THE COUNTRYSIDE INFORMATION SYSTEM: A STRATEGIC-LEVEL DECISION SUPPORT SYSTEM DAVID C. HOWARD and R.G.H. BUNCE Institute of Terrestrial Ecology, Merlewood Research Station, Grange-over-Sands, Cumbria, LA 11 6JU, United Kingdom
Abstract. The Institute of Terrestrial Ecology (ITE) has monitored ecological change in Great Britain (GB) since 1978. The task has been undertaken using a stratified sampling scheme working with a 1 km square as the sample unit. In more recent years, scientific researchers at ITE have been working closely with the policy-makers of the United Kingdom Department of the Environment. The presentation of information to policy advisors and planners was a component within a large project investigating the ecological consequences of land-use change. A simple PC-based decision support system was developed during the project and subsequently has been expanded to produce a marketable product. The system, called the Countryside Information System (CIS), presents and links information at national, regional and thematic levels along with qualifying data describing accuracy and appropriateness of use (i.e., metadata). An integral part of the CIS is the ITE Land Classification, which divides GB into 32 environmental land classes; all 250 000 squares have been classified. The classification allows sampled data to be presented and, as the co-ordinate system is widely used in GB, it allows census datasets to be linked and compared. CIS has been described as a Geographical Information System, but the classification, data held within the system, and the use of metadata to assist in interpretation of results make the system much more decision-support oriented. Indeed, government departments have been involved in directing the development and are now starting to use the system to answer parliamentary questions and formulate, assess and monitor environmental policy. The CIS is an open system, running on a standard PC in Microsoft Windows. Tools for loading and editing new datasets (both sample and census) are incorporated in the suite of programs. The Windows environment and users comments during development have produced a system with an intuitive feel, removing some of the overhead of acquiring specialised technical skills before being able to operate a system. This paper describes the CIS and presents examples of its applications.
1. Introduction In order to formulate and assess effective environmental policy, ecological and environmental information relevant to both the subject and region must be available. Unfortunately, detailed information that is necessary to formulate environmentally sensitive policies is expensive to collect and cannot always be obtained in the time-scale of parliamentary procedures. Part of the Institute of Terrestrial Ecology's (ITE) remit is to monitor land-cover change in the United Kingdom (UK) and identify the interactions between land cover, land use and ecology. A statistically rigorous sampling scheme was devised to collect information that can be interpreted at a national scale. A sample unit of 1 km square was selected as a compromise between the practicable and the desirable. Practically, the unit had to be capable of being surveyed in the field, while the desire was that it be large enough to contain sufficient variation to be representative. Environmental Monitoring and Assessment 39: 373-384, 1996. 1996 Kluwer Academic Publishers. Printed in the Netherlands.
374
DAVID C. HOWARD AND R.G.H. BUNCE
The stratification was produced by applying a multivariate analysis to environmental data derived from published maps. The analysis used was Indicator Species Analysis (ISA), which was originally developed for phytosociology (Hill et al., 1975). Initially, only squares at the intersection of a 15 km x 15 km grid were classified, but these were supplemented by a second grid of squares identified using a dichotomous key (Bunce et al., 1983). The original classification, even including the supplementary squares, only covered about 4% of GB. Different techniques were investigated to classify all squares in GB. The key, produced by ISA, was not ideal; as a dichotomous key it would radically mis-classify if there was an uncharacteristic value in a factor contributing to the early couplets. As there were a number of projects using the classification, it was decided to maintain the original structure as far as possible. Eventually all squares in GB were classified using a combination of techniques. Logistic discrimination divided the population into 8 groups, each of which was split into four groups by discriminant function analysis (Bunce et al., 1991). Sample squares for field survey were drawn at random from each stratum at the intersections of the original grid. National field surveys, called Countryside Surveys, were carded out in 1978, 1984 and 1990 (see Barr et al., 1993) producing figures for changes in land cover and vegetation change. Although, at the time, the original survey was considered to be largely a scientific investigation, the results it produced created interest in a wider audience than anticipated. In the following years, an increased awareness of the value and fragility of the environment has lead to greater consideration of environmental issues in government policy. The Department of the Environment (DOE) sponsored a research project to investigate the ecological consequences of land-use change (Bunce et al., 1993). One aspect of the project was to demonstrate the methods that could be employed to effectively pass information, including field survey results, to policy advisors. Having identified personal computers (PC) as a valuable tool for storing and presenting information, a number of approaches were developed. These ranged from an information system presenting data with spatial selection, through outline processing (holding and linking textual data) to full expert systems (offering diagnosis and suggested action from uncertain data). An information system was thought to hold the greatest potential and so was developed further in a second project commissioned by DOE (Howard et al., 1994). The development was user-driven; a number of potential users from different government departments and agencies were recruited and formed a Steering Group. They were supplied with copies of the software and datasets and were asked to attempt to use them for trial scenarios and applications. Comments and criticisms from the Steering Group were passed back to the development team who modified the programs and circulated new versions. The system, now called the Countryside Information System (CIS) has attracted a lot of interest, not just from government agencies. Non-governmental organisa-
THE COUNTRYSIDE INFORMATION SYSTEM IN GREAT BRITAIN
375
tions (NGO) with a need for environmental data, academic researchers from a range of disciplines, and educators all see a value in the system. The CIS is now available as a commercial software package. Apart from data collected during the Countryside Surveys, any data registered to the Ordnance Survey 1-km grid can be handled.
