Advances in tree species identification from high-resolution aerial imagery and deep learning

Tree species diversity shapes forest functioning, carbon storage, and ecosystem resilience, yet species-level inventories remain limited outside local studies. High-resolution aerial imagery and deep learning now enable individual tree crowns to be mapped at high spatial detail, offering new pathways for biodiversity and climate impact assessments. We synthesize 103 studies (2017–2024), representing 671 deep learning–based species identification tasks using aerial imagery and associated multimodal data across ecosystems and sensor types. Current research is regionally imbalanced and relies heavily on multi-stage classification workflows, with sparse use of multi-temporal or multimodal inputs in end-to-end workflows. We identify core methodological gaps and highlight the need for cross-biome standardized curation, multimodal fusion, automated workflows, and advanced model architectures to achieve scalable species mapping.