Evaluating Camera Traps as Ground-Based Remote Sensing Networks, Linking Snow and Wildlife

Abstract

Seasonal snow covers 31% of the Northern Hemisphere in midwinter, playing an influential role in wildlife communities and water storage for hydrological processes. Monitoring seasonal snow comprises three main approaches, including remote sensing, hydrological modeling, and field techniques. However, current methods can be costly and less accurate in complex terrain, forests, and on cloudy days, and methods present trade-offs in spatial and temporal resolution. Using a large-scale camera network of over 1000+ cameras based in Norway from a project called Scandcam (viltkamera.nina.no), my dissertation sought to determine how camera traps (also referred to as remote cameras, wildlife cameras, or game cameras) can connect on-the-ground and remote sensing snow monitoring techniques for ecological and hydrological applications. Chapter 2 initialized a conceptual and analytical framework for using wildlife cameras to supplement and validate optical remote sensing products for snow cover. I identified advantages and disadvantages and presented wildlife camera traps as a novel method to improve snow monitoring on cloudy days and at high latitudes, regions where satellites are less accurate. In chapter 3, I provided a methodological example in which I used computer vision and machine learning to extract snow depth from remote cameras from NASA and University of Washington snow measurement campaigns, serving as a case study for snow information extraction from imagery as well as a pipeline for future model development. In chapter 4, I demonstrated how information from cameras can answer questions about animal activity in response to changes in snow conditions. I found that information from the camera, including location and time of day, can predict the strength of the snow surface, and subsequently changes in diel activity in roe deer (Capreolus capreolus) but not for mountain hares (Lepus timidus). In chapter 5, I examined how we can leverage wildlife cameras to supplement passive microwave remote sensing from the Advanced Microwave Scanning Radiometer 2 (AMSR2) for rain-on-snow mapping. I developed a model to detect rain in camera trap imagery and demonstrated a 1% improvement in AMSR2-detected rain-on-snow events. In chapter 6, I combined data from passive acoustic recorders (AudioMoths from Open Acoustic Devices) with camera data to improve the ability to track rain-on-snow events. I demonstrated that this technique improves human detection of and neural network predictions for rain-on-snow events. These findings illustrate that wildlife cameras can operate as “mini weather stations” for winter snow monitoring, and that by providing on-the-ground information, cameras demonstrate the ability to improve our snow monitoring techniques and understanding of wildlife communities in a changing climate.