Smart Ranching Project
In 2019, the Property and Environment Research Center (PERC) collaborated with the Upper Yellowstone Watershed Group on a multi-year effort to better understand landowners’ attitudes and challenges with wildlife in Paradise Valley. A detailed report was produced and is available here. Recognizing the common need (and often pain) of livestock producers across Montana to co-exist with wildlife, the Upper Yellowstone Watershed Group (UYWG) and MSU-Extension is now participating in a cooperative project to help local ranchers solve problems like livestock depredation, forage/space depredation and disease transmission using next generation AI and IoT technology.
Beginning spring 2021, preliminary field work using these tools will be conducted on eight Montana ranches to quantify direct and indirect losses of livestock and rangeland due to predation. In 2022-2025, strategies will be tested using both new technology (e.g. smart camera traps, collars, and drones) and old technology (guard dogs and range riders) to mitigate losses. During the same period, ranches will use the technology on their own terms (owning the data for themselves) to determine how it can be used to improve their bottom line.
The Internet of Things and Artificial Intelligence are trillion dollar technology markets changing how people run anything from healthcare, homes, stores, factories, warehouses and cars. Often the media refers to such technology as being Smart. Smart Cars, Smart Homes, etc. Even conservation efforts are jumping on the bandwagon. And, farming, has cautiously been moving into the space. But it is rare to see ranching use cases that have successfully deployed IoT and AI to impact their bottom line.
Photography by Wes Overhold
Karakachan Guard Dogs
Some of the scenarios we are currently working on include:
Livestock depredation (direct and indirect losses) by wild predators (e.g. wolves, grizzlies, mountain lions, and coyotes);
Livestock forage/space depredation by wild ungulates (e.g. elk, deer, feral swine, and wild pocket gophers, prairie dogs); and
Disease transmission between livestock and wildlife (e.g. elk, bison, bighorns, deer, feral swine, and mesopredators) such as brucellosis, tuberculosis, CWD, and pneumonia.
Secondary Benefits for Ranching Operations Include:
“Work from Anywhere” video monitoring of critical ranch operations
Location tracking of livestock, including but not limited to virtual fencing use cases
Security monitoring (e.g. poaching, tresspassing)
Introduction and Justification
Montana is home to one of the largest and richest assemblages of wild ungulates (e.g., bison, elk, mule deer, white-tailed deer, moose, pronghorns, bighorn sheep, and mountain goats) and large predators (e.g., grizzly bears, black bears, coyotes, mountain lions, and wolves) in the western hemisphere (1, 2). However, rapidly expanding elk and grizzly bear populations are causing major agricultural losses and substantial economic damage as well as negative ecosystem effects on biodiversity (3, 4, 5, 6). Livestock can come into direct conflict with wildlife through disease transmission (e.g., bovine brucellosis), predation (direct and indirect effects), and forage competition disrupting operations and causing financial losses which place owners of working land in opposition to wildlife and wildland protection. A central goal of both ecosystem conservation and agricultural production in Montana should be minimizing wildlife conflicts on private land. Here we propose to use newly developed technologies in wildlife conservation to solve heretofore intractable problems in wildlife-livestock conflicts.
Private working lands comprise nearly 65% of Montana’s 93 million acres, and play a critical role in securing future energy, water, food, and fiber for an ever-expanding human population while providing a suite of invaluable ecosystem services (wildlife habitat, water quality, soil health, etc.). Economically viable working lands also serve as a stronghold against the threat of subdivision and development that fragments habitat (4, 7, 8, 9). On many Montana ranches, wildlife-cattle conflicts are greatest in springtime when cattle are relatively concentrated on private lands during or after calving season and before turn-out on summer range. The increasing threat of brucellosis transmission from elk to cattle (see map from 9; Fig. 1) and the increasing threat of cattle depredation due to increasing numbers of grizzly bears (10, 6) is spurring increased interest in emerging technologies that may reduce or mitigate wildlife-livestock interactions and conflicts. Notable among these technologies are recent advancements in artificial intelligence (AI) that can be harnessed in modern, remotely deployed monitoring hardware, including trail cameras and unmanned aerial systems (drones) for conflict applications. In particular, these technologies have the potential to give real-time information on wildlife population occurrence and movements, and the spatiotemporal relationships between wildlife and livestock. Real-time information on when and where wildlife is interacting with livestock would shift the focus of wildlife livestock conflicts from compensation to prevention, which has major implications for production agriculture and wildlife conservation.
