Welcome to Scribd. Sign in or start your free trial to enjoy unlimited e-books, audiobooks & documents.Find out more
Download
Standard view
Full view
of .
Look up keyword
Like this
2Activity
0 of .
Results for:
No results containing your search query
P. 1
Comparación de técnicas de detección de cambios para monitoreo desmonte y regeneración en bosques tropicales en una serie de tiempo.

Comparación de técnicas de detección de cambios para monitoreo desmonte y regeneración en bosques tropicales en una serie de tiempo.

Ratings: (0)|Views: 66|Likes:

More info:

Published by: Armando Rodriguez Montellano on Sep 22, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

01/20/2012

pdf

text

original

 
Comparison of ChangeDetection Techniques forMonitoring Tropical Forest Clearing andVegetation Regrowth in
a
Time Series
Daniel
J.
Hayes
and Steven
A.
Sader
Abstract
The once remote and inaccessible forests of Guatemala's MayaBiosphere Reserve
(MBR)
have recently experienced
high
mtesof deforestation corresponding to human migration andexpansion of the agricultural frontier. Given the importanceof land-cover and land-use change data in conservationplanning, accurate and efficient techniques to detect forestchange from multi-tempoml satellite imagery were desired forimplementation by local conservation organizations. Threedates of Landsat Thematic Mapper imagery, each acquired twoyears apart, were radiometrically normalized and pre-processed to remove clouds, water, and wetlands, prior toemploying the change-detection algorithm. Three change-detection methods were evaluated: normalized differencevegetation index
(NDVI)
image differencing, principalcomponent analysis, and
RGB-NDVI
change detection.
A
technique to generate reference points by visual interpretationof color composite Landsat images, for Kappa-optimizingthresholding and accuracy assessment, was employed. Thehighest overall accuracy was achieved with the
RGB-IVDVI
method
(85
percent). This method was also preferred for itssimplicity in design and ease in interpretation, which wereimportant considerations for transferring remote sensingtechnology to local and international non-governmentalorganizations.
Introduction
With rapid changes in land-cover occurring over large areas,remote sensing technology is an essential tool in monitoringtropical forest conditions. The remote and inaccessible natureof many tropical forest regions limits the feasibility of ground-based inventory and monitoring methods for extensive landareas. Initiatives to monitor land-cover and land-use changeare increasingly reliant on information derived from remotelysensed data. Such information provides the data link to othertechniques designed to understand the human processesbehind deforestation (Lambin, 1994; Rindfuss and Stern,1998).
An
array of techniques are available to detect land-coverchanges from multi-temporal remote sensing data sets Uen-sen, 1996; Coppin and Bauer, 1996). The goal of change detec-tion is to discern those areas on digital images that depictchange features of interest (e.g., forest clearing or land-covertland-use change) between two or more image dates. Onemethod, image differencing, is simply the subtraction of thepixel digital values of an image recorded at one date from thecorresponding pixel values of the second date. The histogram
Maint:ImagoAllalysisLnl~oratory.)cparlment
of
E'oreslManag(:-ment.
5755
Nutting
Hall,Il~iive~.silv
f
Maine.Orono,ME
04469
[~l11ayr!s@imc~1~fa.111:1i11e.e(1~1;
adr:r@~~rnonf~~.~~i~~i~~e~r:~l~~).
of the resulting image depicts a range of pixel values from neg-ative to positive numbers, where those clustered around zerorepresent no change and those at either tail representreflectance changes from one image date to the next (Jensen,1996). This method has been documented widely in change-detection research (Singh, 1986; Muchoney and Haack, 1994;Green et al., 1994; Coppin and Bauer, 1996; Macleod and Con-galton, 1998). Some investigators favor this method for itsaccuracy, simplicity
in
computation, and ease ininterpretation.One difficulty encountered in employing image differenc-ing for change detection is the selection of the appropriatethreshold values in the histogram that separates real and spuri-ous change. The subjectivity of threshold placement may beimproved by the analyst's familiarity with the study area aswell as access to ancillary data such as field information,
GIS
data, and/or matching dates of aerial photography. Fung andLeDrew (1988) tested quantitative methods for developingthese threshold levels using accuracy indices. They recom-mended the Kappa coefficient of agreement in determining anoptimal threshold level, being based on an error matrix ofimage data against known reference data.Image differencing, although mathematically simple,allows for only one band of information to be processed at atime. Other techniques incorporate multiple bands of data forchange detection. Several studies have demonstrated the util-ity of the principal component analysis
(PCA)
technique inmulti-temporal image analysis (Byrne et al., 1980; Fung andLeDrew, 1988; Muchoney and Haack, 1994; Coppin and Bauer,1996; Macleod and Congalton, 1998). The results of using the
PCA
transform on two dates of imagery are contrary to that of itstypical, one-date transformations. In multi-temporal analysis,the first two components tend to represent variation associated
with
unchanged land-cover and overall image noise (i.e.,atmospheric and seasonal variation), while the third and latercomponents are of more interest in identifying change areas(Byrne et al., 1980). Previous studies have confirmed that theminor components have been successful in detecting land-cover changes (Byrne et al., 1980; Fung and LeDrew, 1987)when the areas affected by change of interest occupy a smallproportion of the study area (Fung and LeDrew, 1987; Macleodand Congalton, 1998).Image differencing using band ratios or vegetation indicesis another technique commonly employed for land-cover
I'hotogra~nm(!tric:Ellgil~cering
&
Ktr~l~oteensing
Vol.
67.
No.
$1,
Ssptc:nlt)er2001.
111).
10(i7-1075.
~0~~~-1112/01/670~~-10ti7S3.00/0
6?
2001
A~nc!ric:anSociety
for
Pirotogr~rmmctrv
HII~
io1110t~:ensing
 
