How else can we assess exposure to vegetation and
green spaces?
Alexandra Chudnovsky AIRO Lab
Department of Geography and Human Environment School of Geosciences, Tel-Aviv University
Munich 2016: Exploring potential pathways linking greenness and green spaces to health
• Presentation outline
• Why to use satellite-derived index?
• Some basic RS concepts
• Satellite-derived vegetation data (modelling the regional scale)
• Field in-situ optical observation of the vegetation (modeling from
the local scale to the regional)
GIS Data: still need pre-processing
Evaluation of the building layers. Buildings are delineated in black. On the left image inner spaces marked in cyan. On the right image, greenhouses marked in cyan. All the cyan marked object were removed from the original layer.
GIS Data: still need pre-processing
Categories according to USGS and NLCD
Energy-matter interactions in the atmosphere, at the study area, and at the remote sensor
detector
Jensen 2009
Energy recorded by remote sensing systems undergoes fundamental interactions that should be understood to properly interpret the remotely sensed data.
Discrimination between surfaces vs spectral resolution
1 2 3
1 2 3 4
3 2 1
0.0 20.0 40.0 60.0 80.0 100.0
0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4
Wavelength in Microns
% Reflectance rel. to Halon
5 7
7 6 4 5
HRVIR/HRG TM-7
ASTER
Fe Al-OH
Vegetation C-O kaolinite
grass
carbonate
goethite
8 9 4
Spectral Signature
Beyond the human ability
Visible, NearIR and Middle IR Interactions
RGB vs False color composite
multi-spectral sensors: record energy over several separate wavelength ranges
NDVI
Normalized Difference Vegetation Index
DN4-DN3 is a measure of how much chlorophyll absorption is present, but it is sensitive to cos(i) unless the difference is divided by the sum
DN4+DN3.
3 3 4 4
3 3 4 4
3 3 4
4
3 3 4
4
3 3 3
4 4 4
) cos(
) cos(
) cos(
) cos(
) cos(
);
cos(
r I r I
r I r NDVI I
i r I i r I
i r I i r NDVI I
r i DN I
r i DN I
+
= −
+
= −
=
=
π π π π
Biophysical Variables used in Environmental Studies
• Vegetation: pigment concentration, biomass, foliar water content
• Temperature
• Soil moisture
• Surface roughness
• Evapotranspiration
• Atmosphere: aerosols concentrations, gaseous pollutants, temperature, water vapor, wind speed/direction, energy inputs, precipitation, cloud and aerosol properties
• BRDF
• Ocean: color, phytoplankton, chemistry
• Spatial: x,y, and potentially z
• Temporal: time the image was acquired
• Directional: sensor and sun angle
• Polarization: in RADAR
SATELLITE IMAGERY: different resolutions
• Landscape Scale: Landsat 7/ETM+ (30m)
• Sentinel (20m)
• Aster (30 m)
• Regional Scale: Terra/ Aqua platforms: MODIS (1000m, 500m, and 250m)
http://glovis.usgs.gov/
https://lpdaac.usgs.gov/dataset_discovery/modis/modis_products_table
MODIS 1km (LST, LAI/FPAR, GPP/NPP, Reflectance, NDVI)
MODIS 500m water stress (NDSVI, LSWI)
MODIS 250m EVI, NDVI
L8,30
L8, 90m (LST)
Sentinel 20m
Freely available data
2015
2000 2001 2003 2005
-0.3 0.8
2007 2011
NDVI
MOD13A, First week of July Baghdad
Baghdad
Baghdad Baghdad Baghdad
Baghdad Baghdad
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0 0.05 0.1 0.15 0.2
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
ndvi aod
NDVI AOD
Time series analyses of NDVI averaged over Baghdad and surrounding cities
High resolution data require Image pre-processing (work flow)
• Radiometric correction -
atmospheric correction
• Geometric correction -
rectification & georeferencing
• Display & Enhancement -
Contrast stretching
• Information extraction –
image classification (supervised/unsupervised) data mining, feature extraction, spectral vegetation indices (SVIs)
• Analysis
