1. Field Spectroscopy, Hyperspectral Imaging, Applications in Vegetation and Soils Analysis
Alexander F. H. Goetz
University of Colorado and
Analytical Spectral Devices Inc.
Beijing and NanJing, China
June and July 1-2, 2004
Lecture 2

2. Spectroscopy of Vegetation
Brian Curtiss
Analytical Spectral Devices, Inc.
5335 Sterling Drive, Suite A
Boulder, CO , USA
FAX:

3. Spectral Properties of Vegetation
Unlike minerals, all vegetation is composed of a limited set of spectrally active compounds
The relative abundances of these compounds, including water, are indicators of the condition of the vegetation and of the environment in which the vegetation is growing
Vegetation architecture has a very strong influence on the overall characteristics of the reflectance spectrum
The spatial scale of the reflectance measurement is important in determining the observed reflectance

4. Spectral Properties of Vegetation (cont.)
Reflectance in the visible and near infrared region (350 to 800 nm) is dominated by absorption from chlorophyll and other accessory pigments
Reflectance in the SWIR (800 to 2500 nm) is dominated by absorption from liquid water in the plant’s tissue
Reflectance in the SWIR is modified by minor absorption features associated with C-H, N-H, and CH2 bearing compounds such as starches, proteins, oils, sugars, lignin and cellulose.

5. Scale Dependence of Vegetation Spectra
Important at all scales:
Viewing geometry
Geometry and spectral characteristics of illumination source(s)
Leaf / Needle scale:
relative abundance of biochemicals
cellular structure
Branch scale:
leaf / needle scale reflectance
leaf / needle angle distribution
leaf / needle shape and density

6. Scale Dependence of Vegetation Spectra (cont.)
Crown scale:
branch scale reflectance
branch angle distribution
crown geometry
branch geometry and density
background reflectance
Canopy scale:
crown scale reflectance
crown density
background reflectance (soil, understory, litter)

7. Information Content of Vegetation Spectra
Leaf / Needle scale:
leaf / needle age
water and nutrient availability
evidence of other environmental stress (both biotic and abiotic)
Branch scale:
branch scale architecture (relates to species)
evidence of environmental stress
Crown scale:
tree age
crown scale architecture (relates to species)
Canopy scale:
tree size distribution
canopy scale architecture (relates to species assemblages)

8. Information Extraction from Vegetation Spectra
Limitations of spectral libraries
the multitude of factors affecting vegetation spectra makes it difficult to adequately populate a spectral library
large within-species variability
strong dependence on growing season climate
scale dependence of vegetation spectra
dependence of vegetation spectra on view and illumination geometry
Successful use of spectral libraries
crop type identification
species identification in arid lands
plant community identification

9. Information Extraction from Vegetation Spectra (cont.)
Limitations of model-based approaches
often requires extensive field and laboratory measurements
often are season specific
often are specific to a particular plant community or group of communities (e.g. C4 forage grasses)
Successful use of model-based approaches
quantification of canopy biochemicals
tree crown size > stem diameter > timber volume
stress across an environmental gradient
crop yield

10. Similarity of Vegetation Spectra
single leaf spectra —
all have same set of
absorption features
but with varying depths

11. Major Spectral Features — Vegetation

12. Leaf Structure A Upper leaf cuticle
B Palisade mesophyll cells containing the majority of chloroplasts
C Spongy mesophyll cells with a large area of cell wall interfaces
D Lower cuticle containing stomates
Schematic cross-section of a typical broadleaf showing the basic photon interactions that occur when light strikes the leaf surface. A photon may be (1) specularly reflected, (2) diffusely reflected, (3) absorbed in leaf photosynthetic apparatus, (4) scattered from inside the leaf back in the general direction from which it originally entered, adding to the leaf reflectance, (5) transmitted or scattered out of the leaf in the same general direction as it originally traveled, adding to leaf transmittance.

