Relative Accuracy of Rectifications Using Coordinates Determined from Maps and the Global Positioning System

Similar documents
CanImage. (Landsat 7 Orthoimages at the 1: Scale) Standards and Specifications Edition 1.0

TEMPORAL ANALYSIS OF MULTI EPOCH LANDSAT GEOCOVER IMAGES IN ZONGULDAK TESTFIELD

Image Registration Issues for Change Detection Studies

GEO/EVS 425/525 Unit 9 Aerial Photograph and Satellite Image Rectification

Application of GIS to Fast Track Planning and Monitoring of Development Agenda

Chapter 1 Overview of imaging GIS

Digital Image Processing

Remote Sensing. The following figure is grey scale display of SPOT Panchromatic without stretching.

366 Glossary. Popular method for scale drawings in a computer similar to GIS but without the necessity for spatial referencing CEP

Spatial Analyst is an extension in ArcGIS specially designed for working with raster data.

Remote sensing image correction

Application of Satellite Imagery for Rerouting Electric Power Transmission Lines

Line and polygon features can be created via on-screen digitizing.

Abstract Quickbird Vs Aerial photos in identifying man-made objects

Satellite image classification

DIFFERENTIAL APPROACH FOR MAP REVISION FROM NEW MULTI-RESOLUTION SATELLITE IMAGERY AND EXISTING TOPOGRAPHIC DATA

Determining patch perimeters in raster image processing and geographic information systems

[GEOMETRIC CORRECTION, ORTHORECTIFICATION AND MOSAICKING]

APCAS/10/21 April 2010 ASIA AND PACIFIC COMMISSION ON AGRICULTURAL STATISTICS TWENTY-THIRD SESSION. Siem Reap, Cambodia, April 2010

29 th Annual Louisiana RS/GIS Workshop April 23, 2013 Cajundome Convention Center Lafayette, Louisiana

GGS 412 Air Photography Interpretation

Geometric Quality Assessment of CBERS-2. Julio d Alge Ricardo Cartaxo Guaraci Erthal

NR402 GIS Applications in Natural Resources

Appendix 2: Worked example using GPS

THE HYDROLOGIC IMPACT OF THE VERMONT INTERSTATE HIGHWAY SYSTEM. A Thesis Progress Report. Analeisha Vang. The Faculty of the Geology Department

INTEGRATED DEM AND PAN-SHARPENED SPOT-4 IMAGE IN URBAN STUDIES

White Paper. Medium Resolution Images and Clutter From Landsat 7 Sources. Pierre Missud

Active and Passive Microwave Remote Sensing

HIGH RESOLUTION STEREO SATELLITE ELEVATION MAPPING ACCURACY ASSESSMENT INTRODUCTION

ANNEX IV ERDAS IMAGINE OPERATION MANUAL

USE OF PC.. ERDAS IN SATELLITE MAPPING/GIS EDUCATION

An Introduction to Geomatics. Prepared by: Dr. Maher A. El-Hallaq خاص بطلبة مساق مقدمة في علم. Associate Professor of Surveying IUG

Planet Labs Inc 2017 Page 2

Sommersemester Prof. Dr. Christoph Kleinn Institut für Waldinventur und Waldwachstum Arbeitsbereich Fernerkundung und Waldinventur.

A map says to you, 'Read me carefully, follow me closely, doubt me not.' It says, 'I am the Earth in the palm of your hand. Without me, you are alone

Raster is faster but vector is corrector

Technical Evaluation of Khartoum State Mapping Project

MSB Imagery Program FAQ v1

Lesson 4: Photogrammetry

An Introduction to Remote Sensing & GIS. Introduction

Geo/SAT 2 INTRODUCTION TO REMOTE SENSING

Differential Global Positioning ~ System Techniques For Surveying/ Mapping within Forested Wetlands

Aerial photography: Principles. Frame capture sensors: Analog film and digital cameras

GEOMETRIC RECTIFICATION OF EUROPEAN HISTORICAL ARCHIVES OF LANDSAT 1-3 MSS IMAGERY

Supervised Land Cover Classification An introduction to digital image classification using the Multispectral Image Data Analysis System (MultiSpec )

Suveying Lectures for CE 498

Application of Satellite Image Processing to Earth Resistivity Map

GST 105: Introduction to Remote Sensing Lab 4: Image Rectification

Keywords: Agriculture, Olive Trees, Supervised Classification, Landsat TM, QuickBird, Remote Sensing.

