What is Hyperspectral Imagery? - University of...

1

Application of high spatial resolution, Application of high spatial resolution, hyperspectral imagery to riparian mapping in hyperspectral imagery to riparian mapping in

Yellowstone National ParkYellowstone National Park

Probe1 1 meter data, 8/03/99Probe1 1 meter data, 8/03/99MNF Image by W. A. MarcusMNF Image by W. A. Marcus

W. Andrew MarcusW. Andrew MarcusDepartment of GeographyDepartment of Geography

University of OregonUniversity of Oregon

Collaborators on Image Acquisition,Collaborators on Image Acquisition,Data Collection, and AnalysisData Collection, and Analysis

• Richard Aspinall, Montana State UniversityRichard Aspinall, Montana State University•• Joe Boardman, Analytical Imaging GeophysicsJoe Boardman, Analytical Imaging Geophysics•• Robert Crabtree, Yellowstone Ecosystem StudiesRobert Crabtree, Yellowstone Ecosystem Studies•• Don Despain, U.S. Geological SurveyDon Despain, U.S. Geological Survey•• Earth Search Systems, Inc.Earth Search Systems, Inc.•• Mark Fonstad, Southwest Texas State UniversityMark Fonstad, Southwest Texas State University•• Wayne Minshall, Idaho State UniversityWayne Minshall, Idaho State University•• Chuck Peterson, Idaho State UniversityChuck Peterson, Idaho State University

Field and Research Assistants: Robert Ahl, Kerry Halligan, Carl Legleiter, Jim Rasmussen

Funding Sources:Funding Sources:

The High Tech Landscapes Initiative of The High Tech Landscapes Initiative of Yellowstone Ecosystem Studies, Yellowstone Ecosystem Studies, Bozeman, MontanaBozeman, Montana

The NASA Earth Observations Commercial The NASA Earth Observations Commercial AppliAppli--cationscations Program (EOCAP) Program (EOCAP) -- Hyperspectral, Stennis Hyperspectral, Stennis Space Flight Center, MississippiSpace Flight Center, Mississippi

The EPA bioindicators programThe EPA bioindicators program

Outline of PresentationOutline of Presentation•• Goal of talk:Goal of talk: Review data types and mapping approaches that Review data types and mapping approaches that

have been most successful for us in riparian mappinghave been most successful for us in riparian mapping

•• Concept of high spatial resolution hyperspectral (HSRH) dataConcept of high spatial resolution hyperspectral (HSRH) data

•• Field area and data collectionField area and data collection

•• Research approaches and findings:Research approaches and findings:•• Data DecisionsData Decisions

•• choice of appropriate spatial resolutionchoice of appropriate spatial resolution•• choice of data compression approacheschoice of data compression approaches

•• Classification/Mapping AlgorithmsClassification/Mapping Algorithms•• spectral angle mapping for terrestrial landcoverspectral angle mapping for terrestrial landcover•• maximum likelihood for inmaximum likelihood for in--stream habitatsstream habitats•• regression analysis for stream depthsregression analysis for stream depths•• matched filtering for downed woodmatched filtering for downed wood

•• SummarySummary

Research Objectives and RationaleResearch Objectives and RationaleOverarching goal: determine if HSRH imagery can be used to Overarching goal: determine if HSRH imagery can be used to map key environmental variables in complex stream settings.map key environmental variables in complex stream settings.

Soda Butte Creek, 9ssg68, Oct 10, 1999Soda Butte Creek, 9ssg68, Oct 10, 1999

1999 Probe1 (solid lines) and TM5 spectra1999 Probe1 (solid lines) and TM5 spectra(symbols) for willow, cottonwood and aspen(symbols) for willow, cottonwood and aspen

AspenAspen

WillowWillow

CottonwoodCottonwood

Lamar River, Aug 3, 1999, Image by J. BoardmanLamar River, Aug 3, 1999, Image by J. Boardman

1999 Probe1 1 meter image, Lamar River1999 Probe1 1 meter image, Lamar River

What is Hyperspectral Imagery?What is Hyperspectral Imagery?

