HflNM-$ C H fJK \t ·\AgRISTARS DC-Y1-04069 NSTl/ERl-197 g I -J (0 ~ Domestic Crops and Land Cover Technical Report A Joint Program for Agriculture and Resources Inventory Surveys Through Aerospace Remote Sensing April 1981 AN EVALUATION OF MSS P-FORMAT DATA REGISTRATION Marcellus H. Graham Raymond Luebbe Na tional Aeronautics and Space Ad ministration National Space Technology Laboratories Earth Resources Laboratory NSTL Station, MS 39529 \ I\II\SI\
56
Embed
AgRISTARS - USDA › Education_and_Outreach... · (Latitude, Longitude, ROW2, COL2) Compared with Landsat Coordinates (ROW1, COL1) Determined from P-Format Registration for Data Set
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
HflNM-$ C H fJK\ t
·\AgRISTARSDC-Y1-04069NSTl/ERl-197
g I -J (0~
Domestic Crops and Land Cover
Technical Report
A Joint Program forAgriculture andResources InventorySurveys ThroughAerospaceRemote Sensing
April 1981
AN EVALUATION OF MSS P-FORMAT DATA REGISTRATION
Marcellus H. GrahamRaymond Luebbe
Na tional Aeronautics and Space Ad ministrationNational Space Technology LaboratoriesEarth Resources LaboratoryNSTL Station, MS 39529
\
I\II\SI\
. \
)"
.•
I. REJIIORT NO.
DC-Yl-04069, NSTL/ERL-197
TECHNICAL RE~RT STANDARD TITLE ••AGE
I. GOVERNMENT ACCES.ION NO.
"" A. TITLE AND SUBTITLE
An Evaluation of MSS P-Format Data Registration
7. AUTHOR(S)
1) Marcellus H. Graham and 2) Raymond C. Luebbe9. PERFORMING ORGANIZATION NAME AND ADDRESS
1) NASA/National Space Technology Laboratories/Earth Resources Laboratory
2) USDA/Economics and Statistics Service/Research12. SpONSORI NG AGENCY NAME AND ADDRE.S
National Aeronautics and Space Administrationand
United States Department of Agriculture
I. "I: fOORT DATI:
April 1981I. "E"P'ORMING ORGANIZATION CODE
•• "IERP'ORMING ORGANIZATIONRItJllORT NO.
10. WORK UNIT NO.
II. CONTRACT OR GRANT NO.
DivisiOIlu. TY"E OP' REPORT e. PERIOD
COVERIED
Technical ReportI A. S~NSORING AGENCY CODE
)
15. SUPPLEMENTARY NOTES
II. ABST~ACT
Twelve Landsat scenes of the P-format were analyzed for registration accuracybased on the Hotine Oblique Mercator (HOM) tick marks contained in the annotationrecord. Independently chosen ground control point$ were used to evaluate each scene.
The results indicate that 8 out of the 12 showed either good or fairly good'~)registration and that the registration information provided with the MSS data can.~ be used as a starting point from which to make a more precise registration.
17. KEV WORDS II. DISTRIBUTION STATEMENT
Hotine Oblique Mercator projectionRegistrationGround control pointsP-formatLandsatGeometric correction
NSTL FORM U (JAN 1975) *For sale by the National Technical Information Service,
)
10.
Unlimited
SECURITY CLA.SIF.Coft ••••••••• ) II.
UnclassifiedNO. OF "AGES
57II. PRICE *
Springfield, VA 22151
)
)
)
)
•
AN EVALUATION OF MSS P-FORMAT DATA REGISTRATION
April 1981.
Marcellus H. Grahamand
Raymond Luebbe.
National Aeronautics and Space AdministrationNational Space Technology Laboratories
Earth Resources Laboratory
and
United States Department of -AgricultureEconomics and Statistics Service
Research Division
Acknowledgments
The authors wish to thank June Thormodsgard and KenCrouse of the EROS Data Center for their support in providingtimely information and software for coordinate transformations,Larry Duplessis and Larry Stringer of the Lockheed Engineeringand Management Services Company, Inc., for data processing andtechnical editing support, and Helen Paul of NASA/NSTL/ERL fortyping this report.
