We introduce WildlifeMapper (WM), a flexible model designed to detect, locate, and identify multiple species in aerial imagery. It addresses the limitations of traditional, labor-intensive wildlife population assessments that are central to advancing environmental conservation efforts worldwide. While a number of methods exist to automate this process, they are often limited in their ability to generalize to different species or landscapes due to the dominance of homogeneous backgrounds and/or poorly captured local image structures. WM introduces two novel modules that help to capture the local structure and context of objects of interest to accurately localize and identify them, achieving a state-of-the-art (SOTA) detection rate of 0.56 mAP. Further, we introduce a large aerial imagery dataset with more than 11k Images and 28k annotations verified by domain experts. WM also achieves SOTA performance on 3 other publicly available aerial survey datasets collected across 4 different countries, improving mAP by 42%. Source code and trained models are available at Github (1).

Statistical models use observations of animals to make inferences about the abundance and distribution of species. However, the spatial distribution of animals is a complex function of many factors, including landscape and environmental features, and intra‐ and interspecific interactions. Modelling approaches often have to make significant simplifying assumptions about these factors, which can result in poor model performance and inaccurate predictions. Here, we explore the implications of complex spatial structure for modelling the abundance of the Serengeti wildebeest, a gregarious migratory species. The social behaviour of wildebeest leads to a highly aggregated distribution, and we examine the consequences of omitting this spatial complexity when modelling species abundance. To account for this distribution, we introduce a multi‐latent framework that uses two random fields to capture the clustered distribution of wildebeest. Our results show that simplifying assumptions that are often made in spatial models can dramatically impair performance. However, by allowing for mixtures of spatial models accurate predictions can be made. Furthermore, there can be a non‐monotonic relationship between model complexity and model performance; complex, flexible models that rely on unfounded assumptions can potentially make highly inaccurate predictions, whereas simpler more traditional approaches involve fewer assumptions and are less sensitive to these issues. We demonstrate how to develop flexible spatial models that can accommodate the complex processes driving animal distributions. Our findings highlight the importance of robust model checking protocols, and we illustrate how realistic assumptions can be incorporated into models using random fields.

Mapping wildlife and human distributions are critical to mitigating human and wildlife conflict and conserving wildlife, habitat, and human well-being. Aerial surveys are a critical tool, but new methods for these surveys generate massive image archives. The act of scanning these images is time-consuming and difficult, leading to significant delays in converting data into action. To address this Development Seed and the Tanzania Wildlife Research Institute (TAWIRI), developed an AI-assisted methodology, specifically, Tensorflow based image classification and object detection methods that significantly increase the speed of spotting and counting wildlife, human activities, and livestock after aerial surveys. Our AI-assisted survey method to map wildlife and human distributions and to detect potential conflict areas between wildlife and the human-associated activity for Tanzania. Once the training data quality and model performance of AI-assisted workflow mature and stabilize, we foresee the hours spent on getting accurate human-wildlife proximity maps would only take 19% of current human manual workflow, and potentially reduce the cost of identifying and counting objects over 81,000 aerial images from $20,000 to less than $5,000.

In aerial wildlife counts, human observers often fail to detect animals. We conducted a multi-species sample-count in Tsavo National Park, Kenya, with traditional rear-seat-observers (RSOs) and an automated ‘oblique-camera-count’ (OCC) imaging system to compare estimates of 23 wildlife species derived from these two survey methods. An aerial Total Count of elephant, buffalo and giraffe, conducted a month previously, provided a further comparison. In the Tsavo Core (9560km2), which harbours 80% of Tsavo’s elephants, the OCC system acquired 81 000 images for interpretation, of which 67 000 were obtained in parallel with RSO-counting along 3004km of flight line. The Tsavo outer blocks (24 171km2) were surveyed using the OCC system without RSOs to acquire a further 84 000 images. A random sample of 11 553 images were re-interpreted to derive species-specific probabilities of detection and correction factors. Using ‘Jolly II’, non-parametric and Bayesian analyses, and applying correction factors, we demonstrate that the RSOs did not detect 14% of elephants, 60% of giraffe, 48% of zebra and 66% of the large antelopes. For comparison, the Total Count observers did not detect 27% of elephant, 33% of buffalo, 57% of giraffe and 85% of carcasses. The OCC method raises the elephant population estimate to 16 681±4047 (95% cl) from the 12 722 counted in the Total Count (Z=1.917, p=.0276). These results suggest that RSO-based methods have significantly undercounted wildlife populations. To align with improved counting methods, previous results need to be re-calibrated.

Conservation management is strongly shaped by the interpretation of population trends. In the Serengeti ecosystem, Tanzania, aerial total counts indicate a striking increase in elephant abundance compared to all previous censuses. We developed a simple age-structured population model to guide interpretation of this reported increase, focusing on three possible causes: (1) in situ population growth, (2) immigration from Kenya, and (3) differences in counting methodologies over time. No single cause, nor the combination of two causes, adequately explained the observed population growth. Under the assumptions of maximum in situ growth and detection bias of 12.7% in previous censuses, conservative estimates of immigration from Kenya were between 250 and 1,450 individuals. Our results highlight the value of considering demography when drawing conclusions about the causes of population trends. The issues we illustrate apply to other species that have undergone dramatic changes in abundance, as well as many elephant populations.