2

Rationale for using high spatial resolutionRationale for using high spatial resolution

TM (30 m)TM (30 m)

AVIRIS (17 m)AVIRIS (17 m)

Probe1 (8 m)Probe1 (8 m)


Detection of withinDetection of within--stream features stream features like riffles requireslike riffles requiresHSR imageryHSR imagery


High Spatial ResolutionHigh Spatial Resolution and and HyperspectralHyperspectral combined combined (“HSHR imagery”) provide more information on variables (“HSHR imagery”) provide more information on variables at the stream scale . at the stream scale .

1616pixelspixels

13 pixels13 pixels

Simulated true color TM imageSimulated true color TM image

7 bands7 bands

399 pixels399 pixels

472472pixelspixels

Probe1 1 meter imageProbe1 1 meter image

MNF Bands 7 (invert), 6, 1 MNF Bands 7 (invert), 6, 1 Lamar River, 8/03/99Lamar River, 8/03/99Image by W.A. MarcusImage by W.A. Marcus 128128

bandsbands

Field Area: Northeast YellowstoneField Area: Northeast Yellowstone

Lamar River

9msb109msb10--4, Oct, 4, 19994, Oct, 4, 1999

Soda Butte Creek

9mul89mul8--26, Aug 26, 199926, Aug 26, 1999

33rdrd

orderorder

44thth orderorder

55 thth

orderorder

Landscapes: Cooke CityLandscapes: Cooke City

Landscapes: Round PrairieLandscapes: Round Prairie

Landscapes: Lower Soda Butte CreekLandscapes: Lower Soda Butte Creek

3

Landscapes: Lamar RiverLandscapes: Lamar River

Data Sets: Imagery and Field DataData Sets: Imagery and Field Data


Landsat 7 TM 30m17m

4m

1m

8m

Location of 1999 Hyperspectral Data in NE YNPLocation of 1999 Hyperspectral Data in NE YNP

Red = Hyperspectral LinesYellow = AirSAR Data (AirSAR = Lhv, Chv, Chh))

Upper Soda Upper Soda Butte CreekButte Creek

Lamar Lamar RiverRiver

Backdrop shows 1990 Landsat TM NIR Band 4. AirSAR data acquiredMay 28, 1999.

YNP boundaryYNP boundaryLower Soda Lower Soda Butte CreekButte Creek

11-- m Imagery: Helicopter platformm Imagery: Helicopter platformTo collect 1To collect 1 --m resolution data, the Probem resolution data, the Probe--1 1 sensor was mounted on a helicopter and sensor was mounted on a helicopter and flown 600 m above the ground surface.flown 600 m above the ground surface.

..

J. Boardman, Probe1, misc6, 08/04/99J. Boardman, Probe1, misc6, 08/04/99

The helicopter platform created unique problems:The helicopter platform created unique problems:

•• more vibration, pitch, yaw and roll than fixed wing aircraft more vibration, pitch, yaw and roll than fixed wing aircraft •• gyroscope overcompensation led to serious image swirlgyroscope overcompensation led to serious image swirl

•• rotor blades blocked GPS signal, causing loss of coordinate datarotor blades blocked GPS signal, causing loss of coordinate data•• greater difficulty in coregistering field maps to imagery and GIgreater difficulty in coregistering field maps to imagery and GIS dataS data

Aug 4, 1999, misc5Aug 4, 1999, misc5

99080308b, Lamar River99080308b, Lamar RiverAug3. 1999Aug3. 1999

4

Methods: field mapping forMethods: field mapping forimage training and validationimage training and validation

Morphologic units, stream depth, Morphologic units, stream depth, substrate size, and vegetation communsubstrate size, and vegetation commun--ities and species were mapped directly ities and species were mapped directly to 1 m images to insure coregistrationto 1 m images to insure coregistration

J. Rasmussen, Aug, 1999 (misc12)J. Rasmussen, Aug, 1999 (misc12) C. Legleiter, Aug, 1999 (misc13)C. Legleiter, Aug, 1999 (misc13)

Data DecisionsData Decisions


Two key considerationsTwo key considerations ::•• Different data compression approachesDifferent data compression approaches•• Effects of spatial resolution on feature detectionEffects of spatial resolution on feature detection

