ArticlePDF Available

Effect of different radiation correction methods of Landsat TM data on land-cover remote sensing classification

Authors:
  • institute of remote sensing and digital earth, CAS

Abstract and Figures

Radiometric correction, which is often used as a preliminary step for remote sensing classification, is required in the efficient extraction of land-cover information from remote sensing images, particularly in areas with rough terrain. The objective of this study is to examine the effect of different radiation correction methods of Landsat TM data on land-cover remote sensing classification. Three different radiometric correction methods (ATCOR 3, FLAASH and Look-Up Table (LUT)) were introduced, and the Maximum Likelihood Classifier (MLC) and Support Vector Machines (SVMs) were used for classifiers. The training samples were selected from three corrected images based on geographical coordinates and classifier training to classify the three corrected images mutually. The results indicate the following. (1) The samples selected from the classification image for classifier training significantly improve classification accuracy compared with the samples selected from other images. (2) The classification results of the three radiometric correction images are different, whereas the results of the ATCOR3 and FLAASH methods are similar. (3) The effect of radiometric correction on the classification accuracy of each category is different: radiometric correction affects "Forest" significantly and other categories variably.
Content may be subject to copyright.
1007-4619201402-0320-15 Journal of Remote Sensing 遥感学报
Effect of different radiation correction methods of Landsat
TM data on land-cover remote sensing classification
CHEN Chenxin12HU Changmiao1HUO Lianzhi1TANG Ping1
1. Institute of Remote Sensing and Digital EarthChinese Academy of SciencesBeijing 100101 China
2 Technology and Engineering Center for Space UtilizationChinese Academy of SciencesBeijing 100094China
AbstractRadiometric correctionwhich is often used as a preliminary step for remote sensing classificationis required in the e
fficient extraction of land-cover information from remote sensing imagesparticularly in areas with rough terrain. The objective of
this study is to examine the effect of different radiation correction methods of Landsat TM data on land-cover remote sensing classifi-
cation. Three different radiometric correction methods ATCOR 3FLAASH and Look-Up Table LUT) ) were introducedand
the Maximum Likelihood Classifier MLCand Support Vector Machines SVMswere used for classifiers. The training samples
were s elected from three corrected images based on geographical coordinates and classifier training to classify the three corrected
images mutually. The results indicate the following. 1The samples selected from the classification image for classifier training
significantly improve classification accuracy compared with the samples selected from other images. 2The classification results of
the three radiometric correction images are differentwhereas the results of the ATCOR3 and FLAASH methods are similar. 3
The effect of radiometric correction on the classification accuracy of each category is differentradiometric correction affects
Forestsignificantly and other categories variably.
Key wordsradiation correctionLandsat TMMLCSVMsland-cover classificationaccuracy assessment
CLC numberTP79 Document codeA
Citation formatChen C XHu C MHuo L Z and Tang P. 2014. Effect of different radiation correction methods of Landsat TM d
ata on land-cover remote sensing classification. Journal of R emote Sensing1 82) : 320 - 334 DOI10 . 11834 /
jrs. 20143211
Received2013-08-26Accepted2013-11 -27Version of record first published2013-12-04
FoundationHigh Technology Research and Development Project of China 863 Program) ( No. 2009AA122002
First author biographyCHEN Chenxin 1983) ,maleresearch assistant. He majors in image processing. E-mailchencx@ csu. ac. cn
Corresponding author biographyTANG Ping1968) ,femaleprofessor. Her main research interest is image processing. E-mailtangping@ radi.
ac. cn
1 INTRODUCTION
Landsat TM images have been widely used for land-cover
mapping via image classification and for identifying changes via
change detection Foxet al1983Jensenet al1993
Yuanet al. 2005Sextonet al2013. Howevercon-
straints exist in using Landsat data for land-cover studies. Land-
sat data is affected by many factorsincluding illumination chan-
gesatmospheric effectsand topographic effects in areas with
rough terrain. Given that the electromagnetic signals collected by
satellites in the solar spectrum is modified by the scattering and a
bsorption of gas and aerosols while travelling through the a
tmosphere from the Earth's surface to the sensor Songet al.
2001) ,the image information is often inconsistent with the actu-
al surfacethis discrepancy affects the accuracy of information e
xtraction. Thereforeatmospheric and topographic correction is
used as a primary task before further processing Honkavaara
Hakalaet al. 2012
Radiometric correction is often used for multi-temporal
Zhanget al2006Vicente-Serranoet al2008Tanet
al. 2012Xuet al. 2012 ) ,multi-sensor Xu & Zhang
2013) ,and large-scale remote sensing data analyses Olthofet
al. 2005 Hansen & Loveland2012 Studies have shown
that superior r emote sensing information can be extracted from
radiometric corrected imagessuch as images from ecological ap-
plications Guet al2011and land-use / land-cover classifi-
cation Meyeret al1993 Pons & Solesugranes1994Colby
& Keating 1998Dorrenet al2003Kobayashi & Sanga-
Ngoie2009Habibe t al2011 . Howeverfew studies
have reported the effects of TM radiometric correction on sample
collection strategies and land-cover classification.
This research investigated the effects of the different radia-
tion correction methods of Landsat TM data on land-cover remote
sensing classification. The study was conducted as follows.
Firstradiometric correction was performed on images by using
three different radiation correction methods. Secondsamples
were selected in one corrected imagethe samples were used for
other image classification procedures. Finallythe classification
accuracy was performed. The position and number of training
samples remained unchanged. The classification algorithms used
in this study were the Maximum Likelihood Classifier MLC
and Support Vector Machines SVMs
CHEN Chenxinet alEffect of different radiation correction methods of Landsat
TM data on land-cover remote sensing classification 321
2 STUDY AREA AND DATA
2. 1 Study area
The study area is the north central mountain regions of
Shaanxi Province in northwest China. Fig. 1 shows the location of
the study area The Landsat TM scene is displayed as an RGB
image = NIRG = redB = green The region is loca-
ted in the Loess Plateau and has a cold arid climate or cold semi-
arid climate. The study area covers a wide range of mountain
chains and some small hills. The dominant land covers include
forestsbare soiland farmlands. Farmlands are mixed with
scattered rural villages in flat low areasand steep mountainous
areas are dominated by forest species.
2. 2 Data
The final data set used in this study comprised an original
Landsat TM image and three radiometric corrections images with
spatial resolutions of 30 m. The original image was acquired on
May 122000 with good visibility and without clouds from a
Landsat satellite Path127Row35 The geometric correc-
tion was conducted by the data provider.
2. 3 Classification system and sample selection
The classification system and class definitions of Gonge
t al. 2013were used in this study. According to the land-cov-
er characteristics of the study areathis paper divided land-cover
categories into six categoriesnamelybare soilcropforesti
mpervious areawaterand hill.
Table 1 shows the characteristicsdescriptionand distribu-
tion of the samples in the training and test sets among the six
land-cover classes that characterize the scene under investiga-
tion. These samples were collected by geological image interpre-
tation via Google Earth. The samples were extracted from differ-
ent spatial positions in the study area. Firstthe visual interpre-
tation of the spatial position of the classes was conducted by u-
sing a Landsat TM image. The spatial positions were navigated in
Google earth for a fieldtest. High-quality sample sets were
then obtained. Secondthe sample was optimized and duplicate
sample points were removed. Finallythe samples were randomly
split into training samples and test samples according to a certain
proportion. The total training and testing samples of all six clas-
ses were 2445 and 477 respectively Table 1
3 RADIOMETRIC CORRECTION
In this paperthree different approaches were used for radi-
ation correctionFLAASH 2013 - 08 - 01http/ / www. ex-
elisvis. com / docs / FLAASH. html) ,ATCOR3( [2013 - 08 - 01
http/ /www. rese. ch / pdf / atcor3 _manual. pdf) ,and the Look
Up Table LUTmethod Lianget al1997 ) ,which is wide-
ly used by Landsat TM/Enhanced TM + data irradiation correc-
tion. FLAASH and ATCOR are commercial atmospheric correc-
tion software modules that support the vast majority of multi-spec-
tral and h yper-spectral r emote sensing data for atmospheric cor-
rection processing. FLAASH and ATCOR are both based on ra-
diative transfer theory but differ in radiative transfer equations.