Thus, it is essential to work hand in hand with Montana landowners to further develop and evaluate if these technologies are workable for Montana’s major wildlife-conflict situations. We propose to evaluate if low maintenance, scalable, game cameras equipped with various sensors that use image recognition (smart cameras) and drones equipped with sensors (visual and thermal) for surveying wildlife abundances and distributions can accurately classify images/thermal video and provide real-time alerts when predators or animals that pose a disease risk are present on working lands.
Goals and Objectives
Our goal is to evaluate the ability of smart cameras and thermal drone surveys to be applied in situations to:
reduce commingling of elk with cattle during spring calving and limit interspecies transmission of Brucella; and
reduce cattle depredation (direct and indirect losses) by grizzly bears.
Specifically, our objectives are:
Test new smart cameras leveraging edge AI and internet of things (IoT) integrated devices on the ground and the latest in drone technology in the air to assess broad-scale adaptability and repeatability;
Characterize thermal signatures for large land-dwelling mammal species and training models remotely (i.e., while deployed in the field), with particular emphasis on elk and grizzly bears; and
Integrate artificial intelligence and communication networks to streamline workflow for rapid multispecies detection and quantification as well as real-time alert and notification.
Materials and Methods:
The proposed project is divided into three interacting and mutually reinforcing components: (1) field evaluation of geofenced cattle pastures using smart cameras and field surveys of wildlife abundances and distribution using drone-based thermal and visual imaging; (2) lab-based work focused on modern workflows for data management and analysis, development of methodology and machine learning algorithms for automation of wildlife counts, locations, species identification from camera images and thermal drone imagery; and (3) engagement and outreach activities at the start, during the execution, and at the conclusion of the project structured to provide two-way engagement regarding the technology and its development and application with landowners to ensure tailoring of workflows and tools developed and to engage the broader community of landowners and managers in Montana with the study to ensure the scaling of the methodology and create a larger community of wildlife technology users and practitioners in Montana.
Study area: The focus of this study will be Paradise Valley in Park County, a current hotspot of wildlife-livestock conflicts. Our study area will include 2 ranches, one spring-calving Montana cattle ranch where large numbers of elk typically commingle with cattle during spring, and one spring-calving Montana ranch where large numbers of grizzly bears occupy the landscape. Selected ranches will be located close enough to Bozeman and close enough to each other to enable each ranch to be visited once per week during a 3-month period (March, April, May). At each ranch, cattle will use their normal pasture(s) during spring.
Smart Cameras: We will geofence spring-calving pastures by deploying smart cameras along the perimeter of pastures prior to calving and prior to cattle entering each pasture. Smart cameras will be equipped with edge AI and IoT integrated to allow researchers and ranchers to be alerted when wildlife (e.g., elk, bears) enter the pasture.
Drones: We will use drone-based thermal imaging surveys (e.g., to document animal distributions in relation to calving pastures and estimate elk and grizzly bear abundances. Flights for thermal imaging will follow the protocol our team has developed (11) which uses fixed-wing long-endurance vertical take-off and landing (VTOL) or multirotor drones. Drones will fly at an altitude of 80-120 m imaging with a FLIR Vue Duo Pro R thermal/RGB imagers and MAP-A7R mapping camera. Video and telemetry will be uploaded to ArcGIS Full Motion Video and analyzed by two observers providing redundant georeferenced data for model comparison (11). Our previous work demonstrated the method achieved 92% accuracy when compared with a known population of white-tailed deer.