Hyperspectral data: reducing dimensionalityHyperspectral data: reducing dimensionality

Goals of reducing dimensionality:Goals of reducing dimensionality:•• Data compression and faster processingData compression and faster processing•• Signal enhancementSignal enhancement•• Noise reductionNoise reduction

Approaches examined:Approaches examined:•• Spectral selectionSpectral selection•• Principal Component Analysis (PCA)Principal Component Analysis (PCA)•• Minimum Noise Fraction (MNF)Minimum Noise Fraction (MNF)

Possible future approach: Geostatistical filteringPossible future approach: Geostatistical filtering

Reducing Dimensionality: Selecting SpectraReducing Dimensionality: Selecting Spectra

If a feature has a unique spectral If a feature has a unique spectral signal, selecting specific spectra signal, selecting specific spectra may be an efficient approach for may be an efficient approach for reducing dimensionality. reducing dimensionality.

4 spectral bands shown by yellow dotted lines can be used 4 spectral bands shown by yellow dotted lines can be used to distinguish metal car from other features on 4to distinguish metal car from other features on 4--m m resolution Probe1 image. resolution Probe1 image.

Car

Reducing Data Dimensionality: PCAReducing Data Dimensionality: PCA

True color True color composite using 1composite using 1--m m Probe1 image, Probe1 image, August 2, 1999August 2, 1999

PC1PC1 PC3PC3

PC10PC10 PC40PC40

Reducing Data Dimensionality: MNFReducing Data Dimensionality: MNFMNF 1MNF 1 MNF3MNF3

True color True color composite using 1composite using 1--m m Probe1 image, Probe1 image, August 2, 1999August 2, 1999

MNF10MNF10 MNF40MNF40

5

Reducing Data Dimensionality: PCA Reducing Data Dimensionality: PCA vsvs MNFMNFPC1PC1

PC3PC3

MNF1MNF1

MNF3MNF3

Reducing Data Dimensionality: PCA Reducing Data Dimensionality: PCA vsvs MNFMNFPC10PC10 MNF10MNF10

PC40PC40 MNF40MNF40

Findings: Reducing dimensionalityFindings: Reducing dimensionality

•• Except with spectrally unique targets, spectral band extraction Except with spectrally unique targets, spectral band extraction is is generally not best approach as it discards the power of hypergenerally not best approach as it discards the power of hyper--spectral imagery. Exception is stream depth (to be discussed laspectral imagery. Exception is stream depth (to be discussed later).ter).

•• PC and MNF both reduce data size and processing time by and PC and MNF both reduce data size and processing time by and order of magnitude, but:order of magnitude, but:

•• no no a prioria priori reason for favoring one over the otherreason for favoring one over the other•• each seems to work better for certain applications:each seems to work better for certain applications:

oo PCA best for inPCA best for in--stream classification stream classification oo MNF best for terrestrial land coverMNF best for terrestrial land cover

•• Further work needed to:Further work needed to:•• determine why a given method works better in certain determine why a given method works better in certain

situationssituations•• evaluate alternative approaches evaluate alternative approaches –– e.g., geostatistical filteringe.g., geostatistical filtering

Data decision: spatial resolutionData decision: spatial resolution

Finer scale spatial resolution better discriminates Finer scale spatial resolution better discriminates many features, but at the cost of:many features, but at the cost of:

•• less area coverage for a given sensorless area coverage for a given sensor•• greater data storage cost for a given areagreater data storage cost for a given area

Therefore one wants to choose coarsest pixel resolution Therefore one wants to choose coarsest pixel resolution that still meets mapping needs.that still meets mapping needs.

Entire 1Entire 1--m imagem imageEntire Entire 44--m m imageimage

Effects of spatial resolution: human featuresEffects of spatial resolution: human features

Effects of spatial resolution: StreamsEffects of spatial resolution: Streams

Spatial

resolution (m)

Eddy drop zones (%)

Riffles (%)

Pools (%)

Glides (%)

Overall

accuracy (%)

Kappa 1 84.8 75.9 90.5 86.8 85.5 0.74 Producer's

Accuracy 5 51.4 55.5 50.0 70.0 66.1 0.34

Scour pool

4-m pixel

6

Effects of spatial resolution: Downed woodEffects of spatial resolution: Downed wood

Downed woodDowned wood: Requires 1: Requires 1 --m resolution for detection. m resolution for detection.

