Monitoring wildlife-vehicle collisions in the information age: how smartphones can improve data collection

Bat2Web: A Framework for Real-Time Classification of Bat Species Echolocation Signals Using Audio Sensor Data

Bats play a pivotal role in maintaining ecological balance, and studying their behaviors offers vital insights into environmental health and aids in conservation efforts. Determining the presence of various bat species in an environment is essential for many bat studies. Specialized audio sensors can be used to record bat echolocation calls that can then be used to identify bat species. However, the complexity of bat calls presents a significant challenge, necessitating expert analysis and extensive time for accurate interpretation. Recent advances in neural networks can help identify bat species automatically from their echolocation calls. Such neural networks can be integrated into a complete end-to-end system that leverages recent internet of things (IoT) technologies with long-range, low-powered communication protocols to implement automated acoustical monitoring. This paper presents the design and implementation of such a system that uses a tiny neural network for interpreting sensor data derived from bat echolocation signals. A highly compact convolutional neural network (CNN) model was developed that demonstrated excellent performance in bat species identification, achieving an F1-score of 0.9578 and an accuracy rate of 97.5%. The neural network was deployed, and its performance was evaluated on various alternative edge devices, including the NVIDIA Jetson Nano and Google Coral.

Ungulates and trains - Factors influencing flight responses and detectability

Wildlife-train collisions can have deleterious effects on local wildlife populations and come with high socio-economic costs, such as damages, delays, and psychological distress. In this study, we explored two major components of wildlife-train collisions: the response of wildlife to oncoming trains and the detection of wildlife by drivers. Using dashboard cameras, we explored the flight response of roe deer (Capreolus capreolus) and moose (Alces alces) to oncoming trains and explored which factors, such as lighting and physical obstructions, affect their detection by drivers. In a majority of cases, roe deer and moose fled from an oncoming train, at an average flight initiation distance (FID) of 78 m and 79 m respectively. Warning horns had unexpected influences on flight behaviour. While roe deer initiated flight, on average, 44 m further away from the train when warned, they usually fled towards the tracks, in the direction of danger. FID of moose, however, was unaffected by the use of a warning horn. As train speed increased, moose had a lower FID, but roe deer FID did not change. Finally, detection of wildlife was obstructed by the presence of vegetation and uneven terrain in the rail-side verge, which could increase the risk of collisions. Our results indicate the need for early detection and warning of wildlife to reduce the risk of collisions. We propose that detection systems should include thermal cameras to allow detection behind vegetation and in the dark, and warning systems should use cues early to warn of oncoming trains and allow wildlife to escape the railway corridor safely.

How Can We Do Citizen Science Better? A Case Study Evaluating Grizzly Bear Citizen Science Using Principles of Good Practice in Alberta, Canada

Simple Summary

Citizen science offers an excellent opportunity to engage the public in scientific data collection, educational opportunities, and applied management. However, the practicalities of developing a citizen science program, from generating ideas to developing tools, implementing programming, and evaluating outcomes, are complex and challenging. To address challenges and provide a foundation for practitioners, scientists, and the public, the Government of Alberta developed a set of citizen science principles. Here, we use these principles as an evaluative framework to assess the outcomes of the GrizzTracker program, which was developed to help inform provincial species-at-risk recovery efforts. While the program experienced some successes, we identified challenges, including skepticism from the scientific community about the utility of citizen science and a lack of program leadership, staff capacity, and funding needs for long-term implementation. Reflecting on the principles, we provide policy recommendations that future citizen science programs can consider.

Abstract

Citizen science offers an excellent opportunity to engage the public in scientific data collection, educational opportunities, and applied management. However, the practicalities of developing and implementing citizen science programming are often more complex than considered. Some challenges to effective citizen science include scientists’ skepticism about the ability of public participants to rigorously collect quality data; a lack of clarity on or confidence in the utility of data; scientists’ hesitancy in engaging the public in projects; limited financial commitments; and challenges associated with the temporal and geographic scales of projects. To address these challenges, and provide a foundation upon which practitioners, scientists, and the public can credibly engage in citizen science, the Government of Alberta developed a set of citizen science principles. These principles offer a framework for planning, designing, implementing, and evaluating citizen science projects that extend beyond Alberta. Here, we present a case study using these principles to evaluate GrizzTracker, a citizen science program developed to help inform provincial species-at-risk recovery efforts. While we found that GrizzTracker applied each of the six principles in some way, including successful public engagement, strengthened relationships, and raising public awareness about northwest Alberta’s grizzly bears, we also identified a number of challenges. These included ongoing skepticism from the traditional scientific community about the utility of citizen science and governance challenges related to program leadership, staff capacity, and funding. By using the principles as a guideline, we provide policy recommendations for future citizen science efforts, including considerations for program design, implementation, and evaluation.