LANDSAT-SPOT DIGITAL IMAGES INTEGRATION USING GEOSTATISTICAL COSIMULATION TECHNIQUES

INTERNATIONAL JOURNAL OF ENVIRONMENTAL SCIENCES Volume 6, No 5, Copyright by the authors - Licensee IPA- Under Creative Commons license 3.

EVALUATION OF SATELLITE IMAGE FUSION USING WAVELET TRANSFORM

Basics of Photogrammetry Note#6

Land Cover Change Analysis An Introduction to Land Cover Change Analysis using the Multispectral Image Data Analysis System (MultiSpec )

Orthoimagery Standards. Chatham County, Georgia. Jason Lee and Noel Perkins

A COMPARISON OF COVERTYPE DELINEATIONS FROM AUTOMATED IMAGE SEGMENTATION OF INDEPENDENT AND MERGED IRS AND LANDSAT TM IMAGE-BASED DATA SETS

Active and Passive Microwave Remote Sensing

Geo/SAT 2 MAP MAKING IN THE INFORMATION AGE

NRS 415 Remote Sensing of Environment

INFORMATION CONTENT ANALYSIS FROM VERY HIGH RESOLUTION OPTICAL SPACE IMAGERY FOR UPDATING SPATIAL DATABASE

Landsat 8. Snabba leveranser av bilder till användarna. Lars-Åke Edgardh. tisdag 9 april 13

CHAPTER 7: Multispectral Remote Sensing

ENVI.2030L Topographic Maps and Profiles

Remote Sensing Instruction Laboratory

Land Cover Type Changes Related to. Oil and Natural Gas Drill Sites in a. Selected Area of Williams County, ND

Lab 3 -- Mosaic. Introduction

PHOTOGRAMMETRIC RESECTION DIFFERENCES BASED ON LABORATORY vs. OPERATIONAL CALIBRATIONS

Topographic mapping from space K. Jacobsen*, G. Büyüksalih**

Visualizing a Pixel. Simulate a Sensor s View from Space. In this activity, you will:

COPYRIGHTED MATERIAL. Contours and Form DEFINITION

Land Navigation / Map Reading

CHAPTER 3 MARGINAL INFORMATION AND SYMBOLS

Saturation And Value Modulation (SVM): A New Method For Integrating Color And Grayscale Imagery

RGB colours: Display onscreen = RGB

ASTER GDEM Readme File ASTER GDEM Version 1

Introduction to Remote Sensing

Blacksburg, VA July 24 th 30 th, 2010 Remote Sensing Page 1. A condensed overview. For our purposes

USE OF SMALL FORMAT DIGITAL AERIAL IMAGES FOR CLASSIFICATION OF SATELLITE IMAGES

Addressing Issues with GPS Data Accuracy and Position Update Rate for Field Traffic Studies

GEO/EVS 425/525 Unit 3 Composite Images and The ERDAS Imagine Map Composer

Preparing Remote Sensing Data for Natural Resources Mapping (image enhancement, rectifications )

Lab #10 Digital Orthophoto Creation (Using Leica Photogrammetry Suite)

Using Indexes to Detect the Environmental Changes of Hor Al-Dalmaj and Surrounding Areas in Central Sector of Mesopotamia Plain

7. RECTIFICATION (GEOMETRIC CORRECTION) OF IMAGES AND RESAMPLING

Project Planning and Cost Estimating

Image enhancement. Introduction to Photogrammetry and Remote Sensing (SGHG 1473) Dr. Muhammad Zulkarnain Abdul Rahman

Image Fusion. Pan Sharpening. Pan Sharpening. Pan Sharpening: ENVI. Multi-spectral and PAN. Magsud Mehdiyev Geoinfomatics Center, AIT

Potential of ASTER and LANDSAT Images for Mapping Features in Western Desert

v Introduction Images Import images in a variety of formats and register the images to a coordinate projection WMS Tutorials Time minutes

Lesson 3: Working with Landsat Data

High Precision Positioning Unit 1: Accuracy, Precision, and Error Student Exercise

Lecture 2. Electromagnetic radiation principles. Units, image resolutions.

