Transcript Multi-image manipulation

Vegetation Indices
Lecture 11
prepared by R. Lathrop 3/06
Remote Sensing of the Earth:
Clues to a Living Planet
• Remote sensing scientists measure the amount of energy in
different spectral wavelengths reflected from the earth’s
surface as one means of monitoring the earth’s biosphere.
• Where there are a lot of plants on the earth’s surface, less
red and blue light is measured by the satellite sensor.
Likewise, the more green plants, the more near infrared
energy that is measured.
• By combining measurements in the red and near-infrared
wavelengths, scientists have devised a remotely sensed
vegetation index or what is sometimes referred to as a
‘greenness index’. The more plants, the greener the earth,
the higher the index.
Vegetation Indices
• Linear combination of image bands used to
extract information about vegetation:
biomass, leaf area, productivity
• Most vegetation indices (VI’s) based on the
differential reflectances of healthy green
vegetation, dead/senescent vegetation and
soil in visible vs. near IR wavelengths
Photosynthetically Active
Radiation
•PAR : 0.40-0.70 um, portion of EMR absorbed by plant
pigments and used in photosynthesis
•APAR: PAR energy actually absorbed by a plant canopy
•IPAR: intercepted PAR, probability that photons are
intercepted by plant elements
How plant leaves reflect light
Graphics from http://landsat7.usgs.gov/resources/remote_sensing/radiation.php
How plant leaves reflect light
Sunlight
B
G
R
Incoming
light
Blue & red light
strongly absorbed
by chlorophyll
NIR
NIR
Crosssection of
leaf
Reflected
light
Leaf
Transmitted light
NIR light scattered within
leaf: some reflected back,
some transmitted through
Reflectance from green plant leaves
• Chlorophyll absorbs large % of
red and blue for photosynthesisand strongly reflects in green
(.55um)
• Peak reflectance in leaves in near
infrared (.7-1.2um) up to 60% of
infrared energy per leaf is
scattered up or down due to cell
wall size, shape, leaf condition
(age, stress, disease), etc.
• Reflectance in Mid IR (2-4um)
influenced by water content-water
absorbs IR energy, so live leaves
reduce mid IR return
N
I
R
R
e
f
l
e
c
t
a
n
c
e
Sub-pixel Estimation
Spectral Feature Space
Increasing
Vegetation
Example
Soil Line
Pixel X proportions:
IS: 50%
As green leaf area increases
Grass: 30%
NIR increases
Trees: 20%
red decreases
Red Reflectance
Bi-directional reflectance efects
• Lambertian surface: reflected energy is scattered
equally in all directions; no direction bias = isotropic
• Vegetation canopies are not Lambertian surfaces but
rather demonstrate definite directional bias = anistropic
• Aerial photos of tree canopies often exhibit the bidirectional reflectance differences as a result of the
principal point perspective view
• Tree crowns in the direction of incoming radiation
expose their shady side to the image and those in the
opposite direction show their illuminated sides.
Bidirectional Reflectance
Distribution Function: BRDF
• BRDF is the hemispherical distribution of reflectances
for a feature as a function of illumination geometry
• Bottom line: viewing and illumination angle are
important; a nadir view of the same feature may record
a different reflectance than a side view or look
differently under different sun angles
• Some sensors designed to provide different look angles
at the same feature; e.g., a forward, nadir and
backwards view
• Once quantified, the BRDF can be used to correct for
differential illumination effects
Measuring the BRDF: example
Aircraft-mounted
radiometer used to fly a
closed circle and record
reflectance of a site
Note that the reflectance is not
equally distributed across all
directions
Graphics from http://car.gsfc.nasa.gov/application_brdf.html
Simple method for correcting BR
effect in aerial photographs
• Adjust BR-affected brightness values of the aerial
photographs (AP) with temporally concurrent but coarser
scale imagery (e.g., Landsat TM) on a equivalent bandby-band basis
• Normalize each pixel by multiplying by the ratio of mean
TM over mean AP within a moving window centered on
the AP pixel
• APadj = TMwindow/APwindow * APorig
• See Tuominen and Pekkarinen. 2004. RSE 89:72-82
Vegetation indices
• Simple ratio:
nir/red
• Normalized Difference VI (NDVI):
nir - red
nir + red
NDVI ranges from -1 to + 1
• Transformed VI to eliminate negative
values:
TVI : /NDVI + 0.5
Vegetation Indices: Issues
•VI is a B&W image positively correlated with “greenness”,
as NIR increases and red decreases, VI increases
AVHRR
Landsat TM
Vegetation Indices: Issues
•Soil brightness variations complicating the VI response
•Asymptotic relationship leading to loss in sensitivity at
high vegetation amounts
•Atmospheric interference, especially in the Red band.
