A Framework for Road Network Database Generalization

Sharon Kazemi* Geographic Information Science Center College of Environmental Design University of California Berkeley Samsung Lim School of Surveying and Spatial Information Systems The University of New South Wales

Abstract

This paper presents a generalization methodology to derive multi-scale GEODATA through an evaluation of ESRI ArcGIS software that was used as a testbed based on the principles of generalization. It focuses on integration and utilization of generalization operators in order to generalize a road network database and produce small scale maps at 1:500 000 and 1:1 000 000 from GEODATA TOPO-250K Series 2 data. The derived maps are satisfactory when compared with the existing small-scale road maps such as the Global Map at scale of 1:1000 000. It is suggested that a comprehensive evaluation of generalization systems and their performance is essential to marry the cartographic knowledge from experts and bring this into a generalization framework. Therefore, there is an opportunity to evaluate other generalization systems to derive a multi-scale database from a master database in future investigations to enhance the generalization methodology.
1 Introduction National Mapping Agencies (NMAs), spatial data providers and map producers have been maintaining multiple databases at different scales for different uses. They are also committed to maintaining a set of cartographic data with synchronized updates. Maintaining multiple databases is resource-intensive, time consuming and cumbersome (Arnold and Wright, 2003). This is a major challenge for NMAs and other map producers (Sarjakoski et al., 2002). A seamless master database should offer capabilities to derive various types of maps at different scales, e.g. at scales ranging from 1:250 000 to 1:10 000 000. Derivative mapping has been identified as an active research and development area for NMAs, industry and academia to accomplish the requirements for evaluation and validation of generalization (Muller, 1995; Visvalingam, 1999; Kazemi, 2003; Kazemi and Lim, 2007). Therefore this research has focused on the application of the generalization tool to the development of a detailed generalization framework for deriving multi-scale GEODATA. More particularly, the research has focused on utilising the generalization operators provided by ESRI ArcGIS, combined with the skills of a trained cartographer, to generalize a road network database from GEODATA TOPO-250K Series 2 data and produce smaller scale maps at 1:500 000 and 1:1000 000. To fulfill the need to evaluate and validate existing generalization tools, the authors research has focused on the development of a detailed generalization framework for deriving a multi-scale GEODATA dataset. The research effort concentrates on the integration of generalization operators with the cartographer's intuition/skills and utilization of ArcGIS Generalize to achieve acceptable results. A brief discussion of the generalization framework for deriving multi-scale GEODATA is presented in the next section. It begins with reviews of line and area generalization. Then, the methodology of a conceptual generalization framework and road network generalization with special emphasis on simplification is followed by the analysis of results and evaluation of derived maps based on the Radical Law approach (Pfer and Pillewizer, 1966; Muller, 1995).

Address for correspondence: Sharon Kazemi, Geographic Information Science Center, College of Environmental Design, University of California Berkeley, CA 94720-1820 USA, E-mail: sharon_kazemi@hotmail.com

