idstorical digital orthophotography savannah river ...

IDSTORICAL DIGITAL ORTHOPHOTOGRAPHY SAVANNAH RIVER NATIONAL ENVIRONMENTAL RESEARCH PARK

0 .

.

This article was written and prepared by U.S. Government employees on official time, and it is therefore in the public domain and not subject to copyright.

Rose M. Sumerall F. Thomas Lloyd Colin N. Brooks US Department of Agriculture Forest Service Southeastern Forest Experiment Station Department of Forestry Clemson University Clemson, South Carolina 29634-1003

ABSTRACT In this paper, we report success in using modern softcopy photogrammetry to orthorectify historical aerial photographs. IMAGINE OrthoMAX, a new photogrammetric software developed by Vision lntemational-Autometric, Inc. and marketed by ERDAS, Inc. was instrumental in this success. The historical digital orthophotos are being developed for the Savannah River Site (SRS) in South Carolina. Aerial photographs taken in May of 1951 were scanned to a ground resolution of 0.5 meters. Coordinates for ground control points were derived using Global Positioning Systems (GPS) to an accuracy of 0.25 meters. Triangulation was successful for a block of 52 photos and our test points fell within 3.5 meters of their expected GPS positions. The triangulation solution was used to generate digital orthoimages and then overlaid with other Geographic Information System (GIS) data for the SRS. Although we encountered problems in dealing with hjstorical photography, we should be able to meet our objective and produce digital quarter-quad mosaics matching national map accuracy standards at 1: 12,000 scale.

INTRODUCTION The database is under development at the Southern Research Station in Clemson, South Carolina. Our primary workstation is a SUN Spare lO operating off a SUN 330 Server with approximately six gigabytes of disk space for images. We have completed scanning and collecting ground control d~~a

127

for the entire site, achieved a triangulation solution for a block of 52 images, orthorectified a few sample .images, and experimented with image mosaic king. We hope to complete the project by the end of May 1995. Our final product will be a set of 36.digital quarter-quad mosaics with an image resolution of 1.0 meter. The files will be approximately 55MB, including overedge and reference tic marks and header information that conforms with national spatial data transfer standards. Our goal is to match national map accuracy standards at the 1:12,000 scale. We will also supply originalunrectified images for users with stereo-viewing and 3-D mensuration capabilities, and a set of the orthorectified images with original scanning resolution -and brightness values. This article assumes the reader is acquainted with digital image processing, orthophotography, and principles of differential rectification. We provide details on methods used, report preliminary results, and describe problems related to the historical nature of the photography. The discussion follows general procedural steps: data acquisition, scanning, collection of ground control, geodetic conversions, digital elevation input, and orthorectification.

STUDY AREA The Savannah River Site (SRS), a nuclear weapons facility and National Environmental Research Park, covers 300 square-miles in South Carolina. In terms of Ecoregions. 74% is Coastal Plains and Flatwoods, and 26% is Southern Appalachian Piedmont (Bailey et a!. 1994). Photographic coverage includes parts of Aiken, Barnwell and Allendale counties in South Carolina and Burke county in Georgia. When the SRS was purchased by the Atomic Energy Commission in 1950, the area was heavily farmed. The go.vemment relocated area residents and removed about 6,000 structures , including two small towns (Brooks and Crass 1991). Today, pine plantations and hardwood forests cover approximately 90% of the landscape (Jones et al. 1981 ). The multidisciplinary groups that conduct research on the property are interested in the history of the site and the changes that have occurred, both nalural and man-induced.

PHOTOGRAPHY The black and white photography used in this project was produced in 1951 by Park Aerial Surveys under contract with the USDA Agricultural Stabilization and Conservation Service (ASCS) at a scale of 1:20,000. To cover the entire SRS including stereo overlap, we purchased 251 film diapositives with matching paper prints. The diapositives were printed on 7rnm Kodak 4421 film with a density range of 0.30 -1.20 D. After an unsuccessful attempt to locate winter photography(negatives) for 1951 we selected photographs taken in May. The photographs were taken between May 6 and May 24~ consequently. matchi ng tone and contrast between adjacent images may be difficult because atmospheric conditions and sun angle will vary. Because we located a set of old photos with the modern boundary superimposed, we were able to transfer this boundary to the 1951 photo index to determine which

