Presented at AURISA 2001 - The 29th Annual Conference of AURISA

Grand Hyatt, Melbourne, VIC, 19-23 November 2001

 

 

 

Translation between Multiple Representations of

Spatial Data

 

 

Agus Batara1, Bharat Dave2, and Ian Bishop3

 

1 Faculty of Architecture Building & Planning

The University of Melbourne

3010 Australia

Phone: 03 8344 0400

Email: a.batara@pgrad.unimelb.edu.au

 

2 Faculty of Architecture Building & Planning

The University of Melbourne

3010 Australia

Phone: 03 8344 7259

Email: b.dave@architecture.unimelb.edu.au

 

3 Centre for Geographic Information Systems and Modelling

The University of Melbourne

3010 Australia

Phone: 03 8344 7500

Email: idbishop@unimelb.edu.au

 

 

ABSTRACT

 

aurisa@ausconvservices.com.au

 

 

Urban planning and urban design involve collaboration of diverse participants with multiple agendas and multiple criteria. The participants typically use multiple representations of spatial information to derive inferences and insights about the planning problems, leading to a shared decision-making process. To support such multidisciplinary work, this paper proposes new computational approaches to spatial data representation and techniques for translation between representations. These approaches are designed to support the interrelated interests of planning participants. Prototype implementation and evaluation are conducted to test and validate the proposals.

 

KEYWORDS: urban design and planning, collaboration, negotiation, representations, translation.

 

 

INTRODUCTION

 

Architecture/ urban design is an exploration of trade off between multiple agenda, multiple criteria and at its core, a negotiation mechanism among its multi-participants (Gross et al, 1997). Planning and urban design involves collaboration between multiple disciplines and interests such as urban planners and designers, geographers, sociologists, economists as well as public. Each party comes with a different agenda, design criteria, and its preferred representation of pertinent urban data. Polarization of views becomes a fundamental characteristic of such collaborative work and hence necessitates the use of multiple representations to resolve design decisions from different perspectives.

 

Graphic representations are ubiquitous in planning and urban design. However, the complexity in representing planning and urban design information arises from conflicting agendas in every design decision. In most cases, these complex and interrelated agenda cannot be presented by any single means of representation. A representation may satisfy one agenda while not satisfying the others. As a result, representations cannot be separated from associated agenda, disciplines, and individual preferences.

 

As an example, consider representations of traffic circulation. Is it easier to understand through a diagrammatic map or a walkthrough animation? Such questions cannot be answered meaningfully without knowing the audience and purposes of presentation. For a conceptual understanding, audiences may prefer diagrammatic maps that highlight routes and sequence of interchanges as the most communicative representation. Whereas at other times, diagrammatic maps representing different volumes of traffic loads may not convey much information about variability of traffic without associated time-based visualizations. The same question results in different answers when we take into account the needs of different audiences and different purposes of presentation. Urban designers may choose time-based animation as the most compelling tool in visualizing the aesthetic experience of the street. While developers may find tabulated data and charts to be more useful in demonstrating that a street’s minimal corridor results in maximum sellable land.

 

Despite growing recognition of the importance of multiple representations in planning and urban design, there is a lack of computational support in the collaborative phases of design and planning practices. At present, most of the computational support is mainly concentrated on implementation of database and drafting support that are used in isolation from other related disciplines. While some applications support design collaboration, there still exist key limitations in current planning and design support systems (Dave and Bishop, 2000). Many applications allow data to be viewed in multiple representations such as map, table, chart and 3D model. However editing takes place in one representation only. GIS software and some computer aided design applications allow translation of data between spatial (i.e. graphic) representation and non-graphic databases. In these applications, relations among data traverse in one direction only, e.g. from spreadsheet to graphic representation. To extend bi-directional traversal of data among different representations, there is a need to systematically identify possible translations between representations and consequent loss, if any, of information. 