2 Relevant Research Technological development in the field of generalization is moving very fast, following the trend from manual cartography to computer-based cartography. Map generalization has become a desirable component of today's Geographic Information Systems (GIS). It is an integral part of spatial data collection, representation and access. Most generalization algorithms developed and employed by the GIS and computer science communities have been tailored for map production. During the last decade, there has been rapid development of semi-automatic techniques for map and database generalization. With the increased integration of human knowledge, computer scientists, mapping specialists and cartographers developed generalization operators and algorithms for automation of generalization (e.g. Peschier, 1997; Iwaniak and Paluszynski, 2001). The majority of existing standard GIS software applications support generalization of both line and area features. A comprehensive review of line and area generalization is provided by Kazemi (2003). Since generalization of line features is an interest of this research, some of the most relevant generalization operations are discussed in the Road Generalization section. Many generalization processes exist but the key operators are selection, simplification, classification and symbolization of features. Line feature generalization often involves the use of geometric operators such as selection, elimination, merge, displacement, aggregation and symbolization. However, there is no standard definition of generalization, since it depends on ad hoc applications (AGENT, 1999; Haire, 2001; Kazemi et al., 2004a and 2004b; Kazemi and Lim, 2007). Generalization of linear features is very important because lines represent the majority of map features. Therefore, linear feature generalization plays an important role in GIS (Barrault, 1995; Skopeliti and Tsoulos, 2001). A review on the application of generalization with a special emphasis on road network generalization from 1:250 000 scale to 1:500 000 and 1:1 000 000 scales is provided in Kazemi (2003). Commercial generalization systems for this type of mapping are limited to four major vendors: 1) ArcGIS Generalize (ESRI), 2) DynaGen (Intergraph), 3) Clarity (Laser-Scan), and 4) CHANGE (Institute for Cartography at Hanover University). These GIS software packages support the generalization of line features that motivated this research for a generalization framework to derive multi-scale GEODATA. It focuses on integration and utilization of generalization operators using ArcGIS 8.2 Generalize in order to generalize a road network database from GEODATA TOPO-250K Series 2 and produce small scale maps at 1:500 000 and 1:1 000 000. This research will demonstrate the principle for roads, with the idea of expanding the application to other features. 3 Methodology 3.1 Study Area The central urban area of Canberra, about 64 km, is selected for the study area. The area is chosen because it provides a mixture of different roads, and ease of access that requires minimum resources for data collection purposes. 3.2 Software and Generalization Algorithm ArcGIS is a professional level GIS software developed by Environmental Systems Research Institute (ESRI), Redlands, California. ArcGIS is composed of seven major software applications in the form of ArcMap, ArcCatalog, ArcScene, ArcGlobe, ArcToolbox, ArcCatalog and ArcIMS, and all share a common application framework. The ArcToolbox Generalize tool is used in this research. It has a set of line simplification algorithms such as Pointremove. According to ESRI (2005) ArcGIS Pointremove tool uses Douglas-Peucker (DP) algorithm, which keeps the so-called critical points that depict the essential shape of a line, and removes all other points. The algorithm connects the end nodes of an arc with a trend line. The distance of each vertex to the trend line is measured perpendicularly. Vertices closer to the line than the tolerance are eliminated. The line (arc) is divided by the vertex farthest from the trend line, which makes two new trend lines. The remaining vertices are measured against these lines, and the process continues until all vertices within the tolerance are eliminated (Mroz et al, 1996). The DP algorithm attempts to preserve directional trends in a line using a specified tolerance factor (Taylor, 2005) by removing vertices from the selected lines and simplifies the line bends. The DP algorithm is one of the most popular and accurate generalization algorithms available (Jenks, 1989) that has widely been used for many cartographic

applications such as coastline generalization (Dutton, 1999; Nakos, 1999), river generalization (Rusak and Castner 1990; Moreno et al., 2002), and road generalization (Peschier, 1997; Skopeliti and Tsoulos, 2001). 3.3 TOPO 250K Roads Classification The National Mapping Division (NMD) of Geoscience Australia produces GEODATA TOPO-250K and associated cartographic databases that are used in this research. Roads in GEODATA TOPO-250K have varying widths and types (NMD, 1999). These are Dual Carriageway (DC), Principal Road (PR), Secondary Road (SR), Minor Road (MR), and Vehicle Track (VT). The study area offers all these types. Principal roads (e.g. main roads in urban areas) can be extended in two directions, starting from north to southeast or to southwest. According to GEODATA TOPO-250K specifications, local knowledge and reference datasets, the roads widths range from 5 to 10 metres. Urban streets are concentrated in the central part of the study area. Their widths are between 4 and 8 metres. Minor roads are located mostly in the recreational areas such as parks. These roads include rural roads, access roads and tracks. Their widths are from 2.5 to 5 metres. 3.4 Conceptual Framework In order to gain an appreciation of the conceptual generalization framework, it is worthwhile to look at the main steps of the research methodology in Figure 1. This is linked to the conceptual framework that is presented in Figure 2. To date, many frameworks/workflows have been developed for the generalization of cartographic data (see Kazemi et al., 2004b). It seems that many of these frameworks are generic and do not offer a total solution for the operational environment. The motive of this research is to develop a workflow specifically for derivation of multiple scale maps from a master road network database (Figure 2). A large portion of this framework may be considered as generic too. However, in this paper all the processes are considered specifically for road generalization (refer to Kazemi (2003) for further details).

