Review of Remote Sensing Data Processing Pipelines for Individual Tree Detection and Species Classification

Remote sensing data processed using rule-based and machine learning methods are widely used for Individual Tree Detection (ITD) and Tree Species Classification (TSC), yet the performance of different sensors, platforms and methods remains inconsistent across studies. The ITD and TSC tasks are interconnected, such that the performance of the detector at the ITD stage can substantially affect the subsequent TSC performance. A key factor affecting performance is the choice of data representation (2D, semi-2D or 3D), which is largely influenced by the sensor type (RGB, MSI, HSI and LiDAR) and the data acquisition platform (unmanned aerial vehicle, manned aerial or satellite). Accordingly, this review separately compares methodological configurations reported for the ITD and TSC stages, considering data source, data representation, and method family. The two tasks are treated as consecutive and methodologically related, but their quantitative performance is analyzed separately. The cross-study analysis revealed a general trend toward higher median reported ITD performance for LiDAR-based representations among the methodological groups included in the comparison. LiDAR-derived structural information may also benefit TSC when combined with spectral data. Furthermore, multispectral data appeared to provide a favorable complement to LiDAR by adding species-related spectral information associated with leaf biochemical and physiological properties.