AI/Machine Learning in Wildlife Management. A major drawback to camera trapping is that it generates immense amounts of data, taking months to classify to prepare for analysis. Indeed, a danger from a wide variety of new wildlife technology is that projects can drown in data while being starved for information. AI can be used in workflows to separate images containing data from those that don’t, greatly simplifying analysis, as well as being trained to recognize individuals and species in camera traps and is a rapidly advancing field (12, 13). We will use both images collected by smart cameras as well as pre-classified imagery collected through project partners (e.g., Montana Fish, Wildlife & Parks) to train and test AI algorithms in object recognition and classification from traditional observers. While drone video in RGB (visual) is an active area of study in machine learning and computer vision (14), thermal video as applied to wildlife has received sparse attention in terms of AI for wildlife identification, relying on simple thresholding, and having poor performance in real world conditions (15,16). Given the striking improvement in wildlife detection and abundance in thermal drone 4 imagery, the application of AI to thermal drone video would represent a singular advance in wildlife management and abundance estimation.
Benchmarks: Based on existing methods for classifying occupied vs. unoccupied imagery (is there an animal) and object detection of species (what is in the image and where is it), we will set our smart camera benchmark at 90% accuracy for camera trap images (12). Algorithms that use differencing within a certain camera frame are specific to individual camera locations (e.g., 17), and have slightly higher reported accuracy, and we will test against them to see if we are meeting or exceeding the benchmark, and ask the scientifically interesting question of why? For video analysis there is no benchmark for thermal imagery deployed on drones for machine learning/computer-vision based algorithms. We will benchmark our algorithm to camera-based estimates and use 90% for object localization and 80% for object detection (lower because of species classification at this step). We will update benchmarking throughout the study as the rapidly developing field advances.
Expected Results and Outcomes: The expected outcomes of the project are (1) develop methodology, analysis techniques, and workflow to rapidly and accurately analyze imagery captured by smart cameras and drone-based thermal and video imagery, including spatial analysis, elimination of false positives, and species recognition, (2) illustrate ability to use drones to reliably monitor elk and grizzly bear populations and movement with much lower effort and safety than traditional aerial flights, (3) provide the ability to precisely replicate surveys giving the ability to detect relatively small changes in populations over time, (4) to communicate real-time animal detection along fence perimeters as an early alert system for mitigating wildlife-livestock/human conflict and, (5) publish either a refereed journal article that presents the results of this study. This project has the potential to reduce the cost and effort of data collection and analysis as well as the time required to make collected information available for management by landowners and management agencies. In particular, the proposed technology has great promise for providing reliable multi-species detection and estimation in a rapid and cost effective manner, giving the ability to implement targeting hazing practices for mitigating conflict and depredation while also reliably and cost-effectively estimating wildlife populations.
Relevance and prospective benefits to Montana agriculture and the world
Establishment of a rapid, reliable, and cost-effective means to assess and manage wildlife populations would provide a suite of practical benefits and questions that can be approached for multiple species, including large carnivore-livestock conflict prevention, feral swine detection and control, disease surveillance, and ungulate monitoring. Additionally, combining established methodology for real-time and notification using smart cameras, and reliable and costeffective abundance estimates from thermal drone surveys with a streamlined analysis workflow resulting in automated multi-species detection would serve as a global model that can be scaled to be applied to other managed lands applicable most anywhere in the world where livestock producers live and work in landscapes shared with large predators, and wild ungulates. Accurate information about wildlife populations and movement can also defuse conflicts among landowners and stakeholders about management practices. State and privately managed lands are embedded in a landscape of property owners and stakeholders, and uncertain information about populations can lead to conflict. Finally, the structured interactions and meetings planned, builds a community of landowners and wildlife managers with access to and ability to use the tools, or access those who use the tools, in order to support decisions on their lands. The project presents educational opportunities for the broader public and integrates itself well into Montana State University (MSU) curriculum and MSU Extension programming leading to stronger partnerships, more resilient ranches, and ultimately, better-connected landscapes
1. Wagner, F.H. 2006. Yellowstone’s destabilized ecosystem: elk effects, science, and policy conflicts. New York, NY, USA: Oxford University Press. 392 p.
