11
Adv. Space Ree. Vol.5, No.5, pp.l—11, 1985 0273—1177185 $0.00 + .50 Printed in Great Britain. All rights reserved. Copyright © COSPAR LANDSAT IMAGE DATA QUALITY STUDIES C. F. Schueler* and V. V. Salomonson** 5Electro-Optical Instrumentation Product Line, Santa Barbara Research Center, 75 Coromar Drive, Goleta, CA 93117, U.S.A. **Laboratory for Earth Sciences, NASA/Goddard Space Flight Center, Greenbelt, MD 20771, U.S.A. ABSTRACT The July 1982 launch of Landsat—4 was immediately followed by a two—year comprehensive set of detailed investigations sponsored by the National Aeronautics and Space Administration (NASA) at Goddard Space Flight Center (GSFC). The Landsat Image Data Quality Analysis (LIDQA) research plans for these investigations were specified prior to launch, so that min- imum time would be lost in assessing the performance of the long—awaited Thematic Mapper (TM) sensor that Landsat—4 carried in addition to a fourth Multispectral Scanner (MSS). The LIDQA investigations have been substantially completed, and have shown that the TM is a very good spaceborne nsiltispectral radiometer, and has met or exceeded most of its design goals. TM’s new short—wave infrared (SWIR) spectral capability yielded improved mineral and plant discrimination compared to the MSS, as anticipated by ground—based and airborne TM simulations. Moreover, the improved spatial resolution and geometric accuracy of Landsat—4 and the TM have resulted in satellite image maps exceeding 1:100,000 U.S. map accuracy standards. Finally, based on an information entropy measure, principal component analysis, and classification results, TM data has been shown to approach its theoretical limit in in- formation content per pixel, exceeding the MSS by at least a factor of two. INTRODUCTION This paper discusses the Landsat Image Data Quality Analysis (LIDQA) program, sponsored by the National Aeronautics and Space Administration (NASA) in the United States. This program sought to characterize the Thematic Mapper sensor on board Landsats 4 and 5 relative to de- sign specifications and to evaluate image quality relative to user needs /1/. Sensor evaluation includes radiometric accuracy, spatial blur and geometric fidelity, and spectral performance evaluation. Image evaluation relative to user needs includes cartogra- phy and land use mapping, agricultural and botanical applications, hydrology, and geological applications, including petroleum and mineral exploration as well as land form characteriza- tion. Each of these areas of the LIDQA program was supplemented by investigations conducted as part of the normal NASA research effort. The TM specifications resulted from user needs analysis between 1972 and 1975 /2,3/, fol- lowed by a design analysis program. Several proposals were submitted to build a TM sensor that would meet NASA requirements. A design was selected that was developed by the Hughes Aircraft Company (HAC) and the Santa Barbara Research Center (SBRC) /4/. Three 114 sensors were developed by SBRC: an engineering model for design verification, a protoflight model for Landaat 4 and a flight model for Landsat 5 /5/. Each of the TM sensors was tested prior to launch using a ground testing system comprised of collimating optics and a radiometric and spatial calibration system. Special test patterns established the sensor’s spatial resolution and geometric accuracy characteristics. Radio— metric calibration was provided by integrating spheres calibrated to National Bureau of Standards (NBS) accuracy. Finally, the overall TM spectral response was estimated by com- bining independent spectral measurements of the optical system and the focal plane. The Landsat 4/5 ground—based data processing systems have provided the products that enabled scientists to evaluate sensor performance and image quality. Between July 1982 and July 1983 a set of systems collectively called the “Scrounge” system was used to provide data products for analyses. Following July 1983 to the present the Thematic Mapper image pro- cessing system (TIPS) has provided data products. Aside from minor formatting differences, the products from either of the two systems are the same.

Landsat image data quality studies

Embed Size (px)

Citation preview

Page 1: Landsat image data quality studies

Adv. Space Ree. Vol.5, No.5, pp.l—11, 1985 0273—1177185 $0.00 + .50Printed in Great Britain. All rights reserved. Copyright © COSPAR

LANDSAT IMAGE DATA QUALITYSTUDIES

C. F. Schueler* and V. V. Salomonson**5Electro-OpticalInstrumentationProduct Line, SantaBarbara

ResearchCenter, 75 Coromar Drive, Goleta, CA 93117, U.S.A.**Laboratory for Earth Sciences,NASA/GoddardSpaceFlightCenter, Greenbelt, MD 20771, U.S.A.

ABSTRACT

The July 1982 launch of Landsat—4 was immediately followed by a two—year comprehensive setof detailed investigations sponsored by the National Aeronautics and Space Administration(NASA) at Goddard Space Flight Center (GSFC). The Landsat Image Data Quality Analysis(LIDQA) research plans for these investigations were specified prior to launch, so that min-imum time would be lost in assessing the performance of the long—awaited Thematic Mapper(TM) sensor that Landsat—4 carried in addition to a fourth Multispectral Scanner (MSS). TheLIDQA investigations have been substantially completed, and have shown that the TM is a verygood spaceborne nsiltispectral radiometer, and has met or exceeded most of its designgoals. TM’s new short—wave infrared (SWIR) spectral capability yielded improved mineral andplant discrimination compared to the MSS, as anticipated by ground—based and airborne TMsimulations. Moreover, the improved spatial resolution and geometric accuracy of Landsat—4and the TM have resulted in satellite image maps exceeding 1:100,000 U.S. map accuracystandards. Finally, based on an information entropy measure, principal component analysis,and classification results, TM data has been shown to approach its theoretical limit in in-formation content per pixel, exceeding the MSS by at least a factor of two.

INTRODUCTION

This paper discusses the Landsat Image Data Quality Analysis (LIDQA) program, sponsored bythe National Aeronautics and Space Administration (NASA) in the United States. This programsought to characterize the Thematic Mapper sensor on board Landsats 4 and 5 relative to de-sign specifications and to evaluate image quality relative to user needs /1/.

Sensor evaluation includes radiometric accuracy, spatial blur and geometric fidelity, and

spectral performance evaluation. Image evaluation relative to user needs includes cartogra-phy and land use mapping, agricultural and botanical applications, hydrology, and geologicalapplications, including petroleum and mineral exploration as well as land form characteriza-tion. Each of these areas of the LIDQA program was supplemented by investigations conductedas part of the normal NASA research effort.

The TM specifications resulted from user needs analysis between 1972 and 1975 /2,3/, fol-lowed by a design analysis program. Several proposals were submitted to build a TM sensorthat would meet NASA requirements. A design was selected that was developed by the HughesAircraft Company (HAC) and the Santa Barbara Research Center (SBRC) /4/. Three 114 sensorswere developed by SBRC: an engineering model for design verification, a protoflight modelfor Landaat 4 and a flight model for Landsat 5 /5/.

