Diameter distribution estimation with laser scanning based multisource single tree inventory

Tree detection and tree species recognition are bottlenecks of the airborne remote sensing-based single tree inventories. The effect of these factors in forest attribute estimation can be reduced if airborne measurements are aided with tree mapping information that is collected from the ground. The main objective here was to demonstrate the use of terrestrial laser scanning-derived (TLS) tree maps in aiding airborne laser scanning-based (ALS) single tree inventory (multisource single tree inventory, MS-STI) and its capability in predicting diameter distribution in various forest conditions. Automatic measurement of TLS point clouds provided the tree maps and the required reference information from the tree attributes. The study area was located in Evo, Finland, and the reference data was acquired from 27 different sample plots with varying forest conditions. The workflow of MS-STI included: (1) creation of automatic tree map from TLS point clouds, (2) automatic diameter at breast height (DBH) measurement from TLS point clouds, (3) individual tree detection (ITD) based on ALS, (4) matching the ITD segments to the field-measured reference, (5) ALS point cloud metric extraction from the single tree segments and (6) DBH estimation based on the derived metrics. MS-STI proved to be accurate and efficient method for DBH estimation and predicting diameter distribution. The overall accuracy (root mean squared error, RMSE) of the DBH was 36.9 mm. Results showed that the DBH accuracy decreased if the tree density (trees/ha) increased. The highest accuracies were found in old-growth forests (tree densities less than 500 stems/ha). MS-STI resulted in the best accuracies regarding Norway spruce (Picea abies (L.) H. Karst.)-dominated forests (RMSE of 29.9 mm). Diameter distributions were predicted with low error indices, thereby resulting in a good fit compared to the reference. Based on the results, diameter distribution estimation with MS-STI is highly dependent on the forest structure and the accuracy of the tree maps that are used. The most important development step in the future for the MS-STI and automatic measurements of the TLS point cloud is to develop tree species recognition methods and further develop tree detection techniques. The possibility of using MLS or harvester data as a basis for the required tree maps should also be assessed in the future.