The LUT a pproach is a highly flexible approach that is widely a-
dopted by national researchers.
The FLAASH atmospheric correction package was developed
by Spectral ScienceAmerican Aerodynamics Research Labora-
toryand Spectral Information Technology A pplications Center.
FLAASH works at a 400 nm to 2500 nm wavelength range and u-
ses MODTRAN 4 to generate a series of atmospheric parameter
LUTs. The radiative transfer model of FLAASH considers the
proximity effect but not terrain correction.
322 Journal of Remote Sensing 遥感学报 2014182
ATCOR radiometric correction software was developed by
the German Aerospace Center ATCOR has three versionsA
TCOR2ATCOR3and ATCOR4 Richter2005. ATCOR2
and ATCOR3 has been integrated in remote sensing software ER-
DAS. We adopted ATCOR3 in the current study to process
Landsat TM data. ATCOR3 was mainly used for the atmospheric
and terrain corrections of the remote sensing images of mountains
and regions. ATCOR3 contains MODTRAN4-calculated LUTs of
various atmospheric conditionsatmospheric transmittancepath
radianceand direct and diffuse solar radiation flux. ATCOR3
can use built-in modules to calculate the slopeaspectmoun-
tainous terrain shadingand other data to complete terrain cor-
rection. For areas with large incidence anglesATCOR3 pro-
vides a reliable empirical bi-directional r eflectance distribution
function. ATCOR3 uses different algorithms for visible and infra-
red radiation correctionand provides history-removing modules.
LUT-based atmospheric correction assumes that the surface
is Lambertian. The radiative transfer equation is expressed as
L = Lp+ T Eρ
π1ρS1
where Lis the apparent radianceLpis the atmospheric path ra-
diationρis the surface reflectanceEis the surface irradiance
Sis the atmospheric albedoand Tis the transmittance of the
surface to the top of the atmosphere. Parameters ETLpand
Scan be directly calculated by 6S Second Simulation of Satel-
lite Signal in the Solar Spectrum Vermoteet al. 1997 ρ
can be calculated when Lis known.
The LUT algorithm for Landsat TM data was developed
based on the method of Lianget al. 1997) ,which uses dense
dark vegetation Kaufmanet al1997 to estimate the aerosol
thickness at the vegetation and the moving window interpolation
method to estimate the distribution of the whole aerosol scene.
A semi-empirical smooth terrain correction method was used
after the LUT-based atmospheric correction. Firstthe terrain
slope angle was smoothened. Thereafterthe smoothed c osine
terrain correction was used. Smooth terrain treatment was used to
eliminate the non-Lambertian effects caused by abnormal radia-
tion. Shuttle radar topography mission data Version 4. 1was
used for atmospheric correction. The spatial resolution of the ele-
vation data was 90 m.
4 EXPERIMENT AND RESULT
The effects of different radiometric corrections on the classi-
fication results were assessed by comparing the images in d
ifferent classifications after radiometric corrections. To evaluate
the effect of radiometric correction on classification accuracywe
conducted nine experiments by using two classifiersnamely
MLC and SVMs.
4. 1 Classifier
A number of methods have been used for remote sensing
classification. These methods can be grouped as supervised / un-
supervisedparametric / nonparametrichard / softper pixelper
s ubpixelor per field Lu & Weng2007 We selected two
widely used supervised approachesMLC and SVMsto test the
effects of radiometric correction on the classification process.
MLC is a parametric classifier based on statistical theory
and is one of the most widely used classifiers Huanget al.
2002 Among remote sensing classification methodsMLC can
clearly explain parameter abilitiescan be easily integrated with
prior knowledgeand can be easily implemented. By using the
statistical parameters mean vector and variance-covariance m
CHEN Chenxinet alEffect of different radiation correction methods of Landsat
TM data on land-cover remote sensing classification 323
atrix) ,the probability of a pixel belonging to a predefined set of
classes is calculated. The pixel is then assigned to the class with
the highest probability Tso & Mather2009 SVMs are based
on the structural risk minimization principle and their p opularity
within the remote sensing community is constantly i ncreasing be-
cause of their properties and intrinsic effectiveness Lorenzo &
Lorenzo2006 The Radial Basis Function RBFwas chosen
as the kernel function of SVMsand RBF model p arameters
were established by using the cross-validation method in the mod-
el training process. The data was normalized to 01during the
process by LIBSVM Chang & Lin2011
4. 2 Flowchart
Fig. 2 shows the experimental workflow. Firstby using
three radiometric correction methods to correct the original Land-
sat TM imagecorrected images RC_ARC_Band RC_C were
used in the next process. Secondtraining samples Samp _ a
Samp_band Samp_cwere selected from the above three i mages
by using the same regions of interest and the classifiers MLC and
SVMswere obtained. Finallythe classifiers were used for the
land-cover classification of the four images. The a ccuracy of the
classified image was assessed by using a range of reference data
through the confusion matrix. The producer and user accuracies
along with the overall accuracy and Kappa coefficient statistic for
each classwere calculated from the confusion matrix.
Fig. 2 Experimental workflow
4. 3 Classification result
Fig. 3 shows the classification maps with the best accuracies
obtained by MLC and SVMs. The left image shows the classifica-
tion result of the RC_A image when samples were collected by u-
sing MLC. The right image shows the classification result of the
RC_C image when samples were collected by using SVMs.
Fig. 3 Resultant classification images with the best accuracies obtained by MLC and SVMs
The classification accuracies obtained for the MLC and
SVMs are reported in terms of overall accuracy and Kappa coeffi-
cient in Table 2. These results show that the use of samples from
the classification image provides higher accuracies than other
combinations. F urthermorethe classification results of the
three r adiometric correction image are differentwhereas the r
esults for RC_A ATCOR3and RC_B FLAASHare similar.
The result for RC_C is different from the results of RC _ A and
R C_B.
Table 2 Comparison of classification accuracy on the basis of different sample collection
strategies after radiometric correction MLCSVMs
Samples RC_A RC_B RC_C
OA /% Kappa OA /% Kappa OA /% Kappa
RC_A 93. 29 93. 50 0. 920 0. 922 91. 61 88. 26 0. 899 0. 859 77. 57 88. 05 0. 731 0. 857
RC_B 90. 36 87 . 63 0. 8840 . 851 93. 2993 . 50 0. 9200 . 922 74. 2188 . 05 0. 691 0. 857
RC_C 78. 05 84 . 49 0 . 7020. 814 79. 2085. 95 0 . 7500. 831 91 . 2093. 71 0. 8940 . 925
324 Journal of Remote Sensing 感学报 2014182
Table 3 provides the category classification accuracy of e
xperiments that have high overall classification accuracies for
MLC and SVMsi. e. the samples collected from the classifica-
tion image.
Table 3 Summary of experiment combinations that have high classification accuracies MLCSVMs
Class name RC_ARC_A RC _BRC_B RC_C RC_C
PA /% UA / % PA / % UA / % PA/ % UA /%
Bare soil 90. 0090 . 00 91. 1490 . 00 90. 0088 . 75 87. 8088 . 75 88. 7591 . 25 85. 54
87. 95
Crop 95. 00 95. 00 96. 20 96. 20 93. 75 95. 00 97. 40 98. 70 85. 00 96. 25 98. 55 97. 47
Forest 96. 25 98. 75 100. 0 98. 75 97. 50 98. 75 98. 73 97. 53 97. 50 97. 50 98. 73 98. 73
Hill 91. 25 92. 50 82. 02 84. 09 90. 00 91. 25 83. 72 82. 95 86. 25 90. 00 82. 14 85. 71
Impervious 90. 00 90. 00 92. 31 93. 51 91. 25 92. 50 93. 59 94. 87 92. 50 92. 50 85. 06 93. 67
Water 97. 40 94 . 81 100. 0100 . 0 97. 4094 . 81 100. 0100 . 0 97. 4094 . 81 100. 0 100. 0
Overall accuracy / % 93. 29 93. 50 93. 29 93. 50 91. 20 93. 71
Kappa 0. 920 0. 922 0. 920 0 . 922 0. 894
0. 925
NotePAproduction accuracyUAuse accuracy.