GEO 428: DEMs from GPS, Imagery, & Lidar Tuesday, September 11

Land Cover Change Analysis An Introduction to Land Cover Change Analysis using the Multispectral Image Data Analysis System (MultiSpec )

AT-SATELLITE REFLECTANCE: A FIRST ORDER NORMALIZATION OF LANDSAT 7 ETM+ IMAGES

Lecture 8: GIS Data Error & GPS Technology

Enhancement of Multispectral Images and Vegetation Indices

Hyperspectral Imagery: A New Tool For Wetlands Monitoring/Analyses

Automated GIS data collection and update

8. EDITING AND VIEWING COORDINATES, CREATING SCATTERGRAMS AND PRINCIPAL COMPONENTS ANALYSIS

Transcription:

Practical Paper Relative Accuracy of Rectifications Using Coordinates Determined from Maps and the Global Positioning System Allen E. Cook and John E. Pinder Ill Abstract Because GPS measurements are not subject to digitizing Global positioning systems (GPS) that use code phase receiv- and map production errors, they have the potentid for proers have the potential for providing more accurate coordi- viding more accurate coordinates than 7.5-minute maps. This nates for ground control points (GCPS) than do 7.5-minute advantage of greater accuracy may$ however$ be the quadrangle maps. To evaluate the effect of the greater occu- Feater expense in obtaining GPS receivers and visitmcy of GPS in determining coordinates for satellite images, ing the GCP locations (Clavet et al., 1993). ~lthou~h the use GCP coordinates were measured on maps and by GPS tech- of GPS coordinates has been shown to be comparable or prefniques and were then used to rectify Landsat the ma ti^ Map- erable to map coordinates for orthoimages per, SPOT multispectral, and SPOT panchromatic images from photographs lgg2) and panchromatic data the same area. There were major differences bewen rectifi- (Clavet et d.9 lg93)7 the hproved accuracy GPS may not cations using map and GPS coordinates. For Thematic Map- be worth the additional expense where the pixel sizes are per data, 24 percent of the values assigned to pixels in the large relative to the nominal accuracies of maps and GPS. rectification performed using map coordinates differedfr~m The Purpose of this study is to wmluate the effect of the the values assigned to the pixels in the rectification greater accuracy of GPS coordinates for satellite images comperformed using the GPS coordinates. For SPOT multispectral posed 30-m9 20-m* and pixels. and SPOT panchromatic images, the percentage of pixels with different assignments were greater than 40 percent. The Methods improvements in accuracy when using the GPS data were The effects of using map and GPS coordinates to rectify irnsubstantial for even the relatively large Thematic Mapper ages were compared for (I) a 4096 by 4096 subsection from pixels and warrant the use of GPS where position accuracy is the center of a Landsat Thematic Mapper (TM) scene obessential. tained on 14 April 1988, (2) a SPOT multispectxal scene collected on 17 April 1988 with a viewing angle of + 23.0". and Introduction (3) a SPOT panchromatic scene obtained on 22 October 1987 The coordinates of ground control points (GCPS) used to rec- with a viewing angle of - 16.5'. These scenes have centers tify satellite imagery are usually determined from 7.5-minute between 33" 14' N and 33" 17' N and between 81' 36' W and quadrangle maps published by the U. S. Geological Survey 81" 40' W. These centers are on the U. S. Department of (VSGS). The nominal horizontal accuracy of these maps is Energy's Savannah River Site, a nuclear production facility 12.2 m (i.e., U. S. National Map Accuracy Standards where in Aiken and Barnwell Counties, Carolina. The images 90 percent of the position errors for tested points are less also contain portions of Carolina and Georgia that are than or equal to 12.2 m; Thompson, 1988). Global position- predominantly rural with a mixture of forest and agricultural ing systems (GPs) that use code phase receivers can rapidly land use. The area extends from gently rolling Piedmont termeasure coordinate locations to nominal accuracies of 7 to rain in the northwest to relatively flat Coastal Plain terrain in 12 m (i.e., 90 percent of errors are less than 7 to 12 m; Trim- the southeast. The maximum elevation is < 225 m, and the ble Navigation, 1992). To achieve nominal accuracies less minimum elevation is > 40 m. than 12 m, the GPS position data collected by the receiver at Features in the images that were also depicted on USGS the GCP must be postprocessed and dlerentially corrected 7.5-minute quadrangle maps were chosen as GCPs. A total of using data collected by a second GPS receiver at a base sta- 88 GCPS were established, but differences in image sizes and tion of known location (Trimble Navigation, 1992; Gilbert, resolutions prevented the use of all GCPS in every image. 1993). Although 7.5-minute quadrangle maps and single fre- There were 64 CXPS used in the Landsat TM image. For the quency GPS receivers have similar nominal accuracies, the SPOT panchromatic image, 22 of the GCPS used for the Thecoordinates obtained from maps may possess additional er- matic Mapper image were combined with 24 additional GCPS rors introduced by digitizing and map production. Map pro- for a total of 46. For the SPOT multispectral image, 40 of the duction errors include misplacement of untested features or 46 GCPS for the SPOT panchromatic image were used. Most modifications of feature positions to simplify map printing GCPS were road intersections where the roads joined at an- (Thompson, 1988). gles of 60" or more; however, the rural landscape, the lack of Photogrammetric Engineering & Remote Sensing, Remote Sensing and Geomatics Unit, Savannah River Ecol- Vol. 62, No. 1, January 1996, pp. 73-77. ogy Laboratory, Drawer E, Aiken, SC 29801. 0099-1112/96/6201-73$3.00/0 A.E. Cook is currently with Johnson Control World Services, O 1996 American Society for Photogrammetry P.O. Box 260129, Lakewood, CO 80226-0129. and Remote Sensing PE&RS January 1996 73