SAM rule image

True color

Downed woodDowned wood

4-m pixel

Effects of spatial resolution: Vegetation typeEffects of spatial resolution: Vegetation type44--m resolution provides valley m resolution provides valley wide coverage, while still wide coverage, while still detecting individual large trees. detecting individual large trees.

Individual treesIndividual trees

Probe1 4Probe1 4--m image, m image, Round PrairieRound Prairie

Conifer 100%Conifer 100%

Sedge Sedge 96.6%96.6%

DecidDecid.. 100%100%

ThistleThistle 93.1%93.1%

AverageAverage 97.4%97.4%

% correct% correct

30 sites visited for 30 sites visited for each categoryeach category

Work by Kerry Halligan, YERCWork by Kerry Halligan, YERC

Findings: Spatial resolutionFindings: Spatial resolution

2 to 102 to 1011

11

11

11

Required Required spatial spatial

resolution resolution (m)(m)

8282Downed woodDowned wood9292--9999Vegetation coverVegetation cover

67 to 9967 to 99Stream depthStream depth

76 to 9176 to 91InIn--stream habitatsstream habitats

N.A.N.A.Human featuresHuman features

Accuracy Accuracy at that at that

resolution resolution (%)(%)

FeatureFeature

Appropriate spatial resolution varies with feature in question:Appropriate spatial resolution varies with feature in question:

General Rule: Pixel resolution should be ¼ size of feature of iGeneral Rule: Pixel resolution should be ¼ size of feature of i nterest.nterest.Realistic Rule: Pixel resolution should be approximately same siRealistic Rule: Pixel resolution should be approximately same size as ze as feature of interestfeature of interest

Hyperspectral MappingHyperspectral Mapping

Features of interest for riparian mapping: Features of interest for riparian mapping: •• Terrestrial land cover (vegetation, wetland, human Terrestrial land cover (vegetation, wetland, human

features)features)•• Stream habitats and depthsStream habitats and depths•• Downed woodDowned wood

Classification approaches used for final maps:Classification approaches used for final maps:•• Spectral angle mapper (SAM) Spectral angle mapper (SAM) –– land coverland cover•• Maximum likelihood Maximum likelihood –– inin--stream habitatsstream habitats•• Matched filter (MF) Matched filter (MF) –– downed wooddowned wood•• Regression and hydraulic model Regression and hydraulic model –– stream depthsstream depths

Also evaluated mixture tuned matched filters (MTMF), pixel Also evaluated mixture tuned matched filters (MTMF), pixel purity indices (PPI), and pixel unmixing.purity indices (PPI), and pixel unmixing.

Evaluation site for land cover mappingEvaluation site for land cover mapping

Round Prairie, YNP4-m imagery

7

Material 1Material 1

Material 2Material 2

00 Band 1 ReflectanceBand 1 Reflectance00

Ban

d 2

Ref

lect

ance

Ban

d 2

Ref

lect

ance

Dark PointDark Point

Spectral angleSpectral angle

Similarity between pixels is defined by the angle between their Similarity between pixels is defined by the angle between their vectors. vectors. The advantage of this technique is that illumination differencesThe advantage of this technique is that illumination differencesacross across large landscapes (e.g. different aspects) do not create false dilarge landscapes (e.g. different aspects) do not create false differences fferences between pixels of the same composition.between pixels of the same composition.

Land cover mapping: Spectral Angle MapperLand cover mapping: Spectral Angle Mapper

Data transformations prior to spectral angle mapping:Data transformations prior to spectral angle mapping:•• None None –– use raw untransformed imageryuse raw untransformed imagery•• Principal components (PCA) imagesPrincipal components (PCA) images•• Minimum noise fraction (MNF) imagesMinimum noise fraction (MNF) images

The best classifications occurred with MNF transformed data. The best classifications occurred with MNF transformed data. It is not immediately apparent why this is the case, but it seemIt is not immediately apparent why this is the case, but it seems s likely that MNF does a better job of:likely that MNF does a better job of:

•• Removing imageRemoving image--wide noise prior to data analysiswide noise prior to data analysis•• Isolating signal related to subtle variations in vegetation Isolating signal related to subtle variations in vegetation

covercover

CRITICALCRITICAL to using MNF to using MNF –– using an homogenous dark patch or using an homogenous dark patch or dark bands for defining noise statistics.dark bands for defining noise statistics.

