Wildlife Surveys Research Articles

Using object-based image analysis to detect laughing gull nests

ABSTRACT Remote sensing has long been used to study wildlife; however, manual methods of detecting wildlife in aerial imagery are often time-consuming and prone to human error, and newer computer vision techniques have not yet been extensively applied to wildlife surveys. We used the object-based image analysis (OBIA) software eCognition to detect laughing gull (Leucophaeus atricilla) nests in Jamaica Bay as part of an ongoing monitoring effort at the John F. Kennedy International Airport. Our technique uses a combination of high resolution 4-band aerial imagery captured via manned aircraft with a multispectral UltraCam Falcon M2 camera, LiDAR point cloud data, and land cover data derived from a bathymetric LiDAR point cloud to classify and extract laughing gull nests. Our ruleset uses the site (topographic position of nest objects), tone (spectral characteristic of nest objects), shape, size, and association (nearby objects commonly found with the objects of interest that help identify them) elements of image interpretation, as well as NDVI and a sublevel object examination to classify and extract nests. The ruleset achieves a producer’s accuracy of 98% as well as a user’s accuracy of 65% and a kappa of 0.696, indicating that it extracts a majority of the nests in the imagery while reducing errors of commission to only 35% of the final results. The remaining errors of commission are difficult for the software to differentiate without also impacting the number of nests successfully extracted and are best addressed by a manual verification of output results as part of a semi-automated workflow in which the OBIA is used to complete the initial search of the imagery and the results are then systematically verified by the user to remove errors. This eliminates the need to manually search entire sets of imagery for nests, resulting in a much more efficient and less error prone methodology than previous unassisted image interpretation techniques. Because of the extensibility of OBIA software and the increasing availability of imagery due to small unmanned aircraft systems (sUAS), our methodology and its benefits have great potential for adaptation to other species surveyed using aerial imagery to enhance wildlife population monitoring.

Unmanned aerial systems in search and rescue applications with their path planning: a review

In recent days, the Unmanned Aerial Systems (UAS) is an emerging technology rapidly across many warfare’s and enable new civilian domains which include real-time monitoring, security, border surveillance, wildlife surveys, providing wireless coverage, weather monitoring, smart farming, surveying, search and rescue, products delivery, farming, and civil structure inspection. Initially, there were many challenges in the design and control of UAS as it lacks an onboard pilot for navigation. Now the condition is improved with the implementation of artificial intelligence techniques in path planning and their coordination. Unmanned Aerial vehicles (UAV) in UAS technology open up new potentials in a variety of fields. This paper presents a review of recent literature, starting with the introduction about the UAS and its types, components, developments, potential applications of UAV in search and rescue and the paper completes with a summary of the different path planning methods used in UAS.

A drone-based survey for large, basking freshwater turtle species

Conservation concerns are increasing for numerous freshwater turtle species, including Pseudemys gorzugi, which has led to a call for more research. However, traditional sampling methodologies are often time consuming, labor intensive, and invasive, restricting the amount of data that can be collected. Biases of traditional sampling methods can further impair the quality of the data collected, and these shortfalls may discourage their use. The use of unmanned aerial vehicles (UAVs, drones) for conducting wildlife surveys has recently demonstrated the potential to bridge gaps in data collection by offering a less labor intensive, minimally invasive, and more efficient process. Photographs and video can be obtained by camera attachments during a drone flight and analyzed to determine population counts, abundance, and other types of data. In this study we developed a detailed protocol to survey for large, freshwater turtle species in an arid, riverine landscape. This protocol was implemented with a DJI Matrice 600 Pro drone and a SONY ILCE α6000 digital camera to determine P. gorzugi and sympatric turtle species occurrence across 42 sites in southwestern Texas, USA. The use of a large drone and high-resolution camera resulted in high identification percentages, demonstrating the potential of drones to survey for large, freshwater turtle species. Numerous advantages to drone-based surveys were identified as well as some challenges, which were addressed with additional refinement of the protocol. Our data highlight the utility of drones for conducting freshwater turtle surveys and provide a guideline to those considering implementing drone-mounted high-resolution cameras as a survey tool.