recent map revisions, and efforts to ensure that GCPs were evenly dispersed throughout the images required the use of seven oblique road intersections or alternative features. The pixel coordinates of GCPS were determined to the nearest 0.25 pixel using a 4 by 4 cubic convolution procedure that enhanced pixel resolution (ERDAS, 1990b). To estimate the measuring error for pixel coordinates, each co-author independently determined the subpixel coordinates for each GCP. The Universal Transverse Mercator (UTM) coordinates based on the 1927 North American Datum were measured for GCPS using 1:24,000-scale 7.5-minute quadrangle maps and a large-format digitizing tablet. Almost all of these maps complied with the National Map Accuracy Standards. Standard procedures were employed to ensure the accuracy of digitized coordinates (ERDAS, 1990a). Where maps had not been recently revised and did not have UTM coordinates, the Georgia State Plane coordinates were measured and converted to UTM. The GPS coordinates of each G s were determined using Trimble Pathfnder Professional Systems. Measures of X, Y, and Z coordinates were recorded at 5-second intervals for 2 3 minutes of four-satellite availability at each GCP. Because of vehicle traffic, the coordinates for 54 intersections were calculated by measuring the angle and distance from an antenna position beside the road to the center of the intersection. The GPS position data were corrected by post ~rocessina usina Trimble Pfinder software version 2.20 i~rirnble 61avigation, 1991) and base station data collected at 33" 20' 35" N and 81" 44' 3" W. Coordinates from GPS measures were expressed as 1927 North American Datum UTM. To compare the accuracies of map and GPS methods, coordinates were measured using map and GPS techniques for 20 points of known location. These points were road intersections on the Savannah River Site whose coordinates were available from 1:1200-scale engineering drawings. First-order afhe transformation matrices (Jensen, 1986; ERDAS, 199Ob) were computed from map and GPS coordinates and used to rectify scenes. Those GCPs with large apparent errors were deleted (ERDAS, 1990b) to reduce the root-mean-square error (RMSE) to 0.33-, 0.5-, and 1.0-pixel units for Thematic Mapper, SPOT multispectral, and SPOT panchromatic data, respectively. These RMSE values were selected to approximate 10-m accuracy in the images. To compare the effects of map- and GPS-derived coordinates on rectifications, files (hereafter referred to as grid files) were formed of repeated 8 by 8 grids. Each grid had data values ranging from 11 in the upper left corner to 88 in the lower right corner. The data values were arranged so that the units digit defined the row position and the tens digit defined the column position. Grid files were constructed with the same dimensions as the satellite images, and were rectified using transformation matrices computed from map and GPS coordinates. Nearest-neighbor resampling was used in the rectification procedure. The rectified grid files based on map and GPS coordinates were compared on a pixel-bypixel basis to measure the frequencies of differences in pixel assignments (ERDAS, 199Oc). Because the data values defined row and column positions, the comparisons clearly defined the frequency, direction, and distance of differences in pixel assignments between map based and GPS based rectifications. Results and Discussion Cornparloon of the Accuracy of Map and GPS Coordinates The positions obtained by GPS were more accurate than those obtained from maps for the 20 intersections with known locations on the Savannah River Site (Figure 1). The errors in map coordinates ranged from 2.2 m to 50.1 m, whereas the errors in GPS coordinates ranged from 0.1 to 9.0 m. The mean distance between map and known positions, 19.4 m, was significantly greater than the mean distance of 4.1 m between the GPS coordinates and known positions (t-test of paired observations; t = 5.65; df = 19: P < 0.001; Steel and Torrie, 1960). Before post processing, the uncorrected GPS coordinates had a mean position error of 23.0 m and were not more accurate than map coordinates. The largest error for the map points was due, in part, to a simplified representation of a complex intersection. All of the position errors for the GPS were < 12 m, whereas only six of the map coordinates had position errors < 12 m. The accuracy of the GPS receivers was similar to that observed by Clavet et al. (1993) for comparable GPS receivers and was consistent with the expected accuracy for differentially corrected code phase receivers, which is usually stated as 50 percent of errors being I 5 m (Gilbert, 1993). Because of the errors in map coordinates (Figure I), there were often relatively large discrepancies between the coordinates determined by GPS techniques and by map techniques (Figure 2). The mean distance between map and GPS coordinates was 12.5 m and was comparable to that observed by Perry (1992). Most of the discrepancies, 84 of 88, were less than the 30-m dimensions of the Landsat Thematic Mapper pixels, but only 39 of the distances were less than the 10-m pixel size for SPOT panchromatic data. The Accuracy of Pixel Coordinates The errors involved in measuring pixel coordinates, as determined by comparisons of coordinates between authors, were smaller for the Thematic Mapper and SPOT multispectral images than for the SPOT panchromatic image (Figure 3). The percentages of GBs where pixel coordinates differed by I 0.5 05& 0 10 20 30 40 50 > -I 5 0 15- U 3 a - Lu 10- a - u 5 - GPS 0 - -,,,,,,,,,, 0 10 20 30 40 50 METERS Figure 1. The errors of positions determined by digitizing 7.5-minute quadrangle maps and by GPS for 20 road lntersectlons of known location. January 1996 PE&RS