Growth of the illegal wildlife trade is a key driver of biodiversity loss, with considerable research focussing on trafficking and trade, but rather less focussed on supply. Elephant poaching for ivory has driven a recent population decline in African elephants and is a typical example of illegal wildlife trade. Some of the heaviest poaching has been in Southern Tanzania's Ruaha-Rungwa ecosystem. Using data from three successive aerial surveys and modern spatial analysis techniques we identify the correlates of elephant carcasses within the ecosystem, from which important information about how poachers operate can be gleaned. Carcass density was highest close to wet-season (but not dry season) waterholes, at higher altitudes and at intermediate travel cost from villages. We found no evidence for an ecosystem-wide impact of ranger patrol locations on carcass abundance, but found strong evidence that different ranger posts showed contrasting patterns in relation to carcasses, some being significantly associated with clusters of carcasses, others showing the expected negative correlation and most showing no pattern at all. Despite a spatial change in elephant carcass locations between years, we find little evidence to suggest poachers have changed their behaviour in relation to key modelled covariates. Our maps of poaching activity can feed directly into anti-poaching control measures, but also provide general insights into how illegal harvest of high value wildlife products occurs in the field, and our spatio-temporal analysis provides a valuable analysis framework for aerial survey data from protected areas globally.

Accurate and on-demand animal population counts are the holy grail for wildlife conservation organizations throughout the world because they enable fast and responsive adaptive management policies. While the collection of image data from camera traps, satellites, and manned or unmanned aircraft has advanced significantly, the detection and identification of animals within images remains a major bottleneck since counting is primarily conducted by dedicated enumerators or citizen scientists. Recent developments in the field of computer vision suggest a potential resolution to this issue through the use of rotation-invariant object descriptors combined with machine learning algorithms. Here we implement an algorithm to detect and count wildebeest from aerial images collected in the Serengeti National Park in 2009 as part of the biennial wildebeest count. We find that the per image error rates are greater than, but comparable to, two separate human counts. For the total count, the algorithm is more accurate than both manual counts, suggesting that human counters have a tendency to systematically over or under count images. While the accuracy of the algorithm is not yet at an acceptable level for fully automatic counts, our results show this method is a promising avenue for further research and we highlight specific areas where future research should focus in order to develop fast and accurate enumeration of aerial count data. If combined with a bespoke image collection protocol, this approach may yield a fully automated wildebeest count in the near future.

Aerial surveys are vital to assessing animal populations as part of an effort to understand ecosystem health, a primary component of social and economic development in rural regions of Africa. This paper describes the design, deployment and preliminary performance results of a mobile application that provides visual real-time feedback to assist cockpit crews conducting aerial surveys.

Groups of black and white colobus monkeys, or guerezas (Colobus guereza), in a study site in the Kakamega Forest, Kenya, have declined in number from 18 groups in 1992 to 12 groups in 1998. This decline occurred largely in the eastern half of the study site, and was not offset by an increase in group size. The western half of the study site has areas next to buildings, on the edge of the forest, where guerezas supplement their diet with soil, and it has more human foot traffic which might reduce predation levels; these factors might be partially responsible for holding the number of guereza groups in the western half steady while forest degradation takes its toll on guerezas in the eastern half. This decline in guerezas is of particular concern since it took place in an area of forest that has not decreased in size and because guerezas are among the least sensitive primates to forest degradation.

Groups of black and white colobus monkeys, or guerezas (Colobus guereza), in a study site in the Kakamega Forest, Kenya, have declined in number from 18 groups in 1992 to 12 groups in 1998. This decline occurred largely in the eastern half of the study site, and was not offset by an increase in group size. The western half of the study site has areas next to buildings, on the edge of the forest, where guerezas supplement their diet with soil, and it has more human foot traffic which might reduce predation levels; these factors might be partially responsible for holding the number of guereza groups in the western half steady while forest degradation takes its toll on guerezas in the eastern half. This decline in guerezas is of particular concern since it took place in an area of forest that has not decreased in size and because guerezas are among the least sensitive primates to forest degradation.

This study examined association patterns and reproductive behaviour in a sexually monomorphic potoroid marsupial, the rufous bettong, Aepyprymnus rufescens. A total of 29 individuals was marked, and 22 of these were observed regularly over a six-month period; these 22 individuals accounted for almost all the animals using the study area. Rufous bettongs at this site were predominantly solitary (71% of sightings were of single animals) and groups, when they formed, were generally small (maximum of six individuals). Most groups of two were male-female pairs, and unisex groups occurred significantly less often than expected. These male-female groups formed as a result of sexual investigations of females by males, and were shore-lived. Analysis of the frequency with which particular males and females were seen together (excluding occasions when females were in oestrus) showed that most males did not persistently associate with any particular female. Instead, they appeared to maintain transitory contact with as many females as possible. However, three pairs were regularly seen together, suggesting that some males may maintain especially close contact with particular females. Females as they approached oestrus were followed continually by several males, with one male following very closely and preventing others from approaching. In two well-studied cases, the male who defended priority of access to the female was the same individual who had most often associated with that female when not in oestrus. These males demonstrated intimate knowledge of the nesting locations of the females and were able to join them very early each evening, and defended them against other males with little overt aggression. The mating system in this population appears to be promiscuous, but with a hint of monogamy arising from the tendency of some males to persistently investigate and ultimately to guard sexual access to certain individual females.