Fig. 4 provides the changes in different categories in each
band after three radiometric corrections. The three radiation c
orrection methods have similar results for RC_ A and RC _ B
whereas the result of RC_C is slightly different from the results
of the former The surface reflectance values of each class w
ith the LUT method are significantly lower than the FLAASH
and ATCOR methods after correctionspecifically for the first
band.
Fig. 4 Changes of different categories in each band before and after radiometric correction
CHEN Chenxinet alEffect of different radiation correction methods of Landsat
TM data on land-cover remote sensing classification 325
By comparing the classification results of three radiometric
correction imagesthe classification accuracy shows varying d
egrees of changes for almost all categories. For specific catego-
riesthe classification result of Forestchanges significantly a
fter radiometric correctionthus indicating that this category is
sensitive to any radiometric correction Fig. 5 This result can
be attributed to the use of the dark vegetation method in radio-
metric correction to estimate aerosol thickness. The obtained re-
sults r equire further analysis.
Fig. 5 Classification accuracy of different categories
after radiometric correction MLC
5 CONCLUSION
The radiometric correction of remote sensing imagery is a
prerequisite of data processing for many research applications.
The three different radiometric correction methods described in
Section 3 were applied to correct the original Landsat TM image.
Nine experiments were conducted by using MLC and SVMsi
ncluding selected training samples from three corrected images
and classifier trainingto classify the three corrected images m
utually and examine the effect of different radiation correction
methods of Landsat TM data on land-cover remote sensing classi-
fication.
The following preliminary conclusions are obtained
1The samples selected from the classification image for
classifier training significantly improves the classification accura-
cy compared with the sample selected from other imagese. g.
the overall classification accuracy of RC_ARC_A is 9 3. 29%
by MLCfollowed by RC_BRC_B 93. 29% and RC_CRC
_C 91. 20% These results are significantly higher than the
results obtained by using other combinations.
2The classification results of the three radiometric cor-
rection images are differentwhereas the results of ATCOR3 and
FLAASH are similar. The overall classification accuracy of RC_
C has obvious differences with the overall classification accura-
cies of RC_A and RC _Bthus indicating that ATCOR3 is con-
sistent with the FLAASH method. By contrastthe LUT method
exhibits some deviations probably because of the error in aerosol
distribution estimation Fig. 4
3The effects of radiometric correction on the classifica-
tion accuracy for each category are different. Radiometric correc-
tion affectsForestsignificantly and other categories variably
The SVMs approach produces higher classification accuracy than
MLChoweverboth produce similar r esults.
The study area of this paper is smallthus limiting the
scope of the studyspecifically on global topographies and land-
cover types. The future direction of this study will be directed to-
ward expanding the scope of the study area by using the results of
this study. The results of this paper do not reflect the effects of
terrain correction on classification accuracy. Follow-up studies
will focus on the sample selection and experiment process d
esign.
AcknowledgementsThe authors would like to express
their sincerest appreciations to their laboratory colleagues for
their valuable comments and assistance during this work.
REFERENCES
Chang C C and Lin C J. 2011. LIBSVMa library for support vector ma-
chines. ACM Transactions on Intelligent Systems and Technology2
3) : Article No. 27 DOI10 . 1145 /1961189. 1961199
Colby J D and Keating P L. 1998 Land cover classification using Land-
sat TM/ ETM + imagery in the tropical highlandsthe influence of
anisotropic reflectance. International Journal of Remote Sensing19
8) : 1479 - 1500 DOI10 . 1080 /014311698215306
Dorren L K AMaier B and Seijmonsbergen A C. 2003. Improved Land-
sat-based forest mapping in steep mountainous terrain using object-
based classification. Forest Ecology and Management1831 /3 ) :
31 - 46 DOI10. 1016 /S0378 - 1127 0300113 - 0
Fox LMayer K E and Forbes A R. 1983. Classification of forest re-
sources with landsat data. Journal of Forestry815) : 283 - 287
Gong PWang JYu LZhao Y CZhao Y YLiang LNiu Z G
Huang X MFu H HLiu SLi C CLi X YFu WLiu C XXu
YWang X YCheng QHu L YYao W BZhang HZhu P
Zhao Z YZhang H YZheng Y MJi L YZhang Y WChen H
Yan AGuo J HYu LWang LLiu X JShi T TZhu M H
Chen Y LYang G WTang PXu BGiri CClinton NZhu Z
326 Journal of Remote Sensing 遥感学报 2014182
LChen J and Chen J. 2013. Finer resolution observation and mo-
nitoring of global land coverfirst mapping results with Landsat TM /
ETM + and ETM + data. International Journal of Remote Sensing
347) : 2607 - 2654 DOI10. 1080 /01431161 . 2012. 748992
Gu Z JShi X ZLi LYu D SLiu L S and Zhang W T. 2011 Using
multiple radiometric correction images to estimate leaf area index.
International Journal of Remote Sensing32 24 ) : 9441 - 9454
DOI10 . 1080 /01431161. 2011. 562251
Habib A FKersting A PShaker A and Yan W Y. 2011 Geometric
calibration and radiometric correction of LiDAR data and their im-
pact on the quality of derived products. Sensors119) : 9069 -
9097 DOI10 . 3390 /s110909069
Hansen M C and Loveland T R. 2012. A review of large area monitoring
of land cover change using Landsat data. Remote Sensing of Envi-
ronment12266 - 74 DOI10. 1016 / j. rse. 2011 . 08. 024
Honkavaara EHakala TMarkelin LRosnell TSaari H and Mkynen
J. 2012. A process for radiometric correction of UAV image blocks.
Photogrammetrie Fernerkundung Geoinformation2) : 115 - 127
Huang CDavis L S and Townshend J R G. 2002 An assessment of sup-
port vector machines for land cover classification. International Jour-
nal of Remote Sensing23 4) : 725 - 749 DOI10. 1080 /
01431160110040323
Jensen J RCowen D JAlthausen J DNarumalani S and Weatherbee
O. 1993 An evaluation of the coastwatch change detection protocol
in South-Carolina. Photogrammetric Engineering and Remote Sens-
ing596) : 1039 - 1046
Kaufman Y JWald A ERemer L A and Gao B C. 1997. The MODIS
2. 1-mu m channel-Correlation with visible reflectance for use in re-
mote sensing of aerosol. IEEE Transactions on Geoscience and Re-
mote Sensing355) : 1286-1298 DOI10. 1109 / 36. 628795
Kobayashi S and Sanga-Ngoie K. 2009 A comparative study of radiomet-
ric correction methods for optical remote sensing imagerythe IRC
vs. other image-based C-correction methods. International Journal of
Remote Sensing30 2) : 285 -314 DOI10. 1080 /
01431160802356781
Liang S LFallah-Adl HKalluri S and JáJá J. 1997. An operational at-
mospheric correction algorithm for LTM imagery over the land. Jour-
nal of the Geophysical ResearchAtmospheres102D14) : 17173
- 17186 DOI10. 1029 / 97JD00336
Lorenzo B and Lorenzo C. 2006 A multilevel context-based system for
classification of very high spatial resolution images. IEEE Transac-
tions on Geoscience and Remote Sensing44 9) : 2586 - 2600
DOI10 . 1109 /TGRS. 2006. 875360 .
Lu D and Weng Q. 2007 A survey of image classification methods and
techniques for improving classification performance. International
Journal of Remote Sensing28 5) : 823 - 870 DOI10. 1080 /
01431160600746456
Meyer PItten K IKellenberger TSandmeier S and Sandmeier R.
1993. Radiometric corrections of topographically induced effects on
landsat TM / ETM + data in an alpine environment. ISPRS Journal of
Photogrammetry and Remote Sensing48 4) : 17 - 28 DOI10.
1016 /0924 - 2716 9390028 - L
Olthof IButson C and Fraser R. 2005. Signature extension through
space for northern landcover classificationa comparison of radio-
metric correction methods. Remote Sensing of Environment95
3) : 290 - 302 DOI10. 1016 /j. rse. 2004. 12. 015
Pons X and Solesugranes L. 1994. A simple radiometric correction model
to improve automatic mapping of vegetation from multispectral satel-
lite data. Remote Sensing of Environment48 2) : 191 - 204
DOI10 . 1016 /0034 - 4257 9490141 - 4
Sexton J OUrban D LDonohue M J and Song C H. 2013. Long-term
land cover dynamics by multi-temporal classification across the
Landsat 5 record. Remote Sensing of Environment128246 - 258
DOI10 . 1016 /j. rse. 2012. 10 . 010
Song CWoodcock C ESeto K CLenney M P and Macomber S A.