? 30 - * g 20- W 3 (1 ; 10- L - O -,,. - l - 10 20 SO 40 50 >50 METERS Figure 2. The frequency distribution of the distances between positions determined for ground control points by digitizing 7.5-minute quadrangle maps and by GPS. pixel units were 73 percent for Thematic Mapper, 78 percent for SPOT multispectral, and 48 percent for SPOT panchromatic. The greater error for the SPOT panchromatic data reflects the difficulty in determining which panchromatic subpixel represented the center of the intersection. These data, as well as the results of Clavet et al. (1993), suggest errors of 2 4 m in measuring pixel coordinates. Some large errors of greater than 1 pixel unit occurred due to differences in interpretation of features in the images. These large errors were more numerous for the panchromatic data. Computation of Transformation Matrices The Gps method provided more accurate coordinates for the computation of first-order transformation matrices. The initial RMSEs for GPS data were less than those for the map data for all three images (Figure 4), and fewer GCPS had to be deleted from the GPS data than had to be deleted from the map data to achieve relatively small RMSEs. For all three images, more than 70 percent of the GCPs for the GPS data were retained at RMSES that corresponded to a 10-m accuracy. The large proportion of GCPs retained demonstrates that the differentially corrected GPS data were sufficiently accurate to support rectifications of satellite images at a 10-m resolution. Other studies have demonstrated similar results for GPS data in producing orthoimages (Perry, 1992; Clavet et al., 1993). The transformation matrices computed from GPs coordinates should be considered to be more accurate than those computed from map coordinates for two reasons. First, the GPS coordinates are more accurate than the map coordinates (Figure 1). Second, a greater proportion of the GCPs are retained from the GPS data than are retained in the map data (Figure 4). Thus, the rectifications performed using GPS data are based on a greater number of GCPs whose coordinates have been determined with greater precision. The largest initial RMsE for GPS data occurred for the SPOT panchromatic image. A larger RMsE for the panchromatic data was expected because the pixel size is similar to the errors observed in measuring coordinates with GPS (Figure 1). There were, however, indications that errors involved in measuring pixel coordinates also contributed to the larger RMSE. Ten of the 11 GCPs that were deleted from the GPS data to achieve an RMSE of 1.0 pixel were also deleted from the map data. This large proportion of GCPs deleted from both map and GPS data indicates that errors in determining pixel coordinates were responsible for, or at least contributed to, the inaccuracies of the points. Because the road intersections selected as GCPs had to occur on 7.5-minute quadrangle maps, they were often large enough to encompass several panchromatic pixels. This produced inaccurate measures of pixel coordinates. Smaller and more distinct intersections occurred in the image but not on the maps. If only GPS data were to be used, the GCP selection could have included these more distinct features whose pixel coordinates could have been measured with greater precision. In addition to the greater accuracy of GPS, its use can also contribute to more accurate rectifications by increasing the number, variety, and clarity of features that may be used as GCPs. The Effect of Map and GPS Accuracy on Rectlfled Grid Files Table 1 summarizes comparisons of grid files rectified using map and GPs data for different values of RMSE with different numbers of GCPS retained. The percentage of pixels showing different assigned grid values between map and GPS rectifications ranged from 17.1 percent to 47.7 percent. These differences in assigned values occurred because of small differences in the alignments of map and GPS rectifications. These differences in alignments resulted in a proportion of the pixel assignments to be shifted to neighboring pixel locations. These shifts in assignment were dispersed throughout the files. The direction of the shift often changed as the RMSE was reduced (Table 1). Reductions in RMSE did not reduce the misalignment between map and GPS rectifications (Table 1). As successive, relatively inaccurate GCPS were deleted, the differences between map and GPS rectifications increased. Thus, the map and GPS rectifications did not converge to similar grid structures as accuracy standards increased. The lack of convergence indicates that rectifications based on map data cannot 40 s 0 Z 30 THEMATIC 10 2 0 I SPOT MULTISPECTRAL - W O K LL 30 SPOT 20 PANCHROMATIC 0 0 0.5 1.0 > 1.25 PIXEL DISTANCES Figure 3. The frequency distributions of the distances between pixel coordinates for ground control points determined by each author for Landsat Thematic Mapper, SPOT multispectral, and SPOT panchromatic images. 1 PE&RS January 1996

