Areal Units and the Linking of
Data |
|
|
Areal Units and the Linking of
Data |
|
The following text is from Chapter 2 in Worrall L (ed) Spatial Analysis and Spatial Policy Using Geographic Information Systems (Belhaven Press, London), 1991, 12 - 37.
Contents
|
|
Many users of Geographical Information Systems (GIS) may be developing applications without fully appreciating that the current family of GIS are based on a rather narrow geometric view of spatial entities. Consequently, users may not be able to evaluate whether the methodology offered by a GIS is adequate and appropriate for the problems at hand. The Chorley Report on the handling of geographic information (DOE, 1987) has had a catalytic effect in promoting the awareness and use of GIS in the UK. GIS are advocated as powerful tools for integrating data from different sources. Data from different sources tend to refer to different sets of basic spatial units (BSUs). BSUs are the smallest spatial units for which information is collected and/or made available. The choice of basic spatial units is thus a critical decision in all GIS developments and applications, whether manual or computerised. Digital cartography and GIS are concerned with point, line and areal entities. For the sake of simplicity, this chapter will confine itself to issues relating to areal units. Variations in the geographic scale and focus of applications, the nature of phenomena and the regular character of data collected automatically, have all worked against, and continue to work against, the convergence and unification of areal units into a single standardised system. The Chorley Report recommended two ways forward, namely the use of GIS to link data on demand and the adoption of standards to facilitate the linking of socioeconomic data. These recommendations rely to some extent on the use of space as the common, reference dimension within the current family of GIS. It is important to remember that all phenomena exist within the multidimensional framework of space, time and other descriptive dimensions. Users should be aware of the relative importance of these dimensions in their problem context. We measure phenomena in order to understand and utilise or influence them. The very act of measurement involves a variety of judgements and decisions. The design of the data collecting units involves a consideration of alternative schemes for segmenting the multi-dimensional framework for purposes of observation and recording (Openshaw, 1990). The units for which data are recorded may be derived through the division of the whole or the aggregation of primitive data entities. Implicit in the term, basic spatial unit, is the notion that a data collecting framework can be based on a classification of space. However, data collection and aggregation need not be based exclusively on spatial (or geographic) criteria. The temporal dimension may provide a better framework for some other classes of applications which need either to identify critical events (as opposed to critical areas) or to predict, monitor and control the consequence of events, such as pollution or progressive drought on water supply and outbreaks of fire in forest and park management. In the UK, the Institute of Hydrology has decided to develop its own in-house and hydrologically appropriate GIS partly because of the known limitations of proprietary GIS for modelling the hydrological cycle and for operational purposes. The framework for collecting data may also be determined by some other functional dimension. For example, the monitoring and refinement of policies for encouraging the uptake of primary health services is better served by data aggregated to service points, such as medical general practitioners and treatment centres, rather than relatively gross spatially defined administrative units, such as wards or enumeration districts. Even if the spatial dimension is the most appropriate basis for organising data collection and/or aggregation, the concepts underpinning the manipulation of BSUs within GIS may not be appropriate. GIS functions, such as overlay analysis used for identifying critical areas, reduce data to common spatial denominators to facilitate comparison. This effectively involves a reduction of spatial phenomena to their geometric and geographic levels of description for geometric manipulation. However, areal units are not just different in geometry and geography, they are also different conceptually. Variations in the nature of underlying phenomena and of available data are ignored by GIS. Currently, GIS only provide some geometric tools for spatial analysis. Users are ultimately responsible for determining whether the available tools are adequate and appropriate for their purposes. We have to consider whether the disparate data sources required for policy analysis are compatible and comparable in conceptual and not just geometric terms (see O'Brien, 1990). Although data may be tagged with quality information, describing their source and geometric accuracy (Blakemore, 1990), most systems cannot yet cope adequately with numeric uncertainty, let alone conceptual variations. We then come to the question of the definition of areal units. In what respects do areal units differ from each other? What are the implications of the differences? Standardisation is intended to facilitate the linking of data but will standardisation compromise other objectives? Before we consider these questions we need to consider the conceptual, functional and pragmatic reasons which prompt us to use BSUs as data collecting devices. Classification is a form of simplification and appropriate simplifications serve to further our understanding of complex issues and problems. The physical environment appears to consist of meaningful typologies. The existence of typologies enables us to study samples in a cost-effective manner and to make inferences about the whole. Spatial samples may be based on a number of schemes, including the regular grid. Data are also averaged for unit areas of constant size and shape when data collection is automated as in satellite remote sensing. Spatial averaging also results in smaller, more manageable databases. Spatial classifications are also engineered for functional reasons, for example, for the purposes of administration. Furthermore, in Britain, the reporting of personal data is restricted by a traditional respect for privacy and confidentiality and more recently by the 1984 Data Protection Act. Also, many policy-related and commercially-oriented analyses are concerned with identifying target areas, containing target populations, rather than with identifying individuals or individual households per se. Consequently, data relating to persons and households are released in aggregate form for a set of areal units which already exist for operational purposes or which are specially designed for purposes of reporting data (for example Census enumeration districts). A recent decision to release a sample of anonymised records (Marsh et al., 1988) from the 1991 Census in the UK (OPCS, 1990) marks a major change in policy which will facilitate the more flexible analysis of UK census data. In the UK, there is at present a variety of areal units, which correspond to units of observable phenomena, to functional or administrative units, to measurement units or to units for reporting sensitive data. As with other designed artefacts, the quality of a classification depends upon its fitness for use, especially its primary use, though GIS applications are often not the primary use for which such systems have been designed. Many applications, particularly those concerned with diagnostic and prescriptive analyses, need to devise appropriate data collecting units. Indeed, it is often necessary to utilise some non-spatial dimension as described earlier. Why does the variety of areal units pose a problem? Since the above units are problem-oriented and designed to meet the needs of a disparate array of primary users, they are seldom spatially coincident. Academic and commercial users of government collected data are attracted by their proxy value, particularly when used with data from other sources. For example, the 1981 population census small-area statistics (SAS) have been used very extensively in conjunction with market research and/or a firm's own data in locational analyses, marketing, retailing and advertising (see Brown, 1989, for a review of these applications). There is a growing awareness within government departments of the commercial value of their information. The government's Tradeable Information Initiative was set up to encourage the release of more government information of value to the private sector and to other government departments. However, data from
different sources cannot be compared easily or accurately if they relate
to very different sets of areal units. The Chorley Committee (DOE, 1987)
advocated the need for designing a standard set of spatial units because
it was concerned with increasing the usability of routinely collected
data. The move towards standardisation is unconcerned with the use of
data units within an organisation and is mainly concerned with the
design of reporting units. It may, however, be possible that the
proposed reporting units are smaller than existing operational units.
