This file was created by scanning the printed publication. Errors identified by the software have been corrected; however, some errors may remain. GPS vs Traditional Methods of Data Accuracy Input: Improving Spatial Data Accuracy? Russell Combs, Jr. ', James L. smith2, and Paul V. ~olstad~ Abstract.-Geographic information systems (GIs) have become an integral part of many natural resource organizations. The increased interest and use of these systems for a variety of managerial tasks has resulted in an increased concern about the accuracy of spatial databases. Many natural resource organizations are using their GIs to perform complex analysis and they are becoming very aware of what the cost might be for using inaccurate spatial data. Thus many organizations are involved in developing procedures or methods for improving and maintaining the quality and accuracy of their spatial data. This paper will focus on the issue of spatial data accuracy and how it can be improved or maintained by natural resource organizations. In particular, the paper will concentrate on one of the main areas influencing spatial data accuracy, data input. The paper will review and compare the positional accuracy levels of some of the most common methods of data input (manual digitizing) to newer technology and techniques (GPS, PC-based singlephoto space resection) now available to natural resource organizations. INTRODUCTION Positional accuracy in a natural resource GIs can be defined as a measure, usually in ground distance, of how well the digital object coordinates in the various spatial data layers correspond to the "true" coordinates of an entity on the ground (Bolstad & Smith 1992, Chrisman 1991, and Blakemore, 1984). The accuracy of a spatial database is a direct function of the methods used to input the spatial data, and the sources of data used to collect the spatial information. As with any mapping procedure,whether digital or cartographic, generalizations are made and database coordinates used to represent entities may differ from their "true" ground locations. Thus, the accuracy of a spatial database is inherently limited by the method or sources of data used to construct the database. A natural resource GIs spatial database usually includes multiple data layers, with each layer representing a different theme or group of similar features. 123 1 DatabaseIGPS Forester, Canal Forest Resources, Inc., Charlotte, NC Manager GIs and Remote Sensing Services, Canal Forest Resources, Inc., Charlotte, NC ' Assistant Professor, Department of Forest Resources, University of Minnesota
The individual data layers in a natural resource GIs are not typically developed or created using the same methods or sources of spatial data. Natural resource organizations routinely deal with ground features that are not well mapped or must work with with poorly or undocumented maps. Thus, a wide range of data sources are often used, and often two different sources of data are used within a given data layer. This leads to a combination of methods and sources being used to develop or update the spatial database, and in turn leads to a spatial database with many different data layers having varying levels of positional accuracy. This paper will review the accuracies of the most common data input methods/sources, and propose some suggestions on how natural resource organizations can improve the positional accuracy of their spatial databases. In addition, the paper will discuss what if, any accuracy or benefit may be gained by using some of the new technology becoming available. Traditional MethodslSources of Spatial Data The traditional methods and sources of spatial data input for natural resource organizations can be categorized into two main types: I) manually digitizing from large-scale maps or aerial photographs, and 2) the use of purchased digital data (DLG, TIGER, DEM). While purchased digital data is becoming more widespread throughout the natural resource community, its use is limited to those organizations with established GIs systems, and those who have familiarity with these type of data. Manual digitization is still by far the most common method of data input, and thus the major source of positional error in spatial databases. Positional errors resultant from manual digitization are generally a function of three sources: media limitation, user error, and control point or coordinate registration error (Walsh et al, 1987). While all three of these factors vary as to the affect they have on the positional accuracy of manually digitized data, the first of these factors, source media limitation, has the largest impact on the amount of positional error introduced to spatial data during manual digitization. Manually digitized data in natural resource organizations are usually collected from either aerial photography or large-scale maps. In both cases the source data has inherent limitations that affect the accuracy of the data. Aerial photography is affected by tilt and terrain displacement, while many large-scale natural resource maps contain significant positional errors due to poor drafting, incorrect boundary placement, or drafting from aerial photography. The amount of positional error that can be introduced to spatial data because of these factors has been documented, and by reviewing previous research, an estimate of the positional error caused by manual digitization can be obtained. For example, horizontal positional errors of up to 73.0 meters in steep terrain and