Thermal Sensor Calibration for Unmanned Aerial Systems Using an External Heated Shutter

Uncooled thermal infrared sensors are increasingly being deployed on unmanned aerial systems (UAS) for agriculture, forestry, wildlife surveys, and surveillance. The acquisition of thermal data requires accurate and uniform testing of equipment to ensure precise temperature measurements. We modified an uncooled thermal infrared sensor, specifically designed for UAS remote sensing, with a proprietary external heated shutter as a calibration source. The performance of the modified thermal sensor and a standard thermal sensor (i.e., without a heated shutter) was compared under both field and temperature modulated laboratory conditions. During laboratory trials with a blackbody source at 35 °C over a 150 min testing period, the modified and unmodified thermal sensor produced temperature ranges of 34.3–35.6 °C and 33.5–36.4 °C, respectively. A laboratory experiment also included the simulation of flight conditions by introducing airflow over the thermal sensor at a rate of 4 m/s. With the blackbody source held at a constant temperature of 25 °C, the introduction of 2 min air flow resulted in a ’shock cooling’ event in both the modified and unmodified sensors, oscillating between 19–30 °C and -15–65 °C, respectively. Following the initial ‘shock cooling’ event, the modified and unmodified thermal sensor oscillated between 22–27 °C and 5–45 °C, respectively. During field trials conducted over a pine plantation, the modified thermal sensor also outperformed the unmodified sensor in a side-by-side comparison. We found that the use of a mounted heated shutter improved thermal measurements, producing more consistent accurate temperature data for thermal mapping projects.

Comments on the distribution and population estimation of Neurergus derjugini (Urodela, Salamandridae) in western Iran

Basic knowledge of population status is effective for the conservation of many species, especially threatened and endangered species. In this study, we used the Mark-recapture technique (Linconln-Petersen method) as an established method for estimating population in wildlife surveys. Population surveys of critically endangered newts conducted for Neurergus derjugini in the streams of Kurdistan and Kermanshah Provinces, western Iran, at Sarvabad, Marivan, Baneh, Saghez and Nosoud. The study consists of estimating population size by sampling annually for two consecutive years (2018-2019) with each annual sampling consisting of several surveys between May-August. The species has a restricted distribution, occurring at high mountain streams. Our results also reject the occurrence of this species in Sarvabad and Awraman stream habitats and water ecosystems as was stated by former researchers. Distribution information of this species was updated in the region, and all known records are mapped. The results showed that Miri Sour (Marivan) with 1591-1568 and Asnabad (Marivan) with 14-18 specimens as estimated populations were found as most and less populated and suitable habitats respectively.

RANGE EXTENSION OF CARRIKER’S ROUND-EARED BAT, Lophostoma carrikeri (ALLEN, 1910) (CHIROPTERA: PHYLLOSTOMIDAE) IN THE CERRADO AND CAATINGA

Lophostoma carrikeri is a bat species widely distributed in the northern portion of South America and has been reported in a great variety of forested habitats. However, it is still poorly represented in scientific collections and wildlife surveys. In Brazil, L. carrikeri is known mainly for the Amazon basin and few localities in the Cerrado and Caatinga biomes. Here we report the first records of Lophostoma carrikeri from the states of Minas Gerais and Ceará based on specimens examined in taxonomic collections and collected during fieldwork. These new records represent the easternmost limits on the distribution of L. carrikeri and expand its known geographic range in approximately 488 km southeast and 490 km east of the nearest known records (São Domingos, state of Goiás and Teresina, state of Piauí). We also provide measurements, comments on morphological variation and an overall review of the species geographic distribution.