Land cover mapping: Data ProcessingLand cover mapping: Data Processing

Spectral Angle Mapping of AlgaeSpectral Angle Mapping of AlgaeSAM requires that the user define theSAM requires that the user define thethreshold at which the feature is “found.”threshold at which the feature is “found.”

The user should therefore have The user should therefore have knowledge about the landscapeknowledge about the landscape

for feature ID to work.for feature ID to work.

SAM rule images by W.A. MarcusSAM rule images by W.A. MarcusProbe1 1 m data, Aug 03, 1999Probe1 1 m data, Aug 03, 1999

The algae training siteThe algae training site

Land cover mapping: SAM rule imagesLand cover mapping: SAM rule imagesSedge (marsh) rule

image

Map of sedge in Round Prairie

SAM terrain map of Round Prairie, WY, based on MNF bands 1 -20, Probe1 4-m imagery

bare soil: redsparse veg: orangegrass: yellowmarsh: medium greenconifer: dark greendeciduous: light green

asphalt: graywater: bluegravel: white

Hyperspectral mapping: InHyperspectral mapping: In--stream habitatsstream habitats

GlideGlide

High gradientHigh gradientriffle (whiteriffle (whitewater)water)

Low gradient riffleLow gradient riffle(no white water)(no white water)

Scour poolScour pool

Round Prairie, 9srp104, October, 1999Round Prairie, 9srp104, October, 1999

Eddy drop zone Eddy drop zone (fine sediment(fine sedimentaccumulation)accumulation)

8

Data transformations evaluated prior to mapping:Data transformations evaluated prior to mapping:•• None None –– use raw untransformed imageryuse raw untransformed imagery•• Principal components (PCA) imagesPrincipal components (PCA) images•• Minimum noise fraction (MNF) imagesMinimum noise fraction (MNF) images

The best classifications occurred with PCA transformed data. The best classifications occurred with PCA transformed data. It appears the low reflectance from water causes MNF It appears the low reflectance from water causes MNF transforms to classify signal as noise, thereby reducing the transforms to classify signal as noise, thereby reducing the spectral information available for analysis. spectral information available for analysis.

CRITICALCRITICAL to using PCA in streams to using PCA in streams –– calculate the covariance calculate the covariance statistics using the water portion only of the stream (i.e., masstatistics using the water portion only of the stream (i.e., mask k out remainder of image when calculating statistics). out remainder of image when calculating statistics).

InIn--stream habitats: Data processingstream habitats: Data processing

Image 99080309Image 99080309

InIn--stream habitats: PCA imagesstream habitats: PCA imagesTake raw imageTake raw image

Mask to calculate Mask to calculate covarcovar--ianceiance statistics for waterstatistics for water

Create pc data that highlight Create pc data that highlight withinwithin--stream variationsstream variations

Mask of mapped unitsMask of mapped units

PC bands 8,4,1 PC bands 8,4,1

InIn--stream habitats: Maximum likelihood approachstream habitats: Maximum likelihood approachCalculates the “distance” of a given pixel from the average baseCalculates the “distance” of a given pixel from the average based on d on statistical distribution of training sites in spectral space. statistical distribution of training sites in spectral space.

