Transcript Remote Analysis of Asian Cold Desert Ecosystems

Remote Analysis of Asian Cold Desert Ecosystems
Cheney M. Shreve & Gregory S. Okin: Univ. of Virginia and Univ. of California, Los Angeles
Asian cold desert ecosystems are understudied, poorly
understood and sensitive to the effects of climate change
(Miehe, 1995). The geographical remoteness of these regions
coupled with a lack of in situ data makes remotely derived
satellite datasets the most practical means for studying these
ecosystems. In North American cold deserts precipitation and
temperature are the primary controlling factors on vegetation
productivity (Comstock & Ehleringer, 1992) and in Arctic
environments it has been hypothesized that nutrients may be
the most limiting factor (Tissue & Oechel, 1987; Chapin, 1991).
To investigate the role of climatic variables on vegetation
productivity in Asian cold deserts, specifically portions of the
Iranian and Tibetan Plateaus, and to predict how these
dynamics might change in response to changing climate, it is
necessary to assess the limitations of a remote analysis and to
model climate change scenarios. Snow disasters are also
common in the Tibetan Plateau and pose a threat to humans,
livestock and the economy. Optical snow products have the
potential to outperform microwave based products due to the
spatial distribution (e.g., thin, discontinuous sheets) of snow in
cold deserts. Several optical snow products will be compared
and used to assess snow disasters. Cold deserts also have the
potential for producing large quantities of dust (aka mineral
aerosols), which can potentially affect the apparent surface
reflectance due to additional scattering of light by dust particles
suspended in the atmosphere. Practical limits on linear spectral
mixture modeling in turbid atmospheric conditions using
radiative transfer models (MODTRAN4), a method for spectral
simulation after Okin et al. (2001), and SMA indices will be used
for the evaluation.
Land cover Classification in
Desert Environments
Desert Vegetation
•Often bright mineral soil background
VNIR
Introduction
•weaker “red edge”
•Reduced absorption in the visible
•Wax absorptions ~1720 nm
•Spectral variability within shrub species
Methods:
phenological stages in response to spatially
Top-Down variability analysis
discontinuous precipitation
•evaluate a new multi-temporal linear spectral mixture
algorithm, Relative Multiple Endmember Spectral Mixture
Analysis (RMESMA) designed for use with coarse resolution
satellite data (e.g., MODIS) for land cover classification
•compare optical snow cover products and assess snow
disasters
Okin et al., 2001, RSE
•Open canopies--> poor correlations with LAI
Relative Multiple Endmember Spectral Mixture Analysis (RMESMA): new
multitemporal SMA technique designed for use with coarse resolution (e.g. MODIS)
satellite data (Okin, In Press)
Design of RMESMA:
•apparent sfc. Refl. For a pixel at a reference time in a timeseries of collocated
images is defined as the “baseline” spectra of that pixel
•Spectral components can be added or subtracted from the baseline spectrum,
which allows RSMA to provide information about changes in NPV/litter without
having to separate NPV/litter from soil (e.g., soil is never used explicitly as an
endmember)
Basic SMA Eqn.
RSMA Eqn.

n1
t
t
pixel
 fsoil
soil   fj t j
i
i
i
n 1
ti
pixel
 x    X j  
ti
ps ps
j 0
f
ti
soil
x  X  1
to
ps
  fj  1
ti
to
j
j 0
j 0

j0
ti
j
n1
n1
n 1
ps  f 
f i  0,1
to
soil soil


Investigate the role of climatic variables on vegetation
productivity in Asian cold deserts and to predict how these
dynamics might change in response to changing climate.
Linear spectral mixture analysis (SMA) techniques are commonly
used for remote land cover classification and the high temporal
resolution and global coverage of the MODIS sensor make it
amenable for monitoring Asian cold deserts, however it is necessary
to verify that MODIS derived SMA classifications are representative of
the actual percentage of vegetation and snow cover in a pixel.
•Rapid movement of desert shrubs through

