Artificial Intelligence (AI) has proven powerful for utilizing the increasing amounts of remote sensing (RS) data for applications such as landcover change detection and food security monitoring. However, many AI models require a substantial amount of ground truth labels for training. Collecting these labels often involves pixel-wise image labeling or in-field sample collection, which can demand significant labor and financial resources. Due to domain shift and spatial heterogeneity among regions, models trained in one region often demonstrate poor performance when applied to other regions. Therefore, it is important to develop approaches to address labeled data scarcity and effectively train AI models for RS applications. This session welcomes studies and applications of transfer learning, self-supervised learning, and foundation models in remote sensing that enhance performance under domain shift or limited ground data. It also invites knowledge–guided approaches that reduce reliance on real-world samples, and the development of benchmark datasets.

• B - Biogeosciences