Example of maximum likelihood classification,Example of maximum likelihood classification,((LillesandLillesand and Kiefer, 1994, pages 590 &596)and Kiefer, 1994, pages 590 &596)

Reflectance at 555 nmReflectance at 555 nm

Ref

lect

ance

at

707

nmR

efle

ctan

ce a

t 70

7 nm

Eddy drop zones (red)Eddy drop zones (red)

Riffles (cyan)Riffles (cyan)

Glides (green)Glides (green)

Pools Pools (dark blue)(dark blue)

Raw reflectance values, Probe1 , Lamar River, 1999Raw reflectance values, Probe1 , Lamar River, 1999

Pools (in blue)

Rapids (in white)

Fine grained sediment (in yellow)

Glides (in green) are shallow with gravel substrates

84.8Eddy drop zone

90.5Pool

86.8Glide

75.9Riffle/rapid

Accuracy (%)

TurbulenceCategory

Results of 4 Unit Stream ClassificationResults of 4 Unit Stream Classification

RifflesRiffles

GlidesGlides

PoolsPoolsEddy drop zonesEddy drop zones

1 meter classification1 meter classificationField mapField map

9

Hyperspectral mapping: stream depthsHyperspectral mapping: stream depths

Steps to reach this goal:Steps to reach this goal:

•• Extract pc scores forExtract pc scores fordepth pixelsdepth pixels

•• Conduct multiple regression analysis on depth vs. pc Conduct multiple regression analysis on depth vs. pc scoresscores

•• Error analysisError analysis

•• Create depth image using regression resultsCreate depth image using regression results

Goal: determine if 1 m Goal: determine if 1 m imagery can map imagery can map contincontin--uousuous variations in depth.variations in depth.

Soda Butte Creek, 9ssg53, Oct 10, 1999Soda Butte Creek, 9ssg53, Oct 10, 1999

Multiple Regression Results, Lamar River, WYMultiple Regression Results, Lamar River, WYSummary statistics for multiple regression ofSummary statistics for multiple regression of

depth (y) as a function of principal component scores (xdepth (y) as a function of principal component scores (x11, x, x22… … xxnn))

Cutoff of relation between depth Cutoff of relation between depth and reflectance data at > 60 cm and reflectance data at > 60 cm consistent with consistent with WinterbottomWinterbottomand Gilvear, 1998, but...and Gilvear, 1998, but...

62%62%65%65%0.100.10448686<= 60 cm<= 60 cm

53%53%59%59%0.100.1088105105All depthsAll depths

R2 R2 predictedpredicted

R2R2adjustedadjusted

SignifiSignifi--cant atcant at

Number of Number of pc's included pc's included in regressionin regression

Number Number of sitesof sites

Depth Depth RangeRange

Reflectance is a function of:Reflectance is a function of:•• depthdepth•• andand turbulenceturbulence•• andand substrate substrate

Soda Butte Crk, July, 1999, Untitled 1Soda Butte Crk, July, 1999, Untitled 1--22

Stratifying by inStratifying by in--stream stream habitats should therefore habitats should therefore help to normalize the help to normalize the relation of depth to relation of depth to reflectance. reflectance.

Stream mapping: Depths in the Lamar RiverStream mapping: Depths in the Lamar River

Multiple regression of pc scores (x) vs measured depths (y)

93.149Runs

98.61117Glides

82.5310Rough water runs

72.315Pool

91.7611Low gradient riffle

67.0635High gradient riffle

93.8417Eddy drop zone

Adjusted R2 (%)

Number of pc bands in

regressionNumber of sitesIn-stream habitat

Stream depth mapping: The HABStream depth mapping: The HAB--1 model1 model

Basis for model (developed by Mark Fonstad, SWTU):Basis for model (developed by Mark Fonstad, SWTU):•• Manning equationManning equation•• Jarret’sJarret’s roughnessroughness•• Q, S and W are measuredQ, S and W are measured

Key assumptions:Key assumptions:•• roughness estimate is reasonableroughness estimate is reasonable•• triangle approximation: triangle approximation: DDmaxmax = 2*= 2* DDavgavg

•• avgavg depth depth αα avgavg reflectance signal (implies depth reflectance signal (implies depth driving all variations in reflectance)driving all variations in reflectance)

.