TitleCatalog of Data Sets 1-12Assessment Numbers (0-5) vs. Number ofControl PointsIndependently Chosen Control Points(Latitude, Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROW1, COL1)Determined from P-Format Registration forData Set 1Independently Chosen Control Points(Latitude, Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROW1, COL1)Determined from P-Format Registration forData Set 2Independently Chosen Control Points(Latitude, Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROW1, COL1)Determined from P-Format Registration forData Set 3Independently Chosen Control Points(Latitude, Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROW1. COL1)Determined from P-Format Registration forData Set 4Independently Chosen Control Points(Latitude. Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROW1. CaLl)Determined from P-Format Registration forData Set 5Independently Chosen Control Points(Latitude, Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROW1, COL1)Determined from P-Format Registration forData Set 6Independently Chosen Control Points(Latitude, Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROW1. COL1)Determined from P-Format Registration forData Set 7
vi
Pa~e1719
20
21
22
23
24
25
26
1
.,
(
(
(
C
l
)
)
J
)
Table10
11
12
13
14
15
LIST OF TABLES
TitleIndependently Chosen Control Points(Latitude, Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROWl, COLI)Determined from P-Format Registration forData Set 8Independently Chosen Control Points(Latitude, Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROWl, COLI)Determined from P-Format Registration forData Set 9Independently Chosen Control Points(Latitude, Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROWl, COLI)Determined from P-Format Registration forData Set 10Independently Chosen Control Points(Latitude, Longitude, ROW2, COL2) Comparedwith Landsat Coordinates (ROWl, COLI)Determined from P-Format Registration forData Set 11Independently Chosen Cont~ol Points(Latitude, Longitude, ROW2', COL2) Comparedwith Landsat Coordinates (ROWl, COLI)Determined from P-Format Registration forData Set 12~ummary of Tables 3-14
vii
Page
27
28
29
30
31
32
(
'"
(LIST OF FIGURES (
Figure Title Page1 Distribution of 12 Landsat scenes 33
according to path and row numbers2 Relative distribution of independently 34
chosen control points for data set 13 Relative distribution of independently 35
chosen control points for data set 24 Relative distribution of independently 36
chosen control points for data set 35 Relative distribution of independently 37
chosen control points for data set 46 Relative distribution of independently 38
chosen control points for data set 57 Relative distribution of independently 39 (chosen control points for data set 68 Relative distribution of independently 40 Cchosen control points for data set 79 Relative distribution of independently 41
chosen control points for data set 810 Relative distribution of independently 42
chosen control points for data set 911 Relative distribution of independently 43
chosen control points for data set 1012 Relative distribution of independently 44
chosen control points for data set 1113 Relative distribution of independently 45
chosen control points for data set 1214 Quality assessment number versus 46
relative error of registration value y
l
viii
))
))
)
)
AN EVALUATION OF MSS P-FORMAT DATA REGISTRATION
SummaryThe purpose of this study was to determine the relative
accuracy of the registration of P-format multispectral scanner(MSS) data using independently chosen ground control points.
The study indicated that it will usually be necessaryto improve the local registration of P-format MSS data tomeet the accuracy requirements of the scene-to-map task ofthe AgRISTARS Domestic Crops and Land Cover (DCLC) project.However. it is probable that-any other Landsat full-scene(2983 rows by 3548 columns) registration technique alsosacrifices accuracies at some local-areas. In addition.various sensor or satellite difficulties such as the linestart problem (ref. 1) or the light source problem (ref. 2)can degrade the accuracy of the P-format registrationinformation. Nevertheless. the registration informationprovided with MSS data that have been cor~ected using groundcontrol points in the Master Data Processor can be used as astarting point from which to make a more precise registration.The algorithm that handles the improved registration can indi-cate when problems in MSS data registration are likely to haveoccurred.
IntroductionThis report documents an evaluation of the registration
accuracy of P-format MSS data. which is the first phase of
1
the scene-to-map registration task of the AgRISTARS DCLCproject. (A later phase of the task will deal with theproblem of automatically registering USDA segment datawith Landsat MSS data.)
The Landsat Users Handbook (ref. 3) describes accuraciesassociated with data generated by the Master Data Processor(MDP) at the Goddard Space Flight Center. According to thehandbook, the accuracy of the ground control points used bythe MOP depends on the quality of the reference map fromwhich the points were chosen. When a Landsat scene is auto-matically registered by the MDP, the registration depends onother variables as well, such as the quality of the data.