•Best practice is to convert the original DN values to
radiance (preferably atmospherically corrected) or
reflectance before computing the vegetation index
Vegetation Indices: Issues
•Scaling: ratio of averages (NDVI of larger pixels; e.g.,
AVHRR pixels) is not the same as the average of the ratios
(average NDVI of smaller pixels; e.g., Landsat TM)
Example (sample area from Landsat TM: 30 m pixels vs. km2 composite)
Ratio of averages: Mean red = 28
3 Mean NIR = 73
NDVI = (73-28)/(73+28) = 45/101 = 0.446
Average of ratios: NDVI = 0.416
Vegetation indices:PVI
• Perpendicular VI determines a pixel’s
orthogonal distance from the soil line in
image feature space (X axis: red; Y axis:
NIR)
• The objective is to remove the effect of soil
brightness and isolate reflectance changes
due to vegetation only
Vegetation Indices:SAVI
• Soil Adjusted Vegetation Index (SAVI) is a technique to
minimize soil brightness influences. Involves shifting
the origin of the nir-red feature space to account for 1st
order soil-vegetation interactions and differential red &
nir extinction through vegetation canopies
• SAVI = (1+L) (nir-red) / (nir + red + L)
Where L = 0 to 1.
L = 1 for low veg density,
L = 0.5 for intermed veg density
L = 0.25 for high veg density
From: Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI), Rem. Sens. Environ. 25:295-309.
From: Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI), Rem. Sens. Environ. 25:295-309.
Vegetation indices:MSAVI
•Modified Soil Vegetation Index (MSAVI) employs a
correction factor to reduce sensitivity to soil variation across a
scene
MSAVI0 = ((NIR – RED) (1 + L0)) / (NIR + RED +L0)
MSAVI1 = ((NIR – RED) (2 - MSAVI0)) /
NIR + RED +1 – MSAVI0
Must empirically determine L
MSAVI2 = (2NIR + 1 – ((2NIR +1)2 – 8(NIR – RED)) -1/2 / 2
Vegetation indices: ARVI
• Atmospherically Resistant Vegetation Index
(ARVI) incorporates the blue channel to account
for atmospheric scattering in the red channel by
using the difference between the radiance in the
red and blue channel
•
ARVI = (NIR – RB) / (NIR + RB)
Where RB = Red – g (Blue – Red)
Where Blue is Landsat TM band 1 ,visible blue
wavelengths
g = 1, unless the aerosol model is known a priori
Vegetation indices: EVI
Enhanced Vegetation Index (EVI): developed to optimize the
vegetation signal with improved sensitivity in high biomass regions, a reduction of
sensitivity to the canopy background signal and a reduction in atmosphere
influences.
EVI = G * (NIR – RED) / (NIR + C1*RED – C2*BLUE + L)
Where C1 = atmosphere resistance red correction coefficient = 6
C2 = atmosphere resistance blue correction coefficient = 7.5
L = Canopy background brightness correction factor = 1
G = Gain factor = 2.5
Note: Example coefficients, may vary depending on sensor/ situation
Miura, T., Huete, A.R., Yoshioka, H., and Holben, B.N., 2001, An error and sensitivity analysis of atmospheric
resistant vegetation indices derived from dark target-based atmospheric correction, Remote Sens. Environ.,
78:284-298.