2001. Classification and change detection using Landsat TM /ETM +
datawhen and how to correct atmospheric effectsRemote Sensing
of Environment752) : 230 - 244
Tan K CLim H SMatJafri M Z and Abdullah K. 2012 A comparison
of radiometric correction techniques in the evaluation of the relation-
ship between LST and NDVI in Landsat imagery. Environmental
Monitoring and Assessment184 6) : 3813 - 3829 DOI10.
1007 / s10661 - 011 - 2226 - 0
Tso B and Mather P M. 2009. Classification Methods for Remotely
Sensed DataSecond Edition. New YorkTaylor and Francis Inc
Vermote E FTanre DDeuze J LHerman M and Morcette J J. 1997.
Second simulation of the satellite signal in the solar spectrum6S
an overview. IEEE Transactions on Geoscience and Remote Sens-
ing353) : 675 - 686
Vicente-Serrano S MPerez-Cabello F and Lasanta T. 2008. Assessment
of radiometric correction techniques in analyzing vegetation variabili-
ty and change using time series of Landsat images. Remote Sensing
of Environment11210 ) : 3916 - 3934
Xu H Q and Zhang T J. 2013. Assessment of consistency in forest-domi-
nated vegetation observations between ASTER and Landsat ETM plus
images in subtropical coastal areas of southeastern China. Agricul-
tural and Forest Meteorology168 1 - 9
Xu QHou Z Y and Tokola T. 2012 Relative radiometric correction of
multi-temporal ALOS AVNIR-2 data for the estimation of forest at-
tributes. ISPRS Journal of Photogrammetry and Remote Sensing
6869 - 78
Yuan FSawaya K ELoeffelholz B C and Bauer M E 2005 Land cov-
er classification and change analysis of the Twin Cities Minnesota
Metropolitan Area by multitemporal Landsat remote sensing. Remote
Sensing of Environment982 / 3) : 317 - 328 DOI10. 1016 / j.
rse. 2005. 08 . 006.
Zhang P QYu X CLiu ZLi J S and Wang G. 2006. A study on rela-
tive radiometric correction of multitemporal remote sensing images.
Journal of Remote Sensing103) : 339 - 344 DOI10. 3321 / j.
issn1007 - 4619. 2006 . 03. 009
陈趁新 等: Landsat TM 数据不同辐射校正方法对土地覆盖遥感分类的影响 327
Landsat TM 数据不同辐射校正方法对土地
覆盖遥感分类的影响
陈趁新12,胡昌苗1,霍连志1,唐娉1
1. 中国科学院 遥感与数字地球研究所,北京 100101
2. 中国科学院 空间应用工程与技术中心,北京 100094
本文主要是探索 Landsat TM 数据不同辐射校正方法对土地覆盖遥感分类的影响介绍了使用的 3种不同
辐射校正方法( ATCOR3FLAASH 以及查找表) 和两种分类算法在分类实验部分,根据样本的地理坐标在 3
正影像中分别采集训练样本并训练各自的分类器并交叉用于其他辐射校正影像的土地覆盖遥感分类实验结果
表明: ( 1) 用于分类器训练的样本采集自待分类影像时的分类精度明显高于采集自其他影像的分类精度; ( 23
辐射校正影像的分类结果存在差异,其中使用 ATCOR3 FLAASH 方法校正后影像的分类结果有更相近的精度;
3) 辐射校正对分类类别的影响不同,其中对森林类型影响最大,对裸地等其他类别影响相对较小
关键词: 辐射校正,Landsat TMMLCSVMs土地覆盖遥感分类,精度评价
中图分类号: TP79 文献标志码: A
引用格式: 陈趁新,胡昌苗,霍连志,唐娉. 2014. Landsat TM 数据不同辐射校正方法对土地覆盖遥感分类的影响遥感学报,1
82) : 320 - 334
Chen C XHu C MHuo L Z and Tang P 2014. Effect of different radiation correction methods of Landsat TM d
ata on land-cover remote sensing classification. Journal of R emote Sensing1 82) : 320 - 334 DOI10. 11834 /
jrs. 20143211
收稿日期:2013-08-26修订日期:2013-11-27优先数字出版日期: 2013-12 -04
基金项目:国家高技术研究发展计划( 863 计划) ( 编号: 2009AA122002
第一作者简介:陈趁新( 1983) ,男,助理研究员,研究方向为遥感图像处理E-mailchencx@ csu. ac. cn
通信作者简介:唐娉( 1968) ,女,研究员,现从事遥感图像处理E-mailtangping@ radi. ac. cn
1引 言
近几十年来,基于 Landsat TM 数据的遥感影像
分类和变化检测技术被广泛应用于土地覆盖制图
和变化情况监测研究( Fox 等,1983Jensen 等,
1993Yuan 等,2005 Sexton 等,2013 而,
数据在生成过程中受到多种因素的影响,使得这些
用于土地覆盖研究时存在着各种不确定性
主要的影响包括太阳光照条件变化大气条件差异
以及山区的复杂地形条件等,其中大气中的分子和
气溶胶的散射与吸收,使卫星传感器接收到的地物
反射的电磁波信号在从地球表面经过大气层传输
的过程中发生改变Song 等,2001) ,器记
的影像数据与地表实际情况不相符,信息提取精
度造成了影响因此,在应用前Landsat TM 数据
进行辐射校正,包括大气校正和地形校正以提高
影像的质量( Honkavaara 等,2012) 已经成为通识
用于多时Zhang 等,2006 Vicente - Serrano
等,2008Tan 等,2012 Xu 等,2012多传感器( Xu
Zhang2013) 以及大尺度( Olthof 等,2005Hansen
Loveland2012) 遥感数据分析的辐射校正的方
法较多研究表明,过辐射校正可以提取更加精
确的地物Gu 等,20 11 ) 和
地利用和Meyer 等,1993Pons Sole-
sugranes1994Colby Keating 1998Dorren
2003Kobayashi Sanga-Ngoie2009Habib
2011然而对于遥感影像经过不同辐射校正方
正后对土地覆盖遥感分类结果影响的研究
较少
本文以 Landsat TM 数据为例,初步探索不同辐
328 Journal of Remote Sensing 遥感学报 2014182
射校正方法对土地覆盖遥感分类精度的影响
究的过下: 首3种不同辐射校正方法
ATCOR3FLAASH 以及查找表) 对图像进行辐射
校正然后在一个校正结果上选择样本分别用于其
他校正结果进行土地覆盖分类比较分类的精度
在比较时,练样本和检查样本的位置和数量均保
持不变分类算法分别是最大似然 MLC 和支持向
量机 SVMs
2研究区域与数据
2. 1 研究区域
文选择陕西省中北部山区作为研究区域
1该地区地处土高原,候特为寒
旱或半干研究区域内分布有大量山体及一些
较小的丘陵地貌区域内分布较为广泛的土地覆盖
类型包括森林裸土和农田,在平坦的低洼地区,
田与分散的农村居民点呈现混合分布状态,陡峭的
山区地形主要以森林类型分布为主
2. 2
本文使用的数据包括经过几何校正处理的原
Landsat TM 数据,以及经过 3种辐射校正方法校
正处理的数据,空间分辨率均为 30 m数据轨道条
带号:127行编号: 35获取时间为 2000 - 05 - 12
整个区域内无云,可视性较好
1研究区域及实验数据( GB分别使用近红外红和绿波段显示)
2. 3 分类体系及样本选择
本文的分类体系和类别定义参考了 Gong 等人
2013 使用 30 m 分辨率的 Landsat TM 数据对全球
的土地覆盖进行制图研究中的内容,结合试验区
土地覆盖的特点,将该区域土地覆盖类别分为 6类,
包括: 裸土( bare lands农田( croplands森林( for-
est不透水层impervious areas 水 体 ( water
bodies) 和山体
1给出了6类土地覆盖类别的特征描述,
