Using semi-global matching point clouds to estimate growing stock at the plot and stand levels: Application for a broadleaf-dominated forest in central Europe

Christoph Stepper, Christoph Straub, Hans Pretzsch

Research output: Contribution to journalArticlepeer-review

48 Scopus citations

Abstract

Dense image-based point clouds have great potential to accurately assess forest attributes such as growing stock. The objective of this study was to combine height and spectral information obtained from UltraCamXp stereo images to model the growing stock in a highly structured broadleaf-dominated forest (77.5 km2) in southern Germany. We used semi-global matching (SGM) to generate a dense point cloud and subtracted elevation values obtained from airborne laser scanner (ALS) data to compute canopy height. Sixty-seven explanatory variables were derived from the point cloud and an orthoimage for use in the model. Two different approaches — the linear regression model (lm) and the random forests model (rf) — were tested. We investigated the impact that varying amounts of training data had on model performance. Plot data from a previously acquired set of 1875 inventory plots was systematically eliminated to form three progressively less dense subsets of 937, 461, and 226 inventory plots. Model evaluation at the plot level (size: 500 m2) yielded relative root mean squared errors (RMSEs) ranging from 31.27% to 35.61% for lm and from 30.92% to 36.02% for rf. At the stand level (mean stand size: 32 ha), RMSEs from 14.76% to 15.73% for lm and from 13.87% to 14.99% for rf were achieved. Therefore, similar results were obtained from both modeling approaches. The reduction in the number of inventory plots did not considerably affect the precision. Our findings underline the potential for aerial stereo imagery in combination with ALS-based terrain heights to support forest inventory and management.

Original languageEnglish
Pages (from-to)111-123
Number of pages13
JournalCanadian Journal of Forest Research
Volume45
Issue number1
DOIs
StatePublished - 18 Aug 2014

Keywords

  • Dense image matching
  • Digital aerial images
  • Feature selection
  • Forest attribute estimation
  • Inventory design
  • Linear regression
  • Random forests
  • Sampling density
  • Wall-to-wall mapping

Fingerprint

Dive into the research topics of 'Using semi-global matching point clouds to estimate growing stock at the plot and stand levels: Application for a broadleaf-dominated forest in central Europe'. Together they form a unique fingerprint.

Cite this