Search

Ecological data are collected over vast geographic areas using digital sensors such as camera traps and bioacoustic recorders. Camera traps have become the standard method for surveying many terrestrial mammals and birds, but camera trap arrays often generate millions of images that are time-consuming to label. This causes significant latency between data collection and subsequent inference, which impedes conservation at a time of ecological crisis. Machine learning algorithms have been developed to improve the speed of labelling camera trap data, but it is uncertain how the outputs of these models can be used in ecological analyses without secondary validation by a human. Here, we present our approach to developing, testing and applying a machine learning model to camera trap data for the purpose of achieving fully automated ecological analyses. As a case-study, we built a model to classify 26 Central African forest mammal and bird species (or groups). The model generalizes to new spatially and temporally independent data (n = 227 camera stations, n = 23,868 images), and outperforms humans in several respects (e.g. detecting ‘invisible’ animals). We demonstrate how ecologists can evaluate a machine learning model's precision and accuracy in an ecological context by comparing species richness, activity patterns (n = 4 species tested) and occupancy (n = 4 species tested) derived from machine learning labels with the same estimates derived from expert labels. Results show that fully automated species labels can be equivalent to expert labels when calculating species richness, activity patterns (n = 4 species tested) and estimating occupancy (n = 3 of 4 species tested) in a large, completely out-of-sample test dataset. Simple thresholding using the Softmax values (i.e. excluding ‘uncertain’ labels) improved the model's performance when calculating activity patterns and estimating occupancy but did not improve estimates of species richness. We conclude that, with adequate testing and evaluation in an ecological context, a machine learning model can generate labels for direct use in ecological analyses without the need for manual validation. We provide the user-community with a multi-platform, multi-language graphical user interface that can be used to run our model offline.

Ecological data are increasingly collected over vast geographic areas using arrays of digital sensors. Camera trap arrays have become the ‘gold standard’ method for surveying many terrestrial mammals and birds, but these arrays often generate millions of images that are challenging to process. This causes significant latency between data collection and subsequent inference, which can impede conservation at a time of ecological crisis. To address this, machine learning algorithms have been developed to improve data processing speeds, but these models are not considered accurate enough for fully automated labeling. Here, we present a new approach to building and testing a high performance machine learning model for fully automated labeling of camera trap images. As a case-study, the model classifies 26 Central African forest mammal and bird species (or groups). The model was trained on a relatively small dataset ( c .300,000 images) but generalizes to fully independent data and outperforms humans in several respects (e.g. detecting ‘invisible’ animals). We show how the model’s precision and accuracy can be evaluated in an ecological modeling context by comparing species richness, activity patterns and occupancy derived from machine learning labels with the same estimates derived from expert labels. Results show that fully automated labels can be equivalent to expert labels when calculating these widely-used ecological metrics. We provide the user-community with a multi-platform user interface for running the model offline, and conclude that high performance machine learning models can fully automate labeling of camera trap data. Significance statement Large-scale ecological monitoring can be used to detect ecosystem change. Ecological sensors such as camera traps are deployed across large spatial and temporal scales to monitor species and communities. Camera trap data are often vast (millions of images) and manual processing times cause significant latency between data collection and ecological inference. Existing machine learning models can reduce processing times but are rarely used in fully automated workflows for ecological analyses, mainly because users lack confidence in the model’s precision and accuracy. Here, we show a new, high performance machine learning model can be used to make ecological inference that is equivalent to using manually generated, expert labels. These results pave the way for large-scale, fully automated biodiversity monitoring and forecasting using camera trap arrays. ### Competing Interest Statement The authors have declared no competing interest.

To develop effective conservation planning and mitigate biodiversity loss, standardized metrics for monitoring and assessing biodiversity are needed. This information is particularly vital in Botswana, where knowledge of many of the diverse wildlife populations is lacking. To address this knowledge gap, my dissertation research evaluated the distributions, densities, and ecology of the wildlife community in northern Botswana, with a focus on terrestrial carnivore species given their importance both ecologically and economically. My objectives were threefold: 1) estimate the distributions of the mammal community (n = 44 species) and evaluate community, group, and species-specific responses to anthropogenic and environmental variables, 2) test whether the presence of intraguild species or resource availability had a larger influence on the seasonal distributions of carnivore species, and 3) simultaneously estimate the population densities of 7 carnivore species. To accomplish these objectives, I completed a multi-year camera trap survey in a 1,154-km2 study area and analyzed the data using occupancy models (single and multi- species) and spatially explicit capture-recapture models. Estimates of species richness ranged from 8 to 27 unique species, species had a mean occurrence probability of 0.32 (95% credible interval = 0.21–0.45), and estimated densities ranged from 1.8 aardwolves (Proteles cristata) to 12.7 spotted hyenas (Crocuta crocuta) per 100 km2. The occupancy and richness of terrestrial mammals was negatively associated with human disturbance and in general, positively associated with open grasslands/floodplains. Carnivore species, specifically, tended to have greater species richness and larger population densities in open habitats than in closed. I also found carnivore distributions were positively associated with the detection rate of intraguild competitors and predators, suggesting competitor/predator avoidance did not play a large role in shaping carnivore community structure. My research highlights the pivotal role protected areas and grasslands play in conserving wildlife populations in northern Botswana. Additionally, my research helps progress camera trap analyses from single to multi-species assessments. Broader application of this multi-species approach would likely result in a better understanding of wildlife and carnivore communities which in turn, may help inform management actions aimed at addressing the loss of wildlife populations globally. Ph. D.