Objectives
SWIR
Evaluating appropriate scale Optical Snow Products
for quantifying patches of •May be better for sensing snow in cold deserts because
traditional microwave based products have difficulty sensing
snow with thin, discontinuous spatial patterns as often occurs in
vegetation and snow cover
cold desert environments


n 1

  f jt o  j
j0

ti
ti t0
ti t0
pixel
 x ps
f soil s   x ps
f j  X tji  j  
j0
Gregory S. Okin, Relative Spectral Mixture Analysis-A
multitemporal index of total vegetation cover, RSE (In
Press)
•Repeat the first two steps using multispectral reflectance data
and transferring classes to collocated contemporaneous
hyperspectral reflectance data, calculate semi-variograms for
individual classes and entire image
•Compare (nugget: sill) N:S of classes with N:S ratio of entire
image (at 2.5 m) using kriging leave-one-out cross validation.
•examine how vegetation in the cold desert ecosystem might
respond to changing climate using a model to simulate climate
change
Methods for quantifying spatial patterns of snowfall:
Temporal: annual, seasonal, during snow disaster events
•Connectivity
Preliminary Results: Tibet
•Good agreement (> 80%, p=0.05) between regressions of optical
snow cover products (2 SMA based products and Normalized
Difference Snow Index (NDSI))
•Aspect has greatest effect on agreement
•North facing slopes show poorest agreement
•Spatial distribution of snowfall changes seasonally
Methods:
Bottom-Up variability analysis
•Calculate semi-variogram for multispectral reflectance equivalent
band in high resolution hyperspectral data
•Calculate vegetation (VI) and snow indices (NDSI) for
multispectral equivalent band in high resolution hyperspectral
data
•Block-krig multispectral equivalent bands, VIs and NDSI in high
resolution hyperspectral data to 30 m resolution.
•Calculate VIs, NDSI, and MESMA snow fractional cover of ETM+
data
•Compare co-located ETM+ pixels with block-kriged 30-m ETM+
prediction using paired difference between means t-test, or
equivalent randomization test.
•Repeat for multispectral to coarse resolution (30 m to 1 km)
ArcVeg Diagram
Disturbance
Total Soil
Organic
Nitrogen
Recruitment
Senesce/Mortality
Plant Nitrogen by
Functional Type
•evaluate inter-annual vegetation dynamics
•determine how much dust impedes retrieval of surface
characteristics in arid environments
•Multiple linear regressions
Spatial: aspect, slope, elevation
•Transfer class to collocated contemporaneous multispectral data,
calculate semi-variogram for different classes and entire image
Climate
•Randomization tests
•Data Stratification using Envi Decision Tree Classifier
•Classify 1 km MODIS reflectance data using eCognition and/or kmeans algorithm into natural land-cover classes
Study Sites: Iranian and Tibetan Plateaus
Methods for comparing optical snow products:
Mineralization
Plant-Available
Nitrogen
Epstein et al. (2000)
Plant uptake/Growth
Climate
Current Plant Biomass
Plant Attributes
Plant
Biomass by
Functional
Type
•K-means clustering of green vegetation and snow timeseries
(separately) followed by extraction and comparison of mean annual
signatures from each class suggests spatiotemporal distribution of
snowfall has strong effect on type and spatial distribution of green
vegetation.
Modeling Potential Impact of
Climate Change using a
Transient-Nutrient Based
Model (ArcVeg)
Higher than expected shrub biomass in Northern Tibet, where
rainfall is scarce, suggests that snowfall may be an important
source of water/ insulation for vegetation. Studies in North
American cold deserts suggest temperature and precipitation
are the primary controls on primary productivity (Comstock &
Ehleringer, 1992). Changing climate will alter the equilibrium
between these variables and thus it is necessary to model how
these changes will affect primary productivity.
ArcVeg model will be modified to reflect the climate and
vegetation characteristic of Northern Tibet, water availability and
insulation from snowfall. Fieldwork will be conducted to provide
information on plant functional type, soil nitrogen and
mineralization rates for use in model input and validation
purposes.