Each of the TM sensors was tested prior to launch using a ground testing system comprised ofcollimating optics and a radiometric and spatial calibration system. Special test patternsestablished the sensor’s spatial resolution and geometric accuracy characteristics. Radio—metric calibration was provided by integrating spheres calibrated to National Bureau ofStandards (NBS) accuracy. Finally, the overall TM spectral response was estimated by com-bining independent spectral measurements of the optical system and the focal plane.

The Landsat 4/5 ground—based data processing systems have provided the products that enabledscientists to evaluate sensor performance and image quality. Between July 1982 and July1983 a set of systems collectively called the “Scrounge” system was used to provide dataproducts for analyses. Following July 1983 to the present the Thematic Mapper image pro-cessing system (TIPS) has provided data products. Aside from minor formatting differences,the products from either of the two systems are the same.

Page 2: Landsat image data quality studies

2 C.F. Schueler and V.V. Salomonson

Three digital image products are provided by the ground processing system. The first is theraw data, which is not generally available. The other two products are easily available,and include archival (A) tapes, and product (P) tapes. The A tape consists of data whichhas been corrected radiometrically but not geometrically. The P tape provides both radio—metrically and geometrically corrected data. Color composite or black and white film trans-parencies in 241 x 241 mmformat or paper prints are also available.

SENSOREVALUATION

Three fundamental sets of sensor parameters characterize TM performance. These include ra—diometric, spatial, and spectral performance characteristics. Instrument sensitivity is oneof three fundamental radiometric characteristics, and is often characterized by the user interms of the noise equivalent ground reflectance, or NE~p. A 8econd radiometric parameteris dynamic range, which establishes the sensor’s saturation radiance, above which the sensorwill no longer respond. A third radiometric quality is the calibration accuracy that givesthe user confidence that a digital number (DN) from the sensor corresponds to a specific ra-diance that entered the sensor aperture.

The second set of characteristics involves the sensor spatial resolution, coverage, and geo-metric accuracy. The sensor instantaneous field of view (IFOV), modulation transfer func-tion (MTF), and swath width are the spatial resolution and coverage parameters. The IFOV isthe projection of the detector aperture on the ground, and the MTF defines the total allowa-ble system blur. The swath for Landsat 4 and 5 matches Landsats 1—3 at 185 km across track,for a repeat cycle of 16 days in a 705—kmpolar, sun—synchronous orbit.

The third and final set of characteristics involve the spectral re8olution and coverage, in-cluding spectral band locations and bandpass limits for each, as well as spectral bandpassuniformity from detector to detector. Each of the classes of sensor characteristics justdescribed will now be discussed in terms of the LIDQA program results.

Radiometry

Sensitivity. The TM was designed to meet NASA—specified NE~prequirements of 0.5% in Bands2—4, 0.8% in Band 1, and 1.0% and 2.4% in Bands 5 and 7, respectively. A noise equivalenttemperature (NEAT) requirement of 0.5K was imposed on the thermal Band 6. (This unusualnumbering system designating the 11—am thermal band as number 6 and the

2•2—ii~ short—waveinfrared (SWIR) band as number 7, was created due to the fact that Band 7 was a late addi-tion suggested by the geology specialists.) The dynamic range requirements for each bandaccounted for anticipated high reflectance ground features, and in conjunction with the NE~prequirements, dictated the use of eight—bit quantization to both cover the dynamic range andprovide fine quantization within the noise levels desired.

All reflective bands provide better than specified NE~pfor both the Landsat 4 and 5 114s/5/. The measurements for the thermal band show that the TM NEAT is about four times betterthan specified, as a result of better than anticipated detector performance. This indicatesthat a smaller thermal IFOV (60 meters versus 120 meters) could be had in a TM instrument,and probably still meet the 0.5K NEAT requirement.

Dynamic range. Specifichtions on maximum radiance response were exceeded in every ref lec—tive band, and the specified temperature dynamic range was also exceeded in the thermalband. However, the Thematic Mapper Image Processing System (TIPS) provides processed imageproducts whose dynamic range is approximately as specified.

The satellite radiance to image digital number (DN) relationship has been shown to be linearover the entire dynamic range in both the reflective and thermal bands so that linear con-version from DN to aperture radiance is accurate. Moreover, Dozier /6/ has computed thesaturation radiance in each band, and the calculated saturation reflectance outside the at-mosphere. These latter percentages range from about 25% in the blue Band 1 to about 70% inthe near—IR Band 4.

The low saturation reflectance in 1141 causes saturation in snow images /6/. In general,however, many agricultural and geologic features of interest have low reflectance, and satu-ration is not a problem /7,8/. Features that have low reflectance fill only the low orderbits of the TM dynamic range, and radiometric sensitivity for these features would be im-proved had the sensors been designed with a lower saturation reflectance. However, the TMdynamic range was tailored in each band to produce a sensor that would provide a reasonablecompromise between the conflicting needs of users at the two ends of the dynamic range.Moreover, the TM actually exceeded its specifications on both ends of the dynamic range.

Ground temperature estimation from TN Band 6. TM Band 6 has yielded estimates of surfacetemperature good to approximately ilK /9,10/. Although the TM NE~Fis 0.1K, as measured byLansing /11/, atmospheric effects and, to a smaller extent, calibration errors degrade the

Page 3: Landsat image data quality studies

Landsat Image Data Quality Studies 3

temperature measurements. Engel et al. /9/ Anuta et al. /10/ and Wukelic et al. /12/ eachperformed calculations of water body temperature using Band 6 data; and compared the esti-mates to best water temperature estimates available at test sites.

Engel et al. /9/ assumed a linear relationship of sensor radiance to surface radiance in-cluding atmospheric transmittance and scattering. Thermal data from the TIPS has a dynamicrange from 260K to 320K, and the nonlinear radiance to temperature conversion provided bythe Planck equation yields a nonlinear relationship between surface temperature and TM Band6 DN which can be approximated within 0.2K by a cubic fit. Engel et al. provide a procedure

to calculate ground temperature from sensor DN, given atmospheric transmission and scatter-ing, which can be obtained from radiosonde data.

Reflective band calibration. Measurements at SBRC prior to launch provided detector gainand offset coefficients for all detectors on the TN to meet the calibration requirements.Estimated absolute calibration accuracy was ±5%, or twice as good as was specified /13/.Castle et al. /14/ performed direct ground measurements coincident with TM overflights toproduce in—orbit calibration checks on the Landsat 4 Thematic Mapper. They transferredground radiance measurementsto space with an estimated rms accuracy of ±5%,and comparedthe results to estimates derived from the TM. Selected detectors in Bands 2, 3, and 4 werein agreement with ground—based estimates to within 6.6%, 2.4%, and 12.9%, respectively, foran average agreement of 7.3%.