128

photos to purchase. Obtaining the parameters needed for interior orientation in softcopy photogrammetry was difficult. The oldest camera calibration report we were able to locate for the camera was made in 1963. Unfortunately, older camera calibration reports do not include coordinates for the fiducials and principal point. A photogrammetrist measured one of our diapositives using Erio Technologies Monocomparator Measurement Program with a system accuracy of about one micron. The collimation marker distances measured by the photogrammetrist were shorter than the distances listed on the 1963 camera calibration report. We used these measurements and the focal length from the 1963 report indefining our camera parameters.

SCANNING The film grain size is coarser in older photography, therefore, a scanning density of25 microns, which equals 1,000 dots per inch (dpi), was considered sufficient to capture most of the information content of the film. Scanning 1:20,000 scale photographs at 1,000 dpi yields a pixel size of 0.5 meters and image files of approximately 81 megabytes. Under contract, Intera Corporation of Ottawa, Canada, used a drum-based Optronics C 4100 SP ''Pixel getter" scanner to scan the film diapositives. The scanning system had a dynamic range that accommodated the density of our film diapositives and was operated at an eight bit radiometric resolution, resulting in 256 gray scale values. The contractor scanned a p·recision ruled grid to evaluate the spatial accuracy of the scanner, a density-step wedge to evaluate the spectral accuracy, and a blank screen to illustrate the noise introduced by the scanning process. Visual inspection revealed numerous occurrences of bright scan lines and a brightness gradation partly due to fall-off on the original photographs. The block of 251 photos consists of 13 flight lines flown in a north-south pattern. The scan line direction paralleled the flight direction and all images had the same relative orientation. Because some flight lines were flown toward the south and others toward the north, we had to be alert to the ordering of fiducials in the interior orientation operation. Because the scan line direction is parallel to the elevation profiles in our imported Digital Elevation Models (DEM) (which also run northsouth) (USDI1987), we will probably avoid the problem with artifacts described in the "Standards for Digital Orthophotos" (USDI 1992).

GROUND CONTROL A Global Positioning System (GPS) was used to collect coordinate data on 86 ground control points (GCP's) and seven geodetic control stations during the winter of l993 to 1994. We will reserve ten percent of these points to test the positional accuracy~ of the final product and plan to include some of our strongest points in the test set are photo-identifiable points that can be used to georeference images. The greatest challenge in georeferencing historic images is finding temporally stable points. Control points, difficult to find within the SRS boundary, were easier to find in the surrounding communities where little change had taken place in forty years. Finding points in swampy areas and around highly developed nuclear complexes was especially difficult. ·

129

We selected intersections involving roads, railroads, hedgerows and bridges as our GCP's. We first located candidate points by visually comparing 1951 and 1992 photographs. Next, working in the field, we looked for any evidence that indicated the area had changed and if evidence was found we discarded the point. Generally, we could not establish conclusively that an intersection was unchanged. We did not consult roadway plans or old documents. Geodetic control stations are monumented points, installed as part of the National Geodetic Control Network. The Latitude/Longitude coordinates for these points have been precisely measured and serve to ground control points to a standardized reference system. Using National Geodetic Survey county data sheets, we were able to locate three stations of first order horizontal and vertical accuracy and three stations of second order horizontal accuracy. We used three additional surveyed monuments on the SRS property boundary, only one of which had published coordinates. Survey-grade GPS technology enabled accuracies near 0.25 meters for our GCP's, which is onehalf the ground resolution of an image pixel. The GPS data were collected in a static survey mode using four Trimble series-4000 GPS survey receivers. TRIMVEC-Plus software was used to process data. When the GPS data are processed, vector relationships are computed between the survey stations. Figure 1 is a computer printout of our GPS network for the northeast quadrant of the SRS. For the block of 52 photographs comprising the northeast quadrant, we have 26 GCP' s, 4 geodetic control stations, and two test points. To test the integrity of the data, we ran a loop closure process in the software. In this process, a traverse is run between two known geodetic control stations, similar to a traditional boundary survey. The error of closure on this traverse reflects the quality of the data. Several network vectors are highlighted in Figure 1 to illustrate a traverse. This traverse encompasses a total of nine points, covers a distance of about 20 miles, and the error of closure is two millimeters in the horizontal dimension and one millimeter in the vertical dimension. After the data is reviewed with loop closures, a network adjustment should be performed (SC Geologic Survey 1994). We did not adjust our control to the local network. We plan to study the effects of this omission by comparing the orthophotos with orthophotos from an adjacent county, which are adjusted to the local network.