changedetection.Forexample,thenormalizeddifferenceveg- etationindex(NDVI)asdevelopedforuseinidentifying healthandvigorinvegetation,aswellasforestimates ofgreen biomass.TheNDvI,thenormalizeddifferenceofbrightnessval- uesfromthenearinfraredandvisibleredbands,hasbeen foundtobehighlycorrelatedwithcrownclosure,leafarea index,andothervegetationparameters (Tucker,1979;Sellers, 1985;Singh,1986;Runninget
al.,
1986).Lyonet
al.
(1998)comparedsevenvegetationindicestodetectland-cover changeinaChiapas,Mexicostudysite.Theyreportedthatthe
NDVI
wasleast affectedbytopographicfactorsandwastheonly indexthatshowedhistogramswithnormaldistributions. Changeincanopycoverorvegetationbiomasscanbedetected byanalyzing
NDVI
valuesfromseparatedates(e.g.,
NDvI
imagedifferencing).SaderandWinne(1992)developedatechniquetovisual- izechangeusingthreedatesof
NDVI
imageryconcurrentlyandinterpretationconceptsofcoloradditivetheory.Bysimultane- ouslyprojectingeachdateof
NDVI
throughthered,green,and blue
(RGB)
computerdisplaywritefunctions,majorchangesin
NDVI
(and,hence,greenbiomass)betweendateswillappearin combinationsoftheprimary(RGB)rcomplimentary(yellow, magenta,cyan)colors.Knowingwhichdateof
NDVI
iscoupledwitheachdisplaycolor,theanalystcanvisuallyinterpretthe magnitudeanddirectionofbiomasschangesinthestudyarea overthethreedates.Automatedclassificationcanbeper- formedonthreeormoredates of
NDVI
byunsupervisedclusteranalysis(Saderet
al.,
2001).Changeandno-changecategories arelabeledanddatedbyinterpreteranalysisoftheclustersta- tisticaldataandguidedbyvisualinterpretationof
RGB-~vI
colorcomposites.Study
Area
and
BackgroundSpanningapproximately
2
millionhectaresofnorthernGuate- mala,theMayaBiosphereReserve
(MBR)
isanareaoflowland tropicalforestsandexpansivefreshwaterwetlands,partofthe largestcontinuoustropicalmoist forestremaininginCentral America(Nationset
al.,
1998).The
MBR
isacomplexofdeline- atedmanagementunits,includingfivenationalparks,four biologicalreserves(biotopos),amultipleusezone,andabuffer zone(Figure
I).
Theonceremoteandinaccessibleforestsofthe regionhaveexperiencedhighratesofdeforestationinthelast decade,correspondingtohumanmigrationandexpansionof theagriculturalfrontier(Saderet
al.,
1997).Sader andcolleagues(Saderet
al.,
1997;Saderet
al.,
2001)havemonitoredratesandtrendsofforestclearingusingLand- satThematicMapper
(TM)
mageryfromthemid-1980stolate 1990s.Guatemalangovernmentagenciesandnon-governmen- talorganizations
(NGOS)
relyonregularlyupdatedmapsofthe
MBR
tomonitordeforestationpatternsanddisturbanceinsen- sitiveareasofthereserve.Internationaldonoragenciesrequire theNGOstoquantifyforestclearingratesattwo-yearintervals. Accurateandefficienttechniquesforextractingquantitative forest-changedatafromremotelysensedimagesareneededto supportthe
MBR
forestmonitoringprogram.Furthermore,these dataareneededforanalysiswithcommunitylevelsocio-eco- nomicsurveydataconcerningthedrivingforcesofenviron- mentalchangeinthe
MBR
(Schwartz,1998;Hayes,1999). Thispaperdescribesthetechniquesusedtoprocessand validatemulti-temporalLandsat
TM
imagery(threedates)for obtainingtime-seriesforestclearingandregrowthdatainthe MBR.Threechange-detectionmethodsarecompared:
NDM
imagedifferencing,
PCA
changedetection,andRGB-NDVIlassi- fication.
A
visual interpretation techniquetogeneraterefer- encepointsfromcolorcompositeLandsatimages,forselecting Kappa-optimizinghresholdsandforassessmentofclassifica- tionaccuracy,isdescribed.Thegoalistodeterminethemost accurateandefficientmethodtodetectforestchangeinthe
1068
September
2001
Figure
1.
Locationofthestudyarea(LandsatWRSPath 20/Row
48,1997
TM
band
5
shown)
in
relationtothemanagementunitsoftheMaya BiosphereReserve,
El
Peten,Guatemala.
MBR'S
tropical moist forest and to facilitate the transfer of thistechnology to the local NGOs.Data Acquisltlon and Pre-ProcessingThree dates of Landsat
TM
imagery (1993, 1995, 1997) forWorldwide Reference System path 20, row 48 were acquired.This Landsat scene comprises approximately 90 percent of the
MBR
and buffer zone (Figure
1).
o reduce scene-to-scene varia-tion due to sun angle, soil moisture, atmospheric condition,and vegetation phenology differences, all data were collectedbetween the months of March and May, corresponding to the
MBR'S
dry season. Each scene was georeferenced to a pre-viously rectified 1995
TM
image.
TM
bands
3
(visible red), 4(near inhared), and
5
(mid-infrared) were extracted from theoriginal
TM
data sets to reduce between-band correlation, datavolume, and processing time. Previous studies have shownthat selecting one band each from the visible, near infrared, andmid-infrared spectral regions results in the optimal wavebandcombination for vegetation discrimination (DeGloria, 1984;Horler and Ahern, 1986; Sader, 1989).Bands
3,4,
and 5 wereinput into "isodata" (ERDAS, 1997),an unsupervised classifi-cation module, to produce 200 spectral clusters. Binary imageswere created to isolate water, clouds, and cloud shadowsthrough a combination of analyst definition of cloudlwaterclusters and on-screen editing. A previously developed imageof non-forested wetlands and natural savannas was also added
PHOTOGRAMMETRIC ENGINEERING
&
REMOTE SENSING
 