This requires that organisations hold proprietary data in a relatively
disaggregate form to satisfy both operational and reporting purposes. The aims of this chapter
are to examine why the dispositions of spatial phenomena, of geographic
applications and of spatial data do not always facilitate the automatic
geometric linking of spatial data; to encourage critical user assessment
of the value and limitations of GIS methods in the context of their
primary applications; to discuss the variable properties of areal units
and their implications; and to encourage due consideration of the merits
and drawbacks of alternative systems for deriving BSUs. |
Issues relating to areal units Variations in the geographic scale and focus of applications, the nature of phenomena and the regular character of data collected automatically, have all worked against, and continue to work against, the convergence and unification of areal units into a single standardised system. Areal units are not just different in geometry and geography but they are also different conceptually. This has led to variations in perceptions of the required quality of data and their fitness for use. The design of areal units is influenced by these variations in concepts and requirements and other known properties of areal units. The design of BSUs must, therefore, follow from an analysis of user needs and be determined by the nature of the problem at hand. |
Conceptually different areal units The term BSU is consistently used within all applications to denote the spatial primitive for which data is collected, aggregated and stored within an information system. The term `spatial primitive' represents conceptually different entities within different classes of applications. This is because different applications operate at different levels within the scenario of GIS and are concerned with different levels of abstraction from spatial reality. Even if we restrict ourselves to areal entities, some applications need to manipulate information relating to individuals and/or households and operate at the address or unit postcode level. Other applications are more concerned with properties and their land parcels. Yet others are more concerned with sociospatial structures as revealed by aggregate statistics for small areas, such as enumeration districts or even wards. The same statistics may be studied at grosser levels, such as for larger administrative areas, when the emphasis is on national trends. There is at present considerable confusion over the meaning of the term GIS (Walford et al., 1989; Visvalingam, 1990). Even if we ignore the confusion resulting from over-zealous promotional activities, the term GIS will mean different things to different people. This is not surprising given the variety of geographic phenomena and the diversity of users and uses. Every map is a type of GIS, but the term, GIS, is increasingly associated with computer-based information systems which include facilities for the storage, analysis and display of spatial data from diverse sources. Some GIS, especially those developed in-house by end-users, tend to address specific needs. The so-called state-ofthe-art GIS, on the other hand, seek to be all-singing and all-dancing and to serve a wide range of applications. The term GIS has also come to refer to the technological discipline underpinning a variety of information systems, including Land (and Property) Information Systems (LIS), Environmental Information Systems (EIS), Network Management Systems (NMS), Demographic Information Systems (DIS) and others. In turn, these classes of GIS subsume a variety of problems, which involve conceptually different types of areal units which we will now consider. LIS focus directly or indirectly on land (see Yeh, 1990). All human activity and building work must respect the rights to land and the restrictions upon its use. Land and property information systems tend to operate on suitably defined land parcels. The term `land parcel' is usually associated with spatial units on large-scale maps. It is used in various ways to denote, for example, a unit of landownership by Her Majesty's Land Registry (HMLR), a space subject to tax by the Inland Revenue or a unit of land-use for the purposes of, say, the Annual Census of Agriculture (Dale, 1988). These various usages of the term land parcel correspond to sets of application-defined spatial units which are usually depicted on separate map sheets or coverages. The multi-purpose cadastre, defined as the `framework that supports continuous, readily available and comprehensive land-related information at the parcel level' (Vrana, 1989, p. 34), is missing in the UK. The GISP (General Information Systems for Planning; see HMSO, 1972) initiative examined the requirement for a standard spatial unit and recommended the rating hereditament for urban areas and OS land parcels for rural areas. However, this does not in itself solve the problems of linking data. Property information systems have tended to use a Unique Property Reference Number (UPRN) as an address code which allowed the unique identification of every property within the area of coverage. A UPRN provides the common reference to link together different files containing information relevant to property and planning. The Chorley Report provided some examples of property referencing systems (DOE, 1987, p. 90 and Appendices 5 and 7). Here, the cross-referencing of data is facilitated by the use of consistent names for the same units on the ground. The economic benefits of having a property register is reported to have convinced local politicians to continue with and to expand the Tyne and Wear joint information system (DOE, 1987, p 167). Although utilities and local authorities use common base maps, they tend to operate with different sets of areal units on separate coverages (see Bromley and Coulson elsewhere in this volume). In Britain, the 1950 Public Utilities Streets Works Act requires that public utilities exchange information concerning their mains and plant records. This Act has forced a collaborative approach by the National Joint Utilities Group (NJUG) towards solving the problems of inter-coverage comparisons within computerised systems (Ives and Lovett, 1986). The NJUG approach uses identical copies of base maps, digitised by the Ordnance Survey of Great Britain, but each utility is responsible for entering and managing its own information on a separate layer. Standards have been formulated for the exchange of data between partners. Thus, the linking of data is achieved visually by registering data in the various layers to backdrop base maps. In contrast, the GIS for Northern Ireland (Brand, 1988) is designed to be a single integrated system, facilitating the automatic manipulation of shared data. The administrative structure in Northern Ireland expedites this approach. Also, some 80 per cent of the information required for management of municipal and utility functions and for provision of services is common to more than one organisation. This necessitates procedures for ensuring non-duplication of data conversion and the future integration of data. The topographic database consists of link-and-node structured points, lines, polygons and objects (see Kirby et al.,1989 for an explanation of topologically structured databases). In a multi-source or corporate GIS such as this, the plots of land which result from the intersection of various functional units on different coverages need to be assigned unique names. The term `primitive region' has been used to denote this spatial primitive (Kirby et al., 1989). Within this scenario, user-defined land parcels form higher-level aggregate units consisting of these primitive regions. The identity of the primitive region can then be used to cross-reference data in different files through use of a gazetteer- which is basically a type of codebook or index to codes (see DOE, 1987, p. 166). In Northern Ireland, all spatial entities are given a 12-figure Irish grid reference and this unique key is used to link the spatial and alpha-numeric descriptions of application-oriented objects. The suggestion is that user organisations will develop their own specialised databases and that the structure of the topographic database would enable them to combine the spatial and aspatial descriptions later. In Britain, one would have expected the computerisation of the activities of the HMLR and of other organisations such as the utilities and planning agencies to have stimulated a drive towards a systematic classification, referencing and use of land parcels and primitive regions. However, this involves a level of collaboration and cooperation which seems to run cross-grain to the `separate layer' mentality of British administrative organisations. The central government `Next Steps Initiative' to reorganise the Civil Service may consolidate rather than dilute these partisan interests. Under this initiative Her Majesty's Land Registry became an Executive Agency on 2 July 1990 and