及在 Landsat TM 影像和 Google Earth 中的视觉表
了获取高质量且有代表性的样本数据
取遵循在整个研究区域内均衡分布的原则
首先,基于 Landsat TM 整景影像目视判读地物类别
分布的空间位置再同步定位到具有更高分辨率的
Google Earth 影像中进行验证,据此获取了
大量的高质量样本集; 其次对样本进行了优化,
除重复和非典型的样本点; 最后,按照一定比例将
样本集随机的拆分为训练样本和测试样6个地
类的训练样本共2445 个,检验样本共 477 关于
样本情况详见表1
陈趁新 等: Landsat TM 数据不同辐射校正方法对土地覆盖遥感分类的影响 329
3辐射校正
本文对 Landsat TM 3种不同的
辐射校正方法处理: FLAASH( [2013 - 08 - 01ht-
tp/ / www. eselisvis. com / docs / FLAASH. html AT-
COR3( [2013 - 08 - 01http/ / www. rese . ch / pdf / at-
cor3_manual. pdf) 与传统的查找表 LUTLookUp Ta-
ble) 方法( Liang 等,1997 3种辐射校正是目
Landsat TM / ETM + 数据辐射处理被广泛采用的
其中 FLAASH ATCOR3 是目前有代表性的
大气校正软件模支持目前绝大多数多光
光谱遥的大理,
辐射传输理论存在一些差异; 查找表方法一直被研
究者大量采用,大气校正处理上具有很大的灵活性
FLAASH 模型Spectral Science
空气动力研究实验室AFRL) 与波谱信息技术应用
中心( SITAC) 联合开发的大气校正软件包,工作的
波段范围是 4002500 nmMODTRAN 4 生成
一系列的大气参数查找表,核心的辐射传输模型考
虑了邻近效应,但目前没有考虑地形校正
ATCOR 是由德国宇航中心发展起来的辐射校
正软件,ATCOR2ATCOR3 ATCOR4 3 个版
本( Richter2005 ) ,ATCOR2ATCOR3 已经
成于全球用户群最大的遥感软件 ERDAS 中,
采用 ATCOR3 处理 Landsat TM ATCOR3
要是对山区的卫星遥感影像进行耦合大气和地形
的辐射校正ATCOR3 预先将 MODTRAN 4 计算的
各种大气状况透过程辐射直射和散
射太阳辐射通量存储为查找表ATCOR3 在进行地
形纠正时,利用自带的模块计算坡度坡向天空
可视因子山区阴影等数据对于入射角较大的区
域,ATCOR3 提供了一种可靠的双向反射分布函数
BRDF 经验纠正函数进行地物的 BRDF A
TCOR3采用不同的算法可以同时进行可见光和红外
波段的辐射校正与此同时,它还提供了薄云
去除模块
基于查找表 LUT 的大气校正假设地表为朗伯
体,采用的辐射传输公式如下:
L = L p+ T Eρ
π1ρS1
式中,L为表观辐射亮度,Lp为大气程辐射,
ρ为地
表反射率,
E为地表辐照度,S为大气反照率,T为地
表到大气顶层的透过率使用 6SSecond Simulation
of Satellite Signal in the Solar Spectrum) ( Vermote 等,
1997 可直接计算出参数 ETLpS是,已知 L
情况下便可计算出 ρ
Landsat TM 据的查找表算法参考 Liang
1997) 的做法自主开发的即采用暗植被法( DDV
Kaufman 等,1997) 估算暗植被处的气溶胶厚度,
330 Journal of Remote Sensing 感学报 2014182
用移动窗口内插的方式估算整景数据气溶胶的
分布
查找表大气校正后的地形校正采用的是一种
基于地形抹平的半经验校正方法,该方法首先通过
对地形的坡度角进行抹平处理,然后基于抹平后的
地形采用 COS( 余弦) 校正完成地形辐射校正,简称
SCOS Smoothed COS Correction 地形抹
平处理用来消除地表非朗伯效应造成的校正辐射
常,也有压制因地形数据异常造成的影响的
作用
大气校正采用的高程数据( DEM 90 m 空间
分辨率Shuttle Radar Topography MissionSRTM
数据( 版本 SRTM4. 1
4分类实验与结果分析
了评价不同辐射校正方法对土地覆盖遥感
分类结果的影响,共设计了 9组实验,分别从3
射校正影像中提取训练样本用于 3景影像的分类
每组实验分使MLC SVMs 作为分类器进行
分类实验
4. 1 分类器
遥感影像分类算法较多监督
督分类,参数分类和非参数分类软分类和硬分类
的分类方法,于像亚像元和对象的分类
方法等( Lu Weng2007 本文选用两种目前广
泛应用的分类算法,MLC SVMs作为辐射校正
方法对分类精度评价分类器
MLC 是一种基于统计理论的参数分类器,是广
泛使用的分类器之Huang 等,2002遥感
类方法中,MLC 具有明确的参数解释能力易于集
成先验知识以及易于实现等算法优势根据统计
参数( 均值向量和方差协方差矩阵) ,计算出像元
属于每个预定义类别的概率,并将该像元划分到概
最高的类别完成分类( Tso Mather2009
SVMs 一种基于结构风险最小化原则的分类算
以其独有的特性和高效性使得该分类器在
感分类领域具有较高的知名度( Lorenzo L
orenzo2006为不失一般性本文选用径向基函
Radial Basis FunctionRBF SVMs 的核函
数,在模型训练过程中采用交叉验证的方法选
RBF 的模型参数在处理过程中将数据归一化
01间,使LIBSVM Chang
Lin2011
4. 2 实验流程
2给出了实验流程首先使用3种不同辐射
校正方法对原始 TM 影像进行辐射校正,得到辐射
校正影像 RC_ARC_B RC_C并作为下一步处
理的数据源; 其次,根据同一地理坐标内的样本区
域数据, 3景影像中独立采集样本
Samp_aSamp_bSamp _c) ,分别练得
器( MLC SVMs ; 最后,使用这些分类器分别用于
每景影像覆盖对分类结果的精
度评价通过一系列测试样本的混淆矩阵得出,并以
生产者精度使用者精度总体精度以及 Kappa 系数
等作为分类结果精度评价的统计量
2实验流程
4. 3 类结果
3给出了使MLC SVMs 两种分类器在
各组实验中的部分分类结果其中图3a) 为当
本采集RC_A 影像时训练得到的 MLC RC _ A
影像进行分类的结果; 图 3b) 为当样本采集自 RC_
C影像时训练得到的 SVMs RC_C 影像进行分类
的结果
各组实验的总体精度 OA Kappa 系数统计如
2所示通过表 2可以看出,样本采集策略对分
类精度的影响即当样本采集影像和分类影像为同
一景影像时分类精度最高此外,3景辐射校正影
像之间的分类结果存在一定差异RC_ AAT-
COR3 R C_BFLAASH) 方法校正图像的分类结
果有更相近的精度而辐射校正影像 RC _C 的分类
精度与前两种方法存在差别
陈趁新 等: Landsat TM 数据不同辐射校正方法对土地覆盖遥感分类的影响 331
3部分分类结果图
2辐射校正后不同样本采集策略与分类精度结果统计( MLCSVMs
样本 RC_A RC_B RC_C
OA / % Kappa OA / % Kappa OA/ % Kappa
RC_A 93. 29 93 . 50 0. 9200. 922 91. 61 88. 26 0 . 8990. 859 77 . 5788 . 05 0. 731 0. 857
RC_B 90. 36 87. 63 0 . 8840. 851 93 . 2993 . 50 0 . 9200. 922 74. 21 88. 05 0. 6910 . 857
RC_C 78. 05 84. 49 0 . 7020. 814 79 . 2085 . 95 0 . 7500. 831 91. 20 93. 71 0. 8940 . 925
分类精度较高的实验组合是当样本采集影像
和分类影像相同时的分类结果3给出了当样本 采集自 3景辐射校正影像自身并训练 MLC SVMs
后对各自影像分类时的分类精度
3各组实验中分类精度较高的类别精度统计( MLCSVMs
类名 RC_ARC_A RC_BRC_B RC_CRC_C
生产精度/% 使用精度/% 生产精度/% 使用精度/ % 生产精度/ % 使用精度/ %
裸土 90. 00 90. 00 91. 14 90. 00 90. 0088. 75 87. 80 88. 75 88. 75 91. 25 85. 54 87. 95
农田 95. 00 95. 00 96. 20 96. 20 93. 7595. 00 97. 40 98. 70 85. 00 96. 25 98. 55 97. 47
森林 96. 25 98. 75 100. 0 98. 75 97. 5098. 75 98. 73 97. 53 97. 50 97. 50 98. 73 98. 73
山体 91. 25 92. 50 82. 02 84. 09 90. 0091. 25 83. 72 82. 95 86. 25 90. 00 82. 14 85. 71
不透水层 90. 00 90. 00 92. 3193. 51 91. 25 92. 50 93. 59 94. 87 92. 50 92 . 50 85 . 0693 . 67
水体 97. 40 94. 81 100. 0 100. 0 97. 4094. 81 100. 0 100. 0 97. 40 94. 81 100. 0 100. 0
总体精度/ % 93. 29 93. 50 93. 29 93. 50 91. 20 93. 71
Kappa 0. 920 0. 922 0. 920 0. 922 0. 894 0. 925
从图 4可以看出,各地物类别在经过 3种辐射
校正后的地表反射率( 进行了整型变换) 的变化情
3辐射校正方法得到的结果比较接近尤其
RC_A RC_BRC_C 的结果与前两种校正方
法稍有区别尤其是第 1段,LUT 辐射校正后各
地类的地表反射率值明显低于 FLAASH ATCOR
方法
比较辐射校正后影像的分类结果可以看出,
地覆盖类别的精度均有不同程度的变化对于具
体地物类别,如图 5所示,森林类别在辐射校正后分
类精度变化较为明显,表明辐射校正对该类别的影
响较大这可能与辐射校正均采用暗植被法进行
气溶胶厚度估计的方法相关,但详细原因需要更进
一步分析
