Researchers at the East China University of Technology have developed a novel approach to 3D tree modeling that significantly improves accuracy in ecological data collection. This innovative method, which combines skeleton graph optimization with fractal self-similarity, addresses common errors in existing models, such as misconnected branches and gaps resulting from incomplete data scans. Published on June 1, 2025, in the journal Plant Phenomics, the study demonstrates near-perfect accuracy in modeling trees across various tropical forest sites.
The new method was rigorously tested on 29 trees in diverse ecological environments, including locations in Peru, Indonesia, and Guyana. By utilizing LiDAR (Light Detection and Ranging) technology, the researchers gathered precise point cloud data, which was co-registered with a remarkable accuracy of 1 cm. The results revealed a concordance correlation coefficient of 0.994, indicating a superior alignment with actual tree volumes compared to widely used models like TreeQSM and AdQSM.
Current tree modeling techniques often struggle with data fragmentation and inaccurate branch connections. The limitations of these traditional approaches hinder effective large-scale forest monitoring and biodiversity assessments. Recognizing this gap, the research team aimed to develop a method capable of adapting to varying data quality while maintaining high fidelity in structural reconstructions.
The proposed model, referred to as SfQSM, was validated against destructively harvested trees, allowing researchers to calculate reliable reference values for stem volumes, buttresses, and large branches based on standard forestry formulas. This benchmarking enabled a comprehensive evaluation of the new method’s performance.
Findings showed that SfQSM produced highly accurate volume estimates, with most deviations falling within -1 m3 and 1 m3 across all sites. While smaller-diameter trees from Indonesia resulted in lower deviations, larger trees in Peru exhibited greater discrepancies, highlighting the influence of tree size on modeling precision. The quantitative analysis revealed that SfQSM achieved a mean deviation of 0.162 m3, a root mean square error of 1.023 m3, and relative errors as low as 0.01% and 0.09%. In contrast, TreeQSM produced errors more than twice as high, and AdQSM’s deviations exceeded those of SfQSM by over thirtyfold.
Visual comparisons further underscored the robustness of SfQSM. In contrast to TreeQSM, which often rendered fragmented trunks, and AdQSM, which created overfitted or non-existent branches, SfQSM consistently generated continuous and realistic models that accurately reflected point clouds.
The implications of this research extend beyond mere accuracy in tree modeling. By delivering highly precise individual tree models, SfQSM lays a foundation for critical applications in biodiversity assessments, species classification, and habitat analysis. Furthermore, accurate estimations of tree volume and biomass are essential for calculating carbon stocks and understanding forests’ contributions to the global carbon cycle.
The development of this method is expected to enhance virtual ecological landscapes and inform sustainable forestry practices, reforestation programs, and climate change mitigation strategies. The research was supported by various funding sources, including the National Key Laboratory of Uranium Resources Exploration-Mining and Nuclear Remote Sensing, as well as the National Natural Science Foundation of China.
In conclusion, the introduction of SfQSM represents a significant advancement in the field of ecological modeling. By overcoming the limitations of previous methods, this innovative approach not only enhances the accuracy of tree reconstructions but also contributes to the broader goals of environmental conservation and climate change mitigation.