this has provoked some interesting comments on the need for UPRNs. Rowley, of the Association of Geographic Information (1990, p. 53), expressed the view that he was unsure whether the `market' requires a set of land parcel boundaries in digital form or a coding scheme for UPRNs that users can then quietly abuse in their own GIS. He suspected that today's requirement is merely to have an easier way of building land parcels that suit particular functional needs. He pointed out that the demise of domestic rating, the introduction of the Community Charge and the suggestion of a Local Land Tax all have an effect on appropriate definitions for land parcels. Each land parcel interest group was encouraged to identify its requirements so that compromise solutions could be sought. While these comments may have been contrived to encourage HMLR to take on the responsibility for market research on UPRNs, they do imply that the building of `land parcels that suit particular functional needs' is of higher priority than the provision of an infrastructure for linking data in Britain. Environmental information systems (EIS) based on satellite data, digital terrain models and other surveyed information (such as on soil and geology) - also focus on units of space. However, unlike land and property information systems, these tend to be relatively smaller-scale applications which often utilise areal units with inferred and fuzzy boundaries (these are described in the next section). In contrast to LIS and EIS, Demographic Information Systems (DIS) focus on the socio-economic and socio-demographic characteristics of individuals and/or households rather than on land. They invoke the concept of spatial units as a means of getting around the problems of confidentiality of data and the 1984 Data Protection Act. For this class of applications, the Chorley Report recommended the adoption of the postal address as a standard BSU, since it relates better to individual and household data in multi-occupied properties and it ties in with the postcode system. However, data at this level may still be confidential and may have to be aggregated. Aggregations may be based on different criteria. The Chorley Report recommended the unit postcode as the next level of aggregation. The 1981 Census of Population for Scotland was organised by the General Register Office (Scotland) on such a basis, though this was not the case in England and Wales (Denham and Dugmore, 1989). In England and Wales, census returns on individual households were linked to the census enumeration district (ED) in past population censuses. Systematic code names for the hierarchies of census reporting units were then used for further aggregation of data. The term `basic spatial unit' is misleading in the above context, since the data collecting framework consists of basic population units (BPUs) rather than BSUs and the boundaries representing the BPUs may be highly generalised and even undefined, as in the case of postal addresses and unit postcodes in England and Wales. Area-based targeting of advertising mail and research on socio-spatial structures have in the past relied on sophisticated, but blind, statistical analysis of census and other data for smaller and smaller units. Many applications which are concerned primarily with the analysis of enumerated data (such as population, agricultural, retail and other censuses and surveys) focus on statistical populations rather than containing spaces. They are more concerned with the accuracy of attribute, rather than boundary, data for studying spatial patterns and relationships and statistical clusters and anomalies. Thus BPUs - such as unit postcodes - are derived through manipulation or use of some descriptive, rather than the spatial, dimension. Since their spatial extent is unknown they cannot be linked using overlay analysis. Their common names also take the form of descriptive and nominal locational references which are not usable within geometric spatial analysis. Indeed, many forms of analyses of personal data may require that information on individuals be aggregated to units which do not result in clearly identifiable areas. |
|
Figure 2.1: Enlarged representation illustrating lack of precision when pointing at centroids |
The Chorley Report also recommended the grid referencing of postal addresses and unit postcodes. To avoid ambiguity and inconvenience resulting from the use of different grid references for the same units, the Chorley Committee also recommended that the grid referencing be standardised. The process of digitising grid references on a high resolution digitiser tends to introduce a small random error. Figure 2.1 shows how the extent of the symbol marking the centroid and the digitiser's cursor make it difficult for an operator to point consistently at exactly the same location. Thus, even if the same marked map was digitised twice by an agency (let alone several independent organisations), we would have two non-identical sets of grid references. To avoid confusions arising from such discrepancies, a standard set of grid references have to be provided for all users. The Ordnance Survey and Pinpoint Analysis Limited are collaborating on the creation of the PAC (Postal Address Code) database, which will assign a unique 12-digit grid reference to each address. This will allow all addresses in the UK to be located within an accuracy of one metre. The grid referencing of BPUs effectively means that the attribute data may be further aggregated to other existing systems of areal units based on the point in polygon principle. For example, the 1971 population census small-area statistics were aggregated and released for 100 metre and one kilometre grid square areas. The one-kilometre grid square data provided an acceptable framework for studying national, regional and more detailed patterns even if it was insufficient for studying detailed variations within neighbourhoods (OPCS, 1980). These high resolution data also revealed the impact of the ratio bias in statistical analysis of data for small areas which had remained concealed in many analyses of data for irregular units (Visvalingam, 1983a). Traditional ratio measures tend to place undue importance on spurious extreme ratios in small populations and thereby distort policy-oriented analyses. It is important to remember that standardised grid references for BPUs may be accurate to one metre but they nevertheless are only spatial surrogates (i.e. the points representing areas with extent). Thus, although a grid reference may be treated as a point and be assigned unambiguously to a target BPU whose polygon includes it, the area represented by the grid reference may straddle two or more polygons. Indeed, one of the problems associated with the use of unit postcodes is that the addresses within a unit postcode do not always nest precisely within administrative and ward boundaries. Even though this involves a relatively small proportion of units, it has led to the proposal that postcoded data be aggregated to pseudo-EDs, which best approximate existing ward boundaries (Denham and Dugmore, 1989). From the foregoing discussion it should be apparent that the term `basic spatial unit' refers to conceptually different entities within different classes of applications. Even within a single application different spatial frameworks may be appropriate for various purposes (such as measurement, reporting, monitoring, analyses and other management and operational purposes). For example, the Institute of Hydrology has used a grid-based terrain model, produced by the Military Survey, to extract a complete flow-directed network of rivers and their catchment boundaries. But, its modelling of the hydrological cycle is catchment-based. It would be unrealistic to advocate a standard set of areal units for all tasks, let alone all applications. However, different classes of applications may benefit and progress through adoption and use of standards appropriate to their brief. |