Canadian wildlife health surveillance—patterns, challenges and opportunities identified by a scoping review

The protection and promotion of healthy wildlife populations is emerging as a shared goal among stakeholders in the face of unprecedented environmental threats. Accordingly, there are growing demands for the generation of actionable wildlife health information. Wildlife health surveillance is a connected system of knowledge that generates data on a range of factors that influence health. Canada recently approved the Pan-Canadian Approach to Wildlife Health that describes challenges facing wildlife health programs and provides a path forward for modernizing our approach. This scoping review was undertaken to describe the range of peer-reviewed Canadian wildlife health surveillance literature within the context of the challenges facing wildlife health programs and to provide a quantitative synthesis of evidence to establish baselines, identify gaps, and inform areas for growth. This review describes patterns related to species, location, authorship/funding, objectives, and methodology. Five areas are identified that have the potential to propel the field of wildlife health: representativeness, expanded/diversified collaboration, community engagement, harmonization, and a shift to a solutions-focused and One Health mindset. This scoping review provides a synopsis of 10 years of Canadian wildlife health surveillance, challenges us to envision the future of successful wildlife health surveillance, and provides a benchmark from which we can measure change.

Understanding spatio-temporal patterns of deer-vehicle collisions to improve roadkill mitigation

Vehicles collide with hundreds of thousands of deer on European roads each year. This leads to animal deaths and suffering, economic damage and risks for human safety, making the reduction of road mortality a major field in conservation biology. In order to successfully reduce roadkill, we need improved knowledge regarding spatio-temporal patterns of deer-vehicle collisions (DVCs) on a landscape scale. Here, we analyzed >85,000 DVCs collected over 17 years in Denmark to investigate changes in the number of DVCs over time and to find spatio-temporal patterns of DVC occurrence. We used a use-availability design - originally developed for habitat selection analyses - to compare DVCs involving roe deer (Capreolus capreolus), red deer (Cervus elaphus) and fallow deer (Dama dama) with random road locations on a landscape scale. This approach enabled us to combine temporal (seasonal and diel variation), spatial (land cover, road density and type) and other relevant variables (deer population density, traffic, and deer activity) within the same analysis. We found that factors related to infrastructure and land cover were most important in explaining patterns of DVCs, but seasonal and diel changes, deer activity, and population density were also important in predicting the occurrence of DVCs. Importantly, patterns of DVCs were largely similar between the three deer species, with more DVCs occurring at intermediate traffic density, increasing forest cover, during dusk and dawn, and with increasing deer activity and population density. The strong and consistent patterns found here will allow the development of flexible mitigation measures. We propose that our findings could be used to develop a spatio-temporally flexible warning system for smartphones and navigation systems that is based on existing map providers, making it a widely available and cheap mitigation measure.

A Reprieve from US wildlife mortality on roads during the COVID-19 pandemic

High traffic volume is one of the main contributors to wildlife-vehicle collision (WVC) and wildlife mortality on roads. Government shelter-in-place (SIP) orders have been used to help mitigate the spread of COVID-19, resulting in unprecedented reductions in global traffic volumes. Using traffic and collision data from four US states (California, Idaho, Maine, and Washington), we investigated changes in total WVC, following the state and local SIP orders. From mid-March to mid-April 2020, these orders have resulted in up to 71%, 63%, 73%, and 72% reduction in driving, as measured by vehicle miles traveled (VMT), in CA, ID, ME, and WA respectively. The daily WVC rates from the 4 weeks prior to SIP orders going into effect, to the 4 weeks after, declined 34%, with 21, 36, 44, and 33% declines for CA, ID, ME, and WA, respectively. For mountain lions (Puma concolor) in CA, there was a 58% decline in mortality during the traffic reduction. The changes in WVC from 1 month pre-SIP orders to 1 month post-order only occurred in 2020 and not 2015, 2016, 2017, 2018, or 2019, suggesting that the reductions were associated with the reductions in traffic. The measured declines in WVC reversed in ME and WA during May, June and July 2020, paralleling reversals in traffic volumes. A 34% reduction in WVC would potentially equate to 10s of millions fewer vertebrates killed on US roadways during one month of traffic reduction, representing an unintentional conservation action unprecedented in modern times.

