., Bemowski, R., Dean, M., Townsend, P.A.
Key Findings:
- A new index of diversity called fractional richness is presented, which accounts for differential detection and abundance of different species within a wildlife community.
- Authors compared fractional richness to the Shannon diversity index. They found that fractional richness could be modeled more accurately in a wildlife community where detection rates varied widely between component species.
- Predicted spatial patterns of wildlife diversity across Wisconsin for human-sensitive and human-associated species.
Comparison of In‐Person and Remote Camera Lek Surveys for Prairie Grouse (2023)
Stenglein, J.L., Donovan, E.B., Pollentier, C.D., Peltier, T.R., Lee, S.M., McDonnel, A.B., Kardash, L.H., MacFarland, D.M., Hull, S.D.
Key Findings:
- Compared in-person with trail camera lek surveys to count and estimate the maximum number of male prairie grouse.
- In-person surveys performed better for maximum male counts, and survey types performed similarly when accounting for detection probability with N-mixture models.
- Trail camera lek surveys provide continuous monitoring over the lekking season to understand patterns in activity.
Integrating Harvest and Camera Trap Data In Species Distribution Models (2021)
Gilbert, N.A.*, Pease, B.S., Anhalt-Depies, C., Clare, J. D. J.*, Stenglein, J.L., Townsend, P. A., Van Deelen, T.R., Zuckerberg, B.
Key Findings:
- Snapshot Wisconsin trail camera data were combined with harvest records to improve inference about species occurrence patterns related to environmental predictors like canopy cover and impervious surfaces.
- Community-science camera-trap data can complement existing data sources collected at different scales to provide more precise wildlife population monitoring and support wildlife management decisions.
- Combining data sources led to increased precision and understanding of species-environment relationships.
Abundance Estimation of Unmarked Animals Based on Camera-trap Data (2021)
Gilbert, N.A.*, Clare, J.D.J.*, Stenglein, J.L., Zuckerberg, B.
Key Findings:
- Estimating the abundance of unmarked animals has several challenges, and this paper helps practitioners assess the existing methods and choose ones that will best meet their goals.
- Several methods exist to estimate abundance, and after analyzing each, no one method was optimal for camera-trap data under all circumstances.
- When choosing a method, researchers should consider the focal species' life history, acknowledge what types of data collection are possible and consider whether trends in abundance are acceptable.
Generalized Model-based Solutions to False-positive Error in Species Detection/Non-detection Data (2020)
Clare, J.D.J.*, Townsend, P.A., and B. Zuckerberg
Key Findings:
- Unaccounted-for false-positive detections can create a large bias in models using presence and absence data.
- A generalized model-based solution is described as one that accounts for false-positive error and identifies previous solutions as special cases.
Use of Latent Profile Analysis To Characterize Patterns of Participation In Crowdsourcing (2022)
Anhalt-Depies C., Berland M., Rickenbach M.G., Bemowski R., and A.R. Rissman
Key Findings:
- Crowdsourcing applications tend to have a small number of individuals responsible for the majority of contributions, also called participation inequality.
- In the case of Snapshot Wisconsin's Zooniverse webpage, the top 6% of individuals were responsible for over 20% of classifications.
- Using a technique called latent profile analysis, four distinct patterns of participation were found among crowdsourcers; better understanding these patterns can assist with recruiting and retaining volunteers.
Managing a Large Citizen Science Project to Monitor Wildlife (2019)
Locke, C.M., Anhalt-Depies, C.M., Frett, S., Stenglein, J.L., Cameron, S., Malleshappa, V., Peltier, T., Zuckerberg, B., and P.A. Townsend
Key Findings:
- The Snapshot Wisconsin program is used as a case study to help program managers prepare to launch and run large-scale, long-term citizen science projects.
- User experience is essential, as people are at the heart of citizen science projects.
- Staff requirements vary among the three project phases: planning, growth and maintenance.
Tradeoffs And Tools for Data Quality, Privacy, Transparency, and Trust in Citizen Science (2019)
Anhalt-Depies C., Stenglein J.L., Zuckerberg B., Townsend P.A. and A.R. Rissman
Key Findings:
- Successful citizen-science programs must find a balance between collecting and sharing data while maintaining ethical standards and respecting privacy.
- Before collecting data, program managers should anticipate potential tradeoffs and address them by developing policies and practices.
- We describe the Snapshot Wisconsin program managers' approach to balancing program needs and provide recommendations for others.