outside RS/GIS, data staged in text files/excel from imagery and analyzed in statistics package
Sentinel: spatial and spectral
configuration (data since 2013)
Landsat: since 1972
RGB
False Color Composite (R:7,G:4,B:2-landsat5, R:7,G:5,B:3-landsat8) Landsat 5
1999
Landsat 8 2015 Landsat 8
2013 Landsat 5
2011 Landsat 5
2009 Landsat 5
2006 Landsat 5
2003
NDVI (Vegetation Index)
Halab, Syria :
Time series analyses
Cautions about NDVI
• Saturates over dense vegetation
• Less information than original data
• Any factor that unevenly influences the red and NIR reflectance will influence the NDVI
• such as atmospheric path radiance, soil wetness
• Pixel-scale values may not represent plant-scale processes
• Derivatives of NDVI (FAPAR, LAI) are not
physical quantities and should be used with caution
Other vegetation indices:
• Soil-adjusted Vegetation Index (SAVI)
• Soil and Atmospherically-Resistant Vegetation Index (SARVI)
• Moisture Stress Index (MSI or NDMI)- information on vegetation water content
• Enhanced Vegetation Index (EVI)
L L SAVI
red NIR
red NIR
+ +
+ −
= ρ ρ
ρ ) ρ
1 (
Where L is an adjustment factor for soil. Huete (1988) found the optimal value for L is 0.5
L R
R
R SARVI R
rb NIR
rb NIR
+ +
= −
Huete and Liu, 1994
) 1
(
2 Re
1
Re L
L R
C R
C R
R EVI R
Blue d
NIR
d
NIR +
+
− +
= −
Huete and Justice, 1999
EVI has improved sensitivity to high biomass regions
http://www.harrisgeospatial.com/docs/BroadbandGreenness.html#Infrar
ed
• Aerosol Free Vegetation Index (Karnieli et al. 2001) the atmospheric resistant vegetation index (ARVI)
rb NIR
rb NIR
R R
R ARVI R
+
= − Rrb = Rred −
γ
(Rblue − Rred)Kaufman and Tanre, 1992. Atmospherically Resistant Vegetation Index (ARVI) for EOS-MODIS. IEEE Trans. Geosci. Rem. Sen. 30(2):261-270.
More indexes….
VARI1 = (green – red)/(red + green)
red edge NDVI with two Sentinel-2 bands: 705 and 740nm:
RE NDVI1 = (NIR -705)/(NIR+705)
RE NDVI2 = (NIR -740)/(NIR+740) and certainly NDVI.
It’s also worth to use VARI with blue band (443 nm) that is atmospherically resistant : VARI2 = (green – red)/(red + green-blue)
Select the best – for your study area and application
Landsat-based vegetation indexes: Tel- Aviv and suburbs
SAVI
-0.7 0.5
-0.3 0.45
-0.1 0.4
Brightness 1.3
-0.1 Greenness Index NDVI
What are challenges we face when working
with different spatial resolutions data?
We are able better to estimate different vegetation types with increasing of spectral resolution
Red, blue and green polygons represent
different types of vegetation
¯
0 150 300 600
Meters
¯
0 150 300 600
Meters
Source: Esri, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community
¯
0 12.5 25 50
Meters
Source: Esri, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community
¯
0 12.5 25 50
Meters
Mixed pixel concept
¯
0 12.5 25 50
Meters
NDVI:
0.36 0.33 0.37
Air-Photo
Air-Photo
RGB image
RGB image
NDVI image
0.28
1
0.29
Vegetation 1
A B
C
Mixed Pixel Concept:
Mixed Energies
Build-up 0.050
0.1 0.15 0.2 0.25 0.3 0.35 0.4
0.4 0.9 1.4 1.9 2.4
Wavelength, µm
Reflectance
2
Vegetation:
End-member 1
3
Concrete- end member
0.19
NDVI False color
True color
10 m Sentinel
20 m Sentinel
60 m Sentinel
Spatial Resolution: decreasing the spatial resolution increase the contribution of
mixed pixels
10 m Sentinel
20 m Sentinel
60 m Sentinel
NDVI above the same area sampled at different resolutions: small urban parks
NDVI Count
Chudnovsky, A., and Lugassi, R. in progress 5
15 Mean= 0.69 10 m resolution 6
Mean= 0.64 20 m resolution
Mean= 0.59 Mean= 0.50
1
60 m resolution
30 m resolution
10 m 20 m 60 m 10 m 20 m 60 m
NDVI
Densely populated Neighborhood with high vegetation cover
How we can estimate the contribution of different land
uses/coverages inside of a single pixel?