The EROS Data Center conducted a geometric accuracyevaluation on two 1979 Landsat images and reported a standarderror of about 160m in each of two dimensions (ref. 4). Thetwo images had been system-corrected; that is, correctedwithout the use of ground control points.
Colwell, Davis, and Thomson (ref. 5) conducted an accuracyevaluation of a July 18, 1979, Landsat scene (30500-15352) thatincluded an area around Toledo, Ohio. The scene had beenground-control corrected. When compared with ground control
,
((
(
C
points from 1:24,000-scale topographic maps, Colwell, et al.,reported root mean square (rms) errors of 218.0m in theeast-west direction and 879.8m in the north-south direction. ~
To evaluate the registration accuracy of Landsat MSS ~data, the analyst must be familiar with and be able to retrieve
2
)
)
)
)
information in annotation records of computer-compatible tapes(CCT's) produced after February 1979 at the EROS Data Center.The pertinent registration information is given in terms ofHotine Oblique Mercator (HOM) coordinates and Landsat coordi-nates.' These coordinates are often referred to as tick marksand are indicated on the Landsat image photographs providedby the EROS Data Center as U's and V's, the axes names for theHOM projection (ref. 6). The tick marks appear on the top,bottom, right, and left sides of the data and the number ofticks per ~ide vary from scene-to-scene, depending on where theframe center falls with respect to the HOM projection.
The top edge tick marks always intersect with extendedLandsat row number -17.5; likewise, 'the left side ticks alwaysintersect with extended Landsat column -17.5, the right sideticks with extended Landsat column 3565.5, and the bottomedge "ticks with extended Landsat row 3000.5. The V and Ucoordinate values given in the annotation record should bemultiplied. by 10,000m to obtain the actual coordinate values.Examples of the tick mark information are given in table 1under the heading "Tick Marks."
The conversions between the tick mark information andthe Landsat coordinates and between the Landsat coordinatesand other map projections are not straightforward. For thisreason, the EROS Data Center made available to the Landsatusers software that can be used to convert from one coordinatesystem (Landsat, UTM, HOM, latitude and longitude) to another
3
(ref. 6). This software was obtained from the EROS Data Centerand used for this evaluation.
Also contained in the annotation record of the CCT isan indicator of the type of correction applied and a qualityassessment number for the geometric corrections. The indicatorcan be one of four letters: U for uncorrected, S for systemcorrected, G for corrected based on ground control points, andR for corrected based on relative ground control points.
The quality assessment number is an integer value thatusually ranges from 0 to 5 and indicates, within a range, thenumber of ground control points used for the geometriccorrection. The assessment number is the truncated integerof the expression (N+7)/8, where N is the number of controlpoints used. Table 2 shows a breakout of the number ofcontrol points associated with assessment numbers 0-5.
The purpose of this study was to evaluate the accuracyof the tick mark registration by using independently chosenground control points and to answer the following questions:
1. Is the registration accuracy of the MSS dataadequate for the scene-to-map registration taskof the AgRISTARS DCLC project?
2. Does the registration accuracy depend heavilyupon the quality assessment number?
4
((
((
l
(
)
)
.)
)
..
3. What problems are associated with using theregistration information on the CCT's and canthese problems be "worked around" for thescene-to-map task of the AgRISTARS DCLC project?
Procedure
The procedure used for selecting the independent controlpoints and evaluating the accuracies of these points isoutlined in the following steps:
Step 1Twelve Landsat scenes were used to do the analysis
(table 1). The distribution of the 12 scenes within theU.S., shown in figure 1, was mainly dependent on data readilyavailable to the USDA Economics and Statistics Service (ESS)and the Earth Resources Laboratory (ERL) from associatedresearch efforts.
Step 2Sets of 25 to 38 ground control points were chosen from
each scene (tables 3-14). Each control point consisted offour coordinates--latitude, longitude, ~andsat row, andLandsat column. Sets 1-6 were chosen by ESS and sets 7-12were chosen by ERL.
