1. Remote Sensing of Forest Structure
Van R. Kane
School of Forest Resources

2. Lecture 18 – Forest remote sensing
Wednesday’s lecture
Mars spectroscopy
Today’s lecture:
Forest remote sensing
LECTURES
Jan Intro
Jan Images
Jan Photointerpretation
Jan Color theory
Jan Radiative transfer
Jan Atmospheric scattering
Jan Lambert’s Law
Jan Volume interactions
Feb Spectroscopy
Feb Satellites & Review
Feb Midterm
Feb Image processing
Feb Spectral mixing
Feb Classification
Feb Radar & Lidar
Feb Thermal infrared
Mar Mars spectroscopy (Matt Smith) previous
Mar Forest remote sensing (Van Kane)
Mar Thermal modeling (Iryna Danilina)
Mar Review
Mar Final Exam

3. Today’s Topics Physical measurements Change detection Approaches
Problems
Spectral and LiDAR
Change detection

4. “It is, perhaps, time to draw the conclusion that current satellite sensors are not in general suitable for forestry planning since they contain little relevant information…”
-- Holmgren and Thuresson (1998)

5. Did Someone Miss the Memo?
Remote Sensing of Environment
1990: 5 papers on forest remote sensing (7%)
2000: 32 papers (25%)
2010: 89 papers (36%)
Understates trend – forest ecology papers using remote sensing increasingly common in mainstream ecological journals

6. What Are They Studying? Research goals Map by
Biomass (where’s the carbon?)
Wood volume (when can we take it to the bank?)
Presence (has something removed it?)
Productivity (how much biological activity?)
Fire mapping (where? how bad?)
Map habitat (where can critters live?)
Composition (what kinds of trees?)
Structure (what condition? how old?)
Map by
Space – where?
Time – change?

7. Goal: Map Forest Structure
What is structure?
Vertical and horizontal arrange of trees and canopy
Why structure?
Reflects growth, disturbance, maturation
Surrogate for maturity, habitat, biomass…
We’ll look at just two attributes
Tree size (height or girth)
Canopy surface roughness (rumple)
~ 50 years
~ 125 years
~ 300 years
~ 50 years
~ 125 years
~ 300 years
Robert Van Pelt

8. Spectral Mixture Analysis
Sabol et al. 2002
Roberts et al. 2004
Each pixel’s spectra dominated by a mixture of spectra from dominant material within pixel area

9. Endmember Images Original Landsat 5 image Shade Conifer NPV
(Tiger Mountain S.F.)
Shade
(darker = more)
Conifer
(deciduous is ~ inverse for forested areas)
Lighter = more
NPV
(lighter = more)

10. Physical Model More structurally complex forests produce more shadow
Measure “rumple”
More structurally complex forests produce more shadow
We can model self-shadowing
Use self-shadowing to determine structure

11. Modeled self-shadowing
Test Relationship
Modeled self-shadowing
Rumple
Beer time!
Kane et al. (2008)

12. Reality Check
Topography sucks
Trees!
Kane et al. (2008)

13. No beer… but Chapter 1 of dissertation
One Year Later…
No beer… but Chapter 1 of dissertation

14. New Instrument - LiDAR Systems
Scanning laser emitter-receiver unit tied to GPS & inertial measurement unit (IMU)
Pulse footprint 20 – 40 cm diameter
Pulse density 0.5 – 30 pulses/m2
1 – 4 returns per pulse
Airborne laser scanning systems have four major hardware components: a) a laser emitter-receiver scanning unit, b) GPS (aircraft and ground units), c) a highly sensitive inertial measurement unit (IMU) attached to the scanning unit, and of course, d) a computer to control the system and store data from the first three components.
Laser scanners designed for terrain mapping emit near-infrared (NIR) laser pulses at a high frequency (typically 25,000 to 100,000 pulses per second). The position and attitude of the laser scanner unit at the time each pulse is emitted are determined from flight data collected by the GPS and IMU units. The range or distance between the scanner and an object that reflects the pulse is computed using the time it takes for the pulse to travel from the scanner, to the object, and back to the scanner. A precise coordinate is computed for each reflection point using the position and attitude of the scanner and the direction and distance traveled by the pulse from the scanner to the object.
A swath of terrain under the aircraft is surveyed through the lateral deflection of the laser pulses and the forward movement of the aircraft. The scanning pattern within the swath is established by an oscillating mirror or rotating prism which causes the pulses to sweep across in a consistent pattern below the aircraft (See slide). Large areas are surveyed with a series of swaths that often overlap one another by 20 percent or more. The final pattern of pulse reflection points on the ground and the scanned swath width depend on the scanning mechanism settings and design (e.g. pulse rate, returns per pulse, scanning angle), flying height and speed, and the shape of the topography.

