New satellite remote sensing and machine learning techniques offer untapped possibilities to monitor global biodiversity with unprecedented speed and precision. These efficiencies promise to reveal novel ecological insights at spatial scales which are germane to the management of populations and entire ecosystems. Here, we present a robust transferable deep learning pipeline to automatically locate and count large herds of migratory ungulates (wildebeest and zebra) in the Serengeti-Mara ecosystem using fine-resolution (38-50 cm) satellite imagery. The results achieve accurate detection of nearly 500,000 individuals across thousands of square kilometers and multiple habitat types, with an overall F1-score of 84.75% (Precision: 87.85%, Recall: 81.86%). This research demonstrates the capability of satellite remote sensing and machine learning techniques to automatically and accurately count very large populations of terrestrial mammals across a highly heterogeneous landscape. We also discuss the potential for satellite-derived species detections to advance basic understanding of animal behavior and ecology.

Wildlife counts in Africa and elsewhere are often implemented using light aircraft with ‘rear‐seat‐observer’ (RSO) counting crews. Previous research has indicated thatRSOs often fail to detect animals, and that population estimates are therefore biased. We conducted aerial wildlife surveys in Murchison Falls Protected Area, Uganda, in which we replacedRSOs with high‐definition ‘oblique camera count’ (OCC) systems. The survey area comprises forests, woodlands and grasslands. Four counts were conducted in 2015–2016 using a systematic‐reconnaissance‐flight (SRF) strip‐transect design. Camera inclination angles, focal lengths, altitude and frame interval were calibrated to provide imaged strips of known sample size on the left and right sides of the aircraft. Using digital cameras, 24 000 high‐definition images were acquired for each count, which were visually interpreted by four airphoto interpreters. We used the standard JollyII SRFanalysis to derive population estimates. OurOCCestimates of the antelopes – hartebeest, Uganda kob, waterbuck and oribi – were, respectively, 25%, 103%, 97% and 2100% higher than in the most recentRSOcount conducted in 2014. TheOCCsurveys doubled the 2014RSOestimate of 58 000 Uganda kob to over 118 000. Population size estimates of elephants and giraffes did not differ significantly. Although all fourOCCbuffalo estimates were higher than theRSOestimates – in one count by 60% – these differences were not significant due to the clumped distribution and high variation in herd sizes, resulting in imprecise estimation by sampling. We conclude thatRSOwildlife counts in Murchison have been effective in enumerating elephants and giraffe, but that many smaller species have not been well detected. We emphasize the importance of 60 years ofRSO‐based surveys across Africa, but suggest that new imaging technologies are embraced to improve accuracy.

As the need to accurately monitor key-species populations grows amid increasing pressures on global biodiversity, the counting of large mammals in savannas has traditionally relied on the Systematic-Reconnaissance-Flight (SRF) technique using light aircrafts and human observers. However, this method has limitations, including non-systematic human errors. In recent years, the Oblique-Camera-Count (OCC) approach developed in East Africa has utilized cameras to capture high-resolution imagery replicating aircraft observers’ oblique view. Whilst demonstrating that human observers have missed many animals, OCC relies on labor-intensive human interpretation of thousands of images. This study explores the potential of Deep Learning (DL) to reduce the interpretation workload associated with OCC surveys. Using oblique aerial imagery of 2.1 hectares footprint collected during an SRF-OCC survey of Queen Elizabeth Protected Area in Uganda, a DL model (HerdNet) was trained and evaluated to detect and count 12 wildlife and livestock mammal species. The model’s performance was assessed both at the animal instance-based and image-based levels, achieving accurate detection performance (F1 score of 85%) in positive images (i.e. containing animals) and reducing manual interpretation workload by 74% on a realistic dataset showing less than 10% of positive images. However, it struggled to differentiate visually related species and overestimated animal counts due to false positives generated by landscape items resembling animals. These challenges may be addressed through improved training and verification processes. The results highlight DL’s potential to semi-automate processing of aerial survey wildlife imagery, reducing manual interpretation burden. By incorporating DL models into existing counting standards, future surveys may increase sampling efforts, improve accuracy, and enhance aerial survey safety.

The Queen Elizabeth Protected Area (QEPA) hosts some 60% of Uganda’s elephants and large populations of buffalo, hippopotamus and Uganda kob. However, the area is subjected to continued poaching, livestock incursions, animal disease and invasive plant species. Population estimates derived from aerial observers have shown great variability, and therefore trends are hard to discern. The Uganda Wildlife Authority (UWA) reguires precise wildlife population estimates to guide conservation actions. The aim of this study is to provide accurate and precise baseline estimates for wildlife populations of QEPA using aerial imagery and high sampling intensity, and to improve survey methods for determining future trends. High-resolution cameras, orientated at 45°, captured images along sample strips to left and right of the aircraft. Transects at 1 km separation gave a high sampling intensity of 28% to improve precision. We captured 43000 images for visual enumeration of 13 wildlife species. Population estimates (Ŷ), standard errors (SE) and relative margins of error (RME =95% c.l/Ŷ) were determined using ‘Jolly II’, Marriott and bootstrapping methods. With improved detection using imagery the QEPA elephant population is estimated at 4 711 ± 1106 (95% c. I), which is 62% higher than the previous observer-based estimate of 2900 in 2014, and the highest since counts began in the 1960s. The survey achieved an RME for elephants of 23%, making this one of the most precise counts for any similar-sized elephant area in Africa. The buffalo, Uganda kob and hippopotamus populations have stabilized at ‘safe’ levels but remain well below the high point of the mid-1990s; the high density of bone patches indicate high mortality, attributable to disease and to the continued invasion of the unpalatable grass species Imperata cylindrica, which now covers 50% of the grassland area. Our methods indicate that methods and sample parameters prescribed by international elephant counting standards can be revised and improved using aerial cameras, especially for smaller survey areas. Short Summary Text Uganda’s Queen Elizabeth Protected Area (QEPA) continues its recovery since the decimation of wildlife by militias in the 1970s, but it is challenging to track wildlife trends due to the variability in population estimates. To improve accuracy and precision, we conducted the first aerial count of QEPA using high-resolution imaging. This approach has indicated that the elephant population is at its highest levels since surveys began in the 1960s. Meanwhile, buffalo, topi and Uganda kob are stagnating below previous 1990s levels. We suggest the use of camera systems in future surveys to improve the accuracy and precision of counts, in order to guide effective management.