The selection of a control point requires that theanalyst be able to find the same feature on a map and inthe Landsat imagery; and, after the control point is selected,the analyst must be able to accurately determine the coordinates
5
of the point. ·Therefore, this process hinges on the Landsatimage and the accuracy of the map used. Depending on thetype of terrain (mountains, deserts, agriculture fields, etc.),control points can sometimes be difficult to locate. However,control point selection from Landsat imagery is a processthat has been used many times over the last 6 to 7 years andhence the technique has been repeatedly improved. Also,because it has been used so much, there are data to documentthe accuracy of using this method.
To ensure accurate control point selection, the analystsused large-scale maps, usually the USGS 1:24,OOO-sca1e,7~1/2-minute quadrangle sheets, when these were availablefor the area. For small portions of a few scenes where thesesheets were not available, USGS 1:62,500 or 1:250,000 quad-rangle sheets were used.
The control points chosen were distributed throughoutthe scene. In some cases, clouds or terrain featureshindered the most desirable distribution. Figures 2-13show the relative distributions of the control points basedon their Landsat coordinates.
I.
(
(
Also, to ensure accuracy, the control points as a setwere analyzed against a polynomial model. For this study,ESS used a cubic polynomial model and ERL used a linearpolynomial model. Both models have been used over the past ~several years and have produced good results. These modelsindicated to the analysts which control points yielded large (
6
)
)
)
)
)
)
residual errors. The analyst then tried to determine whythe point had a large residual. For example. were thecoordinates misread either from the map or the Landsatimage? Was the point a bend in a river or a winding roadwhere the map coordinates may be less accurate? In any case.when the analyst was satisfied that the coordinates of thepoint were inaccurate. the coordinates were corrected ifpossible or the point deleted. If the point coordinatesappeared to be accurate and the point had a large residualerror. the point was kept.
Historically. this method of control point selectionand analysis will produce a registration accuracy of + 1Landsat pixel for at least 68 percent of the points forthe area being registered.
Step 3All of the control points chosen at ESS were forwarded
to ERL. where all sets were entered into the computer foranalysis. Software obtained from the EROS Data Center(Subroutine PIXGEO) was used to compute the Landsat row andcolumn for each latitude and longitude. Inputs to thisprogram include the HOM tick marks. the path and row numbers.the sensor type. the projection type. the Worldwide ReferenceSystem longitude. and the type of conversion desired. Forthe purposes of this study the sensor type was always "MSS".the projection type "H" for Hotine. the Worldwide ReferenceSystem longitude 0 (zero), because this value was not used
7
by the program, and the type of conversion was always fromlatitude and longitude to Landsat coordinates. Therefore,the only inputs that changed for each set of control pointswere the tick marks, the path and row numbers, and thelatitude and longitude of the control points.
The EROS software (Subroutine PIXGEO) used to convertlatitude and longitude coordinates to Landsat row andcolumn basically uses two steps. The first step is a con-version from latitude and longitude to the HOM U and Vcoordinates. The equations that relate the two systemsare closed formulas; that is, mathematically there are noapproximations or iterations involved. The second steprelates the U and V coordinates to the Landsat row andcolumn. This is done by using interpolation among the tickmarks given in the annotation record.
Hence, results from both of these steps can be achievedwith the accuracy depending almost entirely· on the accuracyof the machine. Therefore. if it is assumed that the HOMtick marks are exact (although they are expressed only tothe nearest whole row or column number). the error in theconversions from latitude and longitude to Landsat row andcolumn would be much less than ± 1/2 Landsat pixel.·
Step 4
The Landsat row and column numbers computed using theEROS software were compared with their corresponding Landsat
8
(
(
(
C
l
(
)
~ row and column numbers that were chosen manually. Statisticalmeasures used for the evaluation were determined by thefollowing equations. The row bias was given by:
NPRBIAS = ~ (ROWli - ROW2i)
i=1NP
where NP was the number of ground control points chosen,ROWl was the Landsat row determined using the EROS software,and ROW2 was the Landsat row read from the Landsat imagery.Th~ row standard deviation was given by:
Analogous equations were used for the Landsat columnanalysis.
Results
The results of applying the statistical measures givenin Step 4 above to the data sets are shown in tables 3-14,
with a summary given in table 15.
