The Arctic and its associated ecosystems are undergoing rapid change in response to the effects of global climate change. Sentinel species provide a means to track the changes as they unfold. Monitoring of sentinel species, such as marine mammalscan reveal the occurrence and severity of ecological changes when they occur. Population assessments of marine mammals, in particular seals, are often conducted by counting individuals from images collected during aerial surveys using an aircraft (often piloted, but recently Unoccupied Aerial Vehicles, UAVs, are used more frequently). To turn aerial imagery into management advice requires significant effort (i.e. man-hours) to process raw images.
Our project aimed to streamline the process of converting such imagery into abundance estimates by applying state-of-the-art Computer Vision (CV) models on images collected using UAVs. We collected aerial survey data throughout Isfjorden, Van Mijenfjorden and Kongsfjorden over two consecutive seasons, and used these data to train CV models to identify ringed seals hauled out on sea ice. Our models achieved a 99% accuracy in identifying ringed seals and could process over 16,000 survey images in several hours, as opposed to weeks typically taken when manual processing is required. Our project has now generated a time series of ringed seal abundance from key west coast fjords in Svalbard, providing annual assessments between 2023-2025 and an analytical process that can generate management-relevant advice at a fraction of the cost of earlier surveys.