Sensor radiometric stability and anomalous behavior. Relative to TM internal calibrationdata, Barker et al. /15/ noted cyclic gain variations in the cold focal plane (CFPA) withperiods of two months, and peak—to—peak variation of 5 to 7%. Engel et al. /9/ show, how-ever, that relative changes in gain of all detectors on the TM were within about 1% over afour—month period following launch. The apparent inconsistency of Barker’s absolute meas-urements and Engle et al.’s relative gain results has not yet been resolved. If a reasona-ble mechanism could be isolated that would cause Bands 5 and 7 to vary cyclically in gain insuch a way that all detectors vary together, then relative changes would indeed be expectedto be small, in spite of absolute calibration drift.

One other instability, of much higher frequency than bimonthly, is characterized by a scan—to—scan (7 Hz or slower), and somewhat random DC offset level shift that has been describedby Malila et al. /16/. Although these offset level shifts are noticeable in low radiancescene data (e.g., nightime data) by averaging entire scans, the shifts are generally lessthan half a DN, and are well within the allowable sensor performance limits. Moreover, theshifts are systematic enough that they may be correctable by means of an algorithm suggestedby Malila. This algorithm exploits the correlation of calibration shutter data with theoffset level shift, and subtracts one from the other to correct the anomaly. No cause forscan correlated offset shift has yet been determined.

Other anomalies have been studied, including band nonuniformities that cause striping inlow—level imagery such as water /10,16,19/. Efforts by several teams of investigators haveresulted in a thorough characterization of the striping effects in all spectral bands, andmethods of striping removal have been suggested /17,19,20/. Three kinds of striping havebeen noted in TM imagery, including detector—detector striping, scan—to—scan striping, andnultiscan banding /20/.

The most obvious striping effect is detector—to—detector striping. It is noticeable evenafter radiometric correction. Bernstein et al. /19/ show that the detector gain—offset cor-rections do not fully correct detector—detector striping because of rounding errors in theeight—bit quantization process. They also demonstrate an interesting destriping algorithmthey call “probabilistic” destriping that does a good job of correcting the striping thatwould normally remain in radiometrically corrected TM imagery.

Scan—to—scan differences have also been studied. Mauls et al. /16/ report that a west—eastvariation in overall scene brightness, attributable partly to sun angle and reflectance ef-fects, is different in forward and reverse scans. Fischel /20/ noted a bright—target satu-ration effect which also causes scan—to—scan banding to occur. Murphy et al. /17/ includeda striping removal algorithm in the Canadian ground processing procedures which accounts forsome scan—direction dependencies, including an exponential “droop” effect /17,18/ with a de-cay time of about 1,000 pixels. This time period is extremely short compared to the dc res-toration time constant associated with the TM system, and has not yet been explained. Fi-nally, Bernstein et al. /19/ have reported a nziltiscan banding effect consisting of randomchanges in brightness over several scans. The frequency of this effect is very low, andthere is as yet no good explanation of the effect, although it may be related to the scan—correlated shift reported by Malila.

Coherent noise, which is correlated in frequency, or coherent, with specific electronic ormechanical oscillations in the system, also has been studied /10,18,19,21/. Investigatorsmentioned several frequencies, but in particular, they emphasized a small image degradation

Page 4: Landsat image data quality studies

4 C.F. Schueler and V.V. Salomonson

at 0.3 cycle/pixel, purportedly caused by a spacecraft 32—kHz switching power supply /19/.Wrigley et al. /21/ removed this noise from the Image by notch filtering in the spatial fre-quency domain, and Bernstein et al. /19/ demonstrated an interesting spatial domain filter-ing technique based on measuring the phase of the noise pattern.

One detector in the Landsat—4 TN system failed prior to launch, and the corresponding pixelsare replaced with those of a neighboring detector to produce image products. Bernsteinet al. /19/ and Fusco et al. /22/ both examined alternate algorithms for failed detector re-placement. In general, both teams found that an algorithm based on replacement of a faileddetector output with that of the same detector from a different band yielded superior re-sults compared to the standard algorithm that uses a neighboring detector in the sameband. Fusco et al. are examining the possibility of implementing their algorithm in theItalian ground processing system.

The thermal band calibration has been checked by Lansing and Barker /11/. The key result ofthe verification shows that there has been correctable thermal band gain variation overtime. Most of this variation can be traced to suspected water condensation on a radiativecooler window. The average blackbody reference value also decreased by about 15% over thesame period, and relative internal gains among the four Band 6 channels varied by 5% overthe nine months after launch. However, an outgassing cycle in January 1983 was shown to re-turn the thermal band gain to its original value, so that the reliability of the calibrationreference is really quite good.

Spatial Resolution and Geometric Fidelity~

Resolution — Instantaneous Field of View (IFOV) and blur. Landsat’s spatial resolution per-formance is generally characterized by two parameters. One is the instantaneous field ofview (IFOV), which is just the geometrical projection of a detector onto the ground, and ismeasured in meters on a side on the ground. However, a spatial frequency parameter calledthe modulation transfer function (MTF) is more important because it accounts for all systemblur, including the optical system, the detector IFOV, and the electronics response. TheMTF therefore completely characterizes sensor image quality, aside from sampling artifacts,in a spatial sharpness sense. The equivalent spatial blur characteristic is the sensorline—spread—function (LSF), which defines how a linear input radiance (a road is a practicalexample) is blurred by the sensor. Both the sensor MTF and LSF have been characterized ac-curately on the ground prior to launch by SBRC /23/, and approximately on—orbit by severalinvestigators /10,21,24/.

Because the LSF and MTF are essentially equivalent measures of spatial resolution perform-ance, and becauseMTF is conveniently specified at a single spatial frequency (the Nyquistfrequency, with a period equal to twice the center—to—center spacing of adjacent detectorson the focal plane), MTF only was specified, and both the Landsats 4 and 5 TN exceeded theMTF specification. Independent computations /23,24/ of Landsat—4 LSFs were performed basedon available prelaunch TM frequency response measurements made at SBRC. Comparison of theseresults, with an LSF width at half—maximum of 1.2 to 1.3 IFOVs, to on—orbit data, suggests adegradation of on—orbit LSF of 10—20% (1.4 to 1.6 IFOV5) in Markham’s work /24/, as well asthat of Anuta /10/ and Schowengerdt /21/. These degradations seem small enough to be at-tributed to a combination of atmospheric blur and uncertainties, as well as degradations,resulting from the coarse TM sampling rate on orbit. The tentative conclusion to be drawnfrom these efforts is that the TM’s spatial imaging performance appears to be consistentwith prelaunch measurements.