To evaluate the accuracy of each set one has to considerboth the RBIAS and the RSO (or the CBIAS and the CSD) at thesame time. If only the RBIAS (CBIAS) is small, the row
9
(column) differences could still be large in absolute valuebut cancel each other because of opposite signs. But, ifboth the RBlAS (CBlAS) and the standard deviation RSD (CSD)are relatively small, then the registration accuracy is judgedto be good. Such was the case with data sets 1, 2, 3, 7, 8,and 9. For these data sets, the RBlAS, RSD, CBlAS, and CSDvalues were small enough so that one could argue that signifi-cant portions of these measures could be attributed to theselection of the independent control points. For example,point No. 1 in data set 8 (table 10) has relatively largevalues for the row and column differences (ROWl-ROW2.COLl-COL2) when compared with the other control points ofset 8. This suggests that point No. 1 was not a good controlpoint. Such errors can increase the standard deviation measuresubstantially.
The registration of data set 11 was fairly good butshowed a consistent error in the Landsat row direction. withan RBlAS of 2.1 and a standard deviation ofl.s.
The RBIAS (-3.6) and the CBIAS (3.2) of data set 10 are
((
(
C
significantly higher than those of data sets 1, 2, 3, 7, 8, 9,and 11. The data of this set were somewhat hazy and this,coupled with the terrain of the area, hindered good controlpoint selection. However, the row and column differencesseem to be consistent and not random, which may indicate thatthere was a registration problem associated with the tick marks. l
10
))
))
))
Data set 4 had a consistent bias of approximately 16Landsat rows. This data set was affected by the scan line-start problem described in reference 1 and approximately thefirst 30 percent of the data on the left side of the scenecould not be used (figure 5). As a result of this problem,the registration accuracy of the scene may have been affectedwhp-n it was processed through the Master Data Processor. Theindependently chosen control points used for data set 4 pro-duced a good registration when compared with ground truthinformation digitized from maps. Therefore, it was concludedthat the tick mark registration information of the CCT wasinaccurate and not the independently chosen control points.The scene had a quality assessment number of 2, which indi-cates that 9 to 16 control points were used to register theimage.
Data set 12, with a quality assessment number of 4. hada consistent shift in both the Landsat row and column directionof approximately 10 and 9, respectively.
Data sets 5 and 6 showed very poor registration accuracywhen compared with the independently chosen ground controlpoints. The major errors were in the north-to-south directionand show consistent biases of approximately -415 Landsat rowsand -407 Landsat rows for data sets 5 and 6, respectively.As in the case of data set 4, the independently chosen controlpoints were used to register the scene and, when compared withground truth information. the registration was quite accurate.
11
Therefore, it was concluded that the tick mark registrati9nof the CCT's was very inaccurate.
Data set 6 was the only scene of the 12 analyzed that hadan assessment number of O. This means that the scene wassystem-corrected or corrected without the use of ground controlpoints in the MDP. Th~ quoted accuracy for system-correctedscenes is approximately 2.5 km (ref. 3) or 2.7 km (see theappendix). However, the consistent shift of 407 Landsat rowstimes approximately 57m per row yields a shift of well over20 km.
Data set 5 had a quality assessment number of 1, whichindicates that 1-8 control points were used to register thescene. However, it seems to be slightly less accurate thanscene 6, which had an assessment number of O.
In an attempt to relate the registration inaccuracies ofthe CCT's to problems experienced by the sensor system or theMDP, the results of this study were compared with known,documented problems.
During the period from June 14, 1979, to August 1, 1979,a patch in the MDP caused the utilization of ground controlpoints to be bypassed. However, this was not properly indi-cated in the header or the annotation records (see appendix).Therefore, some high-density tapes generated during this timemay not have been corrected with control points, although thequality assessment number may be greater than O.
12
((
(C
l
)~ Of the 12 data sets evaluated. only two were generated
during the period in which the patch was used. These weredata sets 1 and ,2.whose registration was good. The highdensity P-tape numbers of data set 1 (L2MHP79204-08) anddata set 2 (L2MHP79210-05) were not on the list given in theappendix. Therefore. it was assumed that none of the 12 datasets of this study were affected by the patch in the MDP.
))
)
)
As reported in reference 2. a switch from light source Ato light source B in the Landsat 3 MSS sensor system caused ashift in the ground representation of the scan lines of approxi-mately 2.052 to 2.166 meters (36 to 38 pixels).