 

In this paper, we investigate the need for multiple representations in the shared decision-making process. Using planning and urban design scenarios, we look at problems and possibilities in qualitatively improving collaborative aspects of the work. The paper is organized as follows. The next section describes a typical urban design and planning project scenario. It also illustrates the necessity for use of multiple representations to increase understanding between and within discipline(s) in collaborative decision making processes and thus articulates motivations of this project. Based on this scenario and literature review, the next section outlines taxonomy of graphic representations and significant research issues. It is followed by a description of our prototype system, REX, its implementation details and a preliminary evaluation of our project. It is followed by a discussion of various conceptual issues involved in supporting collaboration and identification of limits to computational translation between multiple representations.

 

 

MOTIVATIONS

 

There are few empirical studies of real urban design and planning projects that document in detail how a particular agenda prevails over another and the use of graphic representation in that process. Some of the reasons for this lack include the following. Design process is iterative, many negotiations (if occurs) are undertaken spontaneously or without formal representations, and in-process representations are usually discarded at the end of the project to be replaced with final documentation. Thus much valuable data about how design and planning process unfolds are typically lost.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 1. Examples of representations used in typical urban design and planning projects (from Pyrmont project).

 

Note : Images in Figure 1 and Figure 3 are the work of Master of Urban Development and Design 97 students at the University of New South Wales. One of 3D model images in Figure 1 is taken from Lend Lease Pyrmont Site Master Plan, October 1998.

 

To investigate the relationship between representations and their uses in development of design projects, the following presents a typical project development scenario. The project is based on a design studio in urban development and design in which teams of students worked together. The project brief called for development of a mixed-use community plan on a twelve hectares old sugar refinery site at Pyrmont, Sydney. The project comprised three related tasks: survey and analysis, development of the master plan, and feasibility study.

 

During the first stage of survey and analysis, students worked in groups of two or three, investigating different aspects of site analysis. The issues included natural landscape setting, history and planning control, economic and market, social demographic, traffic and access, and political decision-making. To deal with the diversity of issues, use of multiple representations becomes a necessity during this stage. For example, thematic maps were dominantly used for landscape and traffic analysis whereas economic and social demographic analysis relied upon tabulated data and graphs to accommodate statistical and quantitative data. In contrast, the development of a master plan involved use of measured and scaled graphic representations and physical, scaled-down models at an appropriate level of detail. The feasibility study required both detailed and coarse graphic representations to test spatial feasibility of individual building volumes. Tables and charts were also used to demonstrate its economic feasibility in terms of investment and returns on the property over time.

 

Although there are variations to the organization of project team and development paths, most urban design and planning projects share a number of characteristics.

 

 

 

 

 

 

Figure 2. Multiple representations underpin planning and design inference within and across the disciplines

 

The preceding gross generalization of urban design and planning projects suggests that facilitating multiple representations and translations between them can lead to a number of advantages. First, it will help remove partitions between different disciplines working together on a shared project. It will allow inferences to be made synchronously and dialectically with other design disciplines (indicated by horizontal direction between points 1-1 in Figure 2). Second, it will support inferences within each discipline using multiple representations (indicated by vertical direction between points 2-2 in Figure 2). To support such translation between multiple representations, we introduce a set of concepts in the following sections that are then used to develop and experiment with a prototype computational environment.

 

 

RESEARCH ISSUES

 

Ervin (Ervin, 1992) categorized graphic representations including diagrams, maps, graphs and pictures based on distinctive attributes of each representation as follows.

 

“Diagrams are abstract and schematic and are used to explore structural relationships between parts… Maps involve scaled representations using a consistent system of reference (e.g. coordinate system), and allow inferences about dimensional and spatial relationships… Graphs are concerned with representation of statistical and quantitative data…. Pictures are primarily concerned with impression, expression and realism.” (Ervin, 1992)

 

Jones et al, (1994) subsequently elaborated a similar taxonomy with some variation to that proposed by Ervin.  Recent computational advances augment both these categorization through introduction of 2D and 3D spatial representations that may allow either static or interactive manipulations. Pietsch (2000) has also reviewed issues of representational content and validity. She stresses that factors such as accuracy, abstraction and realism vary in importance through the design process. Analysis of the properties and data attributes contained within each representation allows us to identify the significant research issues to be resolved to permit translation among different representations.

