Application of optical unmanned aerial vehicle-based imagery for the inventory of natural regeneration and standing deadwood in post-disturbed spruce forests

Abstract

ABSTRACTDisturbances caused by the European spruce bark beetle (Ips typographus L.) infestations are amongst the main drivers of forest ecosystem dynamics in stands dominated by Norway spruce (Picea abies [L.] Karst.). Monitoring the post-disturbance stand development including establishment of the new tree cohorts (regeneration) is of particular importance, and is conventionally done by time-intensive field surveys. Efficiency of techniques such as airborne light detection and ranging (LiDAR) or stereo photogrammetry is constrained due to their quality or costs in small-scaled and substantially heterogeneous post-disturbed areas. Small, multi-rotor unmanned aerial vehicles (UAVs) offer alternatives via their lower cost, temporal flexibility and high spatial resolution. We investigated the Digital Surface Models (DSM) derived from the UAV for inventories in post-disturbed sites of the Bavarian Forest National Park, Germany. We compared the numbers and structural attributes of detected living trees and snags from UAV data with standard stereo aerial photogrammetry using conventional field survey as a reference. Moreover, we processed the UAV data both by manual and automated tree recognitions. The results differentiated for individual tree classes (Living Spruce/Standing Deadwood and Individual/Grouped) showed varying performance with best results achieved for Standing Deadwood of moderate height. The UAV products were superior to aerial photography for the height retrieval: UAV-based data showed in average the root mean square error (RMSE) = 1.56 m, coefficient of determination R2 = 0.74 and bias = −0.73 m, compared to the aerial photogrammetry RMSE = 2.71 m, R2 = 0.17 and bias = −1.27 m. In particular, the heights of tall snags were more biased. Furthermore, the UAV data provided good results in crown diameter determination (RMSE = 0.13 m, R2 = 0.87, bias = 0.05 m). The automated recognition method was associated with qualitative and quantitative drawbacks compared to the manual method. Detection rates for trees and regeneration growing individually (60.7% and 39.1% for manual and automated method, respectively) were higher compared to regeneration in groups (28.6% and 17.8%). To conclude, the UAV-based inventory has clear advantages over aerial photogrammetry, especially for inventory of sites dominated by larger individual trees with sparse understorey. However, it cannot fully replace the field survey in post-disturbed sites with dense regeneration, where the performance can be augmented by combining UAVs with reduced fieldwork in different stand structural classes.