Evaluating superpixel algorithms for standing dead tree delineation using aerial orthoimagery

• Adaptels achieved the highest accuracy in standing dead tree crown delineation. • CIELAB color space improved geometric fidelity of delineated crowns. • RGB imagery yielded the strongest detection scores. • Superpixel algorithms performance remained consistent across 2017–2022 imagery. • Reference database of 1,200 SDTs crowns supports future DL-based detection. High-resolution remote sensing data are essential for monitoring forest health and detecting changes such as tree mortality. This study evaluates low-level segmentation methods for delineating standing dead trees (SDTs) using widely available true-color (RGB) aerial orthoimagery, independent of near-infrared (NIR) or LiDAR data. The analysis was conducted across eight forest sites in northern Poland, dominated by coniferous species such as Scots pine ( Pinus sylvestris L.) and Norway spruce ( Picea abies L.). Four representative superpixel-based algorithms were tested — Simple Linear Iterative Clustering (SLIC), its zero-parameter variant (SLIC0), scale-adaptive superpixels (adaptels), and a spatially regularized watershed transform (waterpixels). All methods represent preprocessing approaches designed to reduce image complexity and preserve meaningful spectral–spatial structures prior to object-based image analysis (OBIA). In addition, the impact of converting imagery to the perceptually uniform CIELAB color space was assessed to enhance spectral separability and reduce illumination effects. Segmentation accuracy was evaluated against a manually verified reference dataset of 1,200 SDT crowns using multiple quality metrics. The results indicate that the adaptels algorithm, particularly when combined with the CIELAB color transformation, achieved the most balanced performance across all evaluation metrics, defined by a simultaneous reduction of segmentation fragmentation and boundary generalization errors while maintaining high overall detection accuracy. This combination proved to be an efficient and cost-effective solution for SDT segmentation using standard RGB orthophotos. The findings highlight the potential of perceptually uniform color transformations as practical tools for scalable, reproducible, and low-cost forest monitoring. The study also provides a reference database of standing dead trees to support further research and future integration with deep learning-based detection frameworks.