Thermal imaging is able to detect animals by their thermal signature according to the contrast between the body temperature and the environment, which might reach 30–40 °С. Therefore, winter surveys are more efficient. Unfortunately, the method cannot tell the difference between species of similar mass and shape, e.g. a wolf and a wild boar. Visual analysis of conventional photo and video made it possible to identify animals with the same thermal signature. The simultaneous use of photo and thermal imaging improves the accuracy and reliability of aerial surveys. Figure 2S provides an example of such an analysis. The left image shows three types of objects: the white of the snow background, the numerous translucent round crowns of naked trees and shrubs, and the dark round crowns of coniferous trees. Under this resolution, the patchy background of winter taiga makes it hard to detect heat signatures in the photo image: the silhouettes of trees and shrubs obscure the contours of the animals. However, the thermal image on the right clearly shows the signatures of two elks as their body temperature differs significantly from the fairly uniform temperature background of the winter taiga.
In 2019, it took two flights to survey the territory. Figure 3S (a) shows the flight routes. The dots indicate the centers where RGB images were taken. In 2020, we launched one flight, its route shown in Figure 3S (b). To facilitate the comparison, we placed a fan-shape of glades into the bottom left corner of both Figure 3S (a) and Figure 3S (b). The glades are the system of ski slopes of the Tanay ski resort. Table 4 demonstrates the basic technical information on the flights.
We developed the following algorithm to process the images obtained from the drone planes:
1) We sequenced the infrared video with an interval of ~ 0.6 s.
2) After that, the infrared images were processed using software according to the degree of color intensity and pixel clusters. As a result, we obtained numerous infrared images with thermal extremes, which indicated an object with a higher temperature than that of the snow, e.g. an animal, a human, or a car.
3) We uploaded the RGB photos and telemetry into the Agisoft Metashape Professional software for alignment.
4) The infrared images underwent a visual inspection for the initial screening of “junk” data.
5) The coordinates of the infrared images with extremes were compared in-camera with the aligned RGB photographs, and the presence of large game was determined visually.
6) Finally, we compared the research results at different stages.
Figure 2 illustrates an example of comparing images in the visible and IR spectra. The low-resolution infrared image (Figure 2a, right) shows two thermal signatures. However, the photo image with a similar resolution (Figure 2a, right) provides no reliable identification of the signatures. When the resolution was increased, the body contours of two elks became visible – see the red frame in the photo image (Figure 2b, left).
The analysis employed software developed by the Kemerovo State University which allows jpeg and png image processing. The processing time depended on the number of images: it took the program 25-50 s to process materials of one standard UAV flight that lasted 100-150 min. The software allows for a thermal sensitivity that exceeds the capabilities of a human observer. Taking into consideration the limited flight time, this made it possible to detect even weak thermal anomalies. Figure 4S gives a comparative analysis of the processed results for infrared images taken from a height of 200 m and 400 m. Figure 4S shows a thermal signature that is clearly visible to the human eye. The shot was made from a height of 200 m. When the same area was shot from 400 m, the same thermal signature was almost indistinguishable to the human eye, while the software application was able to detect it.
The survey of 2019 detected 34 objects (numbers 1-34). Figure 3 shows their spatial distribution. Out of 34 objects, numbers 1-25 are elks. Table 1S shows the coordinates of the animals detected by the drone planes in 2019. The detected objects (34) also included untargeted objects not related to wild animals, e.g. a human person and a group of animals contained in the rehabilitation center of the Tanay ski resort.
The Tanai resort caused too many “false positives”. As a result, the contour of the scanning section had to be changed in 2020 to exclude the Tanai resort premises. Figure 4 demonstrates the ratio of the scanned areas in 2019 and 2020. The survey of 2020 revealed 63 objects, of which 55 were elks. Figure 5 shows their spatial distribution. Table 2S specifies the coordinates of the animals detected by the UAV survey in 2020. We failed to calculate the coordinates of numbers 20, 21, and 22 on the RGB images as these objects were too close to the frame. The remaining objects (8) were people. Figure 5S demonstrates a test snapshot of untargeted search objects – some random fishermen that happened to be in the area.
The map of elk distribution (Figure 5) shows two clusters, the largest one being Group 2, which included 15 elks. Figure 6 demonstrates the maximum number of animals recorded in one RGB image – 11 elks. The maximum number of animals fixed in one infrared image was 5 elks (Figure 7). This difference resulted from the different technical characteristics of the thermal infrared camera and the visible spectrum equipment. The bandwidth of the infrared image was 1/3 in the center of the width of the visible spectrum. In Figure 6, the outline of the infrared image is blue. Thus, the shooting area of the infrared image was approximately nine times smaller than the shooting area of the RGB image. All the routes were planned specifically to achieve a transverse overlap of 10%-15% for infrared imaging, in which case the overlap of RGB images was 70%. Figure 7 compares RGB and infrared images of the same surface areas in Group 2. It becomes clear that the distance between the elks was about 50 m
A comparative analysis of the data obtained in 2019 and 2020 revealed that 25 and 55 elks were identified in 2019 and 2020, respectively, where the two studied areas overlapped (15.9 km2). Therefore, the number of animals within the same habitat almost doubled. Figure 8 shows their spatial distribution. The survey of 2020 revealed two clusters of elks. Such uneven distribution could be explained by some behavioral characteristics of the animals. We detected two wolves during the visual analysis of the images obtained in 2020 and the corresponding infrared images with thermal anomalies over an area of 7 km2 (Figure 6S). It was the first time wolves had been detected in the Salair Nature Reserve. No traditional track counts had ever revealed wolves on this territory, and naturalists had always considered the park a wolf-free zone. According to daily track counts, the wolf population had almost disappeared in the Kemerovo region by 2015-2017 as a result of man-induced factors: rangers reported only accidental visits from the neighboring regions [41,42]. Thus, the developed method of digital survey provided a more complete identification of large animals in the given habitat. Accuracy is especially important for monitoring the population of such large predators as wolves. Mistakes can have an extremely negative effect on the managing populations of herbivores.
During the aerial surveys of 2019-2020, the population, distribution, and habitat of the European elk proved to correspond to the data obtained by the traditional method of winter track counts submitted by the Department of Animal Object Protection of the Kemerovo Region. We registered a significant increase in the elk population in the forests of the Salair Ridge. In addition, we detected wolves in the surveyed area. The research justified the combined use of various digital technologies for game animal survey, i.e. photo and thermal imaging. The equipment performance was good even in the harsh winter conditions, which means great prospects for research on larger areas.