Sampling and resampling effects. The spatial performance of the Landsat sensors, both theMSS and the TM, are limited not only by the image blur characterized by the 1~fFF or LSF, butalso by discrete sampling, at approximately one IFOV intervals, of the terrain below thespacecraft. Recent studics of the specific sampling artifacts pertinent to MSS and TM havebeen performed /25/, and show that spatial errors in the TM data can be attributed to imageblur, errors in the nonideal image resampling used to correct geometric errors, and aliasing(Moird) errors due to scene undersampling by the sensor. A key result Is that the cubicconvolution resampling algorithm employed by the original Scrounge geometric processing sys-tem at GSFC produced errors in early scenes which could be reduced by modifying the parame-ters in the algorithm. This “parametric” cubic convolution algorithm /26/ was used to cor-rect later scenes /20/.

TM geometric quality. The appearance, as well as the fundamental accuracy and utility of TMdata, Is founded on a cornerstone of pixel—level data registration to both standard maps(geodetic accuracy) and between successive scenes (temporal registration accuracy), as wellas pixel—to—pixel registration between bands /27/. The key to attaining the required accu-racies In each of these three areas Involves three steps. First, the system, including thesensor and the spacecraft, must be of the highest quality to reduce errors to a minimum.Second, equipment must be available to continuously monitor the errors that exist, and toprovide that information in conjunction with the scene data. Third, a software system mustbe provided to implement the necessary corrections, based on the measured errors.

Page 5: Landsat image data quality studies

Landsat Image Data Quality Studies 5

Geodetic errors were restricted to less than half a pixel after correction processing, andtemporal registration errors were restricted to less than 0.3 pixel. Earth topographic ef-fects were excluded from these error specifications, due to spacecraft ephemeris variationbetween successive overflights, which will cause uncontrollable distortion between differentimages of the same geographic area. Band—to—band registration error was restricted to 0.2pixel between bands on the same focal plane, and to 0.3 pixel between bands on different fo-cal planes.

Early analysis suggested that the TM scan mirror repeatability errors might be the chief er-ror source contributing to problems in all three of the registration performance areas men-tioned above. As a result, 0.23 pixel error was allowed to estimate the probable data qual-ity. In fact, performance was about seven times better than predicted, with only 0.033pixel error being attributed to scan mirror repeatability. Moreover, spacecraft attitudedeviation errors are limited to 0.015 pixel, and the combination of both of these errorsources is about a sixth of the allowable temporal registration error. The result of thishigh—quality system performance has been an overall geodetic performance good enough thatsuccessive scenes that are matched to ground control points, rather than to each other,still are registered to one another to within the temporal registration specifications/28/. Finally, band—to—band registration, after geometric corrections, has been obtained towithin 0.1 pixel, or half the requirement.

The necessary error monitoring and correction system to achieve the successes above involvesa complicated set of interactive measurement, recording, and transmission devices connectedto the sensor and to the spacecraft. The geometric correction system is illustrated in Fig-ure 1. The TM mirror scan motion is monitored and Included in payload correction process-ing, along with angular displacement information. Together, these critical pieces of infor-mation, combined with spacecraft attitude and ephemeris information, allow the TIPS to com-pute the actual point on the ground at which the TM was looking at several points throughouteach mirror scan cycle. As a result, geodetic errors can be corrected, to the accuracy ofthe measurements in the payload correction data stream. A feedback loop to continuously re-duce the measured spacecraft errors is provided through gyroscopes, star trackers, attitudecontrol reaction wheels, and an on—board computer. Finally, the overall ground correctionprocess, involving control point processing and resampling, uses the measured errors to cor-rect the image data.

TM HOUSEKEEPING DRIRU

TM \__(GYROS)

ANGULAR DISPLACEMENT \WIDEBAND DATA SENSOR (ADS) \ ______ STAR• IMACt DATA I •ADS SAMPLES \ TRACKERS• RADIOMETRIC CALIBRATION • ADS TEMPERATURE ___________ ___________

MIRROR • SCAN START TIME ON~BOARD I I ATTITUDESCAN • SCAN DIRECTION FORMATTER 4 COMPUTER j.....j CONTROL

CORRECTION • FIRST HALF SCAN ERROR • DRIRU SAMPLES (OBCI REACTIONDATA • SECOND HALF SCAN ERROR ___________ • DRIRU DRIFT ____________ WHEELS

• PCD ESTIMATE +PAYLOAD • ATTITUDE ESTIMATE L UP LINK

CORRECTION • EPHEMERIS EPHEMERISDATA (PCD) S TM HOUSEKEEPING

FLIGHT SEGMENT

GROUND SEGMENT

REFORMAT MIRROR SCAN CONTROLTO ARCHIVE R TION POINT

IMAGERY LIBRARY PRODUCT___________ DATA ___________ f.~ IMAGERY

I ____ ____ II PAYLOAD SYSTEMATIC CONTROL GEODETIC [GEOMETRIC

I CORRECTION CORRECTION ~] POINT P CORRECTION

I PROCESSING DATA (SCD)1PROCESSING CORRECTION RESAMPLING

[~CHIVE IMAGERY

Fig. 1. Landsat—4 Thematic Mapper geometric correction system (after Beyer /27/).

Key factors that contribute to geodetic error are spacecraft attitude and ephemeris error,as well as pointing errors. Temporal registration accuracy is limited to primarily the sameerrors that affect geodetic accuracy, plus spacecraft jitter and TM scan mirror profile er-rors, nonlinearity, and repeatability. Finally, band—to—band registration accuracy is lim-ited only by TM sensor errors, especially scan mirror nonlinearity and detector electronicstime delay variations. These latter effects were measured prior to launch /23/, and arepart of the ground processing system. Additional band—to—band errors can occur if the tele-scope focal length changes, or the focal planes shift on orbit. In spite of the many errorsources, Colvocoresses /29/ found that 1:100,000 maps can be made with TM data that are as

Page 6: Landsat image data quality studies

6 C.F. Schueler and V.V. Salomonson

accurate as standard map products. In fact, a Thematic Mapper image map of Washington, D.C.is available from the United States Geological Survey (USGS).

Spectral Characteristics

The final set of parameters that characterize the Landsat sensor’s performance are the spec-tral bands, including location, bandpass, and detector—to—detector bandpass uniformity.These specifications represent only the general characteristics, and detailed measurementsof all components that affect sensor response as a function of wavelength were combined toproduce end—to—end characteristics for each channel.