Counting sea lions and elephants from aerial photography using deep learning with density maps

BackgroundThe ability to automatically count animals is important to design appropriate environmental policies and to monitor their populations in relation to biodiversity and maintain balance among species. Out of all living mammals on Earth, 60% are livestock, 36% humans, and only 4% are animals that live in the wild. In a relatively short period, development of human civilization caused a loss of 83% of wildlife and 50% of plants. The rate of species extinction is accelerating. Traditional wildlife surveys provide rough population estimates. However, emerging technologies, such as aerial photography, allow to perform large-scale surveys in a short period of time with high accuracy. In this paper, we propose the use of computer vision, through deep learning (DL) architecture, together with aerial photography and density maps, to count the population of Steller sea lions and African elephants with high precision.ResultsWe have trained two deep learning models, a basic UNet without any feature extractor (Model-1) and another with the EfficientNet-B5 feature extractor (Model-2). We measured the model’s prediction accuracy, using Root Mean Square Error (RMSE) for the predicted and actual animal count. The results showed an RMSE of 1.88 and 0.60 to count Steller sea lions and African elephants, respectively, regardless of complex background, different illumination conditions, heavy overlapping and occlusion of the animals.ConclusionsOur proposed solution performed very well in the counting prediction problem, with relatively low training parameters and minimum annotation. The approach adopted, combining DL and density maps, provided better results than state-of-art deep learning models used for counting, indicating that the proposed method has the potential to be used more widely in large-scale wildlife surveying projects and initiatives.

Assessment and novel application ofN‐mixture models for aerial surveys of wildlife

AbstractAerial surveys are a critical tool for inference of wildlife populations, yet are limited by current available methods to account for imperfect detection without marking animals.N‐mixture models use count data from replicate surveys for estimation of wildlife abundance while accounting for imperfect detection. This statistical framework was recently developed but is rarely applied to aerial surveys of terrestrial wildlife. We applied anN‐mixture model incorporating temporary emigration to a novel aerial survey design to estimate the availability, detection probability, and abundance of an unmarked population of white‐tailed deer (Odocoileus virginianus) in Michigan, USA, using surveys in 2014 and 2016. We assessed the selection of the sample unit size for theN‐mixture model in a post hoc sensitivity analysis and compared all estimates to a hierarchical distance sampling (HDS) approach. Hierarchical distance sampling is an expansion of conventional distance sampling (CDS), a standard method applied to aerial survey designs for wildlife, where HDS allows for modeling spatial variation in abundance and detection probability using environmental covariates and can account for changing population demographics. We foundN‐mixture models produced similar parameter estimates to HDS with both models inferring a population increase between years, which was consistent with independently available estimates for this population (2014N‐mixture = 402 [CI: 303.8–515.27], and HDS = 391 [CI: 297.67–504.63]; 2016N‐mixture = 732 [CI: 572.89–924.22], and HDS = 708 [CI: 552.45–893.89]). Estimates from both models were sensitive to the sample unit size selected in an open population framework. We recommend using ecological (e.g., home range size) and statistical (e.g., sample size) considerations when selecting a sample unit size that reflects the movement ecology of a species and evaluating the sensitivity of parameters to sample unit size. In our study,N‐mixture models and HDS models afforded a practical field approach and analytical methodology that resulted in precise estimation in a population of unmarked deer over time. When used with aerial surveys in terrestrial systems where distance sampling methods may be difficult to implement,N‐mixture models can address observer error and estimate populations for conservation and management goals without marking individual animals.

Conservation Detection Dogs Increase Efficacy for Prey Detection at Carnivore GPS Cluster Sites During Summer

ABSTRACTCombining carnivore movements with site visits to investigate predation events can inform investigators of carnivore diets, kill rates, and risk factors for prey. However, detecting remains of prey can be challenging when prey is small or when vegetation conceals prey. Conservation detection dogs (CDDs) have assisted with wildlife surveys to locate objects of conservation interest, often improving detection over human counterparts. We hypothesized that CDDs would improve efficacy for detection of prey remains at kill locations compared to human searchers. We compared the efficacy of CDD teams (1 dog, 1 human handler) and technician teams (2 humans) for locating prey remains at 729 carnivore cluster (potential predation) sites in the Upper Peninsula of Michigan, USA, during May to August of 2015. Carnivore species included gray wolves (Canis lupus), coyotes (C. latrans), bobcats (Lynx rufus), and American black bears (Ursus americanus). Overall, CDD teams identified prey remains at 2.3 times more clusters, required 42% less search time, and handlers traveled 83% less distance per detected prey remain compared to technician teams. Within carnivore species, CDD teams detected small prey remains at a greater rate than technician teams (4.0–4.3 times greater) for all carnivores except for bobcat which had similar detection rate. Detection of large prey remains (e.g., adult ungulates) was similar between CDD teams and technician teams. Our study supports CDDs ability to outperform humans when searching for objects of conservation interest which are visually cryptic but traceable by scent. Where logistics allow, we recommend use of CDDs over human searchers when it is a priority to locate remains of small prey and when vegetation is dense and may obscure even larger prey items. © 2021 The Wildlife Society.

