Detekcja pojedynczych martwych świerków w Puszczy Białowieskiej na podstawie integracji danych satelitarnych z danymi lotniczego skanowania laserowego

The Białowieża Forest, a unique lowland forest complex in Central Europe, has undergone significant ecological changes in recent years, primarily due to a massive outbreak of the spruce bark beetle (Ips typographus L.). This outbreak, which peaked in 2016, caused extensive Norway spruce (Picea abies (L.) H. Karst) mortality, leading to major changes in forest structure and functioning. Ground-based forest inventory methods remain a fundamental and accurate source of ecological information. However, they are timeconsuming and relatively expensive, particularly when applied across large or inaccessible areas such as protected areas or wetlands. For this reason, alternative approaches are increasingly being explored – especially those leveraging remote sensing techniques, which allow for faster, spatially consistent and repeatable forest condition assessments. Satellite imagery and airborne laser scanning (ALS) data have proven to be powerful tools for monitoring forest condition at different spatial scales. This study investigates the feasibility of integrating high-resolution multispectral satellite imagery (Pléiades, 2 m spatial resolution) with ALS data (11 points/m²) to detect and map individual dead spruce trees in the Polish part of the Białowieża Forest. The methodological framework (Fig. 2) included pre-processing steps such as cloud masking, shadow elimination, canopy gap removal and segmentation of individual tree crowns from ALS data. These procedures aimed to enhance the classification accuracy and spatial alignment of the datasets. A supervised maximum likelihood classification was applied to the satellite imagery, distinguishing three vegetation classes: dead trees, conifers and broadleaves. The classification resulted in a high accuracy for dead trees (producer’s accuracy: 96.8%, user’s accuracy: 94.6%; Tab. 4, Fig. 3). A multistep integration method was used to detect individual dead trees by correlating spectrally classified pixels with tree crown segments derived from ALS data. Each crown was assessed for spatial proximity and spectral coverage of the classified dead pixels. By applying optimized thresholds for proximity and coverage, the algorithm identified individual dead spruce trees with a detection accuracy of 71.1% – true positives and a false positive rate of 22.0% (Tab. 9), validated against manually interpreted reference data. The final detection map indicated over 381,000 individual dead spruce trees in the forest area (Tab. 8, Fig. 5). The integration of ALS and satellite data proved to be effective, especially in forest compartments with lower structural and species complexity. However, limitations were observed in areas such as the Białowieża National Park, where diverse canopy structures, species heterogeneity and occlusion effects significantly reduced detection rates. In such areas, the satellite pixels often represent mixed spectral signals, which reduces the reliability of the classification for individual tree identification. The study emphasizes the utility of combining complementary remote sensing datasets to enable cost-effective, large-scale monitoring of forest health. It also highlights the temporal constraints of integrating data collected in different years, as forest conditions may change due to logging or vegetation regrowth. Although data mismatches can lead to errors, these can be mitigated by excluding known logged areas. This research contributes to the development of practical approaches for operational forest monitoring and supports sustainable forest management under increasing ecological pressures. The proposed method, when adapted to specific forest conditions, can improve forest inventory systems. Future work should explore improvements in tree crown segmentation, the use of multi-temporal satellite data and the integration of additional spectral indices or Deep Learning-based classification algorithms.