Miura, T., Huete, A.R., van Leeuwen, W.J.D., and Didan, K., 1998, Vegetation detection through smoke-filled
AVIRIS images: an assessment using MODIS band passes, J. Geophys. Res. 103:32,001-32,011.
Graphic taken from
http://tbrs.arizona.edu/projects/
modis/figures/Slide5.GIF
MODIS EVI vs. NDVI
Wide Dynamic Range VI (WDRVI)
• NDVI suffers from a decrease in sensitivity at
medium to high leaf areas because the NIR reflectance
continues to increase with increasing LAI while red
absorption tends to stabilize at lower levels
• WDVRI = (a * NIR – RED) / (a * NIR + RED)
where a is a weighting coeff , 0 < a < 1
a < 1, the contribution from the NIR attenuated
a between 0.05 and 0.2 have been found effective in
row crops
For more info: Gitelson. 2004. J Plant Physiology 161:165-173.
Vegetation indices
• Numerous studies have explored the
relationship between remotely sensed
vegetation indices and field measured estimates
of vegetation amount: above-ground biomass,
leaf area
• Goal is to be able to estimate and map these key
variables of ecosystem state
• Best relationships obtained in closed canopy
crops. Woody material complicates but does
not invalidate the relationship
For good review of VI
• NASA Remote Sensing Tutorial
http://rst.gsfc.nasa.gov/Homepage/Homepage.
html
• For specifics on Vegetation Indices
• http://rst.gsfc.nasa.gov/Sect3/Sect3_4.html
Global Biosphere Vegetation
Monitoring
• One of the main satellite systems that have
been used to measure the vegetation index
of the earth over long periods of time is the
AVHRR satellite.
• AVHRR stands for Advanced Very High
Resolution Radiometer
• This system has been largely replaced by
the MODIS AQUA and TERRA systems
Global AVHRR composite
• 1 band in the Red:
.58-.6 um
• 1 band in the NIR:
.72-1.1 um
• Vegetation Index
to map vegetation
amount and
productivity
Global Biosphere Vegetation
Monitoring
NOAA AVHRR used to create global “greenness”
maps based on NDVI. Composited over biweekly
to monthly intervals.
Integrated NDVI: summed over the growing
season to provide index of vegetation
productivity modified from Goward et al. 1985 Vegetation 64:3-14.
Temperate
broadleaf
forest
Int
NDVI
Boreal forest
Desert
Apr
May
June
Jul
Aug
Sept
Oct
Global NDVI summed over an
entire year
Integrated Growing Season NDVI
modified from Goward et al. 1985 Vegetation 64:3-14.
1400
NPP
Moist conifer
g/m2/yr
Dec. broadleaf
Boreal conifer
Grassland
Tundra
0
desert
0
1
2
3
Integrated NDVI
4
5
Global NDVI converted to LAI
(leaf area index m2/m2)
Remote Sensing of the Earth:
Clues to a Living Planet
• You can access these images over the INTERNET
• You can either browse through individual images
or watch an animation
• http://svs.gsfc.nasa.gov/search/Keyword/NDVI.html
Remote Sensing of the Earth:
Clues to a Living Planet
• First, click on the Hologlobe: Vegetation Index
for 1991 on a Flat Earth animation. Open it, and
click on the > button.
• Watch closely, can you observe the Green Wave
in the northern hemisphere?
• What about the Brown Wave?
• Now look at the southern hemisphere. What do
you observe?
Can you see the Green Wave?
NASA/Goddard Space Flight Center
Scientific Visualization Studiohttp://svs.gsfc.nasa.gov/vis/a000000/a001300/a001308/index.html
Remote Sensing of the Earth:
Clues to a Living Planet
• Now take a look at the Northern hemisphere
in greater detail.