Figure 1. Schematic representation of the research methodology.

Figure 2. Framework for road generalization. There are a number of parameters that could be used for generalization of roads to maintain the legibility, the visual identity of each road segment and the pattern. Out of six generalization parameters of Shea and McMaster (1989), the following four are taken into consideration in this research: Congestion/Legibility, Coalescence, Imperceptibility and Length/Distance. Some of the Shea and McMaster parameters have been also used by Peschier (1997). It is known that generalization of a network (for example, roads, rivers, or other linear features) requires at least six key operations/processes: Classification, Selection, Elimination, Simplification, Typification, and Symbolization (Figure 3). They are used in the generalization framework of this research discussed in the next section. The generalization operators of ArcGIS were tested for the generalization of roads by employing the conceptual generalization framework for derivative mapping, but the results showed that ArcGIS algorithms of Douglas-Peucker (Bendsimplify and Pointremove) do not support dynamic generalization. The way to work around this problem is to use Intergraph DynaGen generalization software which enables the derivation of a multi-scale database from a master database (Kazemi and Lim, 2005). Albeit demonstrated that testing and incorporating of the DynaGen generalization system enhanced their conceptual generalization framework, Kazemi and Lim (2005) intend to investigate the capabilities of other systems such Laser-Scan Clarity in future research.

(a) Classification and Selection

(b) Elimination

(c) Simplification

(d) Typification

(e) Symbolization

(f) Elimination

(g) Simplification

(h) Before & after Simplification

(h1) Before & after Simplification

(j) Symbolization (i) Typification Figure 3. Road generalization outputs that illustrate the process of deriving 1:500 000 maps (a-e) from the source data at 1:250 000, and the process of deriving 1:1000 000 maps (f-j) from newly derived 1:500 000 scale data. The area includes roads from the range of road classifications described as part of the GEODATA TOPO-250K specifications. The maps above are necessary not to scale. 3.5 Road Generalization The following operations are required to generalize a road network (Lee, 2003a). The outputs of generalization wokflow applying these operations are shown in Figure 3: Classification: a good classification of roads makes the selection easier and more accurate. This step identifies objects that are placed in groups according to similar properties. It also reduces the complexity and will improve the organization of a map. For example, roads were categorized into 5 classes with widths ranging from 2m to 12m. Selection: only certain road classes are selected for inclusion at the target scale. For instance, all vehicle tracks are dropped from the 1:250 000 road database when deriving 1:500 000 scale roads GEODATA. Elimination: a road branch shorter than a certain length can be eliminated, such as deleting small road segments that can cause conflict in the final map and are not significant for representation on the map. Costello et al., (2001) applied this operation to eliminate small buildings, short roads, and small villages in generalization of 1:250 000 scale data to generate a 1:1 000 000 Global Map. The Global Map Australia 1:1 000 000 (2001) is a digital spatial dataset covering the Australian continent and island territories at 1:1 000 000 scale which consists of eight spatial layers in the vector form (administrative boundaries, drainage, transportation and population centres) and the raster form (elevation, vegetation, land cover and land use). This is created from Geoscience Australia's GEODATA TOPO 250K Series 1

data and the United States Geological Survey (USGS) provided the raster images. This was initiated by the International Steering Committee for Global Mapping (ISCGM). Simplification: selected roads can be simplified to reduce the details. This research used the Pointremove and Bendsimplify operators to remove vertices from the selected lines and simplify extraneous bends of roads. Typification: this function is not directly available in ArcGIS, but a cartographic manual editing approach was used. It is basically to reduce the density of features and simplify the distribution and the pattern of the network. The result should be a connected and less congested network that represents a similar pattern at the smaller scale. Symbolization: the systematic process of creating graphic marks, which represents the objects and features to be mapped. For example, roads are symbolized per their classes.

