Abstract
Forest inventory operations have greatly benefitted from remotely sensed data particularly airborne laser scanning (ALS) which has become a popular technology choice for large-area forest inventories. For remote regions, for fragmented estates or for single stand-level inventories ALS may be unsuitable because of the high cost of data acquisition. Point cloud data generated from digital aerial photography (DAP) is emerging as a cost-effective alternative to ALS. In this study we compared area-based forest inventory attributes derived from point cloud datasets sourced from AIS, small-format and medium-format digital aerial photography (SFP and MFP). Non-parametric modelling approach, namely RandomForest, was employed to model forest structural attributes at both plot- and stand-levels. The results were evaluated using field data collected at 105 inventory plots. At plot-level, the maximum difference among relative RMSEs of basal area (B-top), top height stocking (N) and total stem volume (TSV) of the three datasets was 2.46%, 0.55%, 13.29% and 2.53%, respectively. At stand-level, the maximum difference among relative RMSEs of BA, H-top, N and TSV of the three datasets was 3.86%, 1.25%, 7.85% and 6.04%, respectively. This study demonstrates the robustness of DAP across different sensors, and thus informs forest managers planning data acquisition solutions to best suit their operational needs.
NSTL
Elsevier