LIDAR Based Stand Delineation In Natural Stands

1
LIDAR Based Stand Delineation In Natural Stands

• Alicia Sullivan
• Precision Forestry Cooperative
• University of Washington

2
Outline

• Project background
• Current Methods
• Research Objectives
• Study area and Data
• Methods
• Results
• Discussion

3
Project background

• Funded by the Bureau of Land Management.
• Federal Agency.
• Approximately 2 million acres of forest land in
  western Oregon.
• Generates income for Counties through harvesting
  activities.

4
Background contd

• First step in a LIDAR based inventory.
• Out of date inventory.
• Challenges with large ownership area.
• Time for re-inventory 10 years.
• High cost for field crews.
• Fewer resources available as industry changes.

5
What is LIDAR?

• Active sensor that emits laser pulses.

• GPS and inertial system to accurately identify
  plane location.
• Records X, Y, and Z coordinates for returned
  pulses.

6
Raw LIDAR Data
7
What is a stand?

• A contiguous group of trees sufficiently uniform
  in species composition, arrangement of age
  classes, site quality and condition to be a
  distinguishable unit
• (Smith et al.1997).

8
What is a stand?

• Common parameters considered in stand
  delineation
• Trees per acre
• Percent cover
• Average height
• Basal area
• Stand volume
• Age and species composition
• (Smith Anson 1968, Smelser Patterson 1975,
  Avery 1978).

9
Outline

• Project background
• Current Methods
• Stand delineation
• Research Objectives
• Study area and Data
• Methods
• Results
• Discussion

10
Current methods

• Current stand delineation is inconsistent.
• Varies greatly on company/agency culture.
• Dependant on data quality and resources available.

11
Current Methods contd

• Challenges with current method
• Time and labor intensive.
• Known to be subjective.
• Hard to consistently reproduce.
• Low accuracy expectations (80).

12
Outline

• Project background
• Current Methods
• Research Objectives
• Study area and Data
• Methods
• Results
• Discussion

13
Research Objectives

• Can LIDAR data provide information to develop a
  stand map?
• Develop a repeatable method for forest stand
  delineation from LIDAR data.
• What accuracies can be expected from LIDAR stand
  maps?

14
Outline

• Project background
• Current Methods
• Research Objectives
• Study area and Data
• Blue ridge site
• Lidar data
• Plot data
• Methods
• Results
• Discussion

15
Study Area

• Blue Ridge study site.
• Part of Capitol Forest West of Olympia, WA.
• Second growth, Douglas Fir dominated forest.
• Multiple Silvicutural treatments in area.
• Representative of conditions on BLM lands.

16
Plantation
Mature Forest
Clearcut
Thinning
Patch Cuts
17
LIDAR Data
Specifications for LIDAR flights
18
Plot information

• Plot level information collected in 1999 and
  2003.
• Plot locations set with Javad GPS unit.
• Plots were located across various treatment
  areas.
• Used to create stand classes.

19
Plot Locations
20
Outline

• Project background
• Current Methods
• Research Objectives
• Study area and Data
• Methods
• Method development
• Object based image classification
• Software
• Flow chart and data outputs
• Results
• Discussion

21
Method Development

• Two basic components to the method
• Processing of LIDAR data into data layers.
• Classification of data layers into stand classes.

22
Method Development

• LIDAR derived raster data layers.
• Trees per acre.
• Average height.
• Percent cover.
• Variation on a common technique.
• Object based image classification.

23
Object Based Image Classification

• A technique that was developed for classifying
  images into user-defined classes (Blaschke et al.
  2000).
• Works with image objects (aggregates of pixels),
  vs. a pixel-by-pixel basis
• (Chubey et al. 2006).
• Takes into account spatial relationship of
  pixels.

24
Object Based Image Classification

• Basic steps in object based classification.
• Segmentation of image into objects.
• Training of objects for classes.
• Classification of image based on training.

25
Object Based Image Classification

• Traditionally, images used in this type of
  classification are collected from passive optical
  or hyperspectral instruments.
• Bands correspond to spectral frequency.
• Variation
• Used LIDAR data layers as bands

26
Software

• FUSION
• Developed by the USFS for viewing and analysis of
  LIDAR data.
• SPRING
• Developed by Brazilian Govt. Object based image
  classification.
• ESRI ArcMap-
• Viewing of final products.

27
Raster data layers, 8.9m2 pixels
Flowchart of Method
FUSION analysis
FUSION products, SPRING Inputs
Area of ground (pixel)
Raw LIDAR Data
Percent Cover by pixel
SPRING
Percent Cover Model
Object orientated classification to define stand
types.
of first returns gt2 m.
Bare Earth Model
Stem Density Per pixel
of Maxima points Per pixel
Subtract ground model from points
Canopy Maxima points
Stand type Map
Tree Height Average of pixel
Summarize heights
Identify Local maxima
Digital Canopy Height Model
28
Percent Cover
29
Stem Density
30
Average Height
31
Image Classification

• Stand classes based on plot data.
• Training objects that corresponded to plots used
  when possible.
• Parameters for object minimum size and difference
  are the same for both dates.

32
Stand Classes
33
Outline

• Project background
• Current Methods
• Research Objectives
• Study area and Data
• Methods
• Results
• Stand maps
• Accuracy assessment
• Discussion

34
Blue Ridge
35
1999 Classification
Classes Mature Clear cut/Road Thinned Intermediate
Young 1 Young 2
36
2003 Classification
Classes Mature Clear cut/Road Thinned Intermediate
Young 1 Young 2
37
Accuracy Assessment

• Used an error matrix to evaluate classification.
• Reports users accuracy, producers accuracy,
  overall accuracy.
• KHAT and Z-statistics calculated for each
  classification.
• Congalton Green 1999.

38
Generation of Error Matrix
39
Generation of Error Matrix- 2003
40
1999 Error Matrix
Visual Classification
Computer Classification
41
Accuracy Assessment 1999
42
2003 Error Matrix
Visual Classification
Computer Classification
43
Accuracy Assessment 2003
44
Explanation of Error

• Several sources
• Size of pixels.
• May be too small of pixel, too great of detail
• Definition of classes.
• Classes too similar
• Changes in stand condition.
• Brush, height growth etc.

45
Size of Pixel
46
Size of Pixel/Definition of classes
47
Definition of Classes
48
In-growth
49
Outline

• Project background
• Current Methods
• Research Objectives
• Study area and Data
• Methods
• Results
• Discussion
• Importance of results
• Future work

50
Discussion

• Develop a repeatable method for forest stand
  delineation from LIDAR data.
• Most difficult aspect of project.
• Most important for practical use.
• What accuracies can be expected from LIDAR stand
  maps?
• Will be dependant on many factors, approximately
  78 to 84.

51
Importance of Results

• Demonstrates stand delineation with LIDAR is
  possible.
• Accuracies are similar to photo interpretation.
• Represents significant savings in time to
  delineate stands over large areas.
• Savings in cost for ownerships with LIDAR
  coverage.

52
Future Work

• Increase number of classes and forest area.
• Modify procedure with commercial software like
  eCognition or feature analyst.
• Explore new metrics for stand delineation as
  research advances.

53
Acknowledgements

• Funding provided by U.S. Bureau of Land
  Management.
• DNR, PNW Research Station, and Precision Forestry
  Cooperative.
• University of Washington, College of Forest
  Resources.
• Robert McGaughey, Dr. Peter Schiess, Dr. Monika
  Moskal, Dr. Dave Briggs, Dr. Hans Andersen.

54

• Questions?