Generalization operators are applied in deriving small-scale maps in this investigation. To commence generalization, necessary objects in the map should be chosen; and the next step is to apply simplification, which may include smoothing of curves and straightening paths. The objective of simplification is to increase the legibility of the map by removing unnecessary details of the road network. In this simplification process, the road line features are simplified to remove extra vertices in the road segments. ArcGIS Generalize (Pointremove) is used to simplify the lines for this process. It is based on the DP algorithm that removes vertices from the road linear features with a user-defined tolerance. To simplify the line bends, the bend operator is used to omit extra bends. Tolerances for each scale of the roads were based upon comparison of the generalization results with cartographic products of the same scale. When using the Generalize tool an appropriate weed tolerance (threshold level)) must be determined. For the appropriate tolerance selection Shea and McMaster (1989) argue, the input parameter (tolerance) selection most probably results in more variation, in the final result than either the generalization operator or algorithm selection. Thresholding techniques have been applied in both image processing and map generalization. To decide how much information is to be lost while retaining useful information is a difficult task in line simplification. In fact, it is very a difficult and tedious task to set an appropriate threshold (Zhao et al., 2001). The basic question remains as to what is the best threshold. Forghani (1997 and 2000) mentioned that there is no good answer to this question, and the threshold value must be guided by application. In addition, Mackaness et al., (1998) stated the loss in total line length proceeds at a rate that is approximately linear to the increase in tolerance. Low tolerances produce high rates of change in vertex numbers and low changes in total line length and using higher tolerances increasingly fewer vertices are removed for a rapidly increasing loss in total line length. The threshold value depends mostly on map content, map quality, map geometry, and the amount of noise present in the database. For operational production cartographic environments, generalization capabilities of current GIS software tools such as ArcGIS should take care of automated threshold selection in order to allow the complete cartographic production workflow to be carried out in a uniform environment (Hardy et al., 2004). Line simplification algorithms are mainly based on geometric principles (Cromley, 1992) with the reduction rate affected by a predefined tolerance (Nakos, 1999). The current way of selecting tolerance is by trial and error. Different levels of tolerance, ranging from 10 to 70, were tested in the course of road network generalization. Tolerances of 25 and 35 were acceptable for the 1:500 000 and 1:1 000 000 data, respectively. Similar tolerances have been used in other studies (Visvalingam and Williamson, 1995; Nakos, 1999). In practice, generalization involves thinning out coordinates (vertices) to simplify line strings. In this way, redundant points along lines can be efficiently removed and the changes of line shape are minimized. Approximately 25 percent of vertices in this study were eliminated, while the length of the lines did not change significantly.

The Pointremove operator (based on DP algorithm in ArcGIS) proved to achieve satisfactory line thinning/compression results. It removes vertices quite effectively, but produces angularity along lines that is not very pleasing from an aesthetic point of view. In addition, in some instances, cartographic manual editing is required. The Bendsimplify operator is designed to preserve cartographic quality. It removes small bends along the lines and maintains the smoothness

(Lee, 2003b). Applications of these algorithms are well documented in the literature (e.g. Jenks, 1989; Mroz et al., 1996; Nakos, 1999). It would be interesting to compare the two in future research and find the best use of them. For example, at 1:500 000 the Bendsimplify operation resulted in reduction of approximately 4750 vertices (before generalization) to 760 vertices (after generalization). Bends are defined as sections of curves between two inflection points. Wang and Muller
(1998) introduced the idea of an iterative bend simplification approach that is available in ArcGIS. The ArcGIS Generalize Bendsimplify operator detects bends in lines to drive the line generalization process; it was proposed as a form of structure-based generalization. It reduces the number of vertices significantly through this operation. It was found that generalization of roads requires a very large volume of processing to transform data from 1:250 000 scale to 1:500 000 and 1:1 000 000 scale maps. It eliminates shorter branches and results in simplification of the road network, but also improves the preservation of the characteristics of the structure. An enriched database (GEODATA TOPO-250K) was used as the main reference dataset. Based on quantitative and qualitative assessments (see next section), the result of the semiautomatic generalization is satisfactory. The results were evaluated by experienced cartographers who consider these as good quality, except for lack of proper smoothing on some portions of roads. Previous studies (e.g. Visvalingam and Williamson, 1995) confirmed that the DP algorithm, by applying very low tolerances, resulted in removal of an appropriate number of vertices. 4 Analysis of Results In total, 1202 road segments were present in the GEODATA TOPO-250K database over the study area. Changes in the representation of road features at 1:250 000, 1:500 000 and 1:1 000 000 scales due to generalization were quantified. Road types were considered a major index of importance in different roads. During the transformation of road representation from 1:250 000 to 1:1 000 000, some of the less important road segments were deleted. At each scale the number of road segments classified by road types was counted and the results were graphically illustrated. All vehicle tracks disappeared at the 1:1 000 000 scale. Also, there was a 61 percent reduction of minor roads at 1:500 000 and a further 45 percent reduction of minor roads at 1:1 000 000 (Figure 4).