A common technique to approximate the irregular bandpass data is to pretend that each spec-tral band may be characterized by a perfect rectangular bandpass with area equal to the areaunder the actual bandpass, and cutoffs given by the specified band cutoffs. Palmer andTomasko /30/ show that this approximation works best if the long and short equivalent band—pass cutoffs are selected to be equal to the mean wavelength in the band plus or minus thespectral response shape standard deviation multiplied by the square root of three. Thisresult has proved to be very close to the so—called 50% cutoffs used to specify the TM bands/31/, at which the spectral response equals half its maximum value.

Qualitative conclusions concerning on—orbit detector—detector spectral bandpass uniformitycan be drawn from Engel et al. /9/. They found that detector—detector relative radiometricresponse time variation was less than 1% over four months, using six TM scenes with entirelydifferent spectral characteristics. Spectral nonuniformities among detectors would havecontributed to relative changes in radiometric response, because of the spectral variationin the TM scenes used. Therefore, at least in a qualitative sense, the Engel et al. resultsare evidence of good spectral uniformity.

IMAGE QUALITY

Information Content

A key aspect of TM performance that has emerged from the LIDQA studies beyond the fact thatthe sensor has substantially met its basic engineering performance requirements has been thediscovery that the raw information content of the TM data appears to exceed that of the MSSsensor by at least a factor of two per pixel for the reflective bands.

The LIDQA results were for typical United States scenes. Price /32/ used an information en-tropy (pixel—to—pixel variability) measure to find that the ratio of 114 information contentper pixel to that of MSS for a specific set of scenes was about 1.8 to one. TM’s increasedpixel—to—pixel variability also led to a factor of two improvement in its classificationperformance over the MSS. Anuta et al. /10/ and Haas and Waltz /33/ found a doubling of thenumber of classes separated by an unsupervised clustering algorithm for coincident TM versusMSS data. Bernstein et al. /19/, Anuta et al. /10/, Malila et al. /16/, and Crist andCicone /34/ found that TM has twice as many principal spectral components as MSS does. Aphysically—based example of TM’s increased spectral information content is the “wetness”component reported by Crist and Cicone, which has no counterpart in MSS data. Finally,Welch and Usery /35/, Walker et al. /36/ and Colvocoresses /29/ all found that TM’s spatialresolution and geometric accuracy were within 1:100,000 USGS map accuracy standards, whileMSS offers 1:200,000 accuracy. Again, these results appear consistent with a factor of twoimprovement in information content of TM over MSS.

The LIUQA results summarized above can be better appreciated in the context of a theoreticalupper bound to the TM’s information capacity. Muck et al. /37/ showed that the upper boundon information transmission capacity of a line—scanning sensor in bits per unit area isgiven approximately by the ratio of the number o~bits of quantization (eight for TM versussix for MSS) to the area of a sensor IFOV (900 in for TM versus 6400 in

2 for MSS). There-fore, if the TM has twice as many orthogonal spectral components as the MSS does, the ratioof the TM upper bound information capacity in bits per unit area to that of the MSS is ap-proximately nineteen. Conversion from a ratio of bits per unit area to bits per pixel elim-inates the sensor IFOV area ratio, and results in an upper bound of 2.6 for the ratio of TMinformatIon content in bits per pixel to that of MSS.

The shortfall of LIDQA results (factor of about two) compared to the upper limit of 2.6 canbe understood by noting that the upper bound pertains to the ratio of information capacityof two Information channels (the TM and the MSS) and not to the information content of anyspecific data sets. How much actual information gain is achieved is scenedependent, asdemonstrated by Hyde and Vesper /40/. Using simulated data sets, they showed that MSS andTM classification performance should be essentially the same for fields greater than 40acres In area. When area per field dropped to a range of 10—40 acres, then TM accuracy wasabout 10—15%better than MSS, comparable to LIDQA results on actual data. Finally, forfields smaller than 10 a~~res, simulated TM data offered twice the classification accuracy ofsimulated MSS data.

Page 7: Landsat image data quality studies

Landsat Image Data Quality Studies 7

Image Utility Relative to MSS

The Thematic Mapper was initially designedwith radioinetric, spatial, and spectral require-ments intended to provide some specific improvements in performance relative to MSS forland—use discrimination, crop and mineral identification, and overall land classificationaccuracy. Specifically, the finer quantization was expected to improve classification accu-racy by providing more accurate ground reflectance estimates. The finer spatial resolutionwas motivated by studies that showed that smaller acreage fields than those in the UnitedStates would be classified more accurately by a sensor with a smaller IFOV than MSS, princi-pally due to the reduced number of “mixed” boundary pixels that would otherwise confuse theclassifiers. The extra and narrower spectral bands were intended to yield improved discrim-ination of both minerals and plants due to the spectral differences that exist in groundfeatures in the short—wave infrared that do not exist in the visible and near infrared.

Williams et al. /39/ sequentially degraded TM data to MSS quality so that they could deter-mine what aspects of TM data quality most contribute to its improved classification perform-ance. They found that the radiometric effect was strongest, followed by the spectral per-formance improvement, in conformance with findings by Toll /40/. Notably, they found thatthe TN’s spatial resolution improvement actually degrades classification of fields. This isattributed to the fact that the TM’s smaller IFOV measures within field variability that isignored by the MSS, and this variability tends to confuse the classifier. They suggest thatthe classifiers be improved in such a way as to efficiently use the smaller TM IFOV. Untilthen, one may often do better by aggregating TN pixels prior to running a field classifierthat has been optimized for MSS data, except for fields that are so small as to show littlewithin—field variability even in TM data.

Several investigators have verified Williams’ overall finding that TM exceeds MSS in land—use classification accuracy, with TM providing an average of about 85% versus MSS perform-ance of 75% relative to ground truth. Specifically, Toll found that TM yielded 85% accu-racy, and Quattrochl /41/ about 90%, compared to MSS performance of 75%. Middleton et al./42/ found that TM generally exceeded MSS by 11—13%in land—use classification accuracy.

The expected TM capacity to classify small fields more accurately than MSS has not been ade-quately verified, but available evidence so far is encouraging. In a comparison study be-tween TM data and MSS data, Markham /43/ found that TM can detect objects as small as 16 me-ters, as compared to 40 meters for MSS. Witt et al. /44/ looked for surface mines in Appa-lachia using TM and MSS data. They found that TM offered great improvements over MSS in mapregistration accuracy and in its ability to delineate small irregular surface features suchas narrow contour strip mines. The average improvement in classification accuracy for minecategories was 25%. DeGloria /7/ found that the TM resolution allows improved detection andidentification of fields less than about twelve acres in area. However, there appears to bea paucity of specific investigations into the improved ability of TN to classify smallerfields than MSS, such as those in China.