2. Countryside Information System - Development and System Requirements The CIS (version 5.30) operates in Microsoft Windows TM and will run on any PC which is capable of supporting Windows version 3.1 or later. The software runs better if there is at least 5 Mb of RAM and requires around about 30 Mb of hard-disk space. The disk requirement is flexible, since Countryside Survey data, supplied with the system, are not needed for the system to function. However, it is possible to load new datasets and generate datasets from within the system. The system supplied includes survey data from the Countryside Surveys, (including a satellite-derived Land Cover Map (LCM)) and summary soils and geology data characterising the land classes. The field survey data are from the surveys in 1978, 1984 and 1990 and describe land cover, linear features and vegetation, the information being stored as land class averages. The LCM presents the proportion of each 1 km square covered by each of 17 different land-cover types. The data were derived by interpreting multi-temporal Landsat TM imagery (bands 3, 4 and 5); where possible scenes were from 1990, but problems with cloud cover forced extra scenes to be used from the immediately adjacent years. Details of the methods used in generating the LCM can be found in Fuller et al. (1993). The programming was initially carded out in C, but has now been updated to Microsoft Visual C++. Dart Computing, a commercial software company, wrote the computer code under instruction from the Steering Group and development team. The programs run in a Windows environment and follow the normal conventions for that style of software. The intention throughout the development of the system was that it should be capable of running on a standard office PC, and that users should not need extensive training or high levels of specialised skill to operate it.
3. Countryside Information System - Data Presentation There are obvious risks of inappropriate application of any system or dataset, especially if it is intended for use by people with little formal training. The potential for mis- and over-interpretation increases with the flexibility of data analysis and presentation. Using a sample dataset of 500 squares to predict a population of 250 000 is appropriate so long as the statistical assumptions are not broken and an appropriate level of confidence is placed in the results. Within the system, it is possible to select a single square and request predictions for that from a national
376
DAVID C. HOWARD AND R,G.H. BUNCE
survey. The predictions for that square would not be expected to be correct, so the system warns that the totals are unreliable. However, the figures are still presented, since they allow real measures, collected outside the system, to be compared with a predicted average. Predictions produced from sample data using the CIS are normally presented along with a standard error to assist in evaluation of the results. In a user-friendly system, it is tempting to present confidence intervals or even ranges, but the statistical assumptions of normality cannot be tested satisfactorily, so a standard error is presented. Data are qualified in other ways by associating textual descriptions, definitions and statements of accuracy. The CIS also has a full help system which not only advises on the technical operation of the system, but also offers comments on the data and sources of additional assistance. The development has avoided "re-inventing the wheel" by allowing users to take information from CIS into other packages. Windows offers an open architecture allowing linking to other packages through the clipboard or Dynamic Link Library (DLL). CIS does allow rudimentary graphical presentation of tabular data (as bar or pie charts), but data can easily be moved into spreadsheets and graphics packages allowing further analysis and more complex presentation.
