The Role of Hyperspectral Data in Understanding the Global Carbon Cycle Susan L' Ustin, Pablo J' Zar

1
The Role of Hyperspectral Data in Understanding
the Global Carbon Cycle Susan L. Ustin,
Pablo J. Zarco-TejadaCenter for Spatial
Technologies and Remote Sensing (CSTARS)
Department of Land, Air and Water Resources
(LAWR) University of California, DavisGreg S.
Asner Department of Geological Sciences
University of Colorado, Boulder
2
(No Transcript)
3
Predicting the Consequences of Global Climate
Change

• Significant Feedbacks between Vegetation
• Dynamics and Atmosphere
• carbon, water, nitrogen, sulfur cycles
• mechanistic links via photosynthetic
  regulation
• Major Terrestrial Carbon Sink(s) Exist
• mediated through vegetation dynamics
• magnitude of storage capacity is unknown

4
Why Emphasis on Carbon Cycle?

• Critical Step in Moderating Increased Trace Gases
• Strong Feedbacks to Land Use
• Validation Methods needed for Carbon Credits
• Agricultural Production and Food Security
• Highest National/International Research Priority
• NRC Reports (many) especially Pathways Report
• NASA Science Plan
• Interagency Carbon Science Plan
• IGBP and other International Programs
• Carbon Cycle Initiative

5
Land Use Change

• Changes in Ecosystem
• Structure Function

C source/sink relations nutrient
dynamics hydrology microclimates
altered food web structure
3 band composite
3-endmember Composite

• Requires Spatial Detail
• Scaling Mechanisms

6
Land Cover Characterization

• Land Cover Undergoing Rapid Change
• Land Use and Land Use Change
• - Agriculture
• - Logging and Fuel Extraction
• - Grazing
• Natural Disturbance Regimes
• - Tree Fall to Wildfires and Hurricanes
• Changing Composition of Ecosystems
• - Invasive Species
• - Loss of Biodiversity
• Driven by
• - Human Population Growth
• - Changing Climate and/or Variability

7
Predicting Future CO2 Sequestration and Dynamics

• Global Land Cover Maps have High Uncertainty
• - 30 disagreement on land cover type
  among best maps
• - based on potential climate and not actual
  vegetation
• - inadequate high resolution maps and
  uncertain scaling

8
Methods to Estimate CO2 Dynamics

• Models
• Empirical Data Environmental Inventories
• Flux Measurements Ameriflux, Fluxnet, etc.
• Flux Aircraft Measurements

Ecosystem models (e.g., Simple Biosphere Model
and CASA, etc.) use NDVI to estimate CO2
dynamics
9
Ecosystem Models
Current Limitations

• Models dont agree with field measurements
• Inventory methods and eddy flux measures dont
  agree
• Flux measurements and Mesoscale Climate Models
  dont converge
• Most models use bimonthly NDVI (AVHRR) to
  characterize land cover

• Models
• 8 km pixels (averaged to 1)
• many parameters derived from NDVI

10
Grand Challenges
Information to Improve Ecosystem Model Estimates

• Vegetation Type Maps
• - growth form herbs, shrubs, trees
• - seasonality
• - photosynthetic pathway C4 vs C3
• Leaf Area Index
• Above Below-ground Carbon Stocks
• Environmental Controls on Carbon Sequestration
• Allocation of Biomass Foliage, Stem, Roots
• Turnover of Carbon Stocks

11
What Can Hyperspectral Imagers Contribute?

• Land Cover Characterization
• LAI
• Dry plant litter and woody stems
• Improved Estimates of Photosynthesis and
  Transpiration
• Estimates of Canopy Stress

12
Estimating Vegetation LAI and Dry Carbon
Magnitude and shape of the spectroscopic
signature are sensitive to absolute and relative
amount of green leaf and dry carbon biomass
(Asner et al. 1998)
13
BOREAS Land Cover Classification Using
Hyperspectral Data
Field Methods
Landsat TM
CASI resampled, 30m
Saskatchewan Environment and Resource Management
Hall et al. 1995, 1997
Zarco-Tejada Miller 1999
red edge spectral parameters lp, lo, and s

14
BOREAS Land Cover Classification Using
Hyperspectral Data
AVIRIS -leaf
AVIRIS - index
TM (Hall, 1999)
Jack Pine - Fen
Assumed True (Gruszka,1998)
Old Black Spruce
Fuentes et al. (in press)

15
Mapping Chaparral Ecosystems
AVIRIS 20 m Data October 23, 1996
(Roberts et al.)
16
Land-use Carbon Chemistry in Savannas
AVIRIS Collected in South Texas
Comparison to Field Data
Green Leaf Area Dry Carbon Load
(Asner et al. 1998)
17
Estimates of Carbon Allocation By Type
Contributes to Non-foliar C Estimates Biomass
Estimates

18
Soil Litter Cover Fractions
Jornada and Sevilleta LTER Sites
1
Field Map
2

3
19
Cover Fraction Estimates
Litter C turnover Soil Erosion Overland
flow Canopy Fraction Roughness Flux
rates Successional stage
Asner et al., 2001

20
Comparison between NDVI and HSI Derived Indices

• The NDVI problem Estimates of LAI at LAI gt 3
• Physiological State

Continuum Removal (Canopy Water Content)
Red Edge Parameter ?p (chlorophyll)
NDVI
Ag 2020 Site Cotton Agriculture in the San
Joaquin Valley, CA

21
Leaf Area is a Key Link to Ecosystem Processes
Leaf water provides a better measure of leaf area
than VIs.
Pacific Northwest Conifer Forests
LAI
(?m)
22
Estimating Physiological Variables Using
Hyperspectral Data
Canopy Estimates of Total Photosynthetic
Pigments Minimizes Effects of Understory
Shadows/gaps Potential HSI Measurements
Chlorophyll ? and ?, Xanthophylls,
Carotenes, Derived Estimates of Nitrogen

by RT model inversion (r20 using NDVI)
(Zarco-Tejada et
al., 2001)

23
Estimation of other physiological variables using
hyperspectral remote sensing
Chlab estimation by RT model inversion
Low stressed site Measured 38.8
mg/cm2 Estimated 35.2 mg/cm2
High stressed site Measured 19.1
mg/cm2 Estimated 20.2 mg/cm2
(Zarco-Tejada et al., 2001)

24
Spectroscopy for Land Condition Assessed as
Green Vegetation LAI and Dry Carbon Load
Land-use Carbon Chemistry in Amazonia
Spectroscopy of Pastures
Young Pastures
Comparison to Field Data
Aging Pastures
Old Pastures
(Asner et al. 1999)
25
Mapping Soil Chemistry and Texture

• Organic Matter

• Iron Content

26
Conclusion

• Hyperspectral data will contribute to improved
  understanding of the Global Carbon Cycle
• Improved land cover classifications
• Predictions can be improved by incorporating HSI
  data into models
• Need high spatial resolution data with
  appropriate scaling methods
• Need for routine global HSI data collections