Replicability of Digital Terrain Models (DTMs) and Canopy Height Models (CHMs) derived from drone photogrammetry is important to understand the extent to which time-series are exposed to methodological noise and conceal real environmental changes. Root mean square error (RMSE) distribution metrics (median/IQR) were used as indicators of replicability across seven drone survey setups, three dense matching scales, and 13 ground point filters in a challenging shrubland environment (total of 273 DTMs and CHMs). We conclude that methodological effects have considerable potential to negatively affect replicability. A power-law relationship between point cloud density and dense matching resolution suggested that important dense matching resolution thresholds exist beyond which replicability degrades considerably. For our Arctic study area, replicability of DTMs (median ± 0.1 m RMSE Vegetated Vertical Accuracy) and CHMs (within ±0.05 m of true site-level heights) is most likely when source imagery is collected with ≤1.5 cm spatial resolution and side-lap of >80%, and if classified point clouds are generated using full-scale dense matching and Triangular Irregular Network filtering. Negative biases for maximum shrub height estimates increased from 4–9% to 14–50% with coarser imagery. We advocate for increased attention to drone-derived model replicability to separate real environmental changes from noise during a period of rapid ecological and geomorphic change.
Sluijs et al. (Tue,) studied this question.