Leaf‐wood classification of terrestrial laser scanning data with co‐registered near‐infrared photography

Abstract Due to their importance for climate change monitoring, modelling and adaptation, vegetation structural properties including leaf area index (LAI) are designated essential climate variables (ECVs) by the Global Climate Observing System (GCOS). Terrestrial laser scanning (TLS), which rapidly acquires millions of three‐dimensional point measurements representing the physical environment, is an increasingly popular method for estimating these ECVs. To assess LAI from TLS data collected during leaf‐on conditions, a fundamental requirement is the classification of points as either leaves or wood. Existing intensity‐based leaf‐wood classification methods are confounded by natural variability in the reflectance of leaves and wood, bidirectional reflectance effects and the need for complex radiometric calibration, whilst geometric methods require high point density and are known to misclassify small branches and twigs. A novel leaf‐wood classification approach is presented that avoids these issues by exploiting the spectral transmittance properties of leaves and wood, which, at near‐infrared wavelengths, demonstrate much larger differences than for reflectance. The approach relies on classification of near‐infrared images collected by a co‐registered camera integrated with the TLS instrument and can be directly applied to the whole point cloud without segmentation. The technique is applied for leaf‐on estimation of LAI and wood area index (WAI) at a deciduous broadleaf forest site, and results are benchmarked against reference values derived from leaf‐off scans. Leaf‐on estimates of LAI and WAI demonstrated a small bias (RMSD ≤0.46, bias ≤0.17) but were not significantly different from reference values at the site level. The results provide evidence of the efficacy of the approach, and its use has the potential to reduce uncertainty in ECVs critical to climate change monitoring, modelling and adaptation.