0.8 0.7 U) 0.7 0.0 ; 0.5 = 2.0 a THEMATIC 0.3,.....,..., 20 30 1.2 MULTISPECTRAL 5 10 15 20 25 0 5 10 15 20 POINTS DELETED Figure 4. The initial root-meansquare errors (RMSE) expressed in units of pixels and the reduction in RMSE with the deletion of successive ground control points to obtain RMSE of I 0.33 for The matrc Mapper, 10.5 for SPOT multispectral, and I 1.0 for SPOT panchromat~c images. Computations were performed using the COORDN procedure of ERDAS (1990b). TABLE 1. THE PERCENTAGE OF PIXELS WITH DIFFERENT ASSIGNED VALUES BETWEEN GRID FILES RECTIFIED USING COORDINATES DETERMINED FROM 7.5- MINUTE MAPS AND GPS. Number of GCP Retained Percent Direction of Image Type RMSE Map GPS Different Spatial Shifts Thematic Initial Mapper 0.50 0.33 SPOT Initial Multispectral 0.75 0.625 0.50 SPOT Initial Panchromatic 1.5 1.0 North East North be made as accurate as rectifications based on GPS data by simply imposing a small RMSE on a large number of CXPS. Large differences between map and GPS rectifications occurred for the SPOT panchromatic image. Almost one-half the pixels in the rectified files had different assigned values. A large difference was expected for the panchromatic image due to the small pixel sizes and the RMSE of 1.0, and the results of this study are consistent with those of Clavet et al. (1993) who concluded that cartographically derived coordinates were not sufficiently accurate to produce acceptable rectifications of SPOT panchromatic data. The differences between map and GPS rectifications for the Thematic Mapper and SPOT multispectral images were also large. The differences for the SPOT multispectral data were as large as those for the panchromatic data. Differences > 20 percent occurred for the relatively large Thematic Mapper pixels. The differences for the Thematic Mapper image imply that more than 20 percent of the features in the data rectified by the map coordinates are displaced 30 m from their corresponding positions in the data rectified by the GPS coordinates. This amount of displacement is not consistent with current map accuracy standards. The large differences between map and GPS rectifications for Thematic Mapper, SPOT multispectral, and SPOT panchromatic data in Table 1 indicate the increases in accuracy that may be expected from using coordinates measured by GPS. Whether the benefits of this increased accuracy warrant the use of GPS depends on (1) the accuracy required and (2) the costs of measuring GPS coordinates. In addition to the cost of obtaining the GPS receivers, the major expenses associated with GPS measurements are the time and travel costs involved in visiting the XPs. The costs of visiting the XPS is largely dependent on the size, location, and terrain of the area being rectified (see Clavet et al. (1993) for an example of cost analysis computations). Other types of GPS receivers are available that can provide more accurate position coordinates than code phase receivers. Systems employing p-code receivers can achieve nominal accuracies of I 0.1 m, but their use for rectifying satellite images may not be warranted due to the time and cost required to obtain GCP coordinates (Clavet et al., 19931. Moreover, accuracies as small as 0.1 m are not required when the errors involved in measuring pixel coordinates are 2 4 m. Although the use of p-code receivers may not be warranted, the recent development of GPS receivers that can rapidly measure positions with nominal accuracies of < 2 m (Gilbert, 1993) should prove useful in further increasing the accuracy of satellite image rectifications. Conclusion These comparisons illustrate that using GPS and map coordinates can produce rectified files that differ in more than 40 percent of their pixel assignments. These differences involve generally small spatial shifts, but they can result in more than 20 percent of the features in the rectified images being displaced by 30 m. Where position accuracy in the rectified data is required, the use of GPS may be warranted for even relatively large pixels. Acknowledgments This study was performed under the auspices of Contract No. DE-AC09-76(SR00819) between the U. S. Department of Energy and the University of Georgia. We thank A. Michele May, Terese E. Bertram, and John B. May for their assistance. Graphics were produced using DIAGRAPH, a product of the Computer Support Corporation, Dallas, Texas. Disclaimer Trade and product names are mentioned only for the completeness of the paper, and no endorsement, implied or otherwise, is intended by the authors or the institutions concerned. January 1996 PEfkRS