Accuracy of spatial descriptions It should be apparent by now that there is some considerable variation in the requirements for standards of accuracy of boundary data. The quality of the data must be assessed with respect to its fitness for use which is application and task dependent. The Chorley Report, for example, stated that postal address boundary information is not relevant (DOE, 1987, p. 88). However, many land information systems need accurate data because data is often used in administrative functions such as tax assessment. In Britain there is no national cadastre which defines land parcels for purposes of taxation, valuation and transfer of ownership. Instead, land information systems rely on Ordnance Survey maps at basic scales or even more detailed survey information. There is, at present, considerably more attention being paid to the issues surrounding the creation of spatial databases than to the issues surrounding their subsequent use. In this context, it is vitally important that the creation of a national archive of OS topographic information should address future requirements for accurate data (Haywood, 1987). The Chorley Report has played a significant role in the promotion of GIS in the UK. GIS are described as essential tools for better decision-making and their primary benefits depend upon their ability to link data sets together. Data may be linked using descriptive attributes using proprietary database systems. The distinctive feature of a GIS is its ability to link data on spatial criteria. The most celebrated of the GIS functions is the use of overlay analysis to link data (DoE, 1987, p. 51). Overlay analysis of gridded data is a relatively trivial exercise. Overlay analysis of irregular polygons relies on geometric processing and is computationally more demanding. The polygons may be input or calculated (for example by defining buffer zones around point, line and areal objects). But, as with any other information system, the decision support information obtained from a GIS can only be as good as the input data. Good technology is no remedy for poor data. We have already noted above that boundary data are inevitably of variable quality. Burrough (1986, p. 112) pointed out that most procedures used in GIS assume implicitly that: (a) the source data are uniform; (b) digitising procedures are, infallible; (c) map overlay is merely a question of intersecting boundaries and reconnecting the linework; (d) boundaries can be sharply defined and drawn; (e) all algorithms can be assumed to operate in a fully deterministic way; and (f) class intervals defined for one or other `natural' reason necessarily are the best for all mapped attributes. A great deal of data used within GIS come from cartographic sources. Even when polygons describe geographic phenomena with clearly recognisable boundaries, inherent generalisation errors in source documents and inevitable further generalisations occur during data capture and/or processing which produce spurious polygons on overlay. The automatic `cleaning up' of spurious polygons is not a trivial or reliable process. When the data are captured from analogue sources compiled and generalised at different scales, existing facilities for scaling and intersecting polygons may be inadequate. Saalfeld (1988) has provided a detailed and informative account of advances in conflation, or automatic map compilation. This involves the matching of generalised and displaced features and adjustment of the geometry prior to compilation. However, in general, such pre-processing of the data is not undertaken prior to overlay analysis within GIS. As a result, overlay analysis will produce spurious sliver polygons; the area-based criteria for eliminating sliver polygons cannot cope adequately with these anomalous polygons. The use of map data introduce other problems. Although interpolated boundaries on isopleth (e.g. contour) and proximal (e.g. Thiessen polygon rainfall) maps are objectively derived, the boundaries are hypothetical. Geographic phenomena may have continuous rather than discrete distributions in space and discontinuities, such as buried geological faults, may not be directly observable. Many geological maps for instance are interpolations based on sample values collected at exposure sites and bore holes. Soil maps are also inferred. Consequently, many boundaries are either hypothetical, arbitrary, intrinsically fuzzy or generalised. Manual overlay analysis of environmental information involves judgement and takes into account a number of factors, including the (spurious or lack of) accuracy and reliability of different data, a knowledge of the nature of the underlying phenomena and the purpose and accuracy requirements of the analysis. Thus, experts are able to extract appropriate and usable information from available, but often imperfect, data. Louden (1986) elaborated on this in the context of geological investigations. Automatic overlay analysis does not at present exhibit such expert behaviour through the application of such conceptual and semantic knowledge and judgement. It is possible to quantify the accuracy limits of data but standard deterministic methods for spatial analysis may not be adequate or even appropriate for fuzzy data. For example, Goodchild (1988) reviewed issues relating to accuracy before considering, in a separate section, the need to extend planar spatial algorithms to the spherical global context. He included the Douglas-Peucker line generalisation algorithm within the standard (tried and tested) algorithms without examining the impact of error on this heuristic. In his text, which laid great stress on data quality and errors in spatial data, Burrough (1986) similarly included this algorithm without critical comment. However, it is difficult to accommodate digitising error within this deterministic algorithm in anything other than an arbitrary and questionable way, as explained by Visvalingam and Whyatt (1990). In socio-economic analysis, cross-area comparison of statistics involves assumption-based disaggregation of spatial statistics for source BPUs on one coverage and their re-aggregation for target BPUs on another. The properties of spatial statistics are variable and considerable caution must be exercised in such cross-coverage comparison of spatial statistics. In the past, grids have been.used as a convenient framework for disaggregation but GIS will now disaggregate data for the primitive regions resulting from overlay analysis. Automated disaggregation of statistics is largely based on the areas of the primitive regions. The reliability of the resulting data obviously depends upon the accuracy of the boundary data and we have already indicated that BPUs tend to have undefined or highly generalised boundaries. Also, manual disaggregation of data tends to be based on the principles of dasymetric mapping (Cuff and Mattson, 1982). Dasymetric mapping takes into account other factors which influence the intra-unit distribution of phenomena. For example, when constructing a dot map of population distribution a cartographer would take account of the underlying topography, hydrology and settlement patterns when positioning the dots. This provides a further illustration of the need to link other data, such as environmental data, within statistical analysis. The requirements of cross-area comparisons further contradict expert opinions that the boundaries of BPUs are irrelevant in socioeconomic analysis and that synthesised boundaries in the form of Thiessen polygons are adequate for graphics. Thiessen polygons (Monmonier, 1981; Boots, 1986) are constructed such that all parts of a polygon are closer to its surrogate point than any other data point in that set; hence, this type of mapping is also referred to as proximal mapping. Whilst the accuracy of boundary data is not critical, visualisation is essential for validation of information processing, especially when the required information has to be abstracted from data through statistical, inferential or graphical processes (Visvalingam and Kirby, 1984; Visvalingam, 1985). Spatial surrogates, such as visual centroids, are not always adequate for interactive visualisation. Also, some indication of the relative areas of BPUs is necessary for computing density, for example, for incorporation within measures of social deprivation (Visvalingam, 1983a). |
|
t has already been pointed out that all geographic phenomena exist within the multi-dimensional framework of space, time and other descriptive dimensions. The spatial dimension is used for linking data within GIS and this is useful for some types of applications. Even so, as explained in the previous section, the utility of GIS is limited by a lack of appropriate conceptual and methodological tools for manipulating inaccurate, let alone fuzzy and hypothetical, boundaries. At small scales, geometrically defined boundaries are conceptual devices rather than real entities. These conceptual devices may not be applicable when data units are derived through an aggregative process, based on non-spatial constraints. For example, aggregate health statistics for medical general practitioners and treatment centres are becoming highly pertinent in the context of current policies on health. The data units refer to the functional dimension which is an aspatial concept. Although the distribution of patients is spatially related to the location of these service centres, the catchment areas for the latter overlap and boundaries as such are highly tenuous and probabilistic - perhaps even meaningless. Even if we ignore extreme cases of people travelling across counties to visit their tried and tested opticians and dentists, the catchment areas of service centres (be they patients or shoppers) are highly volatile, amoebic and interwoven in geographic space: they also vary considerably over time. When BPUs are defined on non-geographic criteria, the corresponding `social spaces' may defy precise geometric delimitation. It is possible to use GIS for analysing such data but users must be absolutely clear about the research questions they are seeking to answer and the utility and limitations of the methods they adopt. |
