geog2021 environmental remote sensing lecture 3 spectral information in remote sensing
TRANSCRIPT
GEOG2021Environmental Remote Sensing
Lecture 3
Spectral Information in Remote Sensing
Aim
• Mechanisms variations in reflectance - optical/microwave
• Visualisation/analysis
• Enhancements/transforms
Mechanisms
Reflectance
• Reflectance = output / input• (radiance)• measurement of land complicated by atmosphere
• input solar radiation for passive optical
• input from spacecraft for active systems• RADAR
– Strictly NOT reflectance - use related term backscatter
Mechanisms
Optical Mechanisms
Reflectance
causes of spectral variation in reflectance
• (bio)chemical & structural properties – chlorophyll concentration – soil - minerals/ water/ organic matter
Optical Mechanisms
vegetation
Optical Mechanisms
soil
Optical Mechanisms
soil
RADAR Mechanisms
See: http://southport.jpl.nasa.gov/education.html
RADAR Mechanisms
RADAR Mechanisms
RADAR Mechanisms
Vegetation amountconsider
• change in canopy cover over time
• varying proportions of soil / vegetation (canopy cover)
Vegetation amount
Bare soil
Full cover
Senescence
Vegetation amount
1975 Rondonia See: e.g. http://earth.jsc.nasa.gov/lores.cgi?PHOTO=STS046-078-026
http://www.yale.edu/ceo/DataArchive/brazil.html
Vegetation amount
1986 Rondonia
Vegetation amount
1992 Rondonia
Uses of (spectral) information
consider properties as continuous – e.g. mapping leaf area index or canopy cover
or discrete variable – e.g. spectrum representative of cover type
(classification)
Leaf Area Index
See: http://edcdaac.usgs.gov/modis/dataprod.html
Forest cover 1973
See: http://www.bsrsi.msu.edu/rfrc/stats/seasia7385.html
Forest cover 1985
visualisation/analysis• spectral curves
– spectral features, e.g., 'red edge’
• scatter plot– two (/three) channels of information
• colour composites – three channels of information
• principal components analysis
• enhancements – e.g. NDVI
visualisation/analysis• spectral curves
– reflectance (absorptance) features – information on type and concentration of
absorbing materials (minerals, pigments) • e.g., 'red edge':
increase Chlorophyll concentration leads to increase in spectral location of 'feature'
e.g., tracking of red edge through model fitting or differentiation
visualisation/analysis
visualisation/analysis
http://envdiag.ceh.ac.uk/iufro_poster2.shtm
REP moves to shorter
wavelengths as chlorophyll decreases
Red Edge Position
point of inflexion on red edge
REP correlates with ‘stress’,
but no information on
type/cause
Measure REP e.g. by 1st
order derivative
See also: Dawson, T. P. and Curran, P. J., "A new technique for interpolating the reflectance of red edge position." Int. J. Remote Sensing, 19, (1998),2133-2139.
Consider red / NIR ‘feature space’
Soil line
vegetation
visualisation/analysis• Colour Composites
• choose three channels of information– not limited to RGB– use standard composites (e.g., FCC)
• learn interpretation
visualisation/analysisStd FCC - Rondonia
visualisation/analysisStd FCC - Swanley TM data - TM 4,3,2
visualisation/analysisPrincipal Components Analysis
– PCA (PCT - transform)
• may have many channels of information– wish to display (distinguish)– wish to summarise information
• Typically large amount of (statistical) redundancy in data
visualisation/analysisPrincipal Components Analysis
red NIR
See: http://rst.gsfc.nasa.gov/AppC/C1.html
red
NIR
Scatter Plot reveals relationship between information in two bands
here:
correlation coefficient = 0.137
visualisation/analysisPrincipal Components Analysis
– show correlation between all bands
TM data, Swanley:
correlation coefficients : 1.000 0.927 0.874 0.069 0.593 0.426 0.736 0.927 1.000 0.954 0.172 0.691 0.446 0.800 0.874 0.954 1.000 0.137 0.740 0.433 0.812 0.069 0.172 0.137 1.000 0.369 -0.084 0.119 0.593 0.691 0.740 0.369 1.000 0.534 0.891 0.426 0.446 0.433 -0.084 0.534 1.000 0.671 0.736 0.800 0.812 0.119 0.891 0.671 1.000
visualisation/analysisPrincipal Components Analysis
– particularly strong between visible bands– indicates (statistical) redundancy
TM data, Swanley:
correlation coefficients : 1.000 0.927 0.874 0.069 0.593 0.426 0.736 0.927 1.000 0.954 0.172 0.691 0.446 0.800 0.874 0.954 1.000 0.137 0.740 0.433 0.812 0.069 0.172 0.137 1.000 0.369 -0.084 0.119 0.593 0.691 0.740 0.369 1.000 0.534 0.891 0.426 0.446 0.433 -0.084 0.534 1.000 0.671 0.736 0.800 0.812 0.119 0.891 0.671 1.000
visualisation/analysisPrincipal Components Analysis
– PCT is a linear transformation– Essentially rotates axes along orthogonal axes
of decreasing variance
red
NIR
PC1
PC2
visualisation/analysisPrincipal Components Analysis
– explore dimensionality of data
% pc variance :
– PC1 PC2 PC3 PC4 PC5 PC6 PC7
– 79.0 11.9 5.2 2.3 1.0 0.5 0.1
96.1%
of the total data variance contained within the first 3 PCs
visualisation/analysisPrincipal Components Analysis
– e.g. cut-off at 2% variance– Swanley TM data 4-dimensional
• first 4 PCs = 98.4%
– great deal of redundancy TM bands 1, 2 & 3
correlation coefficients : 1.000 0.927 0.874 0.927 1.000 0.954
0.874 0.954 1.000
visualisation/analysisPrincipal Components Analysis
– display PC 1,2,3 - 96.1% of all data variance
Dull -
histogram equalise ...
visualisation/analysisPrincipal Components Analysis
– PC1 (79% of variance)
Essentially
‘average brightness’
visualisation/analysisPrincipal Components Analysis
stretched sorted eigenvectors
PC1 +0.14 +0.13 +0.28 +0.13 +0.82 +0.12 +0.43
PC2 -0.44 -0.27 -0.60 +2.23 +0.47 -0.49 -0.77
PC3 +1.68 +1.35 +2.45 +1.34 -1.49 -0.67 +0.05
PC4 +0.29 +0.10 -1.22 +1.90 -1.83 +4.49 +2.30
PC5 +0.03 -0.39 -2.81 +0.70 -1.78 -5.12 +6.52
PC6 10.42 +1.10 -6.35 -0.70 +1.64 -0.23 -2.39
PC7 -8.77 28.50 -8.37 -1.43 +1.04 -0.40 -1.75
visualisation/analysisPrincipal Components Analysis
• shows contribution of each band to the different PCs. – For example, PC1 (the top line) almost equal
(positive) contributions (‘mean’)PC1 +0.14 +0.13 +0.28 +0.13 +0.82 +0.12 +0.43
– PC 2 principally, the difference between band 4 and rest of the bands (NIR minus rest)
PC2 -0.44 -0.27 -0.60 +2.23 +0.47 -0.49 -0.77
visualisation/analysisPrincipal Components Analysis
– Display of PC 2,3,4
Here, shows
‘spectral differences’
(rather than ‘brightness’
differences in PC1)
EnhancementsVegetation Indices
– reexamine red/nir space features
EnhancementsVegetation Indices
– define function of the two channels to enhance response to vegetation & minimise response to extraneous factors (soil)
– maintain (linear?) relationship with desrired quantity (e.g., canopy coverage, LAI)
EnhancementsVegetation Indices
– function known as ‘vegetation index’– Main categories:
• ratio indices (angular measure)
• perpendicular indices (parallel lines)
EnhancementsVegetation Indices
RATIO
INDICES
EnhancementsVegetation Indices
– Ratio Vegetation Index • RVI
– Normalised Difference Vegetation Index • NDVI
RATIO
INDICES
EnhancementsVegetation Indices
NDVI
RATIO
INDICES
EnhancementsVegetation Indices
NDVI
RATIO
INDICES
RATIO
INDICES
NDVI over Africa (AVHRR-derived) Tucker et al. - A: April; B: July; C: Sept; D: Dec 1982
EnhancementsVegetation Indices
PERPENDICULAR
INDICES
EnhancementsVegetation Indices
– Perpendicular Vegetation Index • PVI
– Soil Adjusted Vegetation Index• SAVI
PERPENDICULAR
INDICES
PERPENDICULAR
INDICES
And others ...
Summary
• Scattering/reflectance mechanisms
• monitoring vegetation amount
• visualisation/analysis– spectral plots, scatter plots, PCA
• enhancement– VIs