Critters of the Carolinas: A team of researchers and community scientists use trail cameras to study wildlife

Photo 1. A female American black bear (Ursus americanus) and her three cubs walking along an access road in Alligator River National Wildlife Refuge in Dare County, North Carolina, USA. Photo captured on a trail camera used for surveying endangered red wolves. Photo credit: Wildlands Red Wolf Survey. Photo 2. A large American black bear (Ursus americanus) walking along a natural area in Hyde County, North Carolina, USA. Photograph taken by a trail camera set by a wildlife biologist for the North Carolina's Candid Critters camera trapping project. Photo credit: North Carolina's Candid Critters. Photo 3. A gray fox (Urocyon cinereoargenteus) inspecting a trail camera set by primary school students with the Museum of Life and Science in North Carolina, USA, for the Students Discover North Carolina camera trapping project. Photo credit: Students Discover North Carolina. Photo 4. A coyote (Canis latrans) inspecting a trail camera set by a community science volunteer in Dare County, North Carolina, USA. Trail camera set as part of the North Carolina's Candid Critters citizen science project. Photo credit: North Carolina's Candid Critters. Photo 5. A gobbler (male) wild turkey (Meleagris gallopavo) caught on a trail camera set by a community science volunteer in Needmore Game Land, found in the Blue Ridge Mountains of North Carolina, USA. Trail camera set as part of the North Carolina's Candid Critters citizen science project. Photo credit: North Carolina's Candid Critters. These photographs illustrate the article “Carolina Critters: a collection of camera trap data from wildlife surveys across North Carolina” by M Lasky, A Parsons, S Schuttler, B McShea, G Davis, L Kalies, R Sutherland, SR Clark, G Hess, H Boone, L Gatens, C Olfenbuttel, J Shaw, C Dukes, T Menke, J Matthews, and RW Kays published in Ecology. https://doi.org/10.1002/ecy.3372.

Accounting for imperfect detection in data from museums and herbaria when modeling species distributions: combining and contrasting data‐level versus model‐level bias correction

The digitization of museum collections as well as an explosion in citizen science initiatives has resulted in a wealth of data that can be useful for understanding the global distribution of biodiversity, provided that the well‐documented biases inherent in unstructured opportunistic data are accounted for. While traditionally used to model imperfect detection using structured data from systematic surveys of wildlife, occupancy models provide a framework for modelling the imperfect collection process that results in digital specimen data. In this study, we explore methods for adapting occupancy models for use with biased opportunistic occurrence data from museum specimens and citizen science platforms using seven species of Anacardiaceae in Florida as a case study. We explored two methods of incorporating information about collection effort to inform our uncertainty around species presence: 1) filtering the data to exclude collectors unlikely to collect the focal species and 2) incorporating collection covariates (collection type, time of collection and history of previous detections) into a model of collection probability. We found that the best models incorporated both the background data filtration step as well as collector covariates. Month, method of collection and whether a collector had previously collected the focal species were important predictors of collection probability. Efforts to standardize meta‐data associated with data collection will improve efforts for modeling the spatial distribution of a variety of species.

A systematic study of the class imbalance problem: Automatically identifying empty camera trap images using convolutional neural networks