Figure 2.2 The shape of areal units
|
Areal units may be classed into two types, namely regular and irregular units (see Figure 2.2). Regular units, such as grid squares, are the result of a systematic division of space into areas of the same shape, even if not the same orientation or size. Such systematic divisions of space into regular geometric shapes are called tesselations (Peuquet, 1984). Regular tesselations result in units of the same size, while nested tesselations allow the recursive subdivision of a cell into smaller and smaller units. Hybrid tesselations can include more than one type of shape of varying size but these are combined in a systematic way. All tesselations provide a framework for recording data about arbitrary units of space. The units are arbitrary in the sense that cell boundaries do not correspond to phenomena of interest. Information on the character and distribution of spatial phenomena has to be inferred from the data for these arbitrary units. The widely used grid square framework is one form of tesselation. Some practitioners refer to the grid framework as a `raster' system. This is potentially misleading. I prefer to use the word tesselation for a data collecting framework and the term, raster, for a type of data encoding format. The reason for this will be described later. Space may also be divided into irregular units. Each irregular unit is idiosyncratic in the sense that it may have its own peculiar shape with detached parts and holes. As we observed earlier, irregular units may be subclassed into natural, functional and primary units; the latter were called primitive regions in earlier discussions. Natural units define the extent of observable phenomena, such as vegetation, soil, geologic and land-use types, which are described by nominal or discrete categories of data. As noted above, although these typologies may be visible, the boundaries may be hypothetical and fuzzy. Consequently, many small-scale environmental applications use tesselated remote-sensed data of high resolution to study the distributions of natural phenomena and impose their own boundaries. Populations, similarly, are not classified according to application-specific types by data collecting agencies. Instead, a range of descriptive statistics are released for each BPU. Applications classify these populations and their neighbourhoods, using a selection of data and appropriate techniques, although an increasing number of users find it more convenient to work with a proprietary classification, such as ACORN or MOSAIC. Such classifications may be based on data for either high-resolution tesselations or irregular units. Given the growing number of proprietary classifications on offer, users have the added problem of making an informed and considered choice from these (see Brown, 1989). Functional units are specially designed for some specific purpose such as administration, taxation, ownership, targeting services, policing, for the performance of statutory functions or for the delivery of mail. Functional units are almost always synthetic and, given that environmental, population, socioeconomic and several other characteristics are all subject to change, functional units are by nature dynamic: this latter fact poses fundamental problems for time-series analysis. Unit postcodes and Travel to Work Areas are two examples of irregular functional units. Primitive regions are not user-perceived entities. As previously described, they are the fragments of space which result from the overlay of a set of coverages. They form the BSUs within modern spatial information systems and are used to integrate data from different sources and to model and extract automatically complex relationships between areas such as overlaps, hierarchies, holes and detached parts (Kirby et al, 1989). The topological knowledge can be used to check for completeness and consistency in the data supplied and to ensure that any rules about how areas are related (such as in a hierarchy) are obeyed in the data. The
grid square format lends itself to raster representation of data. In
raster format, data values are recorded for each cell in the tesselation.
The raster format is again a user-perceived model. Within a computer,
raster data are held more compactly using run-length encoding schemes (Lauzon
et al., 1985) or linear quadtrees (Gargantini,1982; Samet,1988).
Tesselated data are seldom held in vector form since the shape and
extent of the units may be calculated. Systems for digital cartography
and GIS may use raster and/or vector formats for recording the
boundaries of irregular units. In vector format, the positions of
perceptually significant points on the course of the boundary are
recorded. Vector data may also be held in a variety of forms within the
computer. Many raster-based GIS capture and use boundary data in raster
form. Thus, the term, raster, refers to a representational format for
spatial data whilst the term, tesselation, refers to a more general
system for collecting, recording and indexing spatial data. |
Size of areal units and aggregational flexibility Unless BPUs are aggregated with respect to customer-specified criteria, they should be as small as possible so that they minimise heterogeneous groupings and provide maximum aggregational flexibility - without providing loopholes for the violation of the confidentiality of personal or property-related information. Using these criteria, land parcels and addresses are too small for reporting purposes. If all land parcels and addresses are grid referenced, socio-economic data could be released for any set of spatially defined BSUs. The latter include units with idiosyncratic and regular shapes. Since land parcels are irregular, tesselations may be regarded as artificial agglomerations of address-related information. However, it is reasonable to view the aggregational process as the assignment of addresses to those cells which contain the bulk of that property. The use of visual centroids achieves this. The resulting errors are unlikely to prejudice statistical patterns to any significant extent. These errors have to be seen against the undefined boundaries of postcodes and the deliberate addition of random numbers to census counts to preserve the confidentiality of data. The problem with the uniform grid is that a large number of cells located in rural areas tend to contain very small populations, leading to a suppression of data, while cells in urban areas contain large populations, leading to generalisations that are of limited value. Thus, if tesselations are to satisfy the size criterion, they must be data adaptive; i.e. in rural areas they must be large enough to preserve confidentiality and in urban areas they must be small enough to preserve homogeneity. This can be achieved through the use of nested tesselations. In the case of a grid framework, this effectively means that grid cells can be progressively subdivided within a consistent grid framework. Subdivision need not be limited to the plane and could be extended if appropriate and necessary to the vertical dimension in areas with high-rise property. |
|
All irregular units are susceptible to change and are temporal configurations of space. This is true even of remotely-sensed data since the resolution of scanning has improved with time. But this is especially true of land parcels within LIS and EIS and of functional units. Areal entities may be subjected to extensive across-the-board sweeping revisions at periodic intervals. This occurs, for example, when automated systems for data collection are improved or when there are major changes to administrative or service-delivery structures. Areal entities may also be subjected to continual, piecemeal changes. This is true not only of LIS but also of resource and ecological management systems. Since GIS utilise the geographic dimension as a control variable, they tend to encourage the convenient but limited snapshot approach to recording and studying change. This paradoxically requires unchanged or constant BSUs. Tesselations provide a stable framework for the statistical analysis of change since they provide comparable, even if not constant, units and because they can provide a 100 per cent coverage of space without incurring tremendous overheads for maintaining redundant units, which may be unused at present. They also provide an excellent framework for collecting sample measurements and are used, for example, by the Institute of Terrestrial Ecology for identifying land classes (Vincent, 1987) and by compilers of regional floras; again, for example, the `flora of the East Riding of Yorkshire' was compiled from tetrad (2 x 2 kilometre) samples collected over some forty years (Crackles, 1990). The USA Landsat, French SPOT and other meteorological and microwave satellites generate rectangular or grid square data at various resolutions (Curran, 1985) which are extremely valuable for monitoring and predicting global and even local environmental phenomena. The availability of 5m resolution data from the Russian SOYUZ CARTA satellite also provides some prospects for the use of satellite data for monitoring and mapping topographic change. However, no irregular unit, whether defined in spatial or aspatial terms, can remain constant. The retention of a constant set of functional BPUs would be as inappropriate as the retention of some past standard industrial classification. Changes in the data recording frameworks are inconvenient. For example, such changes necessitated the construction of some slightly larger units to which both the 1971 and 1981 population census enumeration districts could be aggregated in order to study change (DOE, 1987, p. 164). Snapshots only describe fixed states at a given time; they provide an ahistoric view of data. The snapshot recording of phenomena is far from ideal for many applications. The 1981 population census data is now embarrassingly out of date. There is need continually to monitor and maintain a chronological record of change; the currency of information is vital for planning and emergency services. The continual update of BPUs is intrinsically difficult since these units are classified by content rather than form. Approximately 18,000 changes are made at the unit postcode level in Great Britain each year (DOE, 1987, p. 172). The Postcode Address File (PAF), which lists the addresses within postcodes, is updated every three months and is known to include some inconsistencies. Public agencies are required by law to maintain historic archives of land transactions. Such records are used in land searches to study the history of past development, to assess the potential of environmental hazards, to record the outcome of past applications for development and to monitor subsequent usage. Vrana (1989, p. 38) in his paper on historical data in LIS pointed out that `systems that cannot reconstruct a temporally-connected chain of events and states do not adequately replicate pre-computerised paper archives'. Historical information is essential for not only planning and forecasting but also for checking the validity and integrity of input data. However, as in paper archives, the practices of recording date stamps and transaction logs are insufficient because they only record events and specific states. The technique of versioning may be used to derive successive states resulting from, or even preceding, events. But unless the order of updates to the database corresponds to the order of real life events, the database may not reflect true states. Also, update may involve a modification or replacement of user-perceived objects requiring changes both to their spatial and aspatial descriptions. Maintaining the integrity of change to databases is no trivial task. Spatial data processing is facilitated by the use of topologically structured vectors. Langran and Chrisman (1988) proposed common spatial-temporal units which require the maintenance of not only spatial but also temporal topology. This will almost inevitably involve the further subdivision of primitive regions on update. Within a volatile environment, complex corporate databases could become highly fragmented into exceedingly small and ridiculously numerous primitive regions. The management of historic information is still the subject of research. Within the current breed of coverage-oriented systems, the temporal dimension would have to be conceived as similar to the thematic dimension. A snapshot is treated like a theme and encoded on a separate coverage. Similarly for continual update, individual events have to be treated as themes. The user is responsible for managing the versioning of data. |
|
Land parcels and tesselations can provide a total coverage of both space and populations. Unit postcodes do not provide a total coverage of space since they are collections of addresses. Many personal/population data do not relate exclusively to postal addresses. For example, a crime survey can relate information on offenders and victims to their respective postcodes but it is not clear as to which unit the data on the scene of the crime itself will be related if the incident occurred in a public park or out on the moors. Studies which are concerned with activities and incidents need to be particularly careful in their formulation of BSUs. Smith (1988) noted that a postal address based database, even if backed up by National Grid coordinates, would not cope with all situations faced by HMLR, particularly when agricultural property is involved. |
Modifiability, manipulability and ecological fallacy Since tesselations result from the geometric subdivision of space, they are not easily manipulable. However, even a grid is modifiable in the sense that it is possible to vary the geographic extent of units by redefining the origin and interval of the grid. While it is possible to minimise cell heterogeneity by reducing its size, it is impossible to engineer cell homogeneity or to prejudice cell characteristics. In contrast, functional units are inherently modifiable. Land parcels are defined to be homogeneous with respect to a critical set of attributes. Primitive regions are defined to be homogeneous with respect to one or more sets of attributes. BPUs should also contain populations with similar characteristics. But, the characteristic which must remain homogeneous will vary from application to application. Thus, no set of BPUs is ideal for all applications and this explains to some extent the plethora of data collecting, functional and reporting units in current use. It is well known that the results of statistical analysis are influenced to some (usually an unknown) degree by the nature and number of areal units. The design of electoral units for instance can influence electoral results. For statistical comparability, BPUs should be equal population units, which are as homogeneous as possible with respect to some controlling critical attribute. Optimising algorithms are used for the identification of functional units. Thus, all designed units impose some degree of bias to the data to facilitate meaningful analysis or use. The problem occurs when such data are subsequently used for some other purpose, without regard to its inherent bias. Also, all BPUs, whose design involves the optimisation of some critical characteristic and which results in irregular BSUs, are open to covert manipulation. Ecological fallacy refers to the unjustified inference of attributes about individuals from statistical generalisations about BPUs, i.e. groups of people. For example, the statistics may indicate that BPUs with relatively high proportions of immigrants also record above average levels of crime. It would be an error to conclude from this that immigrants are the source of crime, particularly if both immigrants and criminals are minority groups within these areas. Equally, it is difficult to avoid double and multiple counting when formulating statistical indicators; for example, consideration of both one-parent families and female unemployment could lead to double counting of individuals as the female heads of one-parent families of a given age are far more likely to be unemployed than women generally. Openshaw (DOE, 1987, p 166) pointed out that designers of ecological classifications may label areas according to the relative concentration of target groups, which may form a small minority of all residents. Such labels lead to misconceptions about an area's population profile or characteristics. Both examples show that the problems of ecological fallacy are basically problems of misrepresentation or misinterpretation of the results of statistical analyses and area classifications by naive persons. This tendency provides others with opportunities for cynical manipulation of data to support post-hoc justifications. |
Variability in base populations For statistical processing, it is important that BPUs are designed to be near-equal population units. Regular tesselations produce large variations in sample populations. Even irregular units, such as census EDs and unit postcodes, are known to vary in population size. Where land parcels are used to record nominal data, rather than statistical counts, the problem of variability in base populations does not occur.
Variations in base populations produce a ratio bias (Visvalingam,1983a).
In area-based analysis, the relative density of occurrence of phenomena
is an important factor. Regular and nested tesselations provide some
scope for accommodating this ratio bias. Using nested tesselations, it
is possible to subdivide the areal units so as to reduce the variability
in base populations. Also, the area of the BSUs within tesselations is
implicit and may be used in calculations. With irregular units, it is
more difficult to deal with the ratio bias even when data on the area of
BSUs are available. The spatial extent of unit postcodes is not defined.