The light source on Landsat 3 has been sw~tched back andforth between source A and source B several times. Landsat 2was launched with light source A. switched to source B onAugust 24. 1979. and has remained there.
The light source switching is a potential problem forscenes being corrected using the ground control points of theMDP. However. the shift occurs in the scan-line (west-to-east)direction and none of the 12 scenes analyzed" in this studyshowed an error of such magnitude in the west-to-east (COLI-COL2)direction.
The scan line start problem (ref. 1) affected only one ofthe 12 scenes analyzed in this study. As stated earlier. thisproblem seems to have degraded the registration accuracy inthe north-to-south direction.
13
Conclusions
The conclusions are stated as answers to the three questionsasked in the introduction.
1. Is the registration accuracy of the MSS data adequatefor the scene-to-map registration task of the AgRISTARS DCLCproject?
No, because the scene-to-map registration task requiresan error of less than a pixel for each ESS segment within thescene. Therefore, a more precise local fitting will be requiredfor each segment. However, the registration of the P-formatLandsat MSS data that have been registered using the controlpoints in the MDP can be used as a starting point for an algo-rithm that will automatically register the ESS segment data tothe Landsat MSS data. If the algorithm uses a search windowof 10 Landsat columns by 10 Landsat rows, then out of the 12Landsat scenes analyzed in this study, 8 would fall within thewindow search area of this algorithm.
2. Does the registration accuracy depend heavily upon•the quality assessment numbert
The quality assessment number is not necessarily a goodindicator of the registration accuracy. For example, data sets4, 10, and 12 had higher assessment numbers than data set 3 butshowed poorer registration accuracy for this evaluation.Figure 14 shows a plot of the quality assessment number versus
14
((
((
1,I
,.
)
)
)
~)
)
)
the relative errors of registration for the 12 data sets. Thisplot shows that the relative error does not necessarily decreaseas the quality assessment number increases.
3. \~at problems are associated with using the regis-tration information on the CCT's and can these problems be"worked around" for the-scene-to-map task of the AgRISTARSDCLC project?
The potenti-al documented problems of bypassed controlpoints in the MDP, the scan line start problem, and the lightsource switch were enumerated earlier. When there are problemswith the registration defined by the tick marks, such as in thecases of data sets 4, 5, 6, and 12, the algorithm would computelow correlations for all attempts to matCh the segment data withthe Landsat data within the 10 x 10 window. This would be anindication to the analyst that there is a problem with the dataset and that the data set will have to be registered by anothertechnique.
15
TABLE 1. CATALOG OF DATA SETS 1-12
u
TICK MARKS. TOP BOTTOM RIGHT LEFTDATA SET DATE GEN. ASSESS .
NUMBER SCENE ID STATE PATH/ROW BY MDP NUMBER V COL V COL U ROW U ROW40 151 "'5 A71 "'7" Al t..7n 771
)TABLE 2. ASSESSMENT NUMBERS (0-5) VS. NUMBER OF CONTROL POINTS
)
)
)
)
ASSESSMENT NUMBER
o1
2
3
4
5
19
NUMBER OF CONTROL POINTSUSED BY MDP
o1- 89-16
17-2425-3233-40
TABLE 3"(
INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE,ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROWl, COLlYDETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 1 \
",
--------------------------------.--------.------------------------.-------Ll\ T I TUDE LCU'i(; I TunE RO~J1 ROl~2 ROW1-ROW2 COL1 COL2 COLI-COL2----------------.---------------.--------.--------------------------------
) TABLE 4. INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE,ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROWl, COLI)DETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 2
TABLE 5. (INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE( ..ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROW1, COL·DETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 3
) TABLE 6. INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE,ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROWl, COLI)DETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 4
INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE, (ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROWl. COL~DETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 5 \
TABLE 8. INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE,ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROW1, COL1)DETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 6
INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE, (ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROWl, COLl)DETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 7
) TABLE 10. INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE,ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROW1, COLI)DETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 8
INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE,(ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROW1, COLIDETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 9
TABLE 12. INDEPENPENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE,ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROW1, COL1)DETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 10
INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE,_ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROWl, COL(DETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 11 .