4. Countryside Information System - Operation CIS consists of three utilities, each with an independent screen icon. The main body of the system opens with two windows. The left-hand portion of the screen shows a map of GB. The map is merely a pixel presentation of every 1 km square. The right window, or data window, presents information describing the selected area in the map window. The contents of both windows can be altered in different ways. The map window can be magnified by dragging and selecting with the mouse; such a magnification does not alter the information in the data window. Both windows can be altered by refining the region. Refining the region refers to making a selection of a portion of the squares in Great Britain and this operation can be performed in a number of different ways. To select squares a command from the Edit menu is chosen. Methods of choosing the squares are then presented and a button selected for the appropriate way. Methods include screen painting with a mouse, choosing counties or countries, using a file with grid references, or selecting from various data themes. Figure 1 shows the screens after the region has been defined using a file containing all squares in the National Parks of England and Wales. The map window shows the distribution of the squares and the top of the window shows the number of squares selected (15 007). A data window has been selected to show the distribution of squares in each ITE land class in the selected region; there are 26 of the 32 land classes found within the National Parks. Data screens can easily be changed within CIS to show different data or different formats; it is not necessary to see the
377
THE COUNTRYSIDE INFORMATION SYSTEM IN GREAT BRITAIN
File
Edit
View
Supplementaw
Help Landelass Table ICensus. Internal
Current Region [1 5007] --EE-
2
I 34 5 6 7
I
'
8 9 10 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
57 2 11 4 325 847 332 253 319 418 204 6t 508 300 5035 1093 1658 877 5 1207 554 196 174 154 184 229
5700 200 1100 400 32500 84403 19099 16107 31900 41800 20173 5611 50648 29613 503500 109300 165800 87700 500 120700 55400 19600 17400 15400 18400 22900
5700 200 1100 400 32500 84312 19099 16107 31900 41800 20173 5611 50648 29613 503500 109215 165800 87700 500 120700 55400 19600 17400 15239 18400 22900
4579 178 667 282 25666 71542 16881 13194 24426 32144 16948 4683 37503 22388 444238 87874 150395 77040 473 106040 52846 18799 15356 11801 15831 18287
Fig. 1. CountrysideInformationSystem (CIS) screen showingthe land class breakdownof National Parks (right). Columns in the Landclass Table show, for 28 land classes (Lc), the total number of squares, land area (i.e., less sea), rural land area (CS Rural; i.e., less sea and non-surveyed urban land) and open countryside (Open C; land left unclassifiedon a 1:250 000 scale map); areas are in hectares.
land class data. Using the land class areas and field survey data, the CIS produces a prediction for the region. Average values for different landscape features in each land class are held in the system. These are multiplied by the number of squares in the appropriate land classes and summed to produce national or regional totals. The figures for different types of woodland, as seen on the CIS, are shown in Figure 2. The system is capable of calculating values in different ways. In this case, the Countryside Survey 1990 (CS1990) sampled only rural land and, as the screen shows, 1-km squares which were over 75% built up were omitted. The rural part of omitted squares is included in the total and the figures at the bottom report on the proportion which related to that. It would be possible to present results using the means as land area or full square area (including sea). Users have the capability to add and edit both sample and census data using the two other sections of the system, Both the CIS Sample Database and the CIS Census Database have their own independent icons. The land class means for the conifer woodland, worked through to national estimates in Figure 2, are shown in a screen from the CIS Sample Database in Figure 3. Each numbered box in the
378
DAVID C. HOWARD AND R.G.H. BUNCE
File Edit View Supplcmenta W
Help Woodlands (Sample)
Description Conifer woodlands Mixed woodland I~'ood~ayed waods Shrub Feged woodland
% Data Cover 100.00 100.00 100.00 100.O0 100 .IX)
----(ha/sq km rural land)-Mean Std Error 7.198 1.233 0.774 0.233 3.384 0.468 0.351 0.064 0.082 0.031
I~ Total 106200 11410 49940 5176 1210
(ha on rural l a n d ) - - Std Error 18190 3442 6912 943 454
Definition of Rural Land Countryside Survey results have been applied to all land in tl',e current region except the urban areas of squares with more than 75% urban cover The rural parts 01 tl',ese 'urban squares' ware not surveyed either, b~ the results have been applied to this 'tmclessified urban frin_Qe'on the assumption that its content is proba~y similar to the content of the rural squares. Relative proportions are given below: Description Area Area (he) (%) Urban Land excluded 337 0.02 Non-Urban Land (Urban Fringe) 63 0.00 Total Urban square land 400 0.03
Fig. 2. An example screen of a CIS data window showing predictions for the coverage of woodlands, using the ITE Land Classification for National Parks, as shown in Figure ]. Note the error terms qualifying the results and the additional land excluded from the survey because it was built-up.