• Click on the NDVI Animation over
continental United States.
• Can you find where you live? How long
does it stay green?
• Compare Florida with Maine or Minnesota.
North America: Close-up
NASA/Goddard Space Flight Center
Scientific Visualization Studio http://svs.gsfc.nasa.gov/vis/a000000/a002500/a002568/index.html
Remote Sensing of the Earth:
Clues to a Living Planet
• To access more recently acquired AVHRR
imagery go to the National Oceanographic
& Atmospheric Administration (NOAA)
Satellite Active Archive
http://www.saa.noaa.gov/
•36 discrete bands between 0.4 and 14.5 µm
•spatial resolutions of 250, 500, or 1,000 m at nadir.
•Signal-to-noise ratios are greater than 500 at 1-km resolution (at a solar zenith angle of
70°), and absolute irradiance accuracies are < ±5% from 0.4 to 3 µm (2% relative to the
sun) and 1 percent or better in the thermal infrared (3.7 to 14.5 µm).
•MODIS instruments will provide daylight reflection and day/night emission spectral
imaging of any point on the Earth at least every 2 days, operating continuously.
•For more info:
http://eospso.gsfc.nasa.gov/eos_homepage/mission_profiles/instruments/MODIS.php
“Aqua,” Latin for “water,” is a NASA Earth
Science satellite mission named for the
large amount of information that the
mission will be collecting about the Earth’s
water cycle, including evaporation from the
oceans, water vapor in the atmosphere,
clouds, precipitation, soil moisture, sea ice,
land ice, and snow cover on the land and
ice.
Additional variables also being measured by Aqua include
radiative energy fluxes, aerosols, vegetation cover on the land,
phytoplankton and dissolved organic matter in the oceans, and
air, land, and water temperatures.
The AQUA Platform includes the MODIS, CERES and AMSR_E
instruments. Aqua was formerly named EOS PM, signifying its
afternoon equatorial crossing time. AQUA was launched May
2002. For more info: http://aqua.nasa.gov/
Earth Observing 1
• NASA’s New Millennium Program
• Multispectral instrument that is a significant
improvement over the Landsat 7 ETM+ instrument –
Advanced Line Imager (ALI)
• Hyperspectral land imaging instrument – Hyperion
• Low-spatial/high-spectral resolution imager that can
correct systematic errors in the apparent surface
reflectances caused by atmospheric effects, primarily
water vapor - Linear Etalon Imaging Spectrometer
Array (LEISA) Atmospheric Corrector (LAC)
EO-1: Advanced Line Imager (ALI)
• The EO-1 ALI operates in a pushbroom
fashion at an orbit of 705 km, 16 day
repeat cycle. Launched in Nov 2000.
• ALI provides Landsat type panchromatic
and multispectral bands. These bands have
been designed to mimic six Landsat bands
with three additional bands covering
0.433-0.453, 0.845-0.890, and 1.20-1.30
µm.
• The ALI has 30M resolution multispectral 10m panchromatic. 37km swath
width.
• More info: http://eo1.usgs.gov/ali.php
Mt. Fuji Japan
ALI Bands: 6,5,4.
ENVISAT
• In March 2002, the European Space Agency
launched Envisat, an advanced polarorbiting Earth observation satellite which
provides measurements of the
atmosphere, ocean, land,
and ice.
• http://envisat.esa.int/
ENVISAT: primary instruments for
land/sea surface remote sensing
• ASAR - Advanced Synthetic Aperture Radar,
operating at C-band,
• MERIS - is a 68.5 o field-of-view pushbroom
imaging spectrometer that measures the solar
radiation reflected by the Earth, at a ground spatial
resolution of 300m, in 15 spectral bands,
programmable in width and position, in the visible
and near infra-red. MERIS allows global coverage
of the Earth in 3 days.
http://envisat.esa.int/
SPOT Vegetation
• Earth observation sensor on board of the
SPOT satellite in blue, red, NIR & SWIR
• Daily coverage of the entire earth at a
spatial resolution of 1 km
• The first VEGETATION instrument is part
of the SPOT 4 satellite and a second
payload, VEGETATION 2, is now
operationally operated onboard SPOT 5.