References Clavet, D.. M. Lasserre, and J. Pouliot, 1993. GPS control for 1: 50,000-scale topographic mapping from satellite images, Photogrammetric Engineering ' Remote Sensing,.59(1):107-111. ERDAS, 1990a. ERDAS Core, Version 7.4. Volume 2, ERDAS, Inc., Atlanta, Georgia. -. 1990b. Image Processing Module, Version 7.4, ERDAS, Inc., Atlanta, Georgia. -, 1990c. Raster GIS Modeling Module, Version 7.4, Volume 2, ERDAS, Inc., Atlanta, Georgia. Gilbert, C., 1993. Portable GPS systems for mapping: Features versus benefits, Earth Observation Magazine, 2(9):4348. jensen, 1. R., 1986. Introductory Digital Image Processing. Prentice Hall, Englewood Cliffs, New Jersey. Ferry, E. M., 1992. Using GPS in agricultural remote sensing, GPS World, 3(7):30-36. Steel, R. G. D., and J. H. Torrie, 1960. Principles and Procedures of Statistics, McGraw-Hill Book Co., Inc., New York. Thompson, M. M., 1988. Maps for America, Third Edition, U. S. Department of the Interior, Washington, D. C. Trimble Navigation, 1991. Trimble Pfinder", Software Reference Guide, Trimble Navigation, Ltd., Sunnyvale, California. -, 1992. General Reference: GPS Pathj'nder Professional System, Trimble Navigation, Ltd., Sunnyvale, California. (Received 5 May 1993; revised and accepted 1 February 1994; revised fi April 1994) THE GIs Applications Book Examples in Natural Resources: a compendium edited by William J. Ripple The GIs Applications Book: Examples in Natural Resources compendium is a one-stop reference guide to state-of-the-art applications of GIs in the natural resources. The articles were selected from respected GIs, natural resource, and science journals. This book can be used as a textbook supplement for university level courses in GIs. In addition, it can also be used as a reference for geographers, foresters, hydrologists, soil scientist, engineers, wildlife biologists, planners, and ecologists. SECTIONS INCLUDE: Section 1 : Introduction Provides an introduction with information on the role of geographic information systems (GIs) in natural resources, technological developments, and definitions of commonly used GIs terms. Section 2: Environmental Modelling & Assessment Section 3: Soils Section 4: Water Resources Section 5: Forestry Section 6: Landscape Ecology Section 7: Wildlife Section 8: Land Use Section 9: Biological Diversity The GIs A~~lications Book: Exam~les in Natural Resources. 1994. William J. Ripple. 372 pp. 22 color plates. $70 (softcover); ASPRS Members $50.00 Stock # 4634. For ordering information, see the ASPRS Store. PE&RS January 1996