Picking up good vibrations: a systematic review of footfall detection and analysis in the realm of wildlife surveying

Abstract

Exploration of new wildlife surveying methodologies that leverage advances in sensor technology and machine learning has led to tentative research into the application of seismology techniques. This, most commonly, involves the deployment of a footfall trap – a seismic sensor and data logger customised for wildlife footfall. While well-established in fields such as border security and human-intruder detection, the use of seismology in wildlife surveying remains nascent. This systematic review, conducted in accordance with PRISMA reporting guidelines, for the first time, compiles existing research in this developing area and offers a synthesis and discussion to support its progression. Our comprehensive search and screening process identified 34 studies that recorded or analysed wildlife footfall using ground-based sensors. Analysis of publication dates shows a clear upward trajectory in wildlife seismology research: the earliest included study was published in 2000, with 45% of studies appearing in the last five years. The primary driver behind most research was human–wildlife conflict (70%), followed by behavioural studies (15%) and wildlife monitoring (15%). Geophones were used in the majority of studies (70%), valued for their accessibility and affordability, though other sensors, such as pressure plates, vibration sensors, and seismometers, were also represented. Performance synthesis was challenging due to varied parameters. Binary classification tasks reported a median accuracy of 90%, while two recent studies involving four-class classification achieved accuracies of 91.3 and 91.7% using a Mahalanobis distance classifier and a convolutional neural network, respectively. We examine common trends and biases across the studies and conduct a horizon scan to identify key research questions that must be addressed to advance the methodology. High-priority areas include distinguishing closely related species, benchmarking against existing sensor technologies, evaluating generalisation across diverse sites, and investigating how footfall detectability might inform abundance estimates.