332 Journal of Remote Sensing 遥感学报 2014182
4辐射校正后各个地物类别在各波段上 DN 值的变化
5辐射校正后不同地物类别精度变化
( 以 MLC 分类结果为例)
5结 论
对于遥感应用领域来说,辐射校正是开展进一
步工作的前提条件本文在第3节介绍了3种不同
的辐射校正方法并应用于 Landsat TM 数据的辐射
校正文使MLC SVMs 进行了 9组实验的
比较,包括从 3景辐射校正影像中提取训练样本并
分别用于3景影像的分类,以此探索基于不同辐射
方法校正后不同样本采集策略对土地覆盖遥感分
类精度的影响
通过本文实验结果,初步得到以下结论:
1) 样本采集影像与分类影像相同时分类精度
最高如使MLC 时,RC _ ARC _ A
的总体分类度为 93 29% RC _BRC _ B 组合和
RC_CRC _ C 组合总体精度分别为 93. 29% 9
1. 20 % 均明显高于同组其他分类组合的分类精度;
2) 从分类结果来看在不同辐射校正方法的
影像上进行分类结果之间存在差异,中,ATCOR3
FLAASH 法校正影像的分类结果有更相近的
精度值得注意的是RC_C 的分类总体精度与前
两者( RC_A RC _B) 差异较为明显,表明对于本
文实验数据,
ATCOR3 FLAASH 的校正结果较为
一致而自主开发的查找表算法的结果与之有一定
的偏差,这种偏差很可能是由于气溶胶分布估算误
差的影响造成的( 图4) ;
陈趁新 等: Landsat TM 数据不同辐射校正方法对土地覆盖遥感分类的影响 333
3) 辐射校正对每个类别的分类精度影响不
同,受影响最的是类别其他影响
的程度各值得注意的是,SVMs 分类时的
总体精度和稳定性均高于 MLC二者分类精度的变
化趋势类似
本文研究区域较小,相对全球范围内各种复
杂地地貌和土地覆盖类别来说,本文研究存在
一定的局限性下一步将以本文研究结果为基础,
逐步扩大研究范围另外本文的结果没有体现出
地形正对于分类精度的影响,后续的研究将从
选择和实验流程设计等多方面进一步展开
研究
志 谢 本文的样本数据选取得到了中国科
学院遥感与数字地球研究所赵理君丁玲
张爱英等同学的大力协助在此表示衷心的
感谢!
参考文献( References
Chang C C and Lin C J. 2011. LIBSVMa library for support vector ma-
chines. ACM Transactions on Intelligent Systems and Technology2
3) : Article No. 27 DOI10 . 1145 /1961189. 1961199
Colby J D and Keating P L. 1998 Land cover classification using Land-
sat TM / ETM + imagery in the tropical highlandsthe influence of
anisotropic reflectance. International Journal of Remote Sensing19
8) : 1479 - 1500 DOI10 . 1080 /014311698215306
Dorren L K AMaier B and Seijmonsbergen A C. 2003. Improved Land-
sat-based forest mapping in steep mountainous terrain using object-
based classification. Forest Ecology and Management1831 /3 ) :
31 - 46 DOI10. 1016 /S0378 - 1127 0300113 - 0
Fox LMayer K E and Forbes A R. 1983 Classification of forest re-
sources with landsat data. Journal of Forestry815) : 283 - 287
Gong PWang JYu LZhao Y CZhao Y YLiang LNiu Z G
Huang X MFu H HLiu SLi C CLi X YFu WLiu C XXu
YWang X YCheng QHu L YYao W BZhang HZhu P
Zhao Z YZhang H YZheng Y MJi L YZhang Y WChen H
Yan AGuo J HYu LWang LLiu X JShi T TZhu M H
Chen Y LYang G WTang PXu BGiri CClinton NZhu Z
LChen J and Chen J. 2013. Finer resolution observation and mo-
nitoring of global land coverfirst mapping results with Landsat TM /
ETM + and ETM + data. International Journal of Remote Sensing
347) : 2607 - 2654 DOI10. 1080 /01431161 . 2012. 748992
Gu Z JShi X ZLi LYu D SLiu L S and Zhang W T. 2011 Using
multiple radiometric correction images to estimate leaf area index.
International Journal of Remote Sensing32 24 ) : 9441 - 9454
DOI10 . 1080 /01431161. 2011. 562251
Habib A FKersting A PShaker A and Yan W Y. 2011 Geometric
calibration and radiometric correction of LiDAR data and their im-
pact on the quality of derived products. Sensors11 9) : 9069 -
9097 DOI10 . 3390 /s110909069
Hansen M C and Loveland T R. 2012. A review of large area monitoring
of land cover change using Landsat data. Remote Sensing of Envi-
ronment12266 - 74 DOI10. 1016 / j. rse. 2011. 08 . 024
Honkavaara EHakala TMarkelin LRosnell TSaari H and Mkynen
J. 2012 A process for radiometric correction of UAV image blocks.
Photogrammetrie Fernerkundung Geoinformation2) : 115 - 127
Huang CDavis L S and Townshend J R G. 2002 An assessment of sup-
port vector machines for land cover classification. International Jour-
nal of Remote Sensing23 4) : 725 - 749 DOI10. 1080 /
01431160110040323
Jensen J RCowen D JAlthausen J DNarumalani S and Weatherbee
O. 1993 An evaluation of the coastwatch change detection protocol
in South-Carolina. Photogrammetric Engineering and Remote Sens-
ing596) : 1039 - 1046
Kaufman Y JWald A ERemer L A and Gao B C. 1997. The MODIS
2. 1-mu m channel-Correlation with visible reflectance for use in re-
mote sensing of aerosol. IEEE Transactions on Geoscience and Re-
mote Sensing355) : 1286-1298 DOI10. 1109 / 36. 628795
Kobayashi S and Sanga-Ngoie K. 2009. A comparative study of radiomet-
ric correction methods for optical remote sensing imagerythe IRC
vs. other image-based C-correction methods. International Journal of
Remote Sensing30 2) : 285 -314 DOI10. 1080 /
01431160802356781
Liang S LFallah-Adl HKalluri S and JáJá J. 1997. An operational at-
mospheric correction algorithm for LTM imagery over the land. Jour-
nal of the Geophysical ResearchAtmospheres102D14) : 17173
- 17186 DOI10. 1029 / 97JD00336
Lorenzo B and Lorenzo C. 2006. A multilevel context-based system for
classification of very high spatial resolution images. IEEE Transac-
tions on Geoscience and Remote Sensing44 9) : 2586 - 2600
DOI10 . 1109 /TGRS. 2006. 875360 .
