Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

A l'heure d'un indiscutable déclin des populations d'espèces de grande faune, la mise en place de méthodes de suivi régulières et efficaces est indispensable. En raison des difficultés liées aux procédures classiques d'inventaires pédestres et aériens, le développement de la technologie drones'avère être une alternative potentiellement valorisable. Ce travail se concentre sur le monitoring de l'hippopotame commun (Hippopotamus amphibius L.) à l'aide de cet outil novateur. La recherche d'une optimisation des paramètres de vol pour maximiser la détection et la visibilité des animaux amis en évidence une très faible diminution des taux de détection et de certitude de comptages avec la hauteur de vol ainsi qu'une forte variabilité liée aux opérateurs réalisant les observations. Il est recommandé de réaliser les vols à une hauteur de 140 mètres, entre 6h30 et 8h45 en fin de saison sèche, avec une couverture nuageuse et un vent aussi faibles que possible. Trois méthodes de correction des comptages sont proposées pour estimer l'effectif total de la population à partir des seuls animaux émergés observés. Le facteur de correction global moyen est égal à 1,252. Une tentative de description démographique d'un groupe par des mesures de longueurs d'individus a également été mise en œuvre. Elle a permis d'obtenir une répartition moyenne des hippopotames en trois classes d'âge, même si les résultats ne sont pas extrêmement précis. Enfin, un algorithme de comptage semi automatique d'hippopotames sur imagerie infrarouge thermique a été développé, et a conduit à des erreurs moyennes de +3,9%. La prise en compte des recommandations proposées dans cette étude pour la mise en place de protocoles d'inventaires permettrait de remédier aux contraintes pratiques des méthodes classiques de monitoring.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

As the Earth has entered the new Anthropocene Era due to the major human- driven environmental changes, human impact on biodiversity has triggered the planet's sixth mass extinction. Most terrestrial vertebrate populations have shown a sharp decline in abundance and range. The vast (semi-)open ecosystems of Sub- Saharan Africa have among the most incredible mammal richness in the world, including many large ungulates species. Unfortunately, many are threatened by increasing human pressures. Natural ecosystem conservation requires adaptive management, supported by key elements such as regular wildlife abundance surveys. In large areas, aerial surveys with light aircraft generally remain the best alternative for counting large mammals. However, it presents a lot of challenges inherent to plane logistics and heavy costs. In this context, technological progress such as the availability of small Unmanned Aerial Systems (UASs) can be a powerful tool to help preserve and monitor ungulate populations. UASs exhibit high spatial and temporal resolution, and are safer and often less intrusive than manned airplanes. However, the range of small affordable UASs is generally too short for large-scale application of the traditional aerial surveys. The large amount of images collected is also difficult to manage and has impacted developments. In order to enable this emerging technology so that it can become fully operational for large game counts, the aforementioned issues must be addressed. Therefore, this thesis aims to develop and implement new sampling and counting methods with small UASs in order to facilitate regular censuses of large terrestrial African ungulates. The specific objectives are: (i) to design an innovative flight plan and sampling protocol for large ungulates counts (which will be adapted to small UAS constraints, and have relevant statistical performances in regard to the current standard transect sampling method); and (ii) to investigate the potential of UAS imagery to count precisely the often disregarded, partially submerged, hippopotamus populations.First, we identified the opportunities and limits of UASs use in wildlife monitoring based on a review of the available literature (Chapter 2). We describe the range of available models and sensors used by researchers and provided evidence that most studies focused on optical imagery and used small affordable UASs for a wide range of tasks. We focused on detection possibilities and the types of survey plans performed, and the contributions towards anti-poaching surveillance. Our findings indicate that the main drawback preventing UASs from becoming an effective alternative for large-scale censuses is the generally low flight endurance ultimately limiting the area covered. We identified research gaps in terms of innovative sampling methods and the availability of appropriate statistical approaches. To address the issue of managing the large datasets produced by drone flights, and help reviewing thousands of aerial images manually, we developed a user-friendly interface called WiMUAS (chapter 3). An image viewer allows multiple observers to annotate various observations as well as compare counts. The software can generate maps of the projected observations and sampling strips based on telemetry data and payload parameters for any type of flight plan. We tested it on more than 5000 images from flights performed in Garamba National Parc savannahs. We assessed that flying at 100 m is the best compromise between resolution and surface covered to detect accurately the main medium- to large-sized antelopes. Then, we evaluated the relevance of a new sampling method adapted to small UAS limitations to census large African ungulates (chapter 4). We identified that a multiple rosettes flight plan based on the UAS operational range could be more efficient in terms of logistics. We showed by numerical simulations that four aggregated repetitions of the rosette flight plan can give accurate density estimates with a coefficient of variation under 15 % for antelope populations. However, the precision remains low compared to classic transect sampling with manned aircraft. We further identified the impact of gregariousness on density estimates by modeling population distribution for both buffalo herds and the more randomly dispersed antelopes, and concluded the lower sampling rate of the rosette design is unsuited for highly aggregative species.Second, we focused on optimizing the detection of hippos for total counts. We assessed the environmental and flight parameters influencing counts accuracy based on 252 RGB photos taken over two large well-known hippo schools (chapter 5). Eight observers reviewed the images, and the observer's experience had a significant impact on detection probability. Of environmental parameters, sun reflection on water had the worst effect on detection, with cloud cover showing a slight impact and wind speed no influence. Altitude up to 250 m did not have a significant impact on the counts, however it affected observers' confidence in their observations. We calculated correction factors that account for hippos' regular diving behavior and found it similar to previous studies. As counting individuals in dense pods proved tedious and highly impacted by the observers' personal capabilities, we used hippos as a case study to develop an algorithm for an automatic count (chapter 6). TIR imagery provided very clear and contrasting images of hippo schools at several flight heights, ranging from 38 to 155 meters above ground level. The algorithm was based on pixel value thresholding and generation of isolines and polygons, and required images to be cut to show only the group of hippos, as surrounding objects interfered with the detection. Estimated automatic counts showed very similar results to visual counts. However, hippos sometimes appeared cut in multiple polygons as they are partially submerged, which is not always addressed adequately by the algorithm.Finally, we summarized our conclusions of the main results achieved regarding new wildlife census methods with UASs (chapter 7). Following our findings, we discussed the practical implications for using UASs in the field of wildlife monitoring in general, and shared some relevant experiences and points of awareness regarding operations, especially under the challenging context of remote protected areas in tropical environments. We drew attention to the social implications of drones and underlined the importance of stringent legislation. We addressed the remaining crucial endurance limitation of the technology for large- scale wildlife aerial censuses and discussed a potential alternative with sensors mounted on ultralight aircraft to combine low costs and efficiency. Lastly, we developed perspectives to handle the large amount of data produced by drone surveys and concluded that automatic detection with machine learning will likely be one of the most important developments required for the future.