For the same reason, spatial surrogates alone are insufficient for
area-based analysis, as noted earlier. |
|
The very fact that we can design areal units implies the existence of alternatives, trade-offs and choices. To some extent the choices are limited by the nature of the phenomena itself, by the nature of the data collecting process and by the existing frameworks. Although the spatial resolution of recording will continue to improve, satellites will continue to produce tesselated data. Automated sensors `see' rather than `interpret' phenomena. Data for tesselations allow users to impose their own interpretations and classify space as appropriate for their applications. A high resolution spatial tesselation is more appropriate for environmental monitoring systems, studying phenomena with fuzzy and changeable boundaries. On the other hand, many land and property-based information systems tend to record factual information about clearly identifiable spatial objects (although purists would argue that all boundaries are fuzzy to some degree) and areal units which exist by virtue of law. Although different organisations work with different sets of land parcels, past experience of the GISP and Newcastle upon Tyne JIS schemes and the more recent collaborative venture within Northern Ireland demonstrate that closely-related land-based applications can expedite their requirements for information sharing though use of primitive regions. Primitive regions are also more convenient for ensuring the integrity of updates to data within GIS. Standard units are also necessary for linking sets of personal information. Since the census small-area statistics (SAS) often provide the most reliable source of common attribute data, a defined hierarchy of census areas is critical to the linking of data. The Chorley Committee has recommended the use of unit postcodes as an alternative to EDs. In Scotland, the unit postcode is not only used as an areal unit for statistical tabulation and as a base unit for further aggregation, it is also used in the planning of census enumerator workloads. In England and Wales the move is towards the postcoding of addresses to output SAS for postcode sectors, pseudo-EDs and user-defined groups of postcode units, in addition to SAS for EDs. Users have the responsibility for deciding the levels at which the linking of data should take place. The main advantages of the unit postcode are their wide usage and the small size of BPUs. There is some concern that postcodes may be used to work around the 1984 Data Protection Act. Commercial companies have tended to use stereotypes for quick identification of target groups. Socio-economic class was used initially but neighbourhood profiles, based on population census and market research data, have become more popular in the 1980s. Unit postcodes provide a convenient framework for interrelating proxy data and for targeting, especially through direct mail. Recently, the Data Protection Registrar has objected to the inequitable practice of using an address as a cheap and easy predictor of credit-worthiness (The Daily Telegraph, 1988, p. 17). In July 1990, he pronounced the practice as illegal. The use of generalised stereotypes in area-based decision-making is even less discriminating given the problems of ecological fallacy, modifiability and range of competing proprietary classifications which is now available. Stereotypes serve to increase the efficiency, rather than the equity, of targeting (Visvalingam, 1983b). Other deficiencies of unit postcodes have been identified not least within the Chorley Report itself. They suffer from the limitations of idiosyncratic systems for partitioning space - they are not stable, they are not easy to manage as witnessed by the presence of inconsistencies in the system, they do not nest neatly within the hierarchy of census reporting areas and they do not cover all space of interest. They are BPUs, not BSUs. They need to be redesigned and spatially defined to act as BSUs. Will they then correspond to the postman's walk? (i.e., should we still regard them as postcodes?) Or, will they in effect become some other spatial reference to be attached to an address? The Chorley Report not only suggested that the Post Office should take account of topographic features and electoral and administrative boundaries when establishing new postcodes, it also urged the Boundary Commissions to take account of unit postcodes when setting ward boundaries. But we need to consider other possibilities. Now that the street addresses are being assigned unique grid references, will postcodes remain attractive given that a spatial reference offers greater aggregational flexibility than a nominal reference? Already OPCS/GRO(S), the Economic and Social Research Council (ESRC) and ICL Ltd. have funded Birkbeck College for the development of an experimental on-line service for handling unaggregated census questionnaire returns from the 1991 census. Users will be able to extract any aggregate information - for whatever area(s) or groups of people and cross-tabulated as required - unless this is likely to disclose details of an identifiable individual or household or lead to an unreasonable intrusion into privacy (Rhind and Higgins, 1988). If on-line services of this kind are acceptable to the public, then it is quite likely that much personal data, if released, will be disseminated in this way for user-defined units where possible. This will enable users to link personal data (e.g. on health) with environmental data collected in tesselated form. Within this scenario, there may well be less interest in standard BSUs but users should be much more aware of the properties of their own designed areal units. The case for postcodes
rests mainly on the fact that they are widely known and used. This was
convenient for data collection and processing in the past but
information systems of the future should not have to rely on people
remembering and correctly using nominal and/or cadastral codes. Already,
we are carrying more and more cards for one purpose or another (even for
using photocopying machines) and many of these are electronically
processed already. How many of us know, let alone remember, the barcodes
on personal documents? Optical information cards, also known as
LaserCards and smart cards, are already in use for US Army training and
by health insurance companies for personal health cards. Thus, the
information systems of the future may not need the convenience of easily
memorable codes. Even if memorability remains an important requirement,
this does not in itself establish the case for postcodes. Cadastral
addresses and personal identity numbers are used widely in some in some
European and Third World countries respectively (see Boalt and Bernow
elsewhere in this volume). Also, most people have little difficulty in
remembering or recording and using their own and several other telephone
numbers with 10 or more digits. Unique grid references for properties
can themselves be adapted into address codes even in multioccupied
premises, providing the link between personal information systems on the
one hand and land and property information systems on the other. The
design of BSUs for a future national GIS must be based on substantive
rather than pragmatic criteria. |
|
This chapter has not considered in any detail the alternative schemes for spatial referencing. It has merely considered the problems involved in the design of areal units to meet both their primary and secondary uses. Standards for spatial units are intended to facilitate the linking and/or comparison of data. The move to reference data to the national grids for Britain and Ireland is a major step forward in the history of GIS in the UK. The compilation of a standard set of grid-referenced spatial surrogates for postal addresses is to be welcomed and used. But grid references and UPRNs are no more than references. The critical issue is the design of the spatial units themselves. Although it is highly desirable that closely related applications should adopt standard areal units where possible, a universally applicable standard set of areal units is unattainable. Tesselated data will continue to pour out of automated systems for data collection and will be used to model phenomena with continuous distributions. Statutory functions require the management of data for irregular land parcels. Personal information will have to be aggregated for reasons of confidentiality. Only time will tell whether the increasing availability of accurately grid-referenced surrogate and boundary data and the vigilance of the Data Protection Registrar will undermine the use of the convenient but geographically deficient system of postcodes. The problems of linking these various data sets remains a challenge. The postal address is the ultimate BSU within information systems based on personal data. It forms one set of entities within land and property information systems, although these latter applications also require the definition of land parcels corresponding to other types of entities. The postal address thus forms the link