screen holds an average area for one land class. There is a similar screen for loading and editing standard errors, as well as screens for adding textual descriptions of the feature and its accuracy. Both sample and census data can also be imported via ASCII files. The CIS Sample Database also allows the composition of the data tables viewed within the full CIS to be determined and manipulated. Apart from calculating the area of different features for different regions, the CIS can also map predicted distributions and allow simple overlay. The predicted distribution of coniferous woodland across GB, as seen on the CIS, is shown in Figure 4. It is possible to alter the number of categories and the break points of each category. The distribution map is produced by ranking, and then colouring, different land classes; care must be taken not to over-interpret this type of map. The land classes do match an intuitive breakdown of GB, but they are generalisations and are merely used in this instance to predict the probability of finding woodland. The CIS is also designed to manipulate and present census data (i.e., where each 1 km square has an individual data record). CS1990 used two approaches to measure the stock of GB's landscape features. The first, as described above, used the ITE Land Classification to target field survey of a restricted sample; the second used satellite imagery to produce a LCM (Fuller et al., 1993). The CIS can be used to make comparisons between different datasets and an example is given for broad-leaved woodlands (field survey) and deciduous woodlands (LCM). A region was stored which contained the upper quartile of the squares in GB when ranked on the area of deciduous woodland in the LCM. This was overlayed onto the upper quartile of broad-leaved woodlands as predicted from field survey. The result, shown in Figure 5, shows the size and distribution of the four groups; the distributions are somewhat different, but with a broad
"Internal census dataset 0 Square area line sea on coastal sq] 0 Land area ~) ITE Countryside Survey area 0 Open Countryside area 0 ITE Hedgerow Reporl area
Full Title I I Conifer wo~ I Means will be applied to an internal census datasct modified by an optional secondary sample dalaset
per sq km
Units:
Data Umits'
Title: Conifer woodlands "Data by Land Class
-Data Values with special meanings]
Fig. 3. The input screens in the CIS Sample Database utility showing the opening screen (left) and land class averages for conifer woodland surveyed in 1990 (right).
Dense bracken 1 Dense heath Drier northern bog Dune Felled woodland Field beans Hard areas Secondary sample dataset Hard coast l [N o n el Horticulture Selected dataset is used in the I Confidence Limits I "Means representfollowing Tables: 0 Values ~) Asymmetric AJl data 121 0 Symmetric Densities
Short Title Coniler woodlands Datasets in this file: Agrlc. buildings Barley Barry-bush heath Broadlcavcd woods
,.,J
r~ ,-1 m
Z
0
.-]
380
_File
DAVID C. HOWARD AND R.G.H. B U N C E
Edit
View
Help
_Supplementary
egs- 7s4
I:~.|
Analyse Dataset Table Key to Analysis of SAM: Conifer woodlands Range O1o [0.1925] 0.1925 to [1.4155] 1.4155 to [2.1585] 2.1585 to [5.8115] A ~ 5.8115to [8.3975] 8.3975to 29.641 I t No DatafOutof Range Total Squares in range
Squares 21668 32703 50014
55275 26321 41672 12569 227653
Figures in square brackets are not included in the range.
Fig. 4. A distribution map of conifer woodlands in 1990 based upon the national field survey, and presented in 7 categories (grey-tones).
area of overlap. The value of such a comparison is not to reject either estimate (they will both differ from the truth for different reasons), but to help a user to take several estimates and make a value judgement. In the example, deciduous woodland appears to be most common in southern and central England. Differences between datasets do have many causes; some are purely methodological (e.g., survey versus census, or manual versus automatic recording), others are temporal (e.g., recorded over different time-scales or in different years) and yet others reflect the presentation. The terms used to describe features may often suggest recording similar attributes, but these always need to be checked. In the example here, deciduous and broad-leaved woodlands were considered comparable. Although in terms of tree species that may be almost true (exceptions being species such as larch, Larix decidua Mill.), the terms are general and widely used offering no sharp definition. Within the dataset details held on the system are descriptions, covering features such as the relationship to mixed woodland, scrub and orchards, the minimum parcel size recorded, and the area measured (e.g., canopy or base area). Another part of the CIS called Land Use Classification Information and Documentation (LUCID) offers assistance in making comparisons between published datasets (Wyatt et al., 1994). The utility, separately commissioned by DOE, holds
381
THE COUNTRYSIDE INFORMATION SYSTEM IN GREAT BRITAIN
_File _Edit _View Supplementary, Current R e g i o n [ O v e r l a y ]
Help Overlay Analysis Table
I~
S e l e c t D a t a s e t to be a n a l y s e d f o r t h i s o v e r l a y :
iiiiiiiill
ICEN:deciduous woodland Region
Density ha/sq km 1.985 17.34 15.16 1 225
I~ % Data Cover 100.0 100.0 100.O 100.0
Region k,ey Bose Region only - Field survey woodland Overlap (Base 8, Overlay) Overlay Region only - LCM deciduous woodland I : ; 1Neither Region
Total ha
92510 5061 O0 471 go0 163300 Area in sq km 46616 29186 31064 133356
Fig. 5. Areaspredicted to have most deciduous woodland in 1990 (in both cases, the upper quartile of the distribution). The data window shows the predictions for deciduous woodland in each of the four areas (field survey alone, field survey and LCM overlapped, LCM alone, and neither field survey nor LCM). definitions of categories from 17 different surveys; these include both the LCM and the CS 1990 field survey. Apart from giving a text definition there is also a matrix which identifies how a category from one classification would be presented in another. For our example, if deciduous woodland from the LCM is matched with the CS1990 field survey, four categories appear. Broad-leaved woodlands is obvious, but the LCM may also identify cropland with perennial crops (the code including orchards in the field survey), shrub and mixed woodlands as broad-leaved woodlands.
