Remote sensing and deep learning for standing dead-tree detection and mapping: a review of advances, challenges, and future directions

Standing dead trees are visible indicators of tree mortality and an important transitional component linking forest disturbance to future lying deadwood, habitat availability, and carbon storage. As drought, insect outbreaks, pathogens, and climate extremes intensify tree mortality worldwide, scalable methods are needed to detect and map standing dead trees consistently across forest landscapes. Recent advances in high-resolution remote sensing and computer-vision-based deep learning, including object detection, semantic segmentation, instance segmentation, and transformer-based models, have increased the automation and spatial detail of standing dead-tree detection and mapping. This critical narrative review synthesizes 75 studies published or publicly released between 2019 and July 2026 that apply deep learning to remotely sensed data for this purpose. The synthesis shows that high-resolution RGB and extended optical imagery dominate the evidence base. Satellite applications generally support broader observations at coarser target scales, including temporal, fractional, and upscaling analyses. Field-supported validation remains limited, with most studies relying primarily on image-based reference data. Object detection is the most frequently used primary model category, while FCN/U-Net models remain prominent for pixel-level semantic segmentation. Multi-sensor integration can improve robustness where spectral and structural information are complementary, but transferability across sites, biomes, sensors, and acquisition conditions remains weakly demonstrated. Operational deployment is further constrained by limited repeat coverage of very-high-resolution acquisitions, canopy occlusion, class imbalance, inconsistent target definitions and annotation protocols, and uncertain model generalization. Progress toward scalable and reliable monitoring will require shared and transparently documented benchmark datasets, harmonized target definitions and evaluation protocols, transferable learning strategies, and independent cross-site, cross-biome, and cross-sensor validation. By linking ecological monitoring needs with methodological advances, this review identifies priorities for more comparable, transferable, and operationally relevant standing dead-tree mapping.