GEODETIC CONVERSION OF GCP COORDINATES The GPS coordinate data were collected based on the Geodetic Reference System normal ellipsoid (GRS 80), the system best suited for precise GPS measurements. GRS 80 is defined by the horizontal reference North American Datum (NAD) 83(86) and the vertical reference North AmericanVertical Datum (NAVD) 88. We then converted these point coordinates to our output reference system using the government conversion programs Corpscon (v3.0l), Vertcon (vl.O) and Geoid93 (v2.0). We believe these programs are more reliable than similar conversion processes within the GPS and GIS software packages. Because these ellipsoids are modeled surfaces, a small amount of uncertainty is intro-

130

duced in the data conversion processes. For Corpscon, which uses the Nadcon program, the hori~ zontal uncertainty could be as high as 15 em within the contemporaneous U.S. The GIS data of the Savannah River Forest Station is in the Universal Transverse Mercator projection and referenced to the Clarke 1866 ellipsoid. This system is based on the horizontal datum NAD 27 and the vertical datum National Geodetic Vertical Datum (NGVD) 29. IMAGINE OrthoMAX was set to this system and GCP coordinates were converted to it. Because the project used imported Digital Elevation Models (DEM's), the elevation values of the GCP's were input in terms of the orthometric heights used in the DEM' s. We also made sure that the imported DEM's matched the reference system of our control points.

DIGITAL ELEVATION MODEL The DEM's imported into IMAGINE OrthoMA.X and used in ortho~rectification were created and edited in 1991 by Geometronics using the Line-Trace Plus method of data collection from DLG hypsography. These 7.5 minute quadrangle files are of level2 accuracy with a vertical Root Mean Square Error (RMSE) between l and 7 meters. We experienced problems importing these DEM's into IMAGINE OrthoMAX. We solved this problem by passing the DEM's through ARC/INFO software using the DEMLATriCE and GRID IMAGE commands to generate ERDAS ".ian" files . These files were converted to ERDAS IMAGINE format, the projection information was edited, · and the DEM's were imported into IMAGINE OrthoMAX. An interesting problem arises because the original land surface (as visible in the 1951 photography) has been flooded in several areas to create large man-made lakes as deep as seventeen meters. These lakes are represented as flat surfaces at water level in the DEM's (Figure 2). We may edit the DEM's to reflect the subsurface profile before using them for orthorectification. This suggests an interesting application, that historical photography could be used to develop bathymetric data for other man-made lakes.

ORTHORECTIFICATION The input of camera parameters, GCP coordinates, images, and image points was straightforward although it was marred by occasional operator error and bugs in the beta version of the software. Achieving an acceptable triangulation solution on a large block of photos was the most difficult task in this operation. Because we were using historical data, some of our ground control points would probably be unusable. Because we were inexperienced in triangulation, we found identifying bad control points to be difficult. The least squares adjustment distributes the error and so the highest residual is not necessarily the worst point. Also, when a point is dropped, some residuals are improved while others are not, and the net affect is difficult to assess. In one instance, we reached an acceptable solution only to realize we had discarded several points along one side of our block. Therefore, we caution that a good statistical solution may' not be geographically sound. From the beginning, we were able to reach a triangulation solution within 4 iterations, indicating

131

that there were no serious problems with our data. We sought to reduce our image residuals and settled on a solution which disregarded 7 out of 26 GCP's. Based on that triangulation solution our two test points fell within 3.5 meters of their expected GPS coordinates. Once a triangulation solution was formulated and the DEM was imported, the onhorectification operation was a quick process. We were ·able to view the orthoimage and overlay GIS vector coverages (Figures 3 and 4 ). Experiments in mosaic king demonstrate good continuity between the images of a triangulation block.