13. Model-based Analysis of Vegetation Spectra
Curve-fitting of absorption features
more commonly applied to mineral spectra
some success for chlorophyll and water
features from ‘minor’ constituents (e.g. cellulose) found in the residual spectrum
Regression-based chemometrics
commonly used for determination of canopy biochemicals (e.g. lignin, cellulose, & nitrogen)
requires many well characterized field samples collected near the time of over-flight

14. Model-based Analysis of Vegetation Spectra (cont.)
Geometric optics
models canopy as an aggregate of tree crowns
image spectra are modeled as mixtures of sunlit and shaded crown, and, sunlit and shaded background
the model derives the average crown size which is then related to stem diameter and timber volume
Statistical classification
supervised or unsupervised
may be used in conjunction with a spectral library to aid in the definition of classes

15. Quantitative Reflectance Spectroscopy
The Bouguer-Beer law is the fundamental relationship upon which many spectrometric techniques are based:
-log R() = A() = () b() c
R() Transmission as a function of wavelength
A() Absorbance as a function of wavelength b() Absorption coefficient as a function of wavelength () Absorption path length as a function of wavelength
c Concentration of the absorbing compound
Solving for the concentration gives:
c = -log R() / () b()
Reflectance is the quantity that can be measured by remote sensing

16. Reflectance of Multiple Leaf Layers

17. Reflectance of Multiple Leaf Layers (cont.)
Same water concentration, increasing path length

18. Wavelength Dependence of Path Length
The amount of liquid water “seen” varies with wavelength

19. Single Leaf Spectra
The liquid water absorption feature depths are a product of water concentration and path length — Pinyon pine has the lowest liquid water concentration, but the deepest absorption features at both 0.95µm and 1.2µm

20. Liquid Water Band Fitting — Single Leaf Laboratory Spectra

21. Liquid Water Band Fitting — AVIRIS Spectra

22. Liquid Water Band Fitting — AVIRIS Spectra (cont.)
golf course
Ponderosa pine

23. Liquid Water Band Fitting — AVIRIS Spectra (cont.)
cellulose
golf course
Ponderosa pine

24. Chemometric Modeling

25. Chemometric Calibration Techniques
Follows Beers Law
Linear regression-based models
calculations using log 1/Reflectance
Calibration Calibration Image
Samples  Model  Spectra
Develop Apply

26. Chemometrics Model Development

27. Near Infrared Spectroscopy — Data Pre-treatments
Log 1/Reflectance
Spectral Filtering
Smoothing
Multiplicative Scattering Correction
Spectral Transforms
Derivatives

28. Near Infrared Spectroscopy — Chemometric Models
Simple regression
Single Wavelength
Multiple Wavelengths
Multivariate Regression
Principal Component Regression
Partial Least Squares Regression

29. Near Infrared Spectroscopy — Sample Selection
Samples Must:
span the range of variability found in the image
variance for other scene variables must be independent
be collected to be representative of the spatial pixel
For Natural Samples:
many (>100) samples are required for a good calibration
10-20 samples required for proof of concept

30. Calibration for Nitrogen, Lignin & Cellulose in Fresh Leaves

31. Fresh Leaf Calibration — Band Positions

32. Fresh Vs. Dry Predicted Nitrogen

33. AVIRIS Nitrogen PLS Calibration

34. AVIRIS Predicted Vs. Actual Nitrogen

35. Dry Leaf Band Depth Nitrogen Calibration

36. Band Depth Predicted Nitrogen Vs. Actual Nitrogen

37. Spectra of Vegetation Components
Canopy biochemicals
water
woody components: lignin, cellulose
nitrogen rich components: pigments, proteins
other components: sugars, starch, oils, waxes
Non-vegetative canopy elements
wood
bark
Plant litter
dry leaves
decomposing leaves and woody material

38. Spectra of Cellulose & Lignin
The spectra of most woody materials are dominated by lignins
The spectra of leaves/needles show absorption by both lignin and cellulose
Man-made cotton fabrics have no lignin absorption features

39. Spectra of Dry Plant Materials

40. Spectra of Dry Plant Materials (cont.)

41. Spectra of Other Vegetative Components

42. Wavelengths of Plant Biochemical Absorption Features
Graphical plot of the major (boxed) and minor absorption peaks of canopy biochemicals overlaid with the location of the principal water absorption regions of the atmosphere as determined from AVIRIS spectra.