15.0 meters in low terrain have been observed for spatial data digitized directly off of aerial photography (Bolstad, 1992). On the other hand, large scale maps often have linear or areal features that have line widths which vary between 0.25-1.01 mm (Bolstad & Smith, 1992), corresponding to widths on the ground of anywhere between 6.0-24.0 meters if the data were collected from a 1 :24,000 scale map. Another study observed line locations to vary by 0.127 mm of their true positions on a 1 :24,000 scale map, equivalent to 3.04 meters in actual ground distance (Dunn et al, 1990). As can be seen, it does not take much variation in a source document to start adding up to large amounts of inaccuracy on the ground. A factor that is sometimes overlooked when dealing with linear and positional accuracies, is the associated area accuracy that is directly related to the linear accuracy of the spatial database. For instance Chen and Finn (1994) stated that average manually digitized area error is low, with manually digitized data generally underestimating polygon area by 1-3%. In addition, Wiles (1 988) observed acreage errors of up to 10.23% for data digitized directly from aerial photography in varying terrain. The second source of positional error resultant from manually digitization is from operator or user error. Operator error is the error resultant from the operator's inability to trace linear or point features with the digitizing puck. Operator digitizing ground errors of up to 1.15 m in the x-direction and 1.17 m in the y-direction have been documented for manually digitized data from 1 :24,000 scale maps (Combs 1995, Warner & Carson, 199 1). In addition, operator inaccuracies of 0.95 m in the x-direction and 1.26 m in the y-direction have been calculated for manually digitized data collected from NAPP photography (Combs, 1995). In addition to source errors and user errors, positional accuracy during manual digitizing is also affected by control point or coordinate registration. This is an issue that is often overlooked by natural resource organizations, but poor control point registration does affect the overall accuracy of the digitized data. A few studies have highlighted the importance of using accurate control points. Fernandez et al. (1 991) documented map derived control point errors ranging from 2.1 3 to 7.62 meters in ground distance. Bolstad et a2 (1 990) observed positional error attributed to poor control points ranging from 1.52 to 85.04 meters on the ground for data digitized directly from aerial photography in varying terrain. Control point registration and accuracy can be increased by using well known established point locations that have been surveyed or through the use of global positioning systems (GPS). Thus, if the positional errors associated with manual digitization are additive, which they most likely are to some extent, spatial data collected using these methods could be tens to hundreds of meters off in positional accuracy.
Non-traditional Methods/Sources of Spatial Data While manual digitization and the use of purchased digital data are still popular methods and sources of spatial data, the development of new data input methods and techniques may provide natural resource organizations with more accurate and cost-effective alternatives. Two of these new techniques or methods will be discussed: 1) GPS and 2) PC-based single-photo space resection. While both of these alternatives are not necessarily "new", they have not been extensively applied in natural resource organizations. GPS is a satellite-based positioning system which operates using L-band radio signals to provide highly accurate position, velocity, and time data. GPS can provide accuracy levels of up to 30.0 cm (P-code) and 1-5 meters (Course/Acquisition Code or CIA code). Civilian users can achieve accuracies equal to those of the P-code but generally this accuracy is reserved for military use or surveyors. Single-photo space resection is accomplished by measuring photocoordinates, via a digitizing tablet, and applying the collinearity equations to these data to calculate ground coordinates. Since the procedure is based upon the photogrammetric principles of space resection, the calculated