5. Application As described earlier, the user-centred development has lead to the system being widely bench-tested before its development was complete. For example, changes in hedgerow length were analyzed and presented using the CIS during the drafting of legislation to protect hedgerows and encourage better management (Barr et al., 1991). As Steering Group members worked in different government departments,
382
DAVID C. HOWARD AND R.G.H. BUNCE
information could be passed in CIS format. Ministries, departments and agencies could load their own data confidentially to compare with the CS1990. The sharing and availability of data are starting to fulfil one of DOE's aims for the CIS, namely to provide a level playing field of information, on which policies can be discussed and shaped. In part due to the success of the CIS, some government research contracts are now being let with the proviso that information will be suitable for manipulation and presentation in the CIS. The further goal is to define and set standards within a framework, so that information can be compared and integrated with confidence.
6. Discussion Although CIS is now in a form that can be sold as commercial software, there are areas which need further development. The value of the system will grow as more data are available in an appropriate format and structure. DOE has let a contract to identify, acquire and install different datasets onto the CIS; again this is user-driven, and the Steering Group has identified the topics and areas of highest priority. The other aspect of development involves functionality. Some sections of the system can be improved (for example, the control of colour in map production) and as a wider group of users comes into contact with the package, more will, no doubt, be identified. Other developments may take place outside the package. The open architecture that is encouraged in Windows programs means that utilities can be written to pass information back and forth from within the system. A simple model buffering points has already been written which will define a CIS region as the squares within a specified radius of a given point. The datasets within CIS can then be used to map and characterize the region. Other work, which is structured around the ITE Land Classification, can also be loaded and presented on the system. So, for example, the Land Use Allocation Model (Harvey et al., 1986), which models changes in agriculture through agricultural economics, can use the CIS to estimate the environmental consequences of predicted change. The breadth and diversity of application make the system difficult to categorise; it has been compared with expert systems (ES), information systems (IS), decision support systems (DSS) and GIS (Wadsworth, 1992; Stott, 1994). Unfortunately, the categories themselves overlap considerably adding to the confusion. GIS is a utility which can load, edit, manipulate, analyze and present spatial data; CIS can perform those functions to a limited extent. However, CIS is not a GIS in that its limitations in each function make a true GIS a valuable tool to use in tandem. ES and IS differ in the type of information held (knowledge and data) and the approaches to processing (heuristic and algorithmic respectively). CIS, arguably, holds both types of information, although the processing is predominantly algorithmic. DSS falls somewhere between the other categories, being designed to assist decision
THECOUNTRYSIDEINFORMATIONSYSTEMIN GREATBRITAIN
383
makers confront poorly-structured problems by interacting directly with data and analysis models (Sprague and Watson, 1993). For a decision maker to make use of CIS as a DSS requires not only answers to problems, but also information about the quality and reliability of the data. The presentation of error terms may offer some measure of the data quality, but it still requires expertise to interpret. CIS is designed to also carry further qualifiers in the form of alternate datasets (either spatially disaggregated or as simple tables in a supplementary data catalogue) and text descriptions. The latter can be updated by users so that their experience with the system becomes available to them and their colleagues. The production of an information system to carry data, including samples collected using an objective stratification, should increase interest in sampling efficiency. The ITE Land Classification has been used since the 1970s and has acquired a good reputation for providing a means to target ecological and environmental surveys. The CIS will, hopefully, increase familiarity and maintain its reputation.