APPLICATIONS We plan to explore the usefulness of the image base for tasks such as rectifying mullet-date photography and historic vector coverage's, improving existing GIS coverage's, deriving historic data layer themes with on-screen digitizing, and assisting in field location of archaeological sites. We will also experiment with image enhancement to facilitate detection of specific features and demonstrate hardcopy outputs and stereo applications. Several biological projects that ·will use historical images are planned. These include vegetation studies in old-field sites, landcover fragmentation and implications for historical wildlife habitat, and correlations between old field sites and certain chemical residuals.

CONCLUSION The historic orthoimage layer should be a valuable temporal addition to the Geographic Information System at the SRS because high positional acc uracy is achievable in softcopy photogrammetry. We are satisfied with the results we have obtained using IMAGINE OrthoMAX. While working with historic photography, we have encountered some interesting problems but we feel this new technol· ogy can be readily applied to older photography.

RECOM1\1ENDATIONS FOR ERDAS The problems we experienced in earlier versions of llvlAGINE OrthoMAX seem to be resolved in Version 8. 1. However. revising sections on the interpretation of triangulation results in the OrthoMAX User's Manual would significantly improve user operation. We would like to see a comprehensive and consistent explanation of tenns and units for the statistical summary and errC?r propagation reports.

ACKNOWLEDGl\1ENTS This research was funded by the U. S. Department of Energy in cooperation with the Southern .Research Station and the Savannah River Forest Station. We extend a special thanks to William Clerke and Michael Lange of the USDA Forest Service, John Jensen of the University of South Carolina. and the South Carolina Geodetic Survey for their technical advice. GPS-survey equipment, software. and training were provided by the USDA Forest Service Southern Region Lands and Minerals Department in Atlanta, GA. 1 extend personal note of appreciation to my co-workers at the Clemson Lab who assisted in various phases of the work and to employees of ERDAS, Inc. and Autometric Inc. for resolution of problems. 132

~1o•\ln u~uhr

TTL.a.J.• Loep

J t a rt 'l"''• v•n• a c Sudan: MtO stU"t.Lftq Coor-o.t : n'U'U,,,,.., . . .

11 .. 700

...... u n• a ru• ~taa11c taot :ns:.on

rToe Jtntant MIO 'tO 5t.a~on: cn.'•"T O&sunc • Trav.Uri (aJ: tU•.Sl2 C\UT•nc. c:...trd• r n'l:t':tt. l4ltt• • al. •;at'lt.Ot5J)• "

lt.l'U

1Ud1ft 4 J

nu w•u • au t:un .orr rra• s CA t.a.•", CVI.VT

To l tAid one

C ht:aftU Tl'av •Ut4 ( •• ~

C\in'anc:

I099 olU l:l'U'l.t .u•t•• Jf

eo.nr• 1

. . . .111111 ) r u • w...u uu:.n1.oP1" tna Jlt:&Uanl f lJf tlht:anca 1't•"•U..ci t •l ~ ~.nc

c:.ooru

"to St.at.!.ofll: UU.U. J.l1

n'za•ot.an~1·

1

'7...111

1t05t

tl:a-1'lt.onu•

Ill

•••• u,.. rue ua•:

ru

u':t7•ao.)tcn• w

'"·U"

111

4

lSUJIU. .oP'f ,,... U~t:swn: .Oil

To IY'Cti.u:

Ou\Af\Ce "f"nve U .... {II II Ulll..liO r:u&'T• "cteof'd_. • :u·u·Lt.a't6fl&.•'lf

•••• u ... . tll• ..... :

~

OL.a't&f'lc• Tt"avdJ M l e i: 111-t)l.aa l a.n·•f'l ~ C••CU t ll'U•H.t.otGJ. • II

I~UOtH

&a~•• u

c:o.tcd•

1

To

snuon:

c:J ....

.te::.5

l\.~l7')4 .1 414 U• 'I

1'G

ICG'

Dht&nce ha•• lh4 C111: C'UtT• tlt.:

U. \ U

U'!ll'!il . l'll n • u • ll.