The Australian Roadkill Reporting Project-Applying Integrated Professional Research and Citizen Science to Monitor and Mitigate Roadkill in Australia

Simple Summary

Australia has no database of national roadkill. The current research project fills that knowledge gap by developing a roadkill reporting application that enables professional and citizen scientists to record photographs of roadkill with location, time and date. This embodies the concept of ‘One Welfare’ as it affects humans, animals and the environment. Uploaded to a website, these data can identify roadkill hotspots, tabulate species of animals killed and potentially be used for ecological studies of roadkill numbers, species distribution, population trends, animal behaviour and disease. Initial results indicate that mammal roadkill mostly occurs at night and that of birds and reptiles during daytime. Mammals make up three-quarters of the roadkill recorded and this includes endangered species. Two examples of roadkill hotspots are shown in Queensland and Tasmania. These will enable further research to suggest how roadkill mitigation measures may be optimally employed.

Abstract

Australia has no national roadkill monitoring scheme. To address this gap in knowledge, a roadkill reporting application (app) was developed to allow members of the public to join professional researchers in gathering Australian data. The app is used to photograph roadkill and simultaneously records the GPS location, time and date. These data are uploaded immediately to a website for data management. To illustrate the capacity to facilitate cost-effective mitigation measures the article focuses on two roadkill hotspots—in Queensland and Tasmania. In total, 1609 reports were gathered in the first three months of the project. They include data on mammals (n = 1203, 75%), birds (n = 125, 7.8%), reptiles (n = 79, 4.9%), amphibians (n = 4, 0.025%), unidentified (n = 189, 11.8%) and unserviceable ones (n = 9). A significant finding is variance in the distribution of mammals and birds at different times of day. These findings reflect diurnal variation in the activity levels of different species and underline the need for data on a targeted species to be collected at appropriate times of day. By continuing to facilitate roadkill monitoring, it is anticipated that the data generated by the app will directly increase knowledge of roadkill numbers and hotspots. Indirectly, it will provide value-added information on animal behaviour, disease and population dynamics as well as for species distribution mapping.

Animal-vehicle collisions in Texas: How to protect travelers and animals on roadways

Animal-vehicle collisions (AVCs) are a growing problem in the United States, resulting in countless loss of animal life and considerable human injury and death every year, especially to motorcyclists. Due to underreporting, collision data generally provide a very low (highly biased) estimate of actual AVC counts and often lack key details, such as the species of animals involved. However, AVC reports cover entire states and nations, and can illuminate differences in wild versus domestic animal-vehicle collisions through statistical and spatial analysis. 51,522 animal-related crashes were reported to Texas police from 2010 through 2016, at a total cost over $1.3 billion annually to Texas motorists - not including the value of lost animal lives. AVC reports jump twice a day: between 5 and 8 AM and between 5 and 10 PM. Motorists are also significantly more likely to collide with a wild animal during the months of October, November, and December. Wildlife-vehicle collisions (WVC) are 64% of total reports, events involving domestic animals (like dogs and cattle) are 31%, and the remaining 5% of reports are unspecified. Most AVCs in the state occur at night in unlit locations, usually on rural roads with very low traffic volumes. Using ordinary least-squares (OLS) regression analysis across Texas' n = 254 counties, this work finds that less densely populated counties, marked as rural, and those with fewer vehicle-miles traveled (VMT) per capita but more lane-miles per capita, tend to experience the greatest number of AVCs per VMT, after controlling for county average rainfall, share of VMT onsystem roadways, job densities, and vehicle ownership (vehicles per capita). Intervention options for the mitigation of animal-vehicle collisions are numerous and diverse. For wildlife collisions specifically, this work finds that large crossing structures (underpasses and overpasses) at the highway link level return benefit-to-cost ratios near 3.0, while their lower-cost counterparts (wildlife fencing and animal detection systems) deliver ratios up to 30.