ground coordinate accuracy is improved due to the reduction of the tilt and terrain effects inherent in the photograph. Recently studies have been conducted to determine the accuracy of GPS and single-photo space resection in a natural resource setting. Evans et al. (1992) reported average positional accuracy of 2.01 m between GPS positions and actual forest plot center positions during navigation trails to locate known forest plot centers. In other studies, applying differential correction to the mean of 100 to 300 position fixes resulted in positional ground accuracies of between 2.0 to 3.0 m (August et al, 1994) and 2.0 to 4.0 m (Deckert, 1994). A previous study reported positional ground errors from single-photo space resections of two different mountainous study areas to be 11.39 m and 7.07 m respectively (Ran, 1992). That would translate to about 60.0 m more accurate than the data collected directly from aerial photography in steep terrain as reported in the studies previously discussed. A more recent study involved comparing the accuracies of point features manually digitized from topographic maps and NAPP photography versus the accuracies of the same features collected using single-photo space resection and GPS techniques (Combs, 1995). For this study, CIA code level GPS was used to determine the "true" ground or reference location of the test features. While the GPS data collected for this study can't determine the actual ground locations of the test features, previous work has determined CIA code level GPS points to be within 1.O - 3.0 meters of the "true" ground location (Deckert, 1994). For each
test point, 300 position fixes were collected and averaged after differential correction to determine a mean positional location for the test point. The manually digitized and single-photo space resection data were then compared to the GPS value and the difference was determined to be the amount of positional error attributed to that methodlsource. Table 1 highlights the results of this study and shows the average positional accuracy that was determined by using the various techniques and sources. Table 1. Mean positional error (meters) for two study sites in Virginia, GPS used as "trueff ground location. Data Set Space Resection Digitized Aerial Photograph To~oera~hic Mar, High-Relief Study Site 6.27 25.86 11.20 Low-Relief Study Site 5.64 7.14 11.64 The results in Table 1 show how spatial accuracy can be improved by using either GPS or the space resection technique. The positional accuracy of data digitized directly from aerial photography is about what would be expected. As shown in Table 1, the high-relief study site had a mean positional accuracy of more than 3 times greater than the low-relief study site. This difference is attributed to the greater amount of terrain relief and tilt displacement that is present in the high-relief study site aerial photography. In contrast, the singlephoto space resection data sets provided mean positional accuracies that are almost equal between the two study sites. Since the space resection reduced the tilt and terrain effects inherent in the aerial photography (and assuming all other factors are constant), the resultant mean positional accuracies were similar and improved over just digitizing directly from the aerial photography. The digitized topographic map data from both study sites provided similar positional accuracies that were greater than the NMAS (National Map Accuracy Standards), but both were still more than half of what was obtained by the space resection technique. Other results that are not directly shown in Table 1 are the accuracy of the GPS data. While GPS was only used for the "true" ground reference value for this study, the results in Table 1 show how much each of the other data collection techniques differed in positional accuracy from the GPS values. For example, on the high-relief study site the GPS data was on average 25.0 meters more accurate than digitizing directly from aerial photography. In addition, the GPS data proved to be on average 11.0 meters more accurate than digitizing directly from 1 :24,000 scale topographic maps. While the GPS proved to provide the greatest positional accuracy, both it and the single-photo space resection increased the reliability and confidence of the data collected in this study.