Clare, J.D.J.*, Townsend, P.A., Anhalt-Depies, C., Locke, C., Stenglein, J.L., Frett, S., Martin, K.J., Singh, A., Van Deelen, T.R., and B. Zuckerberg
Key Findings:
- Accuracy of crowd-sourced species classification was 93.4% but varied across species with common species generally identified more accurately.
- Screening models or occupancy models accounting for false-positive errors are efficient solutions.
- Data accuracy requirements depend upon the specific research objectives and should be considered in tandem.
Townsend P., Clare, J.D.J.*, Liu, N., Stenglein, J.L., Anhalt-Depies, C., Van Deelen, T.R., Gilbert, N.A.*, et al
Key Findings:
- Techniques that remotely monitor the environment (e.g., satellites, trail cameras, audio recorders) can provide more refined data that can be difficult or expensive to obtain using traditional methods.
- Snapshot Wisconsin successfully shows how natural resource agencies could use trail camera networks to improve wildlife population monitoring and management decisions.
Trail Camera Networks Provide Insights Into Satellite-derived Phenology for Ecological Studies (2021)
Liu, N., Garcia, M., Singh, A., Clare, J.D.J.*, Stenglein, J.L., Zuckerberg, B., Kruger, E.L., and P. Townsend
Key Findings:
- Snapshot Wisconsin time-lapse photos were used to assess the measurement of ground-level phenology data and compare it to phenology data collected via remote sensing.
- Digital photography at ground level is a proven approach for tracking plant phenology, and camera traps could provide another avenue for data collection.
- Different forest types have variations in seasonal changes in the understory (collected from trail cameras) compared to the overstory (collected from remote sensing) vegetation.
Rooney, B., Kays, R., Cove, M.V., Jensen, A., Goldstein, B.R., Pate, C., Castiblanco, P., …, Anhalt-Depies, C., …, Stenglein, J.L., …, Whipple, L.S., …, W. McShea
Key Findings:
- Standardized compilation of the first 5 years of Snapshot USA data, including data from Cove et al. 2021, Kays et al. 2022, and Shamon et al. 2024.
- Data were collected from 263 arrays with 6,712 camera sites across all 50 states. 946,768 detections of wildlife and humans were observed across 371,979 trap nights, with 131 distinct mammal species and 225 bird species detected.
- SNAPSHOT USA has inspired new camera monitoring networks internationally, including SNAPSHOT EUROPE, SNAPSHOT JAPAN and SNAPSHOT GLOBAL.
SNAPSHOT USA 2021: A third coordinated national camera trap survey of the United States (2024)
Shamon, H., Maor, R., Cove, M.V., Kays, R., Adley, J., Alexander, P.D., Allen, D.N., …, Stenglein, J.L., …, Whipple, L.S., …, W. McShea
Key Findings:
- Snapshot Wisconsin data were contributed to a national trail camera data set, which consisted of 1,711 camera sites across 109 camera trap arrays.
- 172,507 sequences of animal observations were captured across 71,519 camera trap nights.
- Increased sampling effort is needed to adequately represent multiple ecoregions across the U.S.
Kays, R., Cove, M.V., Diaz, J., Todd, K., Bresnan, C., Snider, M., ..., Whipple, L.S., …, Stenglein, J.L., Anhalt-Depies, C., ..., W. McShea
Key Findings:
- Snapshot Wisconsin data were contributed to a national trail camera dataset for wildlife monitoring.
- Data collected from camera traps in September and October at 1,485 locations in 43 states, recording 117,415 detections of 78 species of wild mammals.
- This dataset will explore what drives changes in the distribution of wildlife species and the impacts of species interactions, including between humans and wildlife.
SNAPSHOT USA 2019: A Coordinated National Camera Trap Survey of the United States (2021)
Cove M., R. Kays, H. Bontrager, C. Bresnan, M. Lasky, …, Whipple, L.S., …, Stenglein, J.L., Anhalt-Depies, C., …, W. McShea
Key Findings:
- Snapshot Wisconsin data were contributed to a national trail camera data set, which consisted of data gathered in September and October across 50 states at 1,509 camera trap sites from 110 camera arrays.
- 166,036 detections of 83 species of mammals and 17 species of birds were processed through the Smithsonian's eMammal camera trap data repository.
- These data will be helpful to further wildlife research on population dynamics, communities, human-wildlife interactions and conservation action plans.