Linear Spectral Unmixing
Basic Assumptions:
• Spectral variation is caused by a limited number of surface materials (i.e. soil, water, shadow, vegetation)
• The pixel is a linear mixture of endmember constituents
• All endmembers possibly contained in the pixel have been included in the analysis
• A unique solution is possible if the number of spectral components DO NOT exceed the number of spectral bands +1.
F
i= F
1+ F
2+ ... + F
N= 1
i=1 N
∑
DN
λ= F
1DN
λ,1+ F
2DN
λ,2+ ... + F
NDN
λ,N+ E
λ0 0.1 0.2 0.3 0.4 0.5 0.6
0.45 0.95 1.45 1.95 2.45
90%concr10%veg 80%concr20%veg 70%concr30%veg 60%concr40%veg 50%concr50%veg 40%concr60%veg 30%concr70%veg 20%concr80%veg 10%concr90%veg
concr1 veg1
Wavelength, µm
Reflectance
LSU approximation/ Estimation
LSU
0 1
concrete vegetation RMSE
Spectral Angle Mapper
Field in-situ studies
Not only for Validation
Imaging Spectrometer Data of Healthy Green Vegetation in the San Luis Valley of Colorado Obtained on September 3, 1993 Using AVIRIS
224 channels each 10 nm wide with 20 x 20 m pixels
High soot content settled on a leaf
Chemometric approach/ NIRS analyses
n n
X A X
A X
A X
A A
Y = +
1 1+
2 2+
3 3. + ... +
where Y is the chemical constituent, A is an empirical coefficient, and X1-n are wavelengths.
Reflectance response of a single Magnolia leaf
(Magnolia grandiflora) to decreased relative water content
Demonstration of total absorption area of two vegetation samples. The higher area (black) is of sample with low Cl and Na content whereas lower area (red) is of sample with high content.
0.00 0.25 0.50 0.75 1.00
1750 1940 2130 2320
Reflectance (CR)
Wavelength (nm)
2235 nm
1840 nm straight line
absorption line
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
350 850 1350 1850 2350
Wavelength, nm
Reflectance No fertilization
With fertilization
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
350 850 1350 1850 2350
Wavelength, nm
Reflectance
No fertilization With fertilization
Wet Dry
NDI: 550- 686 nm
NDI: 850- 420 nm
NDI: 700- 900 nm
Slope: 686-
775 nm NDI: 550-
686 nm
NDI: 850-
425 nm NDI: 700- 900 nm
Slope: 470- 530 nm
Slope: 686- 775 nm
Shefer S. Israel A., Goldberg A., Chudnovsky, A. Botanica Marina 2016 2 1
1 2
x x
y Slope y
−
= −
Dry1
Dry2Dry4Dry3 Dry5
Dry7Dry9
Dry11
Dry12 Dry13 Dry15 Dry16
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Measured (Log (Glc))
Predicted (Log (Glc))
y=0.87x + 0.16 R2=0.88
-0.00020 -0.00015 -0.00010 -0.00005 0.00000 0.00005 0.00010 0.00015
-0.0010 -0.0005 0.0000 0.0005 0.0010
Dry 16 Dry9 Dry10
Dry11 Dry12 Dry14
Dry13 Dry15
Dry7
Dry5 Dry3 Dry4
Dry1
Dry8
Dry6
Dry2 PC1
PC2
B
-1000 -800 -600 -400 -200 0 200 400 600 800
400 900 1400 1900 2400
Wavelength, nm
A B
C
Regression coefficients
Lantana
Ficus Ibiscus
University
Ironi D
Shai Agnon Str Independence
Garden
Park HaYarkon
Lantana Ibiscus Ficus
1. Sampling vegetation (the same type)
2. Measuring gravimetric weight of “dusty” samples vs cleaned
AISA-ES 2003
Fabric + Chrstolite (asbstos)
Fabric Asbestos
Conclusions
To estimate several indexes for study region and select the best suited
To select the most appropriate sensor for the studied application Field validation is necessary. Validation using existing GIS
layers is also helpful
40% of urban+
60% of vegetation