This academic article introduces a novel active learning method for efficiently estimating the geographic range of a species based on limited on-site observations. The researchers model the range of an unmapped species as a weighted combination of estimated ranges from different species. They utilize models trained on large, weakly supervised, community-gathered observation data. The study highlights the value of active learning and transfer learned spatial representations for species range estimation, emphasizing the significance of large-scale crowdsourced datasets not only for modeling species range but also for active discovery.
Publication date: 3 Nov 2023
Project Page: https://arxiv.org/abs/2311.02061v1
Paper: https://arxiv.org/pdf/2311.02061