Figure 4. Road classification at different scales ranging from 1:250 000, 1:500 000 and 1:1 000 000 and number of road segments in different classes: For example at 1:500 000 scale, there are 10 Minor Road (MR), 8 Secondary Road (SR), 3 Principal Road (PR), and 1 Dual Carriageway, but no Vehicle Tracks (VT) are retained. Results of map derivation have been analysed using two methods: the Radical Law (Pfer and Pillewizer, 1966; Muller, 1995) and an interactive accuracy evaluation method. The Radical Law determines the retained number of objects for a given scale change and the number of objects of the source map (Nakos, 1999). The original GEODATA TOPO-250K roads have 1202 arcs and the derived roads have 856 arcs before the line simplification operation. The equation below computes the number of roads in the map after the line simplification operation:

S s

s S T

Number of lines in 1:500 000 scale roads map => 1202 250 = 849.9 500 T

= 605.3 Number of lines in 1:1 000 000 scale roads map => 856 500 1000 T

The numerical value of 849.9 represents the predicted number of arcs after generating the 1:500 000 scale road map, and the value of 605.3 predicts the number of arcs after generating the 1:1 000 000 scale map. These results have also been analysed visually since evaluation is a critical element of semi-automated road generalization. The cartographic knowledge, such as the presentation of features on the map and the contextual information of features, has been combined with generalization operators (Choi and Li, 2001) to derive multiple scale maps. This includes cartographic knowledge and information about roads and surrounding areas. Thematic properties such as length, width, and connectivity were also used. Although these measures are not directly available in the ArcGIS Generalize tool, but the roads database specification, local/cognitive knowledge, and other ancillary information were employed to characterize the thematic properties. The information was used to determine which road should be deleted or merged in the generalization process in order to retain a certain percentage of roads at a smaller scale. In this regard, Choi and Li (2001) believe that incorporation of thematic information can control geometric operations (e.g. simplification) and topological information (e.g. connectivity). It needs to be used to validate the geometric operation and to control the quality of map generalization. For example, the use of Elimination involves removal of very short branches and unimportant roads, which is mainly based on cognitive knowledge. The generalization processes such as Elimination and Typification result in reduction of the complexity and the density of roads (Figure 3). Qualitative or quantitative assessment of generalization results is necessary. In this study, the derived maps were compared with existing maps. It involved experienced cartographers to carry out a qualitative visual assessment/inspection of the results. From cartographic intuition and perspective through the consultation with experienced GIS operators/cartographers, their evaluation through independent assessments helped to choose the best levels of tolerance ranging from 10 to 70 that produced satisfactory simplification outputs. The tolerance value (positive number greater than zero) determines the degree of line simplification. It reflects the distance that two points in a road segment can be apart and still be considered the same to produce generalized data. It requires setting the tolerance equal to or greater than the threshold of separation (the minimum allowable spacing between graphic elements). The threshold value relates to the map coordinate system. For geodetic data (it is identified by longitude and latitude coordinates), the tolerance value is expressed as a number of metres, whereas for non-geodetic data, the tolerance value is a number of the units that are associated with the coordinate system associated with the data. Various tolerance parameters (values) were tested in the course of road network generalization. Based on several tests, it became possible to optimise the process of map derivation by semi-automated cartographic and database generalization by means of simplification and thresholding to derive small-scale maps from a master database. Two different values based on the introductory experiment demonstrated that threshold parameter values of 25 and 35 were the most useful tolerance levels to be applied for the 1:500 000 and 1:1 000 000 maps, respectively. Similar tolerance levels (20 using Douglas-Peucker and 40 with Reumann-Witkam algorithms) have been reported by Nakos (1999) to generalize a coastline database of Greece. Figure 5 presents the effects of thresholding versus original data where the impact of the Bendsimplify operations with a tolerance of 25 and 35 to produce GEODATA 1:500 000 and 1:1 000 000 respectively is demonstrated. Obviously, using higher threshold levels results in removal of extra vertices in the road segments that are shown in Figure 5b. It required manual editing to correct the over simplification.