to the cloud and water image. These classes, being of no interestto forest clearing and regrowth analysis, were masked for all
and
dates of imagery to avoid confusion in the change-detectionclassification.
Radiometric
Nonnallzatlon
Arelativeradiometriccalibrationtechniquewasappliedto eachbandfromeachdateofimagery.Thetechniqueincorpo- ratedlinearregressionmethodsreportedbyEckhardtetal. (1990),Halletal.(1991),andJensenetal.(1995).The1997
TM
scene,whichwascorrectedforsensorgainandbias,wasused asthereferenceimagetowhichthe1993and1995datawere normalized.First,normalizationtargetswereselectedfromthe wet(e.g.,deep,clearwater)anddry(e.g.,urbanfeatures)non- vegetatedextremesofeachband
(TM
3,4,and5)ateachdate (1993,1995,1997)byvisualinterpretationoftheimageryand queryingthedigitalnumbersofpixelsrepresentingthesefea- tures.Theselectioncriteriawerebasedonproceduresoutlined byEckhardtetal.(1990).Eachtargetconsistedofananalyst- definedareaofinterest(AOI),whichincluded thegreatest numberofpixelscoveringthetarget,whosedigitalnumbers
(DNS)
werelocatedat theextremesoftheimagehistogramand collectivelycontainedlowvariance. ThemeanvalueofthepixelDNswasgeneratedforeachof the normalizationtarget
AOIS
(eachband,eachdate).The parametersusedinthelinearregressionequationwerecalcu- latedbythefollowing"rectificationtransform"(Halletal., 1991):where
Br
is the mean
DN
for the bright target of the referenceimage, Bs is the mean
DN
for the bright target of the subjectimage, Dr is the mean
DN
for the dark target of the referenceimage, and
Ds
s the mean
DN
for the dark target of the subjectimage. Using linear regression, the corrected pixel values forthe subject image
(Y)
were calculated from the original
DN
(XI,
for each band
(11,
by the following equation:
Changehtectlon Methods
Three change-detection methods (NDVIifferencing, PCA, and
RGB-NDW
classification) were independently applied to thecloudlwater-masked and radiometrically normalized time-series
Tbi
data set. A three-date forest change-detection classi-fication of the selected study area was generated from eachmethod. Each method was evaluated and compared with theother methods on its ability to classify temporal states in forestcover(i.e., cleared, regrown, no change) over the three timeperiods. The methods were evaluated and contrasted on thebasis of classification accuracy (Congalton,1991),efficiency ncomputation and processing, and ease in interpretation.NDVI Image DifferencingDifference images were created by first calculating NDvI valuesfor each date (j)of imagery by the following equatioa:Principal ComponentAnalysisThe principal component transformation was performed sepa-rately on two data sets (1993 and 1995,1995 and 1997)usingthree
TM
bands (3,4, and 5) for each date. Each two-date data setcontained six bands. The transformation used the "prince"routine (ERDAS 1997), modified to calculate the transformfrom a correlation matrix of the data set. Several authors havecompared this "standardized" approach to
PCA