Crop identification within agricultural areas depends largely on sensor spectral discrimina-tion capability, and TM outperformed MSS quite dramatically, as was anticipated. Key exam-ples show that the addition of short—wave infrared (SWIK) TM Bands 5 and 7 was the principalreason for the performance improvement, although the narrower visible and near—infraredBands 2, 3, and 4 also contributed. Specifically, DeGloria /7/ found that the SWIR bandsprovide critical leaf moisture content discrimination that leads to the capability to sepa-rate crops such as alfalfa and sugar beets, which are not easily separable with MSS data.

Figure 2 illustrates the TM SWIR crop discrimination capability. Figure 2 shows a portionof the Imperial Valley In California just south of the Salton Sea imaged by TM on 12 Decem-ber 1982. Figure 2(a) illustrates that Band 4 yields essentially no discrimination amongfour fields indicated by a bracket and an arrow. Figure 2(b), on the other hand, demon-strates a dramatic difference in Band 5 reflectance between the fields on the left and thoseon the right. In fact, it has been verified that the two dark fields in the Band 5 Imagewere sugar beet fields with reduced SWIR reflectance due to high leaf moisture content, andthe bright fields were planted with alfalfa.

Pitts et al. /45/ showed that the SWIR bands also enhanced corn and soybean separability,allowing these crops to be identified up to a month earlier in the year than is possiblewith MSS. TM 5 also enhanced separation of meadow and hardwood, as well as soil classifica-tion /46/. Even the TM’s thermal Band 6 has been found useful in renewable resource classi-fication, with tree identification in forests being enhanced /7/, and soybean and sorghumdiscrimination being improved by use of Band 6 /47/.

The water resources necessary to support plants and forestry have not been overlooked.Dozier /6/ found that a critical component in water resources planning that MSS has been un-able to supply is readily available with TM. Especially in California, but in many otherlocations, much of the agricultural water supply is derived from snow runoff. The ability

Page 8: Landsat image data quality studies

8 C.F. Schucler and V.V. Salomonson

~Ii- ~Ei

~NDI!!RIMINABLE ALFALFA

SUGAR BEETS

(a) Band 4 (b) Band 5

Fig. 2. Alfalfa and sugar beet separability is enhancedby Landsat Thematic Mapper SWIK capability.

to estimate snow—pack is critical to estimating snow—melt and subsequent water supplies.Unfortunately, MSS confuses clouds with snow, and much guesswork has been necessary in thepast to distinguish the two In MSS data. TM Bands 5 and 7 make the job of snow—cloud dis-crimination trivial, because snow reflectance is very low in both bands, while cloud reflec-tance remains high. Moreover, Dozier showed that TM has a generally higher saturation levelthan MSS, thus making TN data doubly more useful than MSS for snow hydrology studies.

The TM’s capability extends beyond land and into the aquatic vegetation regions, as well.SpecifIcally, Ackleson and Klemas /48/ found that TM is effective in discriminating sub-merged plants from sand in water depths of up to 2 meters, and Hardisky and Klemas /49/showed that TM Bands 3, 4, and 5 can provide biomass estimates in coastal marshes. Schiebe/50/ found that TM Band 3 can help monitor suspended sediments in lakes. A major ecologicalconcern is the effect of electrical power plant cooling water effluent on the temperature oflarge water bodies. Both Anuta /10/ and Wukelic /12/ studied the application of TM Band 6to temperature measurement of power plant effluents, with encouraging results.

Mineral identification by spectral discrimination has been touted by several geologists asone of the most useful improvements, besides spatial resolution, provided to the geologycommunity by TM. Again, the new SWIR Bands 5 and 7 are the key components in this success,due to significant spectral signature structure possessed by clay minerals in the SWIR spec-tral region. Specifically, Borengasser, et al. /51/ and Podwysocki /52/ have found thatclays and iron oxides are easily discriminable for the first time with TM data using TM 5and 7. Iron oxides provide a unique signature in the visible green and red, and clays inthe SWIR, so that spectral band ratio products allow relatively easy visual identificationof clays and iron oxides in color—coded TM Imagery.

Borengasser,et al. as well as Abrams et al, /53/ and Dykstra et al., /8/ agreed that geo-logical maps made with TM data provide high correlation with existing maps that were pain-fully produced by man—years of field geology. Dykstra found that TM provided detection ofphenomena related to hydrocarbon 8eepage, which may be useful in petroleum exploration. Fi-nally, Borengasser, et al., Dykstra et al., and Short /54/ report that TM’s enhanced spatialresolution provides direct visual identification of likely prospecting sites to an extentsignificantly better than MSS.

CONCLUSION

The LIDQA studies have not only shown that the Thematic Mapper has met its original specifi-cations, but have characterized its performance radiometrically, spatially, and spec-trally. The sensor has been shown to have sensitivity performance of about 0.5% NE~pin thereflective bands, and an NEST of 0.11K. Although raw TM data also exceeds its dynamic rangespecifications, the Thematic Mapper Image Processing System (TIPS) sets the dynamic rangeapproximately equal to specification. Several minor instabilities and anomalous radiometriceffects have been reported, but the overall radiometric performance has been rated excellentby all LIDQA investigators. TM’s spatial performance has exceeded specification both interms of image sharpness and geometric accuracy, yielding mapping accuracy at least twice asgood as MSS. Spectral performance is within specification.

The LIDQA studies have also shown that the TM’s image utility is at least a factor of twohigher than MSS. TM has the capability to separate twice as many land—use classes as MSS,and has demonstrated a 10—15% improvement in land—use classification accuracy in UnitedStates scenes. Finally, TM’s spectral capability in the short-wave IR has been shown to

Page 9: Landsat image data quality studies

Landsat Image Data Quality Studies 9

provide the enhanced plant and mineral discrimination capability that was anticipated, aswell as other improvements, such as snow—cloud discrimination. These results are consistentwith the discovery that TM’s raw information capacity is at least twice as high as MSS, on aper pixel basis, which is close to the upper theoretical limit.