43. Wavelengths of Plant Biochemical Absorption Features (cont.)
Absorption features in the Vis and NIR that have been related to particular foliar chemical concentrations.

44. Effects of Shadows on Vegetation Spectra
The aspect ratio of canopy elements (in this case crowns) determines the relative amounts of illuminated and shaded crown and background.

45. Effects of “Hole” Shape on Vegetation Spectra (cont.)
The aspect ratio of “holes” in the canopy (at all scales) determines the importance of shadows in the measured spectrum.

46. Viewing Geometry Effects — Soil

47. Viewing Geometry Effects — Grass

48. Viewing Geometry Effects — Juniper

49. Bi-Directional Reflectance Distribution Function
illumination
direction
Red (680nm): Infrared (850nm)
squares = pine; circles = broadleaf

50. Influence of View Geometry on Vegetation Spectra
BRDF of a tallgrass prairie grassland community at two wavelengths. Solar zenith angle is 48°. The red band is centered at 662nm; the IR band is centered at 826nm.

51. Effects of Sensor Spatial Resolution on the Spectrum of a Vegetation Canopy
The spatial variance of the amount of shaded canopy is an indicator of the average crown size. Crown size can then be related to trunk diameter and other parameters useful for forest management.

52. Effects of View Geometry on Reflectance of Vegetation Canopies
Mean spectral reflectance of a mixed conifer stand from view angles of 0° and 15°, 30° and 45° in the backscatter direction and in the principal plane of solar illumination.

53. Time Variation of Vegetation Spectra
Conifers retain needles for several years. Spectra of each years needles can be quite different. At the canopy level the number of years growth retained by a tree strongly influences the canopy scale spectral reflectance.

54. Time Variation of Vegetation Spectra (cont.)
left: changes in NDVI for a Indian Grass
below: reflectance spectra for three days in the growing season.

55. The “Red Edge”
The plot to the left shows a “red edge” shift to blue wavelengths for “stressed” vegetation.
Others have reported a shift to the red under “stressed” conditions.

56. Mechanisms for the “Red Edge” Shift
Band broadening can be produced by chloroplast disruption. Band depth can be reduced by either chlorophyll loss or an architectural change that reduces the interaction of illumination with the chlorophyll in the leaf.

57. What Produces the Shift in the “Red Edge”?
Anything that reduces the amount of chlorophyll seen by the sensor at either the leaf/needle scale or at the canopy scale
actual changes in chlorophyll concentration
canopy architecture changes
branch-scale changes in architecture
changes in vegetation cover
Anything that disrupts the chloroplast membranes
These two mechanisms compete under stressful conditions
could get either a red or blue or no shift with stress

58. Effects of Chronic Stress on Vegetation
In a natural ecosystem, chronic stress results in the establishment of individual trees and/or species that have adapted to the stress
The chlorosis observed in seedlings grown under stress is generally not observed in a mature forest
Architectural changes are a common response to chronic stress and result in changes in reflectance that are observable at leaf to canopy scales
Chances in foliar reflectance may also result from chronic stress

59. Vegetation Hyperspectral Remote Sensing - Summary
Step 1: A clearly defined objective; examples:
delineation of dominant species assemblages
mapping of ecosystem productivity
crop type or yield mapping
spatial distribution of heavy metal induced stress
Step 2: Identify spectral changes or differences associated with process of interest; methods:
examination of existing spectral libraries
examination of study-specific field spectral
correlation between spectral and non-spectral field or laboratory measurements

60. Vegetation Hyperspectral Remote Sensing - Summary (cont.)
Step 3: Identify best analytical method based on available image, field, and laboratory resources; methods:
spectral library based
model based
curve-fitting
chemometics
geometric optics
classification