Lu D and Weng Q. 2007. A survey of image classification methods and
techniques for improving classification performance. International
Journal of Remote Sensing28 5) : 823 - 870 DOI10. 1080 /
01431160600746456
Meyer PItten K IKellenberger TSandmeier S and Sandmeier R.
1993. Radiometric corrections of topographically induced effects on
landsat TM / ETM + data in an alpine environment. ISPRS Journal of
Photogrammetry and Remote Sensing48 4) : 17 - 28 DOI10.
1016 /0924 - 2716 9390028 - L
Olthof IButson C and Fraser R. 2005 Signature extension through
space for northern landcover classificationa comparison of radio-
metric correction methods. Remote Sensing of Environment95
3) : 290 - 302 DOI10. 1016 /j. rse. 2004. 12. 015
Pons X and Solesugranes L. 1994 A simple radiometric correction model
to improve automatic mapping of vegetation from multispectral satel-
lite data. Remote Sensing of Environment48 2) : 191 - 204
DOI10 . 1016 /0034 - 4257 9490141 - 4
Sexton J OUrban D LDonohue M J and Song C H. 2013. Long-term
land cover dynamics by multi-temporal classification across the
Landsat 5 record. Remote Sensing of Environment128246 - 258
334 Journal of Remote Sensing 遥感学报 2014182
DOI10 . 1016 /j. rse. 2012. 10 . 010
Song CWoodcock C ESeto K CLenney M P and Macomber S A.
2001. Classification and change detection using Landsat TM / ETM
+ datawhen and how to correct atmospheric effectsRemote Sens-
ing of Environment752) : 230 - 244
Tan K CLim H SMatJafri M Z and Abdullah K. 2012 A comparison
of radiometric correction techniques in the evaluation of the relation-
ship between LST and NDVI in Landsat imagery. Environmental
Monitoring and Assessment184 6) : 3813 - 3829 DOI10.
1007 / s10661 - 011 - 2226 - 0
Tso B and Mather P M. 2009 Classification Methods for Remotely
Sensed DataSecond Edition. New YorkTaylor and Francis Inc
Vermote E FTanre DDeuze J LHerman M and Morcette J J. 1997
Second simulation of the satellite signal in the solar spectrum6S
an overview. IEEE Transactions on Geoscience and Remote Sens-
ing353) : 675 - 686
Vicente-Serrano S MPerez-Cabello F and Lasanta T. 2008 Assessment
of radiometric correction techniques in analyzing vegetation variabili-
ty and change using time series of Landsat images. Remote Sensing
of Environment11210 ) : 3916 - 3934
Xu H Q and Zhang T J. 2013 Assessment of consistency in forest-domi-
nated vegetation observations between ASTER and Landsat ETM
plus images in subtropical coastal areas of southeastern China. Agri-
cultural and Forest Meteorology168 1 - 9
Xu QHou Z Y and Tokola T. 2012 Relative radiometric correction of
multi-temporal ALOS AVNIR-2 data for the estimation of forest at-
tributes. ISPRS Journal of Photogrammetry and Remote Sensing
6869 - 78
Yuan FSawaya K ELoeffelholz B C and Bauer M E 2005 Land cov-
er classification and change analysis of the Twin Cities Minnesota
Metropolitan Area by multitemporal Landsat remote sensing. Remote
Sensing of Environment982 / 3) : 317 - 328 DOI10. 1016 / j.
rse. 2005. 08 . 006.
Zhang P QYu X C Liu ZLi J S and Wang G. 2006. A study on rela-
tive radiometric correction of multitemporal remote sensing images.
Journal of Remote Sensing103) : 339 - 344 DOI10. 3321 / j.
issn1007 - 4619. 2006 . 03. 009
... Through qualitative image interpretation, CZI will aid coastal zone management by providing key information on reclamation [1], wetlands [2], land cover/use [3], mangroves [4], coral reef [5], and sea ice [6] in the coastal zone and islands, where significant changes are usually expected due to intense human activity. At the same time, the quantitative information inversion from CZI images may also be expected to retrieve key parameters of aquatic environments (including turbidity, suspended sediment concentration, transparency, and water depth) [7][8][9], natural resources (such as wetland biomass) [10], and abnormal ecological phenomena [11,12]. ...
Article
Full-text available
The out-of-band (OOB) response is one of the key specifications for satellite optical sensors, which has important influences on quantitative remote sensing retrieval. In this paper, the effect of OOB response on the radiometric measurements made just above the sea surface is evaluated for the three broad visible bands (i.e., blue, green, and red) of the Coastal Zone Imager (CZI) onboard China’s ocean satellite HY-1C to be launched in September 2018. For the turbid coastal (Case 2) waters whose optical properties are mainly dominated by suspended sediment and colored dissolved organic material, the OOB effect can be neglected (<2%) for all three CZI visible bands. For the phytoplankton-dominated (Case 1) waters which are mainly distributed in the clear open ocean, a significant (>2%) OOB effect was found in the green band over oligotrophic waters (chlorophyll a concentration ≤~0.1 mg/m3), and accordingly a model based on the CZI blue-green band ratio is proposed to correct this effect. The OOB influence on the CZI ocean color retrieval is discussed. This research highlights the importance of the comprehensive pre-launch radiometric characterization and the OOB effect correction for the broad band space-borne sensor, in order to achieve a high-quality quantitative ocean product.
Article
Full-text available
Ecological applications of remote-sensing techniques are generally limited to images after atmospheric correction, though other radiometric correction data are potentially valuable. In this article, six spectral vegetation indices (VIs) were derived from a SPOT 5 image at four radiometric correction levels: digital number (DN), at-sensor radiance (SR), top of atmosphere reflectance (TOA) and post-atmospheric correction reflectance (PAC). These VIs include the normalized difference vegetation index (NDVI), ratio vegetation index (RVI), slope ratio of radiation curve (K), general radiance level (L), visible-infrared radiation balance (B) and band radiance variation (V). They were then related to the leaf area index (LAI), acquired from in situ measurement in Hetian town, Fujian Province, China. The VI–LAI correlation coefficients varied greatly across vegetation types, VIs as well as image radiometric correction levels, and were not surely increased by image radiometric corrections. Among all 330 VI–LAI models established, the R 2 of multi-variable models were generally higher than those of the single-variable ones. The independent variables of the best VI–LAI models contained all VIs from all radiometric correction levels, showing the potentials of multi-radiometric correction images in LAI estimating. The results indicated that the use of VIs from multiple radiometric correction images can better exploit the capabilities of remote-sensing information, thus improving the accuracy of LAI estimating.
Article
The NOAA sponsored CoastWatch Change Analysis Project (C-CAP) will utilize remote sensing technology to monitor changes in coastal wetland habitats and adjacent uplands on a cycle of 1 to 5 years. Two study areas in South Carolina were selected to test various C-CAP change detection protocols using near-anniversary Landsat Thematic Mapper data obtained in 1982 and 1988. Fort Moultrie (dominated by salt and brackish marsh) and Kittredge (40 river miles inland and dominated by bottomland hardwoods and riverine aquatic beds) study areas were used to evaluate a modified C-CAP classification scheme, image classification procedures, change detection algorithm alternatives, and the impact of tidal stage on coastal change detection.
Article
The objective of this investigation is to develop and test a radiometric correction process for UAV image blocks. The phases of the process include the laboratory calibration of the sensor and the radiometric correction of the campaign image data. This investigation focuses on developing a process for radiometric correction of the image data collected during a remote sensing campaign. First of all, the orientations for the images are determined using the self-calibrating bundle block adjustment method and an accurate DSM is generated by automatic image matching. The varying radiometric level of images due to changes in illumination and the instability of the sensor are eliminated using a relative radiometric block adjustment technique. Optional reflectance reference observations can be used to adjust the data to absolute reflectance units. The process was demonstrated and evaluated by using two UAV imaging systems: a consumer camera-based system and a novel Fabry-Perot interferometer-based next generation lightweight hyperspectral imaging system. The method improved the homogeneity of the data, but some drift also appeared in the parameters. The first experiment provided 0.003-0.008 reflectance errors in the areas close to the radiometric control points (mostly on the level of 5% of the reflectance value). The presented approach provides a general framework for rigorous radiometric correction of UAV image blocks, and the novel technology provides many possibilities for the further development of the method. Our results also show that hyperspectral stereophotogrammetry is now possible with UAV imaging sensors weighing less than 500 g.
