Amy Davis

Dr. Amy Davis

orcid

Research Focus & Metrics

Academic Career Length

13Years (2014-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
16 September 2020 in Ecology and Evolution
Verified
Journal Article
Open access
PDF file
MTMichael A. Tabak
Michael A. Tabak
MNMohammad S. Norouzzadeh
Mohammad S. Norouzzadeh
DWDavid W. Wolfson
David W. Wolfson
ENErica J. Newton
Erica J. Newton
RBRaoul K. Boughton
Raoul K. Boughton
Ryan S MillerRyan S Miller
Ryan S Miller
Motion‐activated wildlife cameras (or “camera traps”) are frequently used to remotely and noninvasively observe animals. The vast number of images collected from camera trap projects has prompted some biologists to employ machine learning algorithms to automatically recognize species in these images, or at least filter‐out images that do not contain animals. These approaches are often limited by model transferability, as a model trained to recognize species from one location might not work as well for the same species in different locations. Furthermore, these methods often require advanced computational skills, making them inaccessible to many biologists. We used 3 million camera trap images from 18 studies in 10 states across the United States of America to train two deep neural networks, one that recognizes 58 species, the “species model,” and one that determines if an image is empty or if it contains an animal, the “empty‐animal model.” Our species model and empty‐animal model had accuracies of 96.8% and 97.3%, respectively. Furthermore, the models performed well on some out‐of‐sample datasets, as the species model had 91% accuracy on species from Canada (accuracy range 36%–91% across all out‐of‐sample datasets) and the empty‐animal model achieved an accuracy of 91%–94% on out‐of‐sample datasets from different continents. Our software addresses some of the limitations of using machine learning to classify images from camera traps. By including many species from several locations, our species model is potentially applicable to many camera trap studies in North America. We also found that our empty‐animal model can facilitate removal of images without animals globally. We provide the trained models in an R package (MLWIC2: Machine Learning for Wildlife Image Classification in R), which contains Shiny Applications that allow scientists with minimal programming experience to use trained models and train new models in six neural network architectures with varying depths.

10 June 2016 in Ecological Applications
Verified
Journal Article
Open access
PDF file
ADAmy Davis
Amy Davis
MHMevin B. Hooten
Mevin B. Hooten
Ryan S MillerRyan S Miller
Ryan S Miller
MFMatthew L. Farnsworth
Matthew L. Farnsworth
JLJesse Lewis
Jesse Lewis
KPKim M. Pepin
Kim M. Pepin
Evaluation of the progress of management programs for invasive species is crucial for demonstrating impacts to stakeholders and strategic planning of resource allocation. Estimates of abundance before and after management activities can serve as a useful metric of population management programs. However, many methods of estimating population size are too labor intensive and costly to implement, posing restrictive levels of burden on operational programs. Removal models are a reliable method for estimating abundance before and after management using data from the removal activities exclusively, thus requiring no work in addition to management. We developed a Bayesian hierarchical model to estimate abundance from removal data accounting for varying levels of effort, and used simulations to assess the conditions under which reliable population estimates are obtained. We applied this model to estimate site‐specific abundance of an invasive species, feral swine (Sus scrofa), using removal data from aerial gunning in 59 site/time‐frame combinations (480–19,600 acres) throughout Oklahoma and Texas, USA. Simulations showed that abundance estimates were generally accurate when effective removal rates (removal rate accounting for total effort) were above 0.40. However, when abundances were small (<50) the effective removal rate needed to accurately estimates abundances was considerably higher (0.70). Based on our post‐validation method, 78% of our site/time frame estimates were accurate. To use this modeling framework it is important to have multiple removals (more than three) within a time frame during which demographic changes are minimized (i.e., a closed population; ≤3 months for feral swine). Our results show that the probability of accurately estimating abundance from this model improves with increased sampling effort (8+ flight hours across the 3‐month window is best) and increased removal rate. Based on the inverse relationship between inaccurate abundances and inaccurate removal rates, we suggest auxiliary information that could be collected and included in the model as covariates (e.g., habitat effects, differences between pilots) to improve accuracy of removal rates and hence abundance estimates.