Applying the HABApplying the HAB--1 Model:1 Model:•• acquire Q and S from remote sourcesacquire Q and S from remote sources•• estimate estimate DDavgavg and and DDmaxmax at multiple cross sectionsat multiple cross sections•• determine determine avgavg, max, min reflectances at cross sections, max, min reflectances at cross sections•• regress reflectance (x) regress reflectance (x) vsvs depth (y)depth (y)•• apply regression to remainder of remote sensing image apply regression to remainder of remote sensing image

Stream depth mapping: Input dataStream depth mapping: Input data

Smooth water transects Smooth water transects (green)(green)

Rough water transectsRough water transects (yellow)(yellow)

10

Results: Smooth and rough waterResults: Smooth and rough water

Discussion: Reasons for Discussion: Reasons for inaccuracies with ininaccuracies with in--stream mappingsstream mappings

Reflectance from streams may not be directly Reflectance from streams may not be directly proportional to depth due to:proportional to depth due to:

•• variations in substratevariations in substrate

•• turbulence and “false” turbulenceturbulence and “false” turbulence

•• viewing angle relative to sunviewing angle relative to sun

•• variations in background (mixed pixels)variations in background (mixed pixels)

•• turbidityturbidity

Soda Butte Crk, Silver Gate, Oct, 1999 (9ssg57)Soda Butte Crk, Silver Gate, Oct, 1999 (9ssg57)

Reflectance variations due to substrateReflectance variations due to substrate

Soda Butte Crk, Round Prairie, Oct, 1999 (9srp96)Soda Butte Crk, Round Prairie, Oct, 1999 (9srp96)

Reflectance variations from turbulenceReflectance variations from turbulence

Downstream hydraulic riffle

Downstream hydraulic riffle

Up valley wind creating

Up valley wind creating

waves on a glidewaves on a glide

““False” turbulenceFalse” turbulence

Reflectance variations due to viewing angleReflectance variations due to viewing angle

Looking down at same time and Looking down at same time and site (arrows show same transect)site (arrows show same transect)

Looking towards sunLooking towards sun

Soda Butte Crk,Silver Gate, Oct 1999 (9ssg54)Soda Butte Crk,Silver Gate, Oct 1999 (9ssg54)

Soda Butte Crk,Silver Gate, Oct 1999 (9ssg55)Soda Butte Crk,Silver Gate, Oct 1999 (9ssg55)

Reflectance variations due to mixed pixelsReflectance variations due to mixed pixels

Soda Butte Crk, Oct 1999 (9srp94)Soda Butte Crk, Oct 1999 (9srp94)

One pixelOne pixel

11

Reflectance variations due to turbidityReflectance variations due to turbidity

Depth at both sites along Depth at both sites along Soda Butte Creek is ~1.5 mSoda Butte Creek is ~1.5 m

Discussion: Are the depths accurate?Discussion: Are the depths accurate?rr22 = 0.30, seems low, but = 0.30, seems low, but

•• general trend captured (slope general trend captured (slope ≅≅ 1, y int. 1, y int. ≅≅ 0)0)•• cross sections seem realcross sections seem real•• hydraulics are correct hydraulics are correct

Discussion: The power of visualization Discussion: The power of visualization

Hyperspectral mapping: Downed woodHyperspectral mapping: Downed woodData transformations prior to matched filter mapping:Data transformations prior to matched filter mapping:

•• None None –– use raw untransformed imageryuse raw untransformed imagery•• Principal components (PCA) imagesPrincipal components (PCA) images•• Minimum noise fraction (MNF) imagesMinimum noise fraction (MNF) images

The best classifications occurred with MNF transformed data. The best classifications occurred with MNF transformed data.

Classification approaches used for final maps Classification approaches used for final maps –– matched filter (MF)matched filter (MF)Also evaluated mixture tuned matched filters (MTMF), pixel puritAlso evaluated mixture tuned matched filters (MTMF), pixel purity y indices (PPI), pixel unmixing, spectral angle mapper, maximum indices (PPI), pixel unmixing, spectral angle mapper, maximum likelihood classification. likelihood classification.