Article
Relative radiometric correction methods have been widely used to correct ground illumination difference in multi-temporal satellite data. ALOS (Advanced Land Observing Satellite) data starts to play an important role in forest and carbon assessment, such as the REDD (Reducing Emissions from Deforestation and forest Degradation) program. The objective of the study was to compare three relative radiometric correction methods for five multi-temporal ALOS AVNIR-2 (Advanced Visible and Near Infrared Radiometer) images, and to examine the influence of each correction method on the estimation accuracy of forest attributes with auxiliary field inventory plot data. Both spectral features and textural features were extracted before and after radiometric correction and used in estimation procedure. All the radiometric correction methods used improved the estimation accuracy of forest stem volume at plot level, and they were MAD (multivariate alteration detection) transformation-based normalization, PCA (principle component analysis)-based correction and local radiometric correction, among which MAD transformation-based normalization exceeded others by reducing the relative RMSE by 5.75% with the ordinary least square fitting and 6.8% with the K-MSN (K-Most Similar Neighbour) method both after leave-one-out cross-validation. RMSE for only the corrected area is also calculated, in view of the small proportion of plots in that area. The result can be used to improve the visual effect of mosaics of multi-temporal ALOS scenes, and to retrieve more accurate forest estimates for national forest resources and biomass mapping.
Article
An operational atmospheric correction algorithm for Thematic Mapper (TM) imagery has been developed for both sequential and parallel computer environments considering both aerosol and molecular scattering and absorption. The aerosol optical depth is estimated from the image itself using the dark object approach on a moving-window basis, and the surface reflectance is then retrieved by searching lookup tables that are created using a numerical radiative transfer code. The dark object pixels are identified and their surface reflectance estimated using TM channel 7 (2.1 mu m). A variety of techniques are employed to improve computational efficiency. This method is validated by measured aerosol optical depth and extensive visual evaluations accompanied by statistical analysis. Results indicate that the approach is highly stable and useful for both qualitative imagery interpretation (haze removal) and quantitative analysis. Future research activities are also highlighted. The computer codes are available to the general scientific community.
Article
Remotely sensed data have inherent radiometric errors caused by atmospheric and topographic effects. In this paper, the performance of image‐based radiometric correction methods for optical satellite imagery is assessed by comparing variants of the traditional semiempirical C‐correction method with the physically‐based Integrated Radiometric Correction (IRC) method. High‐resolution digital elevation model (DEM) data were used for calculating the topographic parameters needed in both methods. The corrected reflectance obtained by using statistical parameters calculated for the whole of the study area (the general‐c approach) was investigated. In general, high correction accuracy (i.e. low correlation between surface reflectance and topography) was obtained. The IRC method yielded the highest accuracy for any band, indicating that application of this method could improve land‐use/land‐cover (LU/LC) classification. However, the general‐c approach showed a remarkable decrease in accuracy for all methods when applied to specific land‐cover types, particularly in visible bands. Subsequently, higher accuracy was obtained with correction parameters computed by using pixels within each land‐cover type (the specific‐c approach). The difference between these approaches demonstrates the dominant impact of surface features on the correction schemes. In the specific‐c approach, C‐correction gives a higher accuracy in the visible bands (C‐correction parameters tend to be overestimated because of strong haze) whereas IRC provides the best results in infrared bands (less haze).
Article
Vegetation is one of the most important components of the global ecosystem. It plays a vital role for global energy balances and climate modulations. Timely and precisely monitoring vegetation changes has become an increasing need. Among various satellite-based earth observation systems, the Landsat and ASTER sensors are most commonly used in mesoscale vegetation observations. Nevertheless, the quantitative relationship between the two sensors in measuring vegetation has not been investigated in detail. Therefore, this paper aims to investigate how well ASTER vegetation measurements replicate ETM+ vegetation measurements, and more important, how much difference there is in the measurements between the two sensors. The study utilised three date-coincident image pairs of the two sensors, covering the subtropical coastal areas of southeastern China. The approach was achieved by evaluating the consistence of the normalised difference vegetation index (NDVI) data and the NDVI-estimated vegetation area between the two sensors. A further inter-type comparison among forest, paddy and grass has also been carried out to examine whether there is any variation in the consistence between different vegetation types. Results suggest that the ASTER sensor produces similar vegetation measurements to ETM+ in the study areas. Nevertheless, differences have also been observed. The study found an overall lower spectral NDVI measurement of ASTER than ETM+ based on their differences in mean NDVI value and estimated vegetation area. This suggests that the ASTER sensor is less sensitive to vegetation characteristics. Compared with ETM+, the ASTER sensor produces lower NDVI values by up to 8.1% and estimates less vegetation area by up to 4.1%. Inter-type analysis indicates that among the three test vegetation types, forest has the highest degree of agreement in NDVI measurements between the two sensors, whereas grass has the greatest difference in mean NDVI value between both sensors. All these differences are worth considering when the vegetation measurement results of ASTER have to be used together with those of ETM+ for a synergistic scientific application. This often occurs when using the ASTER sensor data for Landsat data continuity in scientific applications in order to fill the gaps in observation. As such, a data conversion between the two sensors is recommended. This paper demonstrated that the conversion using the model equations obtained in this study could reduce the root mean square errors and improve data agreement between the two sensor NDVIs. The study shows that the revealed differences in vegetation measurements between both sensors are due largely to the difference in the relative spectral response functions of the NDVI-related near-infrared bands between the two sensors.
Article
The homogeneity of time series of satellite images is crucial when studying abrupt or gradual changes in vegetation cover via remote sensing data. Various sources of noise affect the information received by satellites, making it difficult to differentiate the surface signal from noise and complicates attempts to obtain homogeneous time series. We compare different procedures developed to create homogeneous time series of Landsat images, including sensor calibration, atmospheric and topographic correction, and radiometric normalization. Two seasonal time series of Landsat images were created for the middle Ebro Valley (NE Spain) covering the period 1984–2007. Different processing steps were tested and the best option selected according to quantitative statistics obtained from invariant areas, simultaneous medium-resolution images, and field measurements. The optimum procedure includes cross-calibration between Landsat sensors, atmospheric correction using complex radiative transfer models, a non-lambertian topographic correction, and a relative radiometric normalization using an automatic procedure. Finally, three case studies are presented to illustrate the role of the different radiometric correction procedures when analyzing and explaining gradual and abrupt temporal changes in vegetation cover, as well as temporal variability. We have shown that to analyze different vegetation processes with Landsat data, it is necessary to accurately ensure the homogeneity of the multitemporal datasets by means of complex radiometric correction procedures. Failure to follow such a procedure may mean that the analyzed processes are non-recognizable and that the obtained results are invalid.
Article
The accuracy of forest stand type maps derived from a Landsat Thematic Mapper (Landsat TM) image of a heterogeneous forest covering rugged terrain is generally low. Therefore, the first objective of this study was to assess whether topographic correction of TM bands and adding the digital elevation model (DEM) as additional band improves the accuracy of Landsat TM-based forest stand type mapping in steep mountainous terrain. The second objective of this study was to compare object-based classification with per-pixel classification on the basis of the accuracy and the applicability of the derived forest stand type maps. To fulfil these objectives different classification schemes were applied to both topographically corrected and uncorrected Landsat TM images, both with and without the DEM as additional band. All the classification results were compared on the basis of confusion matrices and kappa statistics. It is found that both topographic correction and classification with the DEM as additional band increase the accuracy of Landsat TM-based forest stand type maps in steep mountainous terrain. Further it was found that the accuracies of per-pixel classifications were slightly higher, but object-based classification seemed to provide better overall results according to local foresters. It is concluded that Landsat TM images could provide basic information at regional scale for compiling forest stand type maps especially if they are classified with an object-based technique.