Acknowledgements The CIS was commissioned by the Department of the Environment (UK), and throughout its development, members of the Department have been of great assistance. In particular, John Peters, Dr. Terry Parr and Dr. Andrew Stott should be thanked for their ideas, support and hard work. Members of the Steering Group should also be thanked for their active participation and constructive criticism, which helped shape the system.
References Barr, C.J., Howard, D.C., Bunce, R.G.H., Gillespie, M.K. and Hallam, C.J.: 1991, 'Changes in hedgerows in Britain between 1984 and 1991', Institute of Terrestrial Ecology, Grange-overSands, UK, 13 pp. Barr, C.J., Bunce, R.G.H., Clarke, R.T., Fuller, R.M., Furse, M. T., Gillespie, M.K., Groom, G.B., Hallam, C.J., Hornung,M., Howard,D.C. and Ness, M.J.: 1993, 'CountrysideSurvey 1990 Main Report', Countryside 1990 Series, VolumeII, Dep. Environ.,London,UK, 174 pp. Bunce,R.G.H.,Barr,C.J. and Whittaker,H.A.: 1983, 'A stratificationsystemfor ecologicalsampling', In: R. M. Fuller(ed.), EcologicalMappingfrom Ground, Air and Space, ITE SymposiumNo. 10, Grange-over-Sands, UK, pp. 34-46. Bunce, R.G.H., Howard, D.C., Clarke, R.T. and Lane, A.M.J.: 1991, 'The ITE land classification: classification of all 1 km squares in GB', Institute of Terrestrial Ecology,Grange-over-Sands, UK, 74 pp. Bunce, R.G.H., Howard, D.C., Hallam,C.J., Barr, C.J. and Benefield,C.B.: 1993, 'Ecologicalconsequences of land use change', Countryside1990 Series, VolumeI, Dep. Environ.,London,UK, 129 pp. Fuller, R.M., Brown, N.J., Ullyett,J.M., Sanders, M.E., Groom,G.B., Howard, D.C. and Gillespie, M.K.: 1993, 'CountrysideSurvey 1990 - mappingthe land cover of Great Britain using Landsat imagery: a demonstrator project in remote sensing, final report on pattern analysis and GIS', Institute of Terrestrial Ecology,Abbots Ripton, UK, 29 pp.
384
DAVIDC. HOWARDAND R.G.H. BUNCE
Harvey, D.R., Barr, C.J., Bell, M., Bunce, R.G.H., Edwards, D., Errington, A.J., Jollans, J.L., McClintock, J.H., Thompson, A.M.M. and Tranter, R.B.: 1986, 'Countryside implications for England and Wales of possible changes in CAP', Centre for Agricultural Strategy, Reading UK, 313 pp. Hill, M.O., Bunce, R.G.H. and Shaw, M.W.: 1975, 'Indicator species analysis, a divisive polythetic method of classification, and its application to a survey of native pinewoods in Scotland', Journal of Ecology 63, 597-613. Howard, D.C., Bunce, R.G.H., Jones, M. and Haines Young, R.: 1994, 'The development of the Countryside Information System', Countryside 1990 Series, Volume IV, Dep. Environ., London, UK, 55 pp. Stott, A.: 1994, 'The Countryside Information System', In: Countryside Recreation Network News, Vol. I1, University of Cardiff, UK, pp. 6-9. Sprague, R. and Watson, H.J.: 1993, Decision Support Systems, Putting Theory into Practice, Prentice Hall, Englewood-Cliffs, NJ, 437 pp. Wadsworth, R.M.: 1992, 'Software implementation of a decision support system for land use planning', In: M.C. Whitby (ed.), Land Use Change: Causes and Consequences, ITE Symposium No. 27, London, UK. Wyatt, B.K., Greatorex Davies, N., Bunce, R.G.H., Fuller, R.M. and Hill, M.O.: 1994, 'Comparison of land cover definitions', Countryside 1990 Series, Volume III, Dep. Environ., London, UK, 131 pp.