15. Samples of LiDAR Data Point Cloud Canopy Surface Model
400 x 400 ft
400 x 10 ft
Canopy Surface Model
High-density LIDAR data covering a portion of Capitol State Forest located near Olympia, Washington. Douglas-fir is dominant over the area with some Western Hemlock and Red Alder. Return data have been color-coded according to the height above ground.
Old-growth stand Cedar River Watershed

16. What LiDAR Measures x, y, z coordinates of each significant reflection
Accuracies to ~10-15 cm
Height measurements
Max, mean, standard deviation, profiles
Measures significant reflections in vegetation structure not specific tree heights
Canopy density
Hits in canopy / all hits
Shape complexity
Canopy surface model
Intensity (brightness) of return
Near-IR wavelength typically used, photosynthetically active material are good reflectors

17. (# canopy hits/# all pulses) (area canopy surface/area ground surface)
Physical Model
Canopy density
(# canopy hits/# all pulses)
Height
(95th percentile)
Rumple
(area canopy surface/area ground surface)
Calculate for 30 m grid cells

18. Classify Sites by Using LiDAR Metrics
Statistically distinct classes
Distinct groupings of height, rumple, density values
Easy to associate classes with forest development
Class 8 old growth
Class 3 early closed canopy
Beer time!

19. Reality Check #!@^% Trees! But the variation!
Older stands more likely in more complex classes and vice versa
But the variation!
Young and older forests in same classes
Wide range of classes within age ranges
Possible Explanations:
Multiple forest zones, presence or absence of disturbance, site productivity, conditions of initiation…
Trees!

20. Still no beer, but had my dissertation and post-doc funding
Another Year Later…
Still no beer, but had my dissertation and post-doc funding

21. Remotely Sensing Forest Attributes
- 1 m spectral
- 24 m hyperspectral
- 30 m spectral
- LiDAR
From Lefsky et al. 2001
Match your question to the instrument best suited to answer it

22. Change Detection - Fire Severity
Green Vegetation
Burned Vegetation A B A B
B G R
Landsat TM B4
Landsat TM B7
Landsat TM B4
Landsat TM B7 (1984 – 2005)
Landsat MSS B2
Landsat MSS B4
Landsat MSS B2
Landsat MSS B (1974 – 1983)
van Wagtendonk et al. 2004

23. Differenced Normalized Burn Ratio
Pre-fire
Post-fire
High severity
Moderate severity
Low severity
No fire effect
dNBR
Key et al. 2002, Key and Benson 1999, 2002, 2004, 2005, van Wagtendonk 2004, Miller and Fites 2006, Miller 2007

24. Change Detection – Regional Monitoring
LandTrendr
Courtesy Robert Kennedy, OSU

25. Local Detection, Regional Monitoring
Courtesy Robert Kennedy, OSU

26. Some Remote Sensing Thoughts
Remote sensing rarely gives answers
Remote sensing provides data that must be interpreted with intimate understanding of the target system
Interpretation is almost always local
Data must be tied to a physical model of the target system
The more directly the measurement is tied to the physical properties of the system, the easier it is to interpret and apply
In many ways harder than research that collects field data because you must be familiar with both the technical methods of remote sensing and intimately familiar with the target system
You’ll read twice as many papers at a minimum

27. But …
Remote sensing can open up avenues of research at scales impossible with field work alone