Biology subjects

Wich, S.

Publications and source records attributed to Wich, S..

4 recordsLinked to original sources

bioRxiv2026

DroneMegaDetector, A Generalised Model For Detecting Large Mammals In Aerial Imagery

Large-scale drone surveys can generate imagery across substantial geographic areas, but automated analysis remains challenging because animal appearance, target density, imaging conditions and target scale can vary considerably between surveys. We present a general-purpose RF-DETR Medium object detector developed for large-scale aerial drone imagery and evaluate its ability to generalise across taxa, environments and survey conditions. Unlike specialist counting systems that primarily return point detections, the proposed detector produces explicit bounding boxes, providing a spatial representation that can support counting as well as downstream detection, classification and tracking workflows. We evaluated the detector on an internal held-out test set and independent external datasets spanning cattle, birds, caribou and hippopotamuses, together with imagery acquired under substantially different survey conditions. On the comparative Cattle dataset, RF-DETR matched OWL across the reported point-localisation and counting metrics, while on Caribou it achieved almost identical point-based detection performance and similar aggregate counting error. Performance varied across the broader external benchmark, with reduced transfer to unseen avian imagery and differences between models across datasets. At high altitude on SAVMAP, all evaluated approaches experienced substantial performance degradation, demonstrating a shared limitation associated with small target scale. Overall, the results show that a general-purpose detector trained on heterogeneous drone imagery can provide competitive localisation and counting performance across diverse datasets while retaining explicit bounding-box outputs. The trained model weights and inference code are provided online to support reproducible large-scale aerial wildlife monitoring and further development of drone-based computer-vision workflows.

ecology↗

bioRxiv2025

Comparison of a computer vision model to a human observer in detecting African mammals in camera trap images within a safari park

Remote monitoring technologies are increasingly utilized in animal research for their capacity to enhance data collection efficiency. However, they present challenges, and as such researchers have resorted to utilizing deep learning to automatically classify acquired data therefore expediting the review process. While this practice is common in field studies it has been less adopted in zoo monitoring. In this paper we deploy the YOLOv10x model to monitor four species at Knowsley Safari in the UK: African lions (Panthera leo), Southern white rhino (Ceratotherium simum simum), Grevys zebra (Equus grevyi) and Olive baboons (Papio anubis). Camera trap images were processed and classified using the Conservation AI desktop application. The raw images were saved to facilitate the comparative analysis of the models predictions against the findings of human observed images. Processing time for both methods was compared using a subset of 3015 images with Conservation AI, reducing the time required to classify the images by 82% compared to a human analyst. Confusion matrix results showed high accuracy rates for all four species (>0.90). Analysis of count data showed significant differences in three species, where the human observer recorded more observations of each than Conservation AI (lion, rhino, baboon p<0.005). However, no significant difference was seen in zebra (p>0.05). A strong positive correlation in count data between both methodologies was seen in all species; baboon (rho=0.955, p<0.005), lion (rho = 0.969, p<0.005), rhino (rho=0.887, p<0.005) and zebra (rho=0.843, p<0.005). This study highlights the potential for these technologies as a monitoring system in zoos.

animal behavior and cognition↗

bioRxiv2020

Predicting range shifts of African apes under global change scenarios

AimModelling African great ape distribution has until now focused on current or past conditions, whilst future scenarios remain scarcely explored. Using an ensemble forecasting approach, we predicted changes in taxon-specific distribution under future scenarios of climate, land-use and human population changes. LocationSub-Saharan Africa MethodsWe compiled occurrence data on African ape populations from the IUCN A.P.E.S. database and extracted relevant human-, climate- and habitat-related predictors representing current and future (2050) conditions to predict taxon-specific distribution under a best- and a worst-case scenario, using ensemble forecasting. Given the large effect on model predictions, we further tested algorithm sensitivity by considering default and non-default modelling options. The latter included interactions between predictors and polynomial terms in correlative algorithms. ResultsThe future distributions of gorilla and bonobo populations are likely to be directly determined by climate-related variables. In contrast, future chimpanzee distribution is influenced mostly by anthropogenic variables. Both our modelling approaches produced similar model accuracy, although a slight difference in the magnitude of range change was found for Gorilla beringei beringei, G. gorilla diehli, and Pan troglodytes schweinfurthii. On average, a decline of 50% of the geographic range (non-default; or 55% default) is expected under the best scenario if no dispersal occurs (57% non-default or 58% default in worst scenario). However, new areas of suitable habitat are predicted to become available for most taxa if dispersal occurs (81% or 103% best, 93% or 91% worst, non-default and default, respectively), except for G. b. beringei. Main ConclusionsDespite the uncertainty in predicting the precise proportion of suitable habitat by 2050, both modelling approaches predict large range losses for all African apes. Thus, conservation planners urgently need to integrate land-use planning and simultaneously support conservation and climate change mitigation measures at all decision-making levels both in range countries and abroad.

ecology↗

bioRxiv2020

The importance of orangutans in small fragments for maintaining metapopulation dynamics

Orangutans (Pongo spp.) occur at low densities and therefore large areas are necessary to sustain viable metapopulations, defined here as sets of conspecific units of individuals linked by dispersal. Historically, orangutans lived in large contiguous areas of intact rainforest, but are now increasingly found in agricultural and other landscapes modified by people. Here we collate evidence of orangutans utilizing isolated forest fragments (< 500 ha) within multiple-use landscapes dominated by oil palm monoculture across Borneo. Orangutan signs (i.e. nests) were evident in 76 fragments surveyed by helicopter, and in 50 of 70 additional fragments surveyed on the ground; on average 63 ha in size. This includes presence of adult resident females with dependent young confirmed in 40% of the fragments assessed by ground survey. Our study revealed some resident females are raising offspring in isolated forest patches within mature oil palm stands. This not only confirms that some forest patches can sustain orangutans, but indicates migratory males are capable of reaching these fragments scattered throughout the multiple-use landscape. Therefore, orangutans that use or live in even small isolated forest patches are an essential part of the overall metapopulation by maintaining gene flow between, and genetic connectivity within, populations distributed across larger multiple-use landscapes. Orangutan survival is commonly thought to be low in small, isolated forest patches, and the customary management strategy is to remove (translocate) these individuals and release them in larger forests. In some cases, translocations may be necessary, i.e. in case of fire or when the animals are in eminent danger of being killed and have no other refuge. However, the small amount of data available indicates that mortality rates during and after translocations are high, while the impacts of removing animals from spatially dispersed metapopulations are unknown. Therefore, we argue the current policy of routine translocation rather than conserving the species within human-modified landscapes could inadvertently decrease critical metapopulation functionality necessary for long-term viability. It is clear that orangutans need natural forest to survive, but our findings show that fragmented agricultural landscapes can also serve as complementary conservation areas in addition to fully protected areas if they are well designed with ecological connections, and if orangutan killing can be prevented. To achieve this, we call for a paradigm shift from the traditional large single forest model to one that emphasizes metapopulation functionality in the fragmented forest - human use matrix characteristic of the Anthropocene. Version 2 (20 May 2020): this manuscript is a non-peer reviewed preprint shared via the BiorXiv server while being considered for publication in a peer-reviewed academic journal. Please refer to the permanent digital object identifier (https://doi.org/10.1101/2020.05.17.100842). Under the Creative Commons license (CC-BY Attribution-Non Commercial-No Derivatives 4.0 International) you are free to share the material as long as the authors are credited, you link to the license, and indicate if any changes have been made. You may not share the work in any way that suggests the licensor endorses you or your use. You cannot change the work in any way or use it commercially.

ecology↗