The role of digital data entry in participatory environmental monitoring

Many argue that monitoring conducted exclusively by scientists is insufficient to address ongoing environmental challenges. One solution entails the use of mobile digital devices in participatory monitoring (PM) programs. But how digital data entry affects programs with varying levels of stakeholder participation, from nonscientists collecting field data to nonscientists administering every step of a monitoring program, remains unclear. We reviewed the successes, in terms of management interventions and sustainability, of 107 monitoring programs described in the literature (hereafter programs) and compared these with case studies from our PM experiences in Australia, Canada, Ethiopia, Ghana, Greenland, and Vietnam (hereafter cases). Our literature review showed that participatory programs were less likely to use digital devices, and 2 of our 3 more participatory cases were also slow to adopt digital data entry. Programs that were participatory and used digital devices were more likely to report management actions, which was consistent with cases in Ethiopia, Greenland, and Australia. Programs engaging volunteers were more frequently reported as ongoing, but those involving digital data entry were less often sustained when data collectors were volunteers. For the Vietnamese and Canadian cases, sustainability was undermined by a mismatch in stakeholder objectives. In the Ghanaian case, complex field protocols diminished monitoring sustainability. Innovative technologies attract interest, but the foundation of effective participatory adaptive monitoring depends more on collaboratively defined questions, objectives, conceptual models, and monitoring approaches. When this foundation is built through effective partnerships, digital data entry can enable the collection of more data of higher quality. Without this foundation, or when implemented ineffectively or unnecessarily, digital data entry can be an additional expense that distracts from core monitoring objectives and undermines project sustainability. The appropriate role of digital data entry in PM likely depends more on the context in which it is used and less on the technology itself.

Patterns and Composition of Road-Killed Wildlife in Northwest Argentina

Roads have important effects on wildlife, such as natural habitat fragmentation and degradation and direct killing of fauna, which leads to reductions in wildlife population size. We focused on a principal road in Northwest Argentina to test for the effect of seasonality and landscape features on the composition of road-killed wildlife. We conducted regularly scheduled road trips during the dry and wet seasons. We recorded the presence or absence of a vegetation curtain or hedge along the road. We measured land use by remote sensing in a 500 m buffer along the road. We compared the abundance of animals killed between seasons (dry and wet) for different taxonomic groups (mammals, birds and reptiles) and for different origins (domestic and native). We built linear mixed models to test the effect of landscape features on the abundance of killed animals. Two hundred and ninety-three individuals were killed, belonging to 35 species; 75.8 % were native and 24.2 % domestic species. The majority of animals killed were mid-sized mammals. More animals were killed during the dry season. The most important factors to explain the wildlife road-killing were the season and the proportion of agricultural landscape. The composition of the killed animals changed with the season. The proportion of agricultural landscape incremented the number of killed birds and mammals during both seasons, without affecting reptiles. The ratio of wild to domestic animals killed was dependent on the season. This study sets a precedent as the first in road ecology in Northwest Argentina and should be taken into account for road planning and regulation.

Sampling method evaluation and empirical model fitting for count data to estimate densities of Oligonychus perseae (Acari: Tetranychidae) on 'Hass' avocado leaves in southern California

Oligonychus perseae (Acari: Tetranychidae) is an important foliar spider mite pest of 'Hass' avocados in several commercial production areas of the world. In California (USA), O. perseae densities in orchards can exceed more than 100 mites per leaf and this makes enumerative counting prohibitive for field sampling. In this study, partial enumerative mite counts along half a vein on an avocado leaf, an industry recommended practice known as the "half-vein method", was evaluated for accuracy using four data sets with a combined total of more than 485,913 motile O. perseae counted on 3849 leaves. Sampling simulations indicated that the half-vein method underestimated mite densities in a range of 15-60 %. This problem may adversely affect management of this pest in orchards and potentially compromise the results of field research requiring accurate mite density estimation. To address this limitation, four negative binomial regression models were fit to count data in an attempt to rescue the half-vein method for estimating mite densities. These models were incorporated into sampling plans and evaluated for their ability to estimate mite densities on whole leaves within 30-tree blocks of avocados. Model 3, a revised version of the original half-vein model, showed improvement in providing reliable estimates of O. perseae densities for making assessments of general leaf infestation densities across orchards in southern California. The implications of these results for customizing the revised half-vein method as a potential field sampling tool and for experimental research in avocado production in California are discussed.