Downed wood: Matched filter mappingDowned wood: Matched filter mapping

Create pc imageCreate pc imageChoose training sites, create matched filter images Choose training sites, create matched filter images

and assess accuracy at different and assess accuracy at different thresholds.thresholds.

mf > 0.18mf > 0.18mf > 0.50mf > 0.50

mf > 1.00mf > 1.00

Downed Wood: Matched filter mappingDowned Wood: Matched filter mappingField mapField map

Field realityField reality

One pixelOne pixel

Proportion of wood in each pixel,Proportion of wood in each pixel,red = high, green red = high, green -- medium, blue = lowmedium, blue = low

Matched filter imageMatched filter imageby J. Boardmanby J. Boardman

MF classesMF classes

12

Summary: high definition measurementSummary: high definition measurementOverall Accuracies:Overall Accuracies:InIn--stream habitats:stream habitats: 76% 76% –– 91% 91% Depths:Depths: 67% 67% –– 99%99%Woody debris:Woody debris: 82%82%Algae:Algae: 75%75%Wetlands:Wetlands: 100%100%Vegetation:Vegetation: 93% 93% -- 100%100%

low gradient rifflelow gradient riffleglideglidepoolpoolrunrun

high gradient rifflehigh gradient riffle

eddy drop zoneeddy drop zone

Classification mapClassification map

Summary: Sequence of steps & decisionsSummary: Sequence of steps & decisions(1)(1) Data acquisition decisionsData acquisition decisions

•• Location and timeLocation and time•• Spatial scaleSpatial scale

•• Human features: 1 m Human features: 1 m •• General land cover: 2 to 10 mGeneral land cover: 2 to 10 m•• Streams: 1 mStreams: 1 m•• Downed wood: 1 mDowned wood: 1 m

•• Spectral properties:Spectral properties:•• Optical for cover estimates and stream depthsOptical for cover estimates and stream depths•• Radar for biophysical measurements & moisture detection (not Radar for biophysical measurements & moisture detection (not

assessed in this study)assessed in this study)

(2) Data preprocessing(2) Data preprocessing•• Atmospheric (not done for these studies)Atmospheric (not done for these studies)•• Geometric (especially critical if coregistration to be done)Geometric (especially critical if coregistration to be done)•• Data compression/separation of signal from noise:Data compression/separation of signal from noise:

•• Spectral extraction (biomass, unique spectral signatures)Spectral extraction (biomass, unique spectral signatures)•• PCA PCA –– for water features (develop covariance stats for features for water features (develop covariance stats for features

of interest only of interest only –– mask out other features)mask out other features)•• MNF MNF –– for terrestrial landcover features (develop noise stats for terrestrial landcover features (develop noise stats

over dark homogenous features)over dark homogenous features)

Summary: Sequence of stepsSummary: Sequence of steps

(3) Application of Classification/Mapping Algorithms(3) Application of Classification/Mapping Algorithms•• Terrestrial land cover over broad areas: SAM on MNF Terrestrial land cover over broad areas: SAM on MNF

transformed datatransformed data•• Downed wood: MF on MNF transformed dataDowned wood: MF on MNF transformed data•• Streams:Streams:

•• InIn--stream habitats: Maximum likelihood on PCA datastream habitats: Maximum likelihood on PCA data•• Depths:Depths:

•• Multiple regression on PCA if ground data availableMultiple regression on PCA if ground data available•• HABHAB--1 model on raw spectral bands if no depth data 1 model on raw spectral bands if no depth data

from groundfrom ground

(4) Map production(4) Map production•• Assignment of data to classes (e.g. SAM rule images)Assignment of data to classes (e.g. SAM rule images)•• Data overlays to generate comprehensive mapsData overlays to generate comprehensive maps

SummarySummaryThere is great potential for HSRH mapping of riparian environmenThere is great potential for HSRH mapping of riparian environments.ts.

But further work is needed:But further work is needed:•• Rationale for choosing specific data compression and Rationale for choosing specific data compression and

classification algorithms for different applicationsclassification algorithms for different applications•• Evaluation of physical factors driving the signalsEvaluation of physical factors driving the signals•• Alternative modes of assessing accuracyAlternative modes of assessing accuracy•• Search for spectrally driven descriptors of streamsSearch for spectrally driven descriptors of streams•• High precision coregistration for change detectionHigh precision coregistration for change detection•• Incorporation of approaches into automated software sequencesIncorporation of approaches into automated software sequences•• Ethics of remote measurementEthics of remote measurement


What is Hyperspectral Imagery? - University of...

Documents

Transcript of What is Hyperspectral Imagery? - University of...