QSM - Quantitative Structure Model

Previous work

  • Compact tree model with topology and geometry.
  • Branching structure &
    branching order.
  • Volumes, lengths, angles, etc.
  • Hierarchical collection of cylinders.

Single tree reconstruction

Previous work

  • TLS from a few positions around tree.
  • Manually extract single tree.
  • Reconstruct a QSM.
  • Total biomass from QSMs overestimated by 9.7% while local allometric equations show an underestimation between 29.9% and 36.6%. (65 sampled Eucalyptus trees, Calders et al. 2015)

Motivation for plot-level reconstruction

  • Accurate biomass estimates.
  • Ground truth data for airborne measurements.
  • Cheap measurements.
  • Biomass estimation non-destructively.

Forest plot reconstruction

  • Perform TLS from several locations inside a forest plot.
  • Register to single point cloud.
  • Plot-level: tree extraction
    • Remove ground and under-growth.
    • Automatically extract trees based on segmentation.
  • Tree-level reconstruction
    • Segment into branches.
    • Reconstruct each branch with multiple cylinders.

Filtering

Algorithm details

  • Remove points outside plot.
  • Remove some ground and under-story points.
  • Remove low point density regions and small separate clusters.
  • Make high point density regions sparser.
  • Large reduction in point count.
Unfiltered
Filtered

Stem extraction

Algorithm details

  • Partition the point cloud into small subsets corresponding surface patches in tree and other surfaces.
  • Use the patches and heuristics to locate the stems.
    • Patch surface normal approximately horizontal.
    • Large patch components define the stems.
Blue: horizontal normal patches, Red: other
Final stem extraction

Tree extraction

Algorithm details

  • Segment the patches into stems and branches.
  • Restrict the trees inside the plot.
Initial segmentation
Final tree extraction

Test case: Oak plot

  • South-Eastern England
  • Species: Quercus robur L.
  • Five scans, leaf-off
  • Leica HDS-6100
  • Reference biomass from
    allometric equation.

Test case: Eucalyptus plot

  • Victoria, Australia
  • Species: Eucalyptus leucoxylon,
    E. microcarpa and E. tricarpa.
  • RIEGL VZ-400
  • Reference biomass
    from destructive
    harvesting of 27 trees.

Test cases

Oak plotEucalyptus plot
Plot radius 15 m40 m
Number of scans 55
Resolution 0.036°0.060°
Leaves offon
Millions of points 12471
After filtering 3533

Results

2005001,0001,5002,0002,500Reference mass (kg)2005001,0001,5002,0002,5003,000Modeled mass (kg)OakEucalyptus

Results

Oak plotEucalyptus plot
Number of trees 15120 / 27
Modeling time (1 x QSM) 100 min160 min
Modeling time (5 x QSM) 240 min540 min
Average error per tree 24.0%28.5%
Total error per plot 15.0% 8.5%

Computations: MacBook Pro 2.8GHz, 16GB, Matlab.

Conclusion

  • Fast and accurate massive scale tree modelling is possible from TLS data.
    • Tree extraction and modelling in hours.
    • Total plot level biomass were overestimated by 8% and 15% in the test cases.
  • The method seems to be robust.
    • Two different plots with different species, scanners and resolutions.
    • Parameters tested for sensitivity.
  • Multi-hectare forest regions possible.
14 scans, 1000+ trees, automatically extracted and reconstructed.

Questions?

References

Åkerblom2015
Åkerblom, M., Raumonen, P., Kaasalainen, M., Casella, E. Analysis of geometric primitives in quantitative structure models of tree stems 2015 Remote Sensing Submitted
Calders2015
Calders, K., Newnham, G., Burt, A., Murphy, S., Raumonen, P., Herold, M., Culvenor, D., Avitabile, V., Disney, M., Armston, J., Kaasalainen, M. Non-destructive estimates of above-ground biomass using terrestrial laser scanning 2015 Ecology and Evolution 6(2) 198-208
Raumonen2013
Raumonen, P., Kaasalainen, M., Åkerblom, M., Kaasalainen, S., Kaartinen, H., Vastaranta, M., Holopainen, M., Disney, M., Lewis, P. Fast Automatic Precision Tree Models from Terrestrial Laser Scanner Data 2013 Remote Sensing 5(2) 491-520