between a number of human applications based on irregular areal units. The Chorley Report advanced the view that boundaries are irrelevant in socio-economic analysis and recommended the use of surrogates instead. However, accurate boundaries are essential for the disaggregation and reaggregation of statistical data and for overlay analysis. Given that the three dimensional geographic space provides a stable and unifying framework for linking data from different sources, investment in the restructuring and semantic enhancement of accurate vector-digitised large-scale plans seems to provide a flexible route for linking data in the longer term. Such detailed information is also necessary for meeting the long-term requirements for scale-free mapping, which was also recommended in the Chorley Report. Discussion of the design of areal units and of the problems inherent in linking data were stimulated by the Chorley Report, which also campaigned for increasing use of GIS. No doubt, using computerised GIS it is relatively easy to link and cross-reference data with identical names, be they nominal references such as postcodes or spatial grid references. Graphical user interfaces are being evolved to save users the trouble of having to learn the syntax of textual query languages. GIS are also useful for spatial data processing. The potential benefits of computerisation are not in question. But GIS are still in their infancy and many vendors, consultants and data suppliers tend to gloss over the problems involved in linking data when promoting the value of GIS functions such as overlay analysis. This chapter has explained why the extraction of usable information from available, but often imperfect, data is a highly skilled activity based on training, experience and a level of intelligence not evident in current GIS systems. It would be regrettable if such problem-solving skills were to be underrated and - worse still - replaced by skilled operators of computing systems. The inflexible design of the system for the fire brigade in the case of the disaster at Hillsborough soccer ground in Sheffield, where over ninety people were killed, provides a lesson on spatial referencing. There was a delay in sending cutting gear to Hillsborough because the system required the input of the name of the road on which the premise was located and the user did not know this piece of information. Emergency systems, responding to public queries, must provide flexible access to data. We are wiser after an event - a disaster in this case. The foregoing discussions should encourage designers and users of GIS to think about the following issues:
In my view, it is only by rigorously working through such a checklist of issues and questions that GIS will provide appropriate solutions to the research question at hand. |
|
Blakemore, M.J. (1986), Proceedings of Auto Carto London, International Cartographic Association, London. Blakemore, M. (1990), `Quality and GIS - are they compatible?', Proceedings of the Mapping Awareness Conference, pp. 24(1)-24(8). Boots, B.N. (1986), Voronoi (Theissen) polygons, CATMOG 45, Geo Books, Norwich. Brand, M.J.D. (1988), `The geographical information system for Northern Ireland', Mapping Awareness, 2(5), 18-21. Brown, P.J.B. (1989), `Geodemographics: a review of recent developments and emerging issues', Urban Research and Policy Evaluation Regional Research Laboratory, Working Paper 1, Liverpool University, Liverpool. Burrough, P.A. (1986), Principles of geographical information systems for land resources assessment, Clarendon Press, Oxford. Crackles, E. (1990), The flora of the East Riding of Yorkshire, University of Hull and Humberside County Council. Cuff, D.J. and Mattson, M.T. (1982), Thematic maps. their design and production, Methuen, New York. Curran, P.J. (1985), Principles of remote sensing, Longman, London. Daily Telegraph The (1988) `Credit checks get October ultimatum', Saturday 21 May 1988, p 17. Dale, P.F. (1988), `Land information systems and the land parcel', Mapping Awareness, 2(2), 27-9. Denham, C. and Dugmore, K. (1989), `With hindsight - the census ten years on', BURISA, No. 91, pp. 2-7. Department of Environment' (1987), Handling geographic information, HMSO, London. Gargantini, I. (1982), `An effective way to represent quadtrees', Communications of the Association of Computing Machinery, 25(12), 905-10. Goodchild M.F. (1988), `The issue of accuracy in global databases', in H. Mounsey (ed. ), Building databases for global science, Taylor and Francis, Basingstoke. Haywood, P. (1987), `The OS topographic database study', in P. Haywood (ed.), SORSA `87 DURHAM, Ordnance Survey, Southampton. HMSO (1972), General information systems for planning, Report of Joint Local Authority, Scottish Development Department and Department of Environment Study Team, HMSO, London. Ives, M.J. and Lovett, R. (1986), `Exchange of digital records between public utilities digital mapping systems', in M.J. Blakemore (ed.), Proceedings of Auto Carto London, International Cartographic Association, London. Kirby, G.H.,. Visvalingam, M. and Wade, P. (1989), `The recognition and representation of polygons with holes', Computer Journal, 32(6), 554-62. Langran G. and Chrisman, N. (1988), `A framework for temporal geographic information', Cartographica, 25(3), 1-14. Lauzon, J.P., Mark, D.M., Kikuchi, L. and Guevara, J.A. (1985), `Two-dimensional run-encoding for quadtree representation', Computer Vision, Graphics and Image Processing, 30, 56-69. Loudon, T.V. (1986), `Digital spatial models and geological maps', in M.J. Blakemore (ed. ), Proceedings of Auto Carto London, International Cartographic Association, London. Marsh, C., Arber, S., Wrigley, N., Rhind, D. and Bulmer, M. (1988), `The view of academic social scientists on the 1991 UK Census of Population: a report of the Economic and Social Research Council Working Group', Environment and Planning A, 20, 851-89. Monmonier, M.S. (1981), `Trends in atlas development', Cartographica, 18(2), 187213. O'Brien, L.G. (1990), `Small area information systems: problems and prospects', in L. Worrall (ed. ), Geographic information systems. developments and applications, Belhaven Press, London. OPCS, (1980), People in Britain -a census atlas, HMSO, London. OPCS (1990), Census Newsletter, No. 14, Office of Population Censuses and Surveys, London. Openshaw, S. (1990), `Spatial referencing for the user in the 1990s', Proceedings of the Mapping Awareness Conference, pp. 25(1)-25(5). Peuquet, D.J. (1984), `A conceptual framework and comparison of spatial data models', Cartographica, 21(4), 66-113. Rhind, D. and Higgins, M. (1988), 'Customer-selected results from the 1991 census', BURISA, No. 82, pp. 7-9. Rowley J. (1990), in Letters to the Editor, Mapping Awareness, 4(6), 53. Saalfeld, A. (1988), `Conflation. Automated map compilation', Intemational Journal of Geographic Information Systems, 2(3), 217-28. Samet, H. (1988), `Hierarchical representations of collections of small rectangles', ACM Computing Surveys, 20(4), 271-309. Smith, P. (1988), `Developments in the use of digital mapping in HM Land Registry', Mapping Awareness, 2(4), 37-41. Vincent, P. (1987), `Geographical information systems at Lancaster', BURISA, No. 80, Visvalingam, M. (1983a), `The operational definition of area-based social indicators', Environment and Planning A, 15, 831-9. Visvalingam, M. (1983b), `An examination of some criticisms of area-based policies', Census Research Unit Working Paper 22, Department of Geography, University of Durham, Durham. Visvalingam, M. (1985), `Concept refinement through the graphic representation of large data sets', in B. Shackel (ed. ), Interact `84 - Proceedings of the First International Federation for Information Processing Conference on Human Computer Interaction, North Holland, Amsterdam. Visvalingam, M. (1990), `Trends and concerns in digital cartography', Computer-Aided Design, 22(3), 115-30. Visvalingam, M. and Kirby, G.H. (1984), `The impact of advances in IT on the cartographic interface in social planning', Department of Geography Miscellaneous Series No. 27, University of Hull. Visvalingam, M. and Whyatt, J.D. (1991), `The importance of detailed specification, consistent implementation and rigorous testing of cartographic software', M Blakemore (eds) "Mapping the Nations", the 15th Conference of the International Cartographic Association, Bournemouth UK, 22 Sep - 1 Oct, 856 - 859. Vrana, R. (1989), `Historical data as an explicit component of land information systems', International Journal of Geographic Information Systems, 3(1) 33-49. Walford. N., Lane, M. and Shearman, J. (1989), `The rural areas database: a geographical information and mapping system on rural Britain', Transactions of the Institute of British Geographers, NS 14(2), 221-30. Yeh, A.G.O. (1990), `A Land Information System for the monitoring of land supply in the urban development of Hong Kong', in L. Worrall (ed.), Geographic information systems. developments and applications, Belhaven Press, London. |
| © Dr Mahes Visvalingam, University of Hull, 2002 |
Cartographic Information Systems Research Group, University of Hull