(a) Bendsimplify

( b) Bendsimplify

Figure 5. The impact of simplification using Bendsimplify as seen in a portion of the road database; Figure 5b demonstrates over simplification output. In the graph (Figure 5b), vertices in the original line are shown in white dots. Clearly, applying higher threshold (threshold 35) levels resulted in removal of extraneous vertices in the road segments that are shown here. It required manual editing to correct the over simplification, whereas in Figure 5a, the simplification (threshold 25) led to satisfactory results from a cartographic perspective. Despite the fact that results from ArcGISs DP and Bendsimplify are satisfactory, they do not support dynamic generalization. Therefore, there is an opportunity to evaluate other generalization systems such as Intergraphs DynaGen and Laser-Scans Clarity software to derive multi-scale GEODATA in future investigations. It is worthwhile to note that the importance of expert systems application into generalization operations and cartographers experience has often been ignored. Kazemi et al., (2004a; 2004b; 2005) highlighted the importance of integrating cartographers knowledge with the generalization system that facilitates the development of a powerful, flexible and robust expert system capable of composing, editing, exhibiting and demonstrating an innovative method for semi-automated road network generalization. A comprehensive evaluation of generalization systems and their performance is essential to marry the cartographic knowledge from experts and bring this into a generalization framework. 5 Conclusions and Future Research Generalization operators in the ArcGIS software are tested to generalize roads by employing the above methodology for derivative mapping. The method is empirically tested with a reference data set consisting of several roads, which were generalized to produce outputs at 1:500 000 and 1:1 000 000 scales. According to the visual interpretation and quantitative analysis, the results show that the derived maps are satisfactory when compared with the existing small-scale road maps such as the Global Map at scale of 1:1 000 000. The derived 1:1 000 000 map was then compared to 1:1 000 000 Global Map. But there was no existing 1:500 000 reference data in order to compare the derived 1:500 000. As the methodology is only tested on roads it is worthwhile to extend it to other complex cartographic datasets such as drainage networks, power lines and sewerage networks in order to determine the suitability. Additionally, various kinds of linear, areal and point cartographic entities (e.g. coastlines, rivers, vegetation boundaries, administration boundaries, land cover, localities, towers, and so on) should also be studied. Previous studies (e.g. Mroz et al., 1996; Nakos, 1997 and 1999; Costello et al., 2001; Kazemi, 2003; Lee, 2004) have also achieved good results in automatic generalization using ArcGIS; however significant manual work is required since ArcGISs Pointremove generalization algorithm (Douglas-Peucker, 1973) and Bendsimplify algorithm do not support dynamic generalization. Other generalization systems such as Intergraphs DynaGen and Laser-Scans Clarity support such generalization and enable users to derive a multi-scale database from a master database (Watson and Smith, 2004; Kazemi et al., 2005). Therefore, it is intended to enhance the proposed methodology by testing and incorporating such tools in future research.