CONCLUSIONS The accuracy of digital spatial data collected by natural resource organizations is affected by the source of the data and the method use to collect the data. As the use of GIs continues to grow in natural resource organizations, the reliability and accuracy of the data used in GIs analysis is becoming increasingly important. Greater than any other time in recent history, natural resource organizations are being held responsible for their actions and the decisions they make concerning the environment. The research and results reviewed in this paper show that spatial data collected using the traditional methods may adversely affect the accuracy of spatial databases. Research presented in this paper showed that data digitized directly from large-scale maps and aerial photography can provide positional accuracies anywhere between 6.0 to 73.0 meters. On the other hand, spatial data collected using a PC-based single-photo space resection technique can provide positional accuracies between 5.0 to 7.0 meters, depending on the amount of terrain relief. In addition, GPS collected spatial data has been shown to be capable of providing positional accuracies from 1 to 4 times more accurate than some of the traditional techniques discussed in this paper. As the use of GIs technology continues to expand in natural resource organizations, a PC-based single-photo resection or more particular the use of GPS, offers an alternative to traditional methods of data entry and can provide reliable and more accurate spatial data. REFERENCES August, P., J. Michaud, C. Labash, and C. Smith. 1994. GPS for Environmental Applications: Accuracy and Precision of Locational Data. Photogrammetric Engineering & Remote Sensing. 6O(l):4 1-45. Blakemore, M. 1984. Generalization of Error in Spatial Databases. Cartographica. 21: 13 1-139. Bolstad, P.V. 1992. Geometric Errors in Natural Resource GIs Data: Tilt and Terrain Effects in Aerial Photographs. Forest Science. 3 8(2): 367-3 80. Bolstad, P.V., P. Gessler, and T.M. Lillesand. 1990. A Variance Components Analysis of Manually Digitized Map Data. Proc. AC SM-ASPRS. Vol 3. 9-1 8pp. Bolstad, P.V., and J.L. Smith. 1992. Errors in GIs: Assessing Spatial Data Accuracy. Journal of Forestry. 90(11):2 1-29. Chen, 2. Z., and J.T. Finn. 1994. The Estimation of Digitizing Error and Its Propagation with Possible Application to Habitat Mapping. Proc. International Symposium on the Spatial Accuracy of Natural Resource Databases. 57-66 pp. Chrisman, N.R. 1 99 1. The Error Component of Spatial Data. Geographical Information Systems, Volume1 :Principles. John Wiley & Sons. 165-173 pp. Combs, R.G., 1995. Positional Accuracy in a Natural Resource Database: Comparison of a Single-Photo Resection Versus Afline Registration. Master's Thesis. VPI&SU. 148 pp.
Deckert, C. J. 1994. Canopy, Terrain, and Distance Effects on Global Positioning System Position Accuracy. Master's Thesis, VPI&SU. 70 p. Dunn, R., A. R. Harrison, and J.C. White. 1990. Positional Accuracy and Measurement Error in Digital Databases of Land Use: An Empirical Study. International Journal of Geographical Information Systems. 4(4): 3 85-3 98. Evans, D.L., R. W. Carraway, and G.T. Simmons. 1992. Use of Global Positioning System (GPS) for Forest-Plot Location. Southern Journal of Applied Forestry. 1 6(2): 67-70. Fernandez, N., D. F. Lozano-Garcia, G. Deeds, and C. J. Johannsen. 1991. Accuracy Assessment of Map Coordinate Retrieval. Photogrammetric Engineering & Remote Sensing 57(11): 1447-1452. Ran, L., 1992. Single Digital-Photo Correction for a GIs Application and Error Analysis. Master's Thesis, VPI&SU. 139 pp. Walsh, S. J., D. Lightfoot, D. Butler. 1987. Recognition and Assessment of Error in Geographic Information Systems. Photogrammetric Engineering and Remote Sensing. 53(10):1423-1430. Warner, W. S., and W. Carson. 1991. Errors Associated with a Standard Digitizing Tablet. ITC Journal. No. 232-85 pp. Wiles, S. J. 1988. Evaluation of Photographic Properties for Area Estimation. Master's Thesis. VPI&SU. 100 p. BIOGRAPHICAL SKETCH Russell G. Combs, Jr. is a DatabaseIGPS Forester with Canal Forest Resources, Inc. in Charlotte, NC. Russ holds a B.S. in forestry and a M.S. in GIs and GPS from Virginia Tech. Russ has been employed with CFR for the past year where he is responsible for managing the inventory database for CFR's clients as well as heading up Canal's company wide GPS program and assisting on various GIs projects and analysis. James L. Smith holds a B.S. in Forestry and an M.S. in Forest Biometrics from the University of Georgia, and a Ph. D. in Forest Biometrics and Remote Sensing from Virginia Tech. He was a member of the faculty of the Virginia Tech Department of Forestry at Virginia Tech for 13 years, where he specialized in quantitative GIs and remote sensing issues. For the last two years, Jim has been the Manager of the GIs and Remote Sensing Group at Canal Forest Resources, Inc, in Charlotte, NC. Paul V. Bolstad is an Assistant Professor in the Department of Forest Resources at the University of Minnesota. Paul holds a B.S. in Forestry from the University of California at Berkeley, a M. S in Forestry from N. C. State, and a Ph. D. in Forestry from Wisconsin University. Paul was a member of the faculty of the Virginia Tech Department of Forestry at Virginia Tech for 6 years, where he specialized in GIs, Remote Sensing, and GPS issues.