Canadian Forest Service Publications
Estimating individual conifer seedling height using drone-based image point clouds. 2020. Castilla, G.; Filiatrault, M.; McDermid, G.J.; Gartrell, M. Forests 11(9):924.
Issued by: Northern Forestry Centre
Catalog ID: 40172
Availability: PDF (download)
Available from the Journal's Web site. †
† This site may require a fee
Research Highlights: This is the most comprehensive analysis to date of the accuracy of height estimates for individual conifer seedlings derived from drone-based image point clouds (DIPCs). We provide insights into the effects on accuracy of ground sampling distance (GSD), phenology, ground determination method, seedling size, and more. Background and Objectives: Regeneration success in disturbed forests involves costly ground surveys of tree seedlings exceeding a minimum height. Here we assess the accuracy with which conifer seedling height can be estimated using drones, and how height errors translate into counting errors in stocking surveys. Materials and Methods: We compared height estimates derived from DIPCs of different GSD (0.35 cm, 0.75 cm, and 3 cm), phenological state (leaf-on and leaf-off), and ground determination method (based on either the DIPC itself or an ancillary digital terrain model). Each set of height estimates came from data acquired in up to three linear disturbances in the boreal forest of Alberta, Canada, and included 22 to 189 surveyed seedlings, which were split into two height strata to assess two survey scenarios. Results: The best result (root mean square error (RMSE) = 24 cm; bias = −11 cm; R2 = 0.63; n = 48) was achieved for seedlings >30 cm with 0.35 cm GSD in leaf-off conditions and ground elevation from the DIPC. The second-best result had the same GSD and ground method but was leaf-on and not significantly different from the first. Results for seedlings ≤30 cm were unreliable (nil R2). Height estimates derived from manual softcopy interpretation were similar to the corresponding DIPC results. Height estimation errors hardly affected seedling counting errors (best balance was 8% omission and 6% commission). Accuracy and correlation were stronger at finer GSDs and improved with seedling size. Conclusions: Millimetric (GSD
Plain Language Summary
Monitoring regeneration success in disturbed forests involves costly ground surveys of tree seedlings taller than a minimum height. Could drones replace boots on the ground for this? Drones have been shown to detect conifer seedlings reliably, but how good are they at estimating their height? This study employed three drones to estimate the height of nearly 200 conifer seedlings growing in linear disturbances in Alberta’s boreal forest. Each seedling had up to 15 height estimates automatically derived from drone-based image point clouds (3D models of the seedlings made of tiny points extracted from overlapping overhead photos). The best results were achieved for seedlings taller than 30 cm based on photos where the size of a pixel on the ground was less than a centimetre; the height estimation error was 5 cm or less for half of the seedlings. Results for seedlings shorter than 30cm were unreliable due to the presence of nearby short vegetation. Height estimation errors led to few seedling counting errors (counting a seedling shorter than the minimum height, or vice versa). This study is the first to demonstrate that the height of an individual conifer seedling can be accurately measured from a drone.