Acknowledgements The lead author would like to thank the School of Surveying and Spatial Information Systems and the Faculty of Engineering of the University of New South Wales for sponsorship of this research through a PhD Scholarship. Professor Graeme Wright of Department of Spatial Sciences, Curtin University of Technology contributed constructive and meticulous review of manuscript of this paper. We are also grateful to Dr Dan Lee of ESRI Redlands California and Dr Alan Forghani of ACRES Geoscience Australia, and Professor Gary Hunter of University of Melbourne for participating in various discussions on the topic of generalization and reviewing this manuscript, and we wish to thank other two anonymous referees for their critical and valuable comments on the paper. The data are provided by Geoscience Australia. The authors gratefully appreciated the provision of these recourses. We acknowledge that the content of this submission has been accepted for publication in the Journal of Spatial Science (paper accepted 20 January 2007). References AGENT C 1999 Selection of Basic Algorithms. Technical Annex, AGENT Consortium, pp. 1-31. Arnold L A and Wright G L 2003Analysing product-specific behaviour to support process-dependent updates in a dynamic spatial updating model. In Proceedings of Spatial Sciences 03 Conference, Canberra, ACT, September, pp. 1-12. Barrault M 1995 An automated system for linear feature name placement which complies with cartographic quality criteria. In Proceedings of Autocarto 12 in Charlotte, NC, ACSM/ASPRS, pp. 321-330. Choi Y H and Li Z 2001 Correlation of Generalization Effects with Thematic Attributes of Cartographic Objects. Department of Land Surveying and Geo-Informatic, the Hong Kong Polytechnic University, Hong Kong, pp. 1-26. Costello S Broers R and Swift A 2001 Helping the global community to resolve global problems through spatial information: AUSLIGs contribution to the Global Map project, In Proceedings of AURISA 2001 - the 29th Annual Conference of AURISA, Grand Hyatt, Melbourne, VIC, 19-23 November, pp. 1-11. Cromley R C 1992 Digital cartography. Prentice Hall, Englewood Cliffs, New Jersey. Douglas D H and Peucker T K 1973 Algorithms for the reduction of the number of points required to represent a digitized line or its caricature. Canadian Cartographer, Vol. 10, No. 2, pp. 112-122. Dutton D 1999 Scale, sinuosity and point selection in digital line generalization. Cartography and Geographic Information Science. 26(1), 33-53. ESRI 2005 ArcGIS Documentations. ESRI Press, 2005, Redlands, CA, USA. Forghani A 2000 Semi-automatic detection and enhancement of linear features to update GIS files, In Proceedings of XIXth ISPRS Congress, July, Amsterdam, Netherlands, pp. 1-8. Forghani A 1997 A knowledge-Based Approach to Mapping Roads from Aerial Imagery Using a GIS Database. PhD Dissertation, Department of Surveying and Spatial Information Science, The University of Tasmania, Hobart, Tasmania, November 1997, Australia, pp 1-180. Hardy P Briat M O Eicher C and Kressmann T 2004 ICA Database-driven cartography from a digital landscape model, with multiple representations and human overrides. In Proceedings of 04 ICA Workshop on Generalization and Multiple Representation, Leicester, UK, pp. 1-6. Hardy P Briat M O Eicher C and Kressmann T 2004 Database-driven cartography from a digital landscape model, with multiple representations and human overrides, In Proceedings of ICA Workshop on Generalization and Multiple Representation, August, Leicester, UK, pp. 1-6. Haire L 2001 An optimization Approach to Cartography Generalization. Doctoral Thesis, Department of Surveying, Lund Institute of Technology, Lund University, Sweden. Iwaniak A and Paluszynski W 2001 Generalization of Topographic Maps of Urban Areas. Institute of Geodesy and Photogrammetry, Wroclaw Academy of Agriculture, Wroclaw, Poland, pp. 1-7. Jenks G F 1989 Geographic logic in line generalization. Cartographical, Volume 26 Number I (Spring): p. 27-42. Kazemi S 2003 A Generalization framework to derive multi-scale geodata. In Proceedings of 03 Spatial Sciences Conference, September, Canberra, Australia, pp. 1-12.