Camera traps, which are widely used in wildlife surveys, often produce massive images, and many of them are empty images not contain animals. Using the deep learning model to automatically identify the empty camera trap images can reduce the workload of manual classification significantly. However, the performance of deep learning models is easily affected by the class imbalance problem of training datasets, which is a common problem for actual wildlife survey projects. Almost all previous studies on empty image recognition used down-sampling or oversampling methods to eliminate the effect of class imbalance on the performance of deep learning classifiers. The class imbalance problem has been systematically studied in the field of traditional image recognition, yet very limited research is available in the context of identifying camera trap images taken from highly cluttered natural scenes. This study systematically studied the impact of class imbalance on model performance when using a deep learning model to identify empty camera trap images. Then we proposed the construction method of training sets of the deep learning model when the data set has different class imbalance levels. Based on results from our experiments we concluded that (i) the class imbalance showed little effect on the performance of the model when the empty image ratio (EIR) in the data set was between 10% and 70%, so the training sets can be randomly built without changing the class distribution; (ii) we recommended using oversampling to partially eliminate class imbalance to reduce omission errors when the EIR of the data set exceeded 70%; (iii) when the EIRs of the training set and the test set were close, the overall error, omission error, and commission error of the model were relatively smaller, and the model tended to achieve a better overall performance; (iv) the omission and commission errors can be adjusted by changing the percentage of empty images in the training set.

Carolina critters: a collection of camera-trap data from wildlife surveys across North Carolina.

Camera trap surveys are useful to understand animal species population trends, distribution, habitat preference, behavior, community dynamics, periods of activity, and species associations with environmental conditions. This information is ecologically important, because many species play important roles in local ecosystems as predators, herbivores, seed dispersers, and disease vectors. Additionally, many of the larger wildlife species detected by camera traps are economically important through hunting, trapping, or ecotourism. Here we present a data set of camera-trap surveys from 6,043 locations across all 100 counties of North Carolina, USA from 2009 to 2019. These data come from 26 survey initiatives and contain 215,108 records of 36 mammal species and three species of terrestrial birds. This large data set increases the geographical distribution data for these 39 mammal and bird species by >500% over what is available for North Carolina in the Global Biodiversity Information Facility (GBIF). These data can be used to conduct inquiries about species, populations, communities, or ecosystems, and to produce useful information on wildlife behavior, distribution, and interactions. There are no copyright restrictions. Please cite this paper when using the data for publication.

A global review of the threats of mining on mid-sized and large mammals

Mineral exploitation threatens wildlife by the contamination of soil and water sources, vegetation suppression, and due to changes in landscape configuration. The large and mid-sized forest-dependent mammals need large areas to support their ecological needs, and are frequently impacted by these landscape alterations. Hitherto, there have not been comprehensive reviews of scientific papers reporting on the impacts of the legal mineral exploitation on these animals worldwide. We used the Web of Science and Scopus to select peer-reviewed research articles reporting on those impacts on the large and mid-sized mammals with terrestrial and arboreal habits. Thus, we i) classified the scientific articles according to their methodological approach, ii) examined the main methods of data collection, iii) identified the main impacts recorded, iv) associated the number of studies published by country to the measures of social welfare and economic growth of countries, and v) evaluated the density of mineral resources within the species extent of occurrence, as a proxy to the threat level of mining activities on these species. We found only 39 peer-reviewed articles describing the impacts of mining activities on these animals. These researches mostly quantify the population reduction associated to mining activities, based on line-transect surveys of wildlife. Forest suppression around mining zones is the most frequent threat assessed, and is reported as the main impact on these mammals. We found 153 mammal species to be affected by mining activities, but mining is not mentioned as a threat in the IUCN Red list for 84% (128 spp) of these. Thus, we suggest the inclusion of mining threats in the IUCN assessments of these species, and we urge researchers to intensify the studies of the direct and indirect impacts of mining on large and mid-sized mammals across all ecosystems.

StatEcoNet: Statistical Ecology Neural Networks for Species Distribution Modeling

This paper focuses on a core task in computational sustainability and statistical ecology: species distribution modeling (SDM). In SDM, the occurrence pattern of a species on a landscape is predicted by environmental features based on observations at a set of locations. At first, SDM may appear to be a binary classification problem, and one might be inclined to employ classic tools (e.g., logistic regression, support vector machines, neural networks) to tackle it. However, wildlife surveys introduce structured noise (especially under-counting) in the species observations. If unaccounted for, these observation errors systematically bias SDMs. To address the unique challenges of SDM, this paper proposes a framework called StatEcoNet. Specifically, this work employs a graphical generative model in statistical ecology to serve as the skeleton of the proposed computational framework and carefully integrates neural networks under the framework. The advantages of StatEcoNet over related approaches are demonstrated on simulated datasets as well as bird species data. Since SDMs are critical tools for ecological science and natural resource management, StatEcoNet may offer boosted computational and analytical powers to a wide range of applications that have significant social impacts, e.g., the study and conservation of threatened species.

