Large-scale tree-level mapping of forest structure including species type with remote sensing data and ground measurements

Remote-sensing based tree maps can be used to calculate various diversity indices, but the detection probability of trees depends on size and species. We propose a novel approach combining individual tree detection (ITD) with resampling corrections (+R) which aims to simultaneously correct the size, species, and spatial distribution of trees using scalable algorithms. Using airborne laser scanning, optical data, and ground measurements, we demonstrate the compatibility of ITD+R with two different types of forests and ITD algorithms, as well as its scalability to areas exceeding 3000 km 2 . The tree maps were evaluated using plot-level variables and benchmarked against area-based k nearest neighbors ( k -NN). The ITD+R improved ITD results for most studied metrics, with the Shannon index being an exception, and even outperformed k -NN in predicting dominant height in managed stands, though k -NN still outperformed for stem density and volume. The ITD+R approach was shown to be adaptable to various diversity indices which it has not been specifically trained on, with 254 m 2 plot-level predictions correlating at r =0.42–0.91. While ITD trees could be classified with OA=82.0%–86.6% to pine, spruce, and deciduous, further research is needed to account for rare tree species, as low prevalence results in a large number of false detections which cannot be sufficiently addressed with classification alone. • Structurally representative maps were created using RS data and ground truth. • Species, DBH, height, volume, and location were predicted for each tree. • Height, DBH, and volume distributions by species were accurately reproduced. • The created maps can be reliably used to calculate biodiversity proxies.