 

 

1. Data Content

a. Abstraction

The first significant dimension of graphic representation is encapsulation of an appropriate level of abstraction. Ervin used abstract and schematic variables in differentiating diagram from scaled or measured map. Level of abstraction reflects the level of detail or grain-size of information contained in a particular representation. Level of abstraction tends to decrease as the design progresses although this may not hold true in all circumstances. It also varies in accordance with the purposes of presentation. Higher level of abstraction requires representation of only key information whereas lower abstraction contains much more information.

 

 

 

Figure 3. Examples of representation with different levels of abstraction and realism (from Pyrmont project).

 

b. Realism

Another dimension of graphic representations, which is closely related to but different from the level of abstraction, is degree of realism. It is a measure of proximity of representation to associated real world situation. Photograph is an example where representation has a relatively high realism. According to Ervin (Ervin, 1992), realism together with expression and impression make pictures better suited at encoding visual renditions of phenomena.

 

c. Data domain

The nature of domain knowledge constrains what may or may not be represented as well as how it can be represented. From a computational point of view, managing translation between different domains of data content can be seen as a matter of documenting, coding, querying and retrieving of information that may depend upon one or more discipline knowledge bases.

 

There are many potential problems in translations between representations. Sometimes, problems arise while translating representations within one discipline domain; at other times, problems arise when translations require mapping of information from one discipline domain to another. As an example, translating any change from table to graph does not raise any problems but moving in the other direction may require additional domain knowledge. Similarly, one may extract a table of doors and windows schedule from a 3D model, however it would not be easy to traverse in reverse. At other times, translations may cross various domains, for example, when an architectural layout requires a companion electrical or plumbing layout. In general, these translation bottlenecks arise from possible relations between different representations and they include: one-to-one, many-to-one, many-to-many, and one-to-many.

 

2. Data Types

The third research issue is managing interpretation across different data types. It encompasses interpretations between spatial and non-spatial representations, and quantitative and qualitative representations. A graphic representation such as map, for example, carries spatial information while others such as graphs, tabulated data or diagrams may not represent spatial information at the same level of detail. Between such different representations, only limited translations may be possible. When translation is from spatial to non-spatial representations, for example, spatial information may be lost in the process and it may be impossible to reverse that translation in an unambiguous way since there may be one-to-many possibilities.

 

Another example in translating different data types can be found in the translation between quantitative and qualitative representation. Ervin used the quantitative variable to distinguish graphs and tabulated data representation from other representations. In the context of translation between these different data types, the problems are similar to those involved in translation between spatial and non-spatial representations.

 

3. Context of Use

Planning is an iterative process comprising many generative and evaluative cycles. During the early stages of project development, the tentative nature of information can lead any changes and translations into generation of many planning and design alternatives. However as information becomes more definite and detailed with progressive development of the project, any changes and translations are most likely to be more constrained and localized. This is where manipulation of data attributes results in planning and design variants.

 

These two contexts set different constraints and translation options. Translations from a tentative representation may require vast numbers of interpretation. This can be problematic if required interpretation cannot be supported through existing dataset content or user’s input. However translations at a more constrained environment are likely to require less interpretation. This is since more definitive constraints and interpretations have been acquired in the existing data content.

 

The preceding discussion highlighted significant problems associated with translation of information across different representations that although sharing the same design context may originate from and be informed by entirely different knowledge bases of different collaborators. Our research is aimed at developing computational techniques that facilitate and support such translation between representations. A prototype was implemented to investigate these issues and is described next.

 

 

THE PROTOTYPE

 

 

 

Figure 4: REX User Interface consists of four windows representing Map, 3DView, Table

and Graph Representations

 

The prototype application, called REX (Representation EXchange), demonstrates translations between four different representations: Map, Table, Graph, and 3DView. REX combines two separate prototypes developed previously, which are Map-Graph (Dave and Bishop, 2000), and Map-3DView. In REX, we start with map/table representation in the form of an ArcView shape file, and reflect these two representations into graph and 3D representations. Manipulation can be done either through the table, graph or 3D windows and changes will be reflected onto other representations. For instance, while working in the 3D representation, it is possible to drag a building envelope up and down to change its height. The change feeds back to the attribute table for the corresponding polygon. The change is reflected onto both the map representation (though changes in polygon color) and the land use distribution shown in the chart representation. The same result can also be achieved by manipulation of the commercial/ offices/ residential floor columns in the spreadsheet window.