ACKNOWLEDGEMENTS

The authors express appreciation to the entire LIDQA research team for valuable work thathas contributed to verifying the high quality of the Thematic Mapper sensor and to continu-ing recognition of the important role that land remote sensing can play in resource manage-ment. The authors also express gratitude to members of the Santa Barbara Research Centertechnical staff who directly supported the organization of the review, including Mss. AlisonCall, Helen Hardenbergh, and Barbara Marks. and Mssrs. Jack Lansing, Gary Later, PaulMaynard, Mark Stegall, and Andrew Stevens. Finally, thanks go to Carolyn Darga and KathyHudson at SBRC and to Joan Wentz at GSFC for manuscript typing.

REFERENCES

1. Goddard Space Flight Center, Application Notice for Participation in the Landsat—D Im-age Data Quality Analysis Program, National Aeronautics and Space Administration,Greenbelt, MD, U.S.A. (October 1981).

2. J. Harnege and 0. Landgrebe, Landsat—D Thematic Mapper Technical Working Group FinalReport, JSC—09797, National Aeronautics and Space Administration, Houston, TX, U.S.A.

(June 1975).

3. V.V. Salomonson and H. Mannheimer, An overview of the evolution of Landsat—4, Proceed-ET1 w466 495 m501 495 lSBTings of the PECORAVIII Symposium, Augustana Research Inst., Sioux Falls, SD, p. 32—44(1984).

4. Hughes Aircraft Company Space and Communications Group, Proposal for a Thematic MapperInstrument System, RFP No. S—79160/229, MAC Ref. 76(44)—11728/B4596—SCG6O347P,El Segundo, CA, U.S.A. (August 1976).

5. J.L. Engel and 0. Weinstein, Thematic Mapper — an overview, IEEE Trans. Geosci. RemoteSensing, GE—21, #3, 258—265 (1983).

6. J. Dozier, Snow reflectance from Landsat—4 Thematic Mapper, IEEE Trans. Geosci. RemoteSensing, GE—22, #3, 323—328 (1984).

7. S. DeGloria, Spectral variability of Landsat—4 TM and MSS, IEEE Trans. Geosci. RemoteSensing, GE—22, #3, 303—311 (1984).

8. J. Dykstra, J. Everett, C. Sheffield, Evaluation of Landsat—D TM performance as appliedto hydrocarbon exploration, in: Land8at Early Results Symposium Proceedings, ed.J.L. Barker, NASA GSFC, May 1984.

9. J. Engel, J. Lansing, D. Brandshaft, and B. Marks, Radiometric performance of the The-matic Mapper, in: Proceedings of the Seventeenth Simposium on Remote Sensing of Envi-ET1 w203 295 m502 295 lSBTronment, Environmental Research Institute of Michigan, Ann Arbor, MI, May 1983.

10. P. Anuta, L. Bartolucci, M. Dean, D. Lozano, E. Malaret, C. McGillem, J. Valdes, and C.Valenzuela, Landsat 4 MSS and TM data quality and information content analysis, IEEETrans. Geosci. Remote Sensing, GE—22, #3, 222~236 (1984).

11. J. Lansing and J. Barker, Thermal band characterization of TM, in: Landsat Early Re-ET1 w422 230 m498 230 lSBTsults Symposium Proceedings, ed. J.L. Barker, NASA GSFC, May 1984.

12. C. Wukelic, H. Foote, L. Martussi, F. Braun, and S. Blair, Landsat—4 image data qualityanalysis for energy related applications, in: Landsat Early Results Symposium Proceed—~ ed. J.L. Barker, NASA GSFC, May 1984.

13. J. Barker, Prelaunch absolute radiometric calibration of the reflective bands on theLandsat—4 Protoflight Thematic Mapper, in: Landsat Early Results SymposiumProceed—

ed. J.L. Barker, NASA CSFC, May 1984.

14. K.R. Castle, R.G. Holm, C.J. Nastner, J.M. Palmer, P.N. Slater, M. Dinguinard, C.E.Ezra, R.D. Jackson, and R.K. Savage, In—flight absolute radiometric calibration of theTM, IEEE Trans. Geosci. Remote Sensing, GE—22, #3, 251—255 (1984).

15. J. Barker, R. Abrams, D. Ball, and K. Leung, Characterization of radiometric calibra-tion of Landsat—4 TM reflective bands, in: Landsat Early Results SymposiumProceed—~iig~, ed. J.L. Barker, NASA GSFC, May 1984.

Page 10: Landsat image data quality studies

10 C.F. Schueler and V.V. Salomonson

16. W.A. Malila, M.D. Metzler, D.P. Price, and E.P. Crist, Characterization of Landsat—4MSS and TN digital image data, IEEE Trans. Geosci. Remote Sensing, CE—22, #3, 177—191(1984).

17. J. Murphy, T. Butlin, P. Duff, and A. Fitzgerald, Revised radiometric calibration tech-nique for Landsat—4 TM data, IEEE Trans. Geosci. Remote Sensing, CE—22, #3, 243—250(1984).

18. H. Kieffer, E. Eliason, P. Chavez, Thematic Mapper intraband and radiometric perform-ance, in: Landsat Early Results Symposium Proceedings, ed. J.L. Barker, NASA GSFC, May1984.

19. R. Bernstein, J. Lotspeich, H. Myers, H. Kolsky, and R. Lees, Analysis and processingof Landsat—4 sensor data using advanced image processing techniques and technologies,IEEE Trans. Geosci. Remote Sensing, GE—22, #3, 192—221 (1984).

20. D. Fischel, Validation of the TM radiometric and geometric correction algorithms, IEEETrans. Geosci. Remote Sensing, GE—22, #3, 237—242 (1984).

21. R. Wrigley, 0. Card, C. Hlavka, J. Hall, F. Mertz, C. Archwamety, and R. Schowengerdt,TM image quality: registration, noise, and resolution, IEEE Trans. Ceosci. RemoteSensing, GE—22, #3, 263—271 (1984).

22. L. Fusco, P. Treverse, TM failed detector replacement, in: Landsat Early Results Sym-ET1 w393 518 m510 518 lSBTposium Proceedings, ed. J.L. Barker, NASA GSFC, May 1984.

23. C. Schueler, TM protof light model line spread function, in: Proceedings of the 17thInternational Symposium on Remote Sensing of Environment, Environmental Research Insti-tute of Michigan, Ann Arbor, MI, May 1983.

24. 8. Markham, Characterization of the Landsat sensor’s spatial response, in: LandsatEarly Results Symposium Proceedings, ed. J.L. Barker, NASA CSFC, May 1984.

25. 5. Park and R. Schowengerdt, Image sampling, reconstruction, and the effect of sample—

scene phasing, Applied Optics, 21, #17, 3142—3151 (1982).

26. S. Park and K. Schowengerdt, Image reconstruction by parametric cubic convolution,Comp. Vision, Graphics, and Image Proc., 23, 258—272 (1982).

