Benchmarking acquisition parameters for digital aerial photogrammetric data for forest inventory applications: Impacts of image overlap and resolution

Abstract

Listen

Translate article

Research has demonstrated the utility of digital aerial photogrammetry (DAP) for area-based predictions of forest inventory attributes. To date, studies have used DAP data acquired with a range of spatial resolutions and image overlaps. The systematic benchmarking of DAP acquisition parameters remains an outstanding research and operational gap for forest applications. While the impact of along-track overlap on point cloud metrics and area-based attribute estimates can be readily simulated, the impact of image resolution or across-track overlap requires purpose-acquired data. Moreover, although increases in along-track overlap are enabled by digital camera systems, costs for increasing across-track overlap can be substantial and may negate the cost-effectiveness of DAP for forest inventory. Hence, determining the impacts of varying acquisition parameters is of practical value for inventory programs. Researchers and practitioners have often assumed that more overlap will result in better DAP data, and that minimal overlaps often associated with historic airborne image campaigns are inadequate to support DAP processing. In our study, we found no marked difference among 15 and 20 cm spatial resolutions and overlap scenarios unless across-track overlap was reduced to 40%. Mean differences between DAP metrics and the ALS reference generally increased with decreasing overlap, and mean differences were larger for lower height percentiles (p10). Estimates of canopy height using the p90 metric varied by a root mean squared difference (RMSD) of approximately 5% between 15 cm and 20 cm datasets when along-track overlap was greater than 40%. Lower height percentiles were more strongly impacted by overlaps and resolution. Cover metrics varied by 2% RMSD across all overlap scenarios and resolutions. Comparisons between forest types (conifer, deciduous, mixed), terrain slope and aspect, and ALS-derived canopy cover were conducted to determine whether significant mean differences existed between DAP and the ALS reference. Although some significant differences were found by forest type and terrain variables, significant differences were most commonly associated with canopy cover. Based on the results reported herein, along and across-track overlaps ≥ 60% result in DAP metrics that were more similar to ALS. Increasing across-track overlap from 60% to 80% did not consistently improve the level of agreement between DAP metrics and ALS reference metrics. Conversely, DAP metrics generated using across-track overlaps <60% resulted in metrics with greater differences from the ALS reference, and a greater range of variability in DAP metric values. Image acquisitions for forest inventory must consider a broad range of factors and herein we have quantified that increasing along- or across-track overlap beyond 60% does not improve agreement with ALS area-based point cloud metrics commonly used to model forest inventory attributes. Likewise, overlap that is <60% does result in greater differences with ALS reference. Other applications beyond forest inventory may have different overlap requirements.