against transfor-mations based on the covariance matrix (Conese et al., 1988;Eastman and
Falk,
1993; Rencher 1995). Reported advantagesof the standardized approach include improved interpretabil-ity, the isolation of seasonal effects and variability due to noise,better statistical control, and more precise classification. Foreach data set, the "standardized"
PCA
routine output includedsix component images, a table of eigenvalues quantifying theproportion of variance explained by each component, and amatrix of eigenvectors (weights or factor loadings) depictingbetween-date correlation for each band with each component.Components that represent change typically show an ibsenceof correlation amone bands between dates IBvrne et al.. 19801.The component that'best highlights the chGge of interest is
'
chosen for thresholding, using visual interpretation of compo-nent images and analysis of the eigenvector matrix.Image interpretation was based on the assessment of spa-tial continuity, by seeking out the components that express thedifferences in the changes of interest as spatially discontinuousareas within the image. The eigenvector analysis examined thealgebraic signs on the weights. Differences between dates areexpressed by the weight of one band at one date having anopposite sign to that of the same band of the other date. Basedon these criteria, two of the six components (components3 and4 for each two-date data set) were selected from the
PCA
forthresholding of no-change and change areas.
Of
these two com-ponents, the one that showed the highest ability to thresholdforest
clearinglno-changelregrowth
i.e., the highest estimatedKappa according to the reference sample points) was chosen forfinal classification.
RGB-NDVZ
lassification
NDVI
valuesfromthreedates(ascalculatedbyEquation3)were classifiedinto50spectralclusters.Foreachclusterclass,the mean
NDVI
valuesateachdate(1993,1995,1997)werecatego- rizedasveryhigh,high,medium-high,medium,medium-low, low,orverylow,basedonthedistributionof
~~vr
aluesoverthestudyarea.Theselevels of
NDVI
wereestablishedonthe observationthat,becausemostofthestudyareaiscomposedof undisturbedforest,valueswithin
r
0.5standarddeviationsfromthemeanrepresented
high
greenbiomass(highmean NDVI).The other
NDVI
levelsweresetatintervalsof0.5stan- darddeviationsoutwardfromthemean.Eachclusterwas examinedforchangesin
NDVI
levelsovertime.Clusterswere namedaccordingtotypeofchange(clearing,regrowth,orno change)andthecorrespondingtimeperiod(s)ofchange accordingtothe
NDW
levelsastheyrelatedtothree-date
RGB-NDVI
interpretation(Plate
1).
(TM4
-
m3)
NDw[jl
=
(TMI
+
TM~)
(3)
ClasslfylngtheChangeImages
Boththe
NDVI
differencingandthe
PCA
methodsresultinimageswithan8-bit(0to255)datarange.Thresholdsmust be Twodifferencemageswerecreatedby subtractingonedateof identifiedalongthehistogramstoseparate change(bothclear-
NDVI
valuesfrom thoseofthepreviousdate,sothat ingandregrowth)fromnochange.Thresholdlevelswereset
PHOTOGRAMMETRICENGINEERING
81
REMOTE
SENSING
September
200
1089

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->