Abstract. The Institute of Terrestrial Ecology (ITE) has monitored ecological change in Great Britain (GB) since 1978. The task has been undertaken using a stratified sampling scheme working with a 1 km square as the sample unit. In more recent years, scientific researchers at ITE have been working closely with the policy-makers of the United Kingdom Department of the Environment. The presentation of information to policy advisors and planners was a component within a large project investigating the ecological consequences of land-use change. A simple PC-based decision support system was developed during the project and subsequently has been expanded to produce a marketable product. The system, called the Countryside Information System (CIS), presents and links information at national, regional and thematic levels along with qualifying data describing accuracy and appropriateness of use (i.e., metadata). An integral part of the CIS is the ITE Land Classification, which divides GB into 32 environmental land classes; all 250 000 squares have been classified. The classification allows sampled data to be presented and, as the co-ordinate system is widely used in GB, it allows census datasets to be linked and compared. CIS has been described as a Geographical Information System, but the classification, data held within the system, and the use of metadata to assist in interpretation of results make the system much more decision-support oriented. Indeed, government departments have been involved in directing the development and are now starting to use the system to answer parliamentary questions and formulate, assess and monitor environmental policy. The CIS is an open system, running on a standard PC in Microsoft Windows. Tools for loading and editing new datasets (both sample and census) are incorporated in the suite of programs. The Windows environment and users comments during development have produced a system with an intuitive feel, removing some of the overhead of acquiring specialised technical skills before being able to operate a system. This paper describes the CIS and presents examples of its applications.
1. Introduction In order to formulate and assess effective environmental policy, ecological and environmental information relevant to both the subject and region must be available. Unfortunately, detailed information that is necessary to formulate environmentally sensitive policies is expensive to collect and cannot always be obtained in the time-scale of parliamentary procedures. Part of the Institute of Terrestrial Ecology's (ITE) remit is to monitor land-cover change in the United Kingdom (UK) and identify the interactions between land cover, land use and ecology. A statistically rigorous sampling scheme was devised to collect information that can be interpreted at a national scale. A sample unit of 1 km square was selected as a compromise between the practicable and the desirable. Practically, the unit had to be capable of being surveyed in the field, while the desire was that it be large enough to contain sufficient variation to be representative. Environmental Monitoring and Assessment 39: 373-384, 1996. 1996 Kluwer Academic Publishers. Printed in the Netherlands.
374
DAVID C. HOWARD AND R.G.H. BUNCE
The stratification was produced by applying a multivariate analysis to environmental data derived from published maps. The analysis used was Indicator Species Analysis (ISA), which was originally developed for phytosociology (Hill et al., 1975). Initially, only squares at the intersection of a 15 km x 15 km grid were classified, but these were supplemented by a second grid of squares identified using a dichotomous key (Bunce et al., 1983). The original classification, even including the supplementary squares, only covered about 4% of GB. Different techniques were investigated to classify all squares in GB. The key, produced by ISA, was not ideal; as a dichotomous key it would radically mis-classify if there was an uncharacteristic value in a factor contributing to the early couplets. As there were a number of projects using the classification, it was decided to maintain the original structure as far as possible. Eventually all squares in GB were classified using a combination of techniques. Logistic discrimination divided the population into 8 groups, each of which was split into four groups by discriminant function analysis (Bunce et al., 1991). Sample squares for field survey were drawn at random from each stratum at the intersections of the original grid. National field surveys, called Countryside Surveys, were carded out in 1978, 1984 and 1990 (see Barr et al., 1993) producing figures for changes in land cover and vegetation change. Although, at the time, the original survey was considered to be largely a scientific investigation, the results it produced created interest in a wider audience than anticipated. In the following years, an increased awareness of the value and fragility of the environment has lead to greater consideration of environmental issues in government policy. The Department of the Environment (DOE) sponsored a research project to investigate the ecological consequences of land-use change (Bunce et al., 1993). One aspect of the project was to demonstrate the methods that could be employed to effectively pass information, including field survey results, to policy advisors. Having identified personal computers (PC) as a valuable tool for storing and presenting information, a number of approaches were developed. These ranged from an information system presenting data with spatial selection, through outline processing (holding and linking textual data) to full expert systems (offering diagnosis and suggested action from uncertain data). An information system was thought to hold the greatest potential and so was developed further in a second project commissioned by DOE (Howard et al., 1994). The development was user-driven; a number of potential users from different government departments and agencies were recruited and formed a Steering Group. They were supplied with copies of the software and datasets and were asked to attempt to use them for trial scenarios and applications. Comments and criticisms from the Steering Group were passed back to the development team who modified the programs and circulated new versions. The system, now called the Countryside Information System (CIS) has attracted a lot of interest, not just from government agencies. Non-governmental organisa-
THE COUNTRYSIDE INFORMATION SYSTEM IN GREAT BRITAIN
375
tions (NGO) with a need for environmental data, academic researchers from a range of disciplines, and educators all see a value in the system. The CIS is now available as a commercial software package. Apart from data collected during the Countryside Surveys, any data registered to the Ordnance Survey 1-km grid can be handled.