 

Navigation allows pan, zoom and rotate function to be performed in the 3D model window. This is done through execution of left, right or both mouse clicks. Zoom in, zoom out, pan, and window’s reset button are provided for the 2D map window. At present development, we have implemented manipulation of building height’s components (commercial, office, and residential) and related this manipulation into floor-space calculations. In future stages, we plan to include a wider area of applications such as economic feasibility, environmental impact assessment, or policy development.

 

 

1. Data Content

The problem in translating between different levels of data content is that a one-to-one relationship cannot always be established in each translation. Therefore interpretation has to be made in order to resolve ambiguities. This is the typical situation in any translation towards lower levels of abstraction, higher realism or general to specific domain. Unless the context of use provides adequate information, there are two possible options in carrying out interpretation in such one-to-many relations. First, is to define knowledge-based rules of generation to automate the translation. Second, is to require user input to resolve ambiguities.

 

In REX, there are four representations and many possible translations. Because no editing is available in the map view and this is tightly linked to the map attributes though the attribute table, this can be treated in some respects as a single representation: the ‘GIS view’. Here however we have treated them as separate entities because the potential exists to allow map editing though movement of vertices, this would then have clear consequences for all other representations. The approach to all possible translations is summarized in Table 1. To cope with one-to-many relations, for example, moving from the Map and Table to the 3Dview, we have incorporated knowledge-based interpretations to resolve ambiguity. In REX, we took the general assumption that each additional floor is represented by a 3m height’s increase in its model representation. This assumption could however be made explicit or varied through additional attribute columns. At present, in REX no manipulation can be initiated from Graph. For example, it is not possible to add commercial floors as the Graph is adjusted. However this capability, based on planning assumptions and user intervention has been successfully implemented in our previous Map-Graph prototype. This can be seen in the Figure 5.

 

 

Figure 5: Map-Graph allows many-to-one translation, through graph manipulation and additional user input.

 

 

 

To:

From:

MAP

TABLE

GRAPH

3D VIEW

 

MAP

 

-

 

Changes in map vertices will change values of area entity which will flow through to other views

 

Conventional abstraction of data as available in most GIS

 

Shape of extruded polygons changes accordingly

 

TABLE

 

Changes in polygon colour reflecting some chosen attribute

 

-

 

Conventional abstraction of data as available in most GIS

 

3D drawing based on explicit (number of floors) and implicit (floor height) attributes

 

GRAPH

 

Occurs via Table

 

Ambiguity in allocation of land use changes. Resolved by user choice of random, lowest cost or proximity based reallocation

 

-

 

Occurs via Table

 

3D VIEW

 

Occurs via Table

 

Explicit if floor types edited individually. Ambiguous if whole building height changed. Allocation can be based on dominant use, pro rate distribution, lowest cost, highest return

 

Occurs via Table

 

-

 

Table 1. Approach to possible translation requirements in REX prototype. (Plain text – not yet available, Bold text – available in REX, Italics– developed in earlier prototype) Extension to further representations – such as Diagram – can be accommodated.

 

 

2. Type of Data

In REX, data types can be differentiated as spatial or non-exclusive quantitative data (Map and 3DView) and non-spatial or quantitative data (Table and Graph). 3DView-Map and Table-Graph are two examples of translation within the same types of data. While 3Dview-Table or 3DView-Map-Table (and visa versa) are examples of translation between those two different data types. Table-Map, for example, passes the update of height table attribute so that it changes the color of the corresponding polygon.

 

The risk of data loss does not apply in REX as translations take place within the unified system.  Thus, manipulation of data attributes in table representation does not result in the loss of spatial data in Map or 3DView representations. This allows re-generation of each representation based on inference between the existing dataset and the current manipulated attributes.