10

Kazemi S Lim S and Rizos C 2004a A review of map and spatial database generalization for developing a generalization framework. In Proceedings of XXth Congress of 04 the Int. Society for Photogrammetry & Remote Sensing, Istanbul, Turkey, in DVD Commission 4, pp. 1221-1226. Kazemi S Lim S and Ge L 2004b Is automated generalization there yet? In Proceedings of 04 12th Australasian Remote Sensing and Photogrammetry Association Conference, Fremantle, Western Australia, pp. 1-11. Kazemi S Lim S and Ge L 2005 Integration of cartographic knowledge with generalization algorithms. In Proceedings of 05 IGARSS, Seoul, Korea, pp. 1-4. Kazemi S Lim S 2005 Generalization of road network using Intergraph DynaGen system. In Proceedings of the 05 SSC Spatial Intelligence, Innovation and Praxis: The national biennial conference of the Spatial Sciences Institute, Melbourne: Spatial Sciences Institute. ISBN 0-9581366-2-9, pp. 1-11. Kazemi S and Lim S 2005 Deriving multi-Scale GEODATA from TOPO-250K road network data, Journal of Spatial Science, 2007, pp. 1-12 (paper accepted 20 January 2007). Lee D 2003a Personal communication. ESRI, Redlands, USA. Lee D 2003b GEOPRO: Generalization within a geoprocessing framework. In Proceedings of 03 GEOPRO Workshop, Mexico City, November, pp. 1-10. Lee D 2004 Geographic and cartographic contexts in generalization. In Proceedings of ICA 04 Workshop on Generalization and Multiple representations, August, Leicester, pp. 1-9. Mackaness W Edwardes A and Urwin T 1998 Self evaluating generalization algorithms for the derivation of multi-scale products from UKBORDERS data of the census, In Proceedings of the 3rd International Conference on GeoComputation, University of Bristol, United Kingdom September, pp. 1-6. McMaster R B 1989 The integration of simplification and smoothing algorithms in line generalization. Cartographica, Vol. 26, No. 1, pp. 101-121. Moreno M Torres M Levachkine S and Fajard I 2002 The Automatic Generalization of the Multiscale Geographic Information. WWW document tgis.esri.com/library/userconf/proc02/pap0173/p0173.htm. Mroz M Berger K Bechere R Packman J 1996 The Defence Mapping Agency (DMA) World Vector Shoreline Plus (WVSPLUS) Revision and Conversion Project. WWW document http://gis.esri.com/library/userconf/proc96/TO250/PAP237/P237.htm. Muller J C 1995 GIS and Generalization: Methodology and Practice. Taylor & Francis. Nakos B 1999 Comparison of manual versus digital line generalization. In Proceedings of 99 Workshop on Generalization, Ottawa, Canada, pp. 1-8. NMD 1999 Topographic Data and Map Specifications. Technical Specifications, Version 3.4, National Mapping Division, Geoscience Australia. Peschier J 1997 Computer Aided Generalization of Road Network Maps. M.Sc Thesis Department of Computer Science, Utrech University, pp.1-25. Pfer T and Pillewizer W 1966 The principles of selection. The Cartographic Journal, Vol. 3, No. 1, pp. 1016. Rusak E and Castner H W 1990 Hortons ordering scheme and the generalization of river networks. Cartographic Journal, 27, pp. 104-112. Sarjakoski T L Lehto M Sester A Illert F Nissen B. Rystedt and Ruotsalainen R. 2002 Geospatial infomobility services: a challenge for national mapping agencies. In Proceedings of 02 International Symposium on Geospatioal Theory - Processing and Applications, Ottawa, Canada, pp. 1-6. Shea K S and McMaster R B 1989 Cartographic generalization and how to generalize. In Proceedings of Auto-Carto 9, Baltimore, Maryland, March, pp. 1-6. Skopeliti A and Tsoulos L 2001 A knowledge based approach for the generalization of linear features. In Proceedings of 20th International Cartography Conference, Beijing, China, pp. 1-10. Taylor G 2005 Line simplification algorithms. WWW document http://www.comp.glam.ac.uk/pages/staff/getaylor/papers/lcwin.pdf. Visvalingam M. and Williamson P J 1995 Simplification and generalization of large-scale data for roads - a comparison of two filtering algorithms. Cartography and geographic information systems, Vol. 22, No. 4, pp. 264-275. Visvalingam M. 1999 Aspects of line generalization. pp. 1-15. WWW document www2.dcs.hull.ac.uk/CISRG/ica-oldhtm.

11

Wang Z Muller J C 1998. Line Generalization Based on Analysis of Shape. Cartography and Geographic Information Systems, 25, pp 3-15. Watson P and Smith V 2004 ICA Interoperability of agent-based generalization with open, geospatial clients. ICA Workshop on Generalization and Multiple representation 20-2, Leicester, pp. 1-5. Zhao H Li X and Jiang L 2001 A modified Douglas-Pecuker simplification algorithm. In Proceedings of 01 International Geoscience and Remote Sensing Symposium, Sydney, Australia Volume:2, pp. 1-3.

12