27. E. Beyer, Thematic Mapper geometric correction processing, in: Proceedings of the 17thInternational Symposium on Remote Sensing of Environment, Environmental Research Insti-tute of Michigan, Ann Arbor, MI, May 1983.

28. J. Brooks, A. Jai, T. Keller, E. Kimmer, and J. Su, TM geometric correction performanceand evaluation, to appear in: Proceedings of the 10th International Symposium on Ma-ET1 w263 326 m506 326 lSBTchine Processing of Remotely Sensed Data, Purdue University, West Lafayette, IN (1984).

29. A. Colvocoresses, Mapping of Washington, D.C. and vicinity with the Landsat—4 TM, in:Proc. of the 50th Meeting of the American Society of Photogrammetry, p. 757—7b4 (1984).

30. J. Palmer, Effective bandwidths for Landsat—4 MSS and TM subsystems, IEEE Trans. Geo

sci. Remote Sensing, GE—22, #3, 336—338 (1984).

31. J. Palmer and M. Tomasko, Broadband radiometry with spectrally selective detectors, ~tics Letters 5, #5, 208—210 (1980).

32. J. Price, Comparison of the information content of data from the Landsat—4 TM and MSS,IEEE Trans. Geosci. Remote Sensing, GE—22, #3, 272—280 (1984).

33. K. Haas and F. Waltz, Evaluation of TM data for natural resource assessment, PECORAVIII Proceedings, Augustana Research Inst., Sioux Falls, SD, p. 122—133 (1984).

34. E. Crist and K. Cicone, A physically—based transformation of TM data — the TM tasseledcap, IEEE Trans. Geosci. Remote Sensing, GE—22, #3, 256—262 (1984).

35. K. Welch and E. Usery, Cartographic accuracy of Landsat—4, MSS and TM image data, IEEETrans. Geosci. Remote Sensing, GE—22, #3, 281—287 (1984).

36. R. Walker, A. Zobrist, M. Bryant, B. Gokham, S. Friedman, and T. Logan, An analysis ofLandsat—4 TM geometric properties, IEEE Trans. Geosci. Remote Sensing, GE—22, #3, 288—293 (1984).

Page 11: Landsat image data quality studies

Landsat image Data Quality Studies 11

37. F. Muck, S. Park, N. Halyo, and S. Staliman, Application of information theory to thedesign of line—scan imaging systems, NASA Tech. Paper 1897, 1981.

38. R. Hyde, M. Spencer, and N. Vesper, An estimate of cost effectiveness of machine pro-cessing increased spatial resolution MSS data, Holcomb Research Institute, ButlerUniv., National Oceanic and Atmospheric Administration, P.O. No. NA83AAAO2O79, 50pages, April 1983.

39. D. Williams, J. Irons, B. Markham, R. Nelson, D. Toll, R. Latty, and M. Stouffer, Astatistical evaluation of the advantages of Landsat TM data in comparison to MSS data,IEEE Trans. Geosci. Remote Sensing, GE—22, #3, 294—302 (1984).

40. D. Toll, Preliminary study of information extraction of TM data for a suburban/regionaltest site, in: Landsat Early Results Symposium Proceedings, ed. J.L. Barker, NASAGSFC, May 1984.

41. D. Quattrochi. An Initial analysis of Landsat—4 TM data for the discrimination of agri-cultural, forested wetland, and urban land cover, in: Landsat Early Results SymposiumProceedings, ed. J.L. Barker, NASA GSFC, May 1984.

42. E. Middleton, Y. Lu, R. Witt, and R. Sekhon, Relative accuracy assessment of Landsat—4MSS and TM data for Level I land cover inventory, in: Landsat Early Results SymposiumProceedings, ed. J.L. Barker, NASA GSFC, May 1984.

43. B. Markham, Preliminary comparisons of the information content and utility of TM versusMSS data, in: Landsat Early Results Symposium Proceedings, ed. J.L. Barker, NASA GSFC,May 1984.

44. R. Witt, H. Blodget, and R. Marcell, Final comparison of TM and MSS for surface mineassessment in Logan County, West Virginia, in: Landsat Early Results Symposium Pro-ET1 w337 466 m497 466 lSBTceedings, ed. J.L. Barker, NASA CSFC, May 1984.

45. D. Pitts, G. Badhwar, D. Thompson, K. Henderson, S. Chen, C. Sorensen, and J.Carnes,Evaluation of corn/soybean separability using TM data and TM simulator data, IEEETrans. Ceosci. Remote Sensing, CE—22, #3, 312—318 (1984).

46. D. Thompson and K. Henderson, Evaluation of TM for detecting soil properties undergrassland vegetation, IEEE Trans. Ceosci. Remote Sensing, GE—22, #3, 319—322 (1984).

47. R. MacDonald, Preliminary evaluation of TM image data quality, in: Landsat Early Re-ET1 w426 376 m502 376 lSBTsults Symposium Proceedings, NASA GSFC, May 1984.

48. S. Ackleson and V. Kiemas, Assessing Landsat TM and MSS data for detecting submergedmast communities, in: Landsat Early Results Symposium Proceedings, ed. J.L. Barker,NASA GSFC, May 1984.

49. ii. Hardisky and V. Klemas, Remote sensing of coastal wetlands biomass using TM wave—bands, in: Landsat Early Results Symposium Proceedings, ed. J.L. Barker, NASA GSFC,May 1984.

50. F. Schiebe, A first evaluation of TM data to monitor suspended sediments in lakes,in: Landsat Early Results Symposium Proceedings, ed. J.L. Barker, NASA GSFC, May 1984.

51. H. Borengasser, D. Brandshaft, and J. Taranik, Geological Applications of EnhancedLandsat 4 TM Imagery of South—Central Nevada, in: Proceedings of the InternationalSymposium on Remote Sensing of Environment, Third Thematic Conference, Remote Sensingfor Exploration Geology, Environmental Research institute of Michigan, ColoradoSprings, CO, April 1984.

52. M. Podwysocki, L. Bender, A Preliminary Evaluation of Landsat—4 TM Data for Geometricand Radiometric Performance, in: Landsat Early Results Symposium Proceedings, ed.J.L Barker, NASA GSFC, May 1984.

53. M. Abrams, Geologic utility of Landsat—4 TM data, in: Landsat Early Results SymposiumProceedings, ed. J.L. Barker, NASA GSFC, May 1984.

54. N. Short, Assessment of computer-based geologic mapping of rock units in the Landsat—4scene of northern Death Valley, CA, in: Landsat Early Results Symposium Proceedings,ed. J.L. Barker, NASA GSFC, May 1984.

4555 5/5—5