19 June 2019 in Functional Ecology
Verified
Journal Article
Closed access
PDF file
BGBrian D. Gerber
Brian D. Gerber
MHMevin B. Hooten
Mevin B. Hooten
CPChristopher P. Peck
Christopher P. Peck
MRMindy B. Rice
Mindy B. Rice
JGJames H. Gammonley
James H. Gammonley
ADAmy Davis
Amy Davis

25 July 2019 in Journal of Applied Ecology
Verified
Journal Article
Closed access
PDF file
ADAmy Davis
Amy Davis
JKJordona D. Kirby
Jordona D. Kirby
RCRichard B. Chipman
Richard B. Chipman
KNKathleen M. Nelson
Kathleen M. Nelson
TXTatiana Xifara
Tatiana Xifara
KPKim M. Pepin
Kim M. Pepin

18 December 2018 in Wildlife Research
Verified
Journal Article
Closed access
PDF file
ADAmy Davis
Amy Davis
BLBruce Leland
Bruce Leland
MBMichael Bodenchuk
Michael Bodenchuk
KVKurt C. VerCauteren
Kurt C. VerCauteren
KPKim M. Pepin
Kim M. Pepin
Context: Management of overabundant or invasive species is a constant challenge because resources for management are always limited and relationships between management costs, population density and damage costs are complex and difficult to predict. Metrics of management success are often based on simple measures, such as counts, which may not be indicative of impacts on damage reduction or cost-effectiveness under different management plans. Aims: The aims of this study were to evaluate the effectiveness of aerial gunning for the management of wild pigs (Sus scrofa), and to evaluate how cost-effectiveness would vary under different relationships between levels of damage and densities of wild pigs. Methods: Repeated reduction events were conducted by aerial gunning on three consecutive days at three study sites. Using a removal model, the proportion of the population removed by each flight was estimated and population modelling was used to show the time it would take for a population to recover. Three possible damage–density relationships were then used to show the level of damage reduction (metric of success) from different management intensities and levels of population recovery, and these relationships were expressed in terms of total costs (including both damage and management costs). Key results: Populations were typically reduced by ~31% for the first flight, ~56% after two flights and ~67% after three flights. When the damage relationship suggests high damage even at low densities, the impact of one, two or three flights would represent a reduction in damage of 2%, 19% and 60% respectively after 1 year. Different damage relationships may show considerable damage reduction after only one flight. Removal rates varied by habitat (0.05 per hour in open habitats compared with 0.03 in shrubby habitats) and gunning team (0.03 versus 0.05). Conclusions: Monitoring the efficacy of management provides critical guidance and justification for control activities. The efficacy of different management strategies is dependent on the damage–density relationship and needs further study for effective evaluation of damage reduction efforts. Implications: It is critically important to concurrently monitor density and damage impacts to justify resource needs and facilitate planning to achieve a desired damage reduction goal.

18 March 2019 in Ecosphere
Verified
Journal Article
Open access
PDF file
KPKim M. Pepin
Kim M. Pepin
DWDavid W. Wolfson
David W. Wolfson
RMRyan S. Miller
Ryan S. Miller
MTMichael A. Tabak
Michael A. Tabak
NSNathan P. Snow
Nathan P. Snow
ADAmy Davis
Amy Davis
Success of large‐scale control programs for established invasive species is challenging to evaluate because of spatial variability in expansion rates, management techniques, and the strength of management intensity. For a well‐established invasive species in the spreading phase of invasion, a useful metric of impact is the magnitude by which control slows the rate of spatial spread. The prevention of spatial spreading likely results in substantial benefits in terms of ecosystem or economic damage that is prevented by an expanding invasive species. To understand how local management actions could impact the spatial spread of an established invasive species, we analyzed distribution and management data for feral swine across contiguous United States using occupancy analysis. We quantified changes in the rate of spatial expansion of feral swine and its relationship to local management actions. We found that after 4 yr of enhanced control, invasion probability decreased by 8% on average relative to pre‐program rates. This decrease was as high as 15% on average in states with low‐density populations of feral swine. The amount of decrease in invasion rate was attributed to removal intensity in neighboring counties and depended on the extent of neighboring counties with feral swine (spatial heterogeneity in local invasion pressure). Although we did not find a significant overall increase in the probability of elimination, increased elimination probability tended to occur in regions with low invasion pressure. Accounting for spatial heterogeneity in invasion pressure was important for quantifying management impacts (i.e., the relationship between management intensity and spatial spreading processes) because management impacts changed depending on the strength of invasion pressure from neighboring counties. Predicting reduction in spatial spread of an invasive species is an important first step in valuation of overall damage reduction for invasive species control programs by providing estimates of where a species may be, and thus which natural and agricultural resources would be affected, if the control program had not been operating. For minimizing losses from spatial expansion of an invasive species, our framework can be used for adaptive resource prioritization to areas where spatial expansion and underlying damage potential are concurrently highest.