Monitoring Exotic Beetles with Inexpensive Attractants: A Case Study.

Simple SummaryDetecting and monitoring exotic and invasive beetles is a complex activity, and multiple species still manage to evade controls. Citizen science can be an important adjunct in alien species monitoring programs, but to have a greater chance of success, it must employ traps and attractants that are easy to gather and use. Bottle traps baited with food products are successfully used during long term faunistic surveys, and the same methodology can be adapted to alien species detection and monitoring. In this article, we tested the use of bottles baited with apple cider vinegar, red wine, and 80% ethyl alcohol in capturing exotic and invasive beetles in the surroundings of Malpensa Airport (Italy). The traps proved effective, and in the traps with vinegar as an attractant, they captured four out of five invasive Nitidulidae, as well as the only invasive Scarabaeidae present in the area. Popillia japonica’s response to apple cider vinegar is documented for the first time and suggests the use of this attractant in monitoring surveys for this species, especially if supported by citizen science programs. The substantial reduction in the activity time of the traps seems to have considerably reduced collateral catches of native fauna.Detecting and monitoring exotic and invasive Coleoptera is a complex activity to implement, and citizen science projects can provide significant contributions to such plans. Bottle traps are successfully used in wildlife surveys and can also be adapted for monitoring alien species; however, a sustainable, large scale trapping plan must take into account the collateral catches of native species and thus minimize its impact on local fauna. In the present paper, we tested the use of bottles baited with standard food products that can be purchased in every supermarket and immediately used (apple cider vinegar, red wine, and 80% ethyl alcohol) in capturing exotic and invasive beetles in the area surrounding Malpensa Airport (Italy). In particular, we reduced the exposition type of the traps in each sampling round to three days in order to minimize native species collecting. We found a significant effect of the environmental covariates (trap placement, temperature, humidity, and forest type) in affecting the efficiency in catching target beetles. Nearly all invasive Nitidulidae and Scarabaeidae known to be present in the area were captured in the traps, with apple cider vinegar usually being the most effective attractant, especially for the invasive Popillia japonica.

An Adaptive Automatic Approach to Filtering Empty Images from Camera Traps Using a Deep Learning Model

ABSTRACTCamera traps are widely used in wildlife surveys because they are non‐invasive, low‐cost, and highly efficient. Camera traps deployed in the wild often produce large datasets, making it increasingly difficult to manually classify images. Deep learning is a machine learning method that provides a tool to automatically identify images, but it requires labeled training samples and high‐performance servers with multiple Graphics Processing Units (GPUs). However, manually preparing large‐scale training images for training deep learning models is labor intensive, and the high‐performance servers with multiple GPUs are often not available for wildlife management agencies and field researchers. Our study explores an adaptive deep learning method to use small‐scale training sets and a commonly‐available, desktop personal computer (PC) to achieve automatic filtering of empty camera images. Our results showed that by using 29,192 training samples, the overall error, commission error, and omission error of the proposed method on a PC were 2.69%, 6.82%, and 6.45%, respectively. Moreover, the accuracy of our method can be adaptively improved on PCs in actual ecological monitoring projects, which would benefit researchers in field settings when only a PC is available. © 2021 The Wildlife Society.

Modelling species distribution from camera trap by‐catch using a scale‐optimized occupancy approach

