Two-phase forest inventory using very-high-resolution laser scanning

Abstract

In this study, we compared a two-phase laser-scanning-based forest inventory of stands versus a traditional field inventory using sample plots. The two approaches were used to estimate stem volume (VOL), Lorey's mean height (HL), Lorey's stem diameter (DL), and VOL per tree species in a study area in Sweden. The estimates were compared at the stand level with the harvested reference values obtained using a forest harvester. In the first phase, a helicopter acquired airborne laser scanning (ALS) data with >500 points/m2 along 50-m wide strips across the stands. These strips intersected systematic plots in phase two, where terrestrial laser scanning (TLS) was used to model DL for individual trees. In total, phase two included 99 plots across 10 boreal forest stands in Sweden (lat 62.9° N, long 16.9° E). The single trees were segmented in both the ALS and TLS data and linked to each other. The very-high-resolution ALS data enabled us to directly measure tree heights and also classify tree species using a convolutional neural network. Stem volume was predicted from the predicted DBH and the estimated height, using national models, and aggregated at the stand level. The study demonstrates a workflow to derive forest variables and stand-level statistics that has potential to replace many manual field inventories thanks to its time efficiency and improved accuracy. To evaluate the inventories, we estimated bias, RMSE, and precision, expressed as standard error. The laser-scanning-based inventory provided estimates with an accuracy considerably higher than the field inventory. The RMSE was 17 m3/ha (7.24%), 0.9 m (5.63%), and 16 mm (5.99%) for VOL, HL, and DL respectively. The tree species classification was generally successful and improved the three species-specific VOL estimates by 9% to 74%, compared to field estimates. In conclusion, the demonstrated laser-scanning-based inventory shows potential to replace some future forest inventories, thanks to the increased accuracy demonstrated empirically in the Swedish forest study area.