--------------------------------------~--~---------------------------------L 1\ T r T1.J [: F LU (jITlIL·E RG~~l R(lW2 R()W1-ROW2 COLl COL2 COLI-COL2------------------------------------------.--------------------------------1 41.r~b3l) -CJ&+.~(!OO 92.4 90.0 2.&+ 466.8 &+69.0 -2.2(.~
) TABLE 14. INDEPENDENTLY CHOSEN CONTROL POINTS (LATITUDE, LONGITUDE,ROW2, COL2) COMPARED WITH LANDSAT COORDINATES (ROWl, COLI)DETERMINED FROM P-FORMAT REGISTRATION FOR DATA SET 12
Figure 14. Quality assessment number versus relative error ofregistration value Y (Y = y(RBIAS)2+(RSD)2+(CBIAS)2+(CSD)2where RBIAS, RSD, CBIAS, and CSD are values fromtable 15)
46
I/
)
..I ~
). )
)
)
)
APPENDIX
47
". ,-
)
)J:'S .,~ (c;.
•January 16, 1980
TO: 415/0PS/A Project Manager•
,.)
y,.
)
FR~1: 563/HeFd, Ima~e Processing BranchSUBJECT: Geouctric Correction Errors t;oted by the EROS Data Center (EDC)
As noted by the EOC there is a discrepancy between the number of G~oundControl ?oiuts (GCPs) indicated tv be involved in the geometric correctionalgorithms and the actual count.During the period from June 14.1979~ to August ·1, 1979,. a 'patch was usedon the M~ster Data Processor (MOP) to resolve an infre~~ent timing probiem.When this patch is used, all GCPs are bypassed, howe~~r, this is not properlyindicated in the header or annotaticn records.It shou1d be assumed that all data processed with ~,is patch has teen processedusing systerratic cor •.ectior-s (accurate to 2.7 lc:nj net Wit., GC?s; t.lis '11;:1.eliminata t.ie possibility of error •..
A listing of all P"-tapes produced usinS the special ~.Cl'patch is attached.All other p-tapes are properly annotated •.
~:11e~Attachmentcc: Or. Freden/902
Mr. Holmes/563Mr. Tinsley/563Mr. I':iomson/EDC
49I
J
,••
J
CSTA•
COMPUTER SCIENCES - TZCHNICOLOR ASSOCIATES
INTEROFFICE CORRESPONDENCE
)
)
•to M. Selig from N•. Vred~uFg/D: Hq1?
. IIV · p. tf .subject Landsat P-Tapes with CC? Errors in Header Records
date JanUAry 15 t . 1980
A patch 1Jsed to correct an "error return from RSS" on the ¥.i)? which caused GC?error c~~nt in the HUT-PM header records was used in production from June 14, 1979to Auguse 1, 1979.The following dD~~?~ tapes were ~.de wit~
1. "MSS Line-Start Problem," Landsat Data Users Notes,Issue No. 12, U.S. Department of the Interior, U.S.Geological Survey, EROS Data Center, Sioux Falls, SD,May 1980, pp. 5-6.
2. "Landsat 3 Light-Source Problem," Landsat Data Users Notes,Issue No. 14, U.S. Department of the Interior, U.S.Geological Survey, EROS Data Center, Sioux Falls, SD,September 1980, p. 7.
3. Landsat Data Users Handbook, Revised Edition, U.S. Departmentof the Interior, U.S. Geological Survey, Arlington, VA,1979, pp. 7-19.
4. "EDIPS Image Accuracy Tests," Landsat Data Users Notes,Issue No.9, U.S. Department of the Interior, U.S.Geological Survey, EROS Data Center, Sioux Falls, SD,November 1979, p. 4.
5. Colwell, J., G. Davis, and F. Thomson: Detection and Measure-ment of Changes in the Production and Quality of RenewableResources, Environmental Research Institute of Michigan,Report No. 145300-4-F, Ann Arbor, MI, October 1980, pp. 73-84.
6. "The Hotine Oblique Mercator Projection," Landsat Data UsersNotes, Issue No. 11, U.S. Department of the Interior, U.S.Geological Survey, EROS Data Center, Sioux Falls, SD,March 1980, pp. 4-5.