AbstractHabitat selection is strongly scale‐dependent, and inferring the characteristic scale at which an organism responds to environmental variation is necessary to obtain reliable predictions. The occupancy framework is frequently used to model species distribution with the advantage of accounting for imperfect observation, but occupancy studies typically do not define the characteristic scale of the modelled variables. We used camera trap data from winter wildlife surveys in the Swiss part of the Jura Mountains to model occupancy of wild boar (Sus scrofa) and roe deer (Capreolus capreolus). We used a three‐step approach: (1) first, we identified factors influencing detectability; (2) second, we optimized the characteristic scale of each candidate explanatory variable; and (3) third, we fit multivariable, multiscale occupancy models in relation to land cover, human presence and topography. Wild boar occupancy was mainly influenced by the interaction between elevation within 2500 m and the proportion of forested areas within a 2500 m, with a nonsignificant additional effect of the interaction between ruggedness within 1900 m and the proportion of forested areas within 2500 m as well as the distance to urban areas. Roe deer occupancy was mainly associated with the interaction between ruggedness within 900 m and the proportion of open landscape within 900 m, with an additional nonsignificant effect of the interaction between elevation within 1500 m and the proportion of open landscape within 900 m as well as the distance to urban areas. Incorporating scale optimization in occupancy modelling of camera trap data can greatly improve the understanding of species‐environment relationships by combining the possibility of occupancy models to correct for detection bias and simultaneously allowing to infer the characteristic scale at which certain factors influence the distribution of the organisms studied.

A Novel Method for Using Small Unoccupied Aerial Vehicles to Survey Wildlife Species and Model Their Density Distribution

There is growing interest from research and conservation groups in the potential for using small unoccupied aerial vehicles (UAVs; <2 kg) to conduct wildlife surveys because they are affordable, easy to use, readily available and reliable. However, limitations such as short flight endurance, and in many situations, aviation regulations, have constrained the use of small UAVs in survey applications. Thus, there is a need to refine survey methods adapted to small UAVs that conform to standard operations within aviation law. We developed a novel survey approach based on a grid sampling design using two multirotor UAVs (Phantom 4 Pros) flying simultaneously, within visual line of sight, from our vessel base-station. We used this approach to assess the fine-scale distribution and abundance of dugongs (Dugong dugon) in the remote waters of the Pilbara, Western Australia during three field seasons across 2 years. We surveyed 64 non-overlapping survey cells in random order one or more times and obtained complete image coverage of each surveyed cell of our 31 km2 survey area. Our sampling design maximizes sampling effort while limiting survey time by surveying four cells, two at a time, from one location. Overall, we conducted 240 flights with up to 17 flights per day (mean = 14 flights per day) and could obtain complete coverage of up to 11.36 km2per day. A total of 149 dugongs were sighted within the 50,482 images which we manually reviewed. Spatially-explicit models of dugong density distribution (corrected for availability and perception bias) were produced using general additive models to identify areas more or less used by dugongs (range of corrected dugong densities across all field season = 0.002–1.79 dugongs per 0.04 km2). Dugong abundance estimates ranged from 47 individuals in June 2019 (CV = 0.17) to 103 individuals in May 2018 (CV = 0.36). Our method, which proved convincing in a real-word application by its feasibility, ease of implementation, and achievable surface coverage has the potential to be used in a wide range of applications from community-based local-scale surveys, to long-term repeated/intensive surveys, and impact assessments and environmental monitoring studies.

Evaluating new technology for biodiversity monitoring: Are drone surveys biased?

Drones and machine learning‐based automated detection methods are being used by ecologists to conduct wildlife surveys with increasing frequency. When traditional survey methods have been evaluated, a range of factors have been found to influence detection probabilities, including individual differences among conspecific animals, which can thus introduce biases into survey counts. There has been no such evaluation of drone‐based surveys using automated detection in a natural setting. This is important to establish since any biases in counts made using these methods will need to be accounted for, to provide accurate data and improve decision‐making for threatened species. In this study, a rare opportunity to survey a ground‐truthed, individually marked population of 48 koalas in their natural habitat allowed for direct comparison of the factors impacting detection probability in both ground observation and drone surveys with manual and automated detection. We found that sex and host tree preferences impacted detection in ground surveys and in manual analysis of drone imagery with female koalas likely to be under‐represented, and koalas higher in taller trees detected less frequently when present. Tree species composition of a forest stand also impacted on detections. In contrast, none of these factors impacted on automated detection. This suggests that the combination of drone‐captured imagery and machine learning does not suffer from the same biases that affect conventional ground surveys. This provides further evidence that drones and machine learning are promising tools for gathering reliable detection data to better inform the management of threatened populations.