2. Mosley, J.C., J. Fidel, H.E. Hunter, P.O. Husby, C.E. Kay, J.G. Mundinger, and R.M. Yonk. An ecological assessment of the Northern Yellowstone Range: introduction to the special issue. 2018. Rangelands 40:173-176.
3. Côté, S. D., T. P. Rooney, J. Tremblay, C. Dussault, and D. M. Waller. 2004. Ecological impacts of deer overabundance. Annual Review of Ecological and Evolutionary Systems 35:113–147.
4. Haggerty, J.H., and W.R. Travis. 2006. Out of administrative control: absentee owners, resident elk and the shifting nature of wildlife management in southwestern Montana. Geoforum 37:816-830.
5. Yonk, R.M., J.C. Mosley, and P.O. Husby. 2018. Human influences on the Northern Yellowstone Range. Rangelands 40:177-188.
6. Wells, S.L., L.B. McNew, D.B. Tyers, F.T. Van Manen, and D.J. Thomson. 2019. Grizzly bear depredation on grazing allotments in the Yellowstone ecosystem. Journal of Wildlife Management 83:556-566.
7. Gosnell, H., J.H. Haggerty, and W.R. Travis. 2006. Ranchland ownership change in the Greater Yellowstone Ecosystem, 1990-2001: Implications for conservation. Society and Natural Resources 19:743-758.
8. Crane, K.K., J.C. Mosley, T.K. Brewer, W.L.F. Torstenson, and M.W. Tess. 2016. Elk foraging site selection on foothill and mountain rangeland in spring. Rangeland Ecology and Management 69:319- 325.
9. Mosley, J.C., and J.G. Mundinger. 2018. History and status of wild ungulate populations on the Northern Yellowstone Range. Rangelands 40:189-201.
10. Morehouse, A.T., and M.S. Boyce. 2011. From venison to beef: seasonal changes in wolf diet composition in a livestock grazing landscape. Frontiers in Ecology and the Environment 9:440-445.
11. Beaver, J.T., R.W. Baldwin, M. Messinger, C. Newbolt, S.S. Ditchkoff, and M.R. Silman. 2020. Evaluating the use of drones equipped with thermal imaging technology as an effective method for estimating wildlife. Wildlife Society Bulletin 44:434–443.
12. Tabak, MA, M.S. Norouzzadeh, D.W. Wolfson, S.J. Sweeney, K.C. Vercauteren, N.P. Snow, et al. 2010. Machine learning to classify animal species in camera trap images: Applications in ecology. Methods Ecol Evol 10:585-590
13. Willi M, R.T. Pitman, A.W. Cardoso, C. Locke, A. Swanson, A. Boyer, et al. 2019. Identifying animal species in camera trap images using deep learning and citizen science. Methods Ecol Evol 10:80-91.
14. Carrio A, C. Sampedro, A. Rodriguez-Ramos, P. Campoy. 2017 A review of deep learning methods and applications for unmanned aerial vehicles. Journal of Sensors 3296874.
15. Christiansen P, Steen KA, Jørgensen RN, Karstoft H. 2014. Automated detection and recognition of wildlife using thermal cameras. Sensors (Basel) 14:13778-13793.
16. Lešetický J, Matějka P, Olmr M. Detection of Animal Occurrence Using an Unmanned System. 2019. Scientia Agriculturae Bohemica 50:203-210.
17. Janzen M, A. Ritter, P.D. Walker, D.R.Visscher. 2019. EventFinder: a program for screening remotely captured images. Environ Monit Assess 191:406