 

3. Context of Use

We generate the initial dataset, map and spreadsheet, as an ArcView shape file. We manipulate the data in REX while receiving feedback in the four representations. Manipulation is REX’s context of use, while generation is undertaken outside of the prototype, using ArcView (polygon and table), CAD (polygons only), and Excel (table only) applications. The benefit of working in the prototype with a highly constrained environment is that translations are facilitated within the existing dataset. There is a minimum risk of data loss resulting from translations between different data types or data contents. The deficiency of this environment is that it limits manipulation outside the predefined variables. Within REX, for example, we have to regenerate the map and attribute table in ArcView shape file to deal with changes in geometry of the plan or an additional column for new attributes. This is one of the impediments that exist at the current stage of development. REX is therefore a good environment to generate design variants rather than design alternatives.

 

 

DISCUSSION

 

REX addresses several problems in translation between representations.  Some of the benefits of application and prospects for further extension include:

 

 

Consider the following application scenario. An architect proposes a development plan in his drawings then a financial planner comes with her development scenario. This does not meet the architect’s plan. How does the architect translate his plan to that of the financial planner? Of course, one of them may be able to compromise and redo the work, but should not the architect just initiate his plan after the financial plan is finalized or the other way around? The sequential and segmented nature of traditional decision making in design development results in fragmented or cyclical iterations which could be avoided with appropriate tools.

 

If we assume that different representations present the preferences of different disciplines, then the benefit in using an application such as REX in a multi-disciplinary work environment can be indicated as:

 

REX assists communication by disseminating understanding across the disciplines while the knowledge-based interpretation that is either embedded in, or used in, executing users’ input operations still requires the expertise of the disciplines. Therefore REX opens new possible inferences to audiences outside the discipline without taking over the role of the discipline itself.

 

 

CONCLUSION

 

There are two findings that emerge from the immediate experimentation with multi-representation prototypes such as REX.

 

First, the prototype demonstrates that the 3Dview-Table translation has generated a possibility of two-way communication. This is a breakthrough in planning computation support, bearing in mind that GIS currently implements one-way translation only. Furthermore the two-way communication prospectively allows multiple iterations in planning methods, which is previously constrained in the GIS application. This supports the recognition that planning is a trade-off mechanism between its many participants, agendas and criteria. Hence multi-iterative editing becomes a feature that has to be addressed in any negotiation process.

 

Second, by using multiple representations in planning, inference may be dynamically seen through various levels of abstraction and types of data.  For those disciplines that are previously limited in their representation options, this opens a new way of looking and working beyond their established conventional practice. Some of the areas of the planning process that multiple representations can prospectively enhance are feasibility study, policy development, environmental impact assessment, community participation, as well as visualization.

 

Proposed further research and development includes:

 

Although it is outside this research aim, the ‘ideal’ would be for GIS and/or CAD software to extend their functionality to permit both generation and manipulation of design proposal through a range of representations to support both professional collaboration and public participation in the process.

 

 

ACKNOWLEDGEMENTS

 

This research is supported by the Australian Research Council Large Grants program. The prototype system, REX, described in the paper was implemented by John Bird. Earlier versions of the prototype and some other components were developed earlier by a group of students in Software Engineering program at the University of Melbourne.

 

REFERENCES

 

Dave, B. (1993) CDT: A Computer-Assisted Diagramming Tools. In Flemming and Wyk (ed.), CAAD Futures ’93, Elsevier Science Publishers, pp. 91-109.

 

Dave, B. and Bishop, I. D. (2000) Multiple representations for diverse perspectives: collaboration in urban design. Proceedings of Spatial Data Handling, Beijing, 10-12 August, IGU Study Group on Geographical Information Science, pp. 3b.37-3b.49.

 

Ervin, S. (1992) Intra-Medium and Inter-Media Constraints. In G. Schmitt (ed.), CAAD Futures ’91, Branuschweig: Vieweg, pp. 365-380.

 

Gross, M., Parker, L., and Elliot, A. (1997) MUD: Exploring Trade-Offs In Urban Design. In R. Junge (ed.), CAAD Futures 1997, Kluwer Academic Publishers, Netherlands, pp. 373-387.

 

Jones, R., Edmonds, E.A., and Branki, N.E. (1994) An analysis of media integration for spatial planning environments. Environment and Planning B, v.21, pp. 121-133.

 

Pietsch, S. (2000) Computer visualization in the design control of urban environments: a literature review. Environment and Planning  B, v.27, pp.  521-536.