• http://www.spot-vegetation.com/
SPOT Vegetation Spectral Bands
http://www.spot-vegetation.com/
Global Annual Changes of Vegetation Productivity
http://www.spot-vegetation.com/
SPOT Vegetation
• Free products are :
• extracts from ten day global syntheses.
• available 3 months after insertion in the
VEGETATION archive.
• in full resolution (1km).
• in plate carrée projection.
• available on 10 predefined regions of interest.
• in the standard VEGETATION product format.
• http://free.vgt.vito.be/
•3 visible/NIR(VNIR: 0.5 and 0.9 µm) with 15-m resolution
•3 mid IR (SWIR: 1.6 and 2.43 µm) with 30-m res.
•5 TIR (8 and 12 µm) with 90-m resolution
•60- km swath whose center is pointable cross-track ±8.55° in the SWIR and TIR,
with the VNIR pointable out to ±24°. An additional VNIR telescope (aft pointing)
covers the wavelength range of Channel 3. By combining these data with those for
Channel 3, stereo views can be created, with a base-to-height ratio of 0.6.
•Overpass every 16 days in all 14 bands and once every 5 days in the three VNIR
channels.
For more info:
http://eospso.gsfc.nasa.gov/eos_homepage/mission_profiles/instruments/ASTER.php
Vegetation Water Stress Indices
• Moisture Stress Index (MSI) contrast water absorption
in the MIR with vegetation reflectance (leaf internal
structure) in the NIR
• MSI: MIR / NIR or R1600/R820
• Normalized Difference MSI :
NDMIS: (NIR – MIR) / (NIR + MIR)
• Normalized Difference Water Index
NDWI:
(R860-R1240) / (R860+R1240)
Other Vegetation Indices: NBR
• Normalized Burn Ratio (NBR) contrasts
NIR(TM4) which decreased after fire and
MIR(TM7) which increased after fire
• NBR : (TM4 – TM7) / (TM4 + TM7)
Differencing of pre- vs. post-fire NBR images
has been found to be an effective measure
of burn severity
For more info: http://nrmsc.usgs.gov/research/ndbr.htm
Subpixel Analysis: Unmixing mixed pixels
Spectral endmembers: signature of
“pure”land cover class
endmember1
Unknown pixel represents some
proportion of endmembers based
on a linear weighting of spectral
distance. For example:
60% endmember 2
Band j
20% endmember 1
unknown pixel
20% endmember 1
endmember3
endmember2
Band i
Subpixel Analysis: Unmixing mixed pixels
Linear mixture modeling assumes that a pixel’s spectral signature
is the results of a linear mixture of the spectra from the component classes.
X = Mf + e
where
X = mixed pixel spectral signature
M = n x c matrix of endmember spectra
f = c x 1 vector of land cover class proportions
N = number of bands
C = number of classes, c must be < n + 1
E = noise term
Simple least-squares approach can then be used to “unmix” and
calculate the proportions of land cover in each pixel.
Sub-pixel analysis: linear mixture modeling
Least squares approach
selecting f that minimizes
the following
(x – Mf)T(x-Mf)
Can be unconstrained or
constrained such that
0 <= f <= 1
With no constraints can be
simplified
Subpixel analysis of urban land cover
Landsat TM pixel boundaries on IKONOS image backdrop
Subpixel analysis of urban land cover
Sub-pixel analysis of urban land cover
Study Area 1
IKONOS for
reference
Landsat TM
output
Blue = IS
Green – lawn
Red = tree
Study Area 2