2. Countryside Information System - Development and System Requirements The CIS (version 5.30) operates in Microsoft Windows TM and will run on any PC which is capable of supporting Windows version 3.1 or later. The software runs better if there is at least 5 Mb of RAM and requires around about 30 Mb of hard-disk space. The disk requirement is flexible, since Countryside Survey data, supplied with the system, are not needed for the system to function. However, it is possible to load new datasets and generate datasets from within the system. The system supplied includes survey data from the Countryside Surveys, (including a satellite-derived Land Cover Map (LCM)) and summary soils and geology data characterising the land classes. The field survey data are from the surveys in 1978, 1984 and 1990 and describe land cover, linear features and vegetation, the information being stored as land class averages. The LCM presents the proportion of each 1 km square covered by each of 17 different land-cover types. The data were derived by interpreting multi-temporal Landsat TM imagery (bands 3, 4 and 5); where possible scenes were from 1990, but problems with cloud cover forced extra scenes to be used from the immediately adjacent years. Details of the methods used in generating the LCM can be found in Fuller et al. (1993). The programming was initially carded out in C, but has now been updated to Microsoft Visual C++. Dart Computing, a commercial software company, wrote the computer code under instruction from the Steering Group and development team. The programs run in a Windows environment and follow the normal conventions for that style of software. The intention throughout the development of the system was that it should be capable of running on a standard office PC, and that users should not need extensive training or high levels of specialised skill to operate it.
3. Countryside Information System - Data Presentation There are obvious risks of inappropriate application of any system or dataset, especially if it is intended for use by people with little formal training. The potential for mis- and over-interpretation increases with the flexibility of data analysis and presentation. Using a sample dataset of 500 squares to predict a population of 250 000 is appropriate so long as the statistical assumptions are not broken and an appropriate level of confidence is placed in the results. Within the system, it is possible to select a single square and request predictions for that from a national
376
DAVID C. HOWARD AND R,G.H. BUNCE
survey. The predictions for that square would not be expected to be correct, so the system warns that the totals are unreliable. However, the figures are still presented, since they allow real measures, collected outside the system, to be compared with a predicted average. Predictions produced from sample data using the CIS are normally presented along with a standard error to assist in evaluation of the results. In a user-friendly system, it is tempting to present confidence intervals or even ranges, but the statistical assumptions of normality cannot be tested satisfactorily, so a standard error is presented. Data are qualified in other ways by associating textual descriptions, definitions and statements of accuracy. The CIS also has a full help system which not only advises on the technical operation of the system, but also offers comments on the data and sources of additional assistance. The development has avoided "re-inventing the wheel" by allowing users to take information from CIS into other packages. Windows offers an open architecture allowing linking to other packages through the clipboard or Dynamic Link Library (DLL). CIS does allow rudimentary graphical presentation of tabular data (as bar or pie charts), but data can easily be moved into spreadsheets and graphics packages allowing further analysis and more complex presentation.
4. Countryside Information System - Operation CIS consists of three utilities, each with an independent screen icon. The main body of the system opens with two windows. The left-hand portion of the screen shows a map of GB. The map is merely a pixel presentation of every 1 km square. The right window, or data window, presents information describing the selected area in the map window. The contents of both windows can be altered in different ways. The map window can be magnified by dragging and selecting with the mouse; such a magnification does not alter the information in the data window. Both windows can be altered by refining the region. Refining the region refers to making a selection of a portion of the squares in Great Britain and this operation can be performed in a number of different ways. To select squares a command from the Edit menu is chosen. Methods of choosing the squares are then presented and a button selected for the appropriate way. Methods include screen painting with a mouse, choosing counties or countries, using a file with grid references, or selecting from various data themes. Figure 1 shows the screens after the region has been defined using a file containing all squares in the National Parks of England and Wales. The map window shows the distribution of the squares and the top of the window shows the number of squares selected (15 007). A data window has been selected to show the distribution of squares in each ITE land class in the selected region; there are 26 of the 32 land classes found within the National Parks. Data screens can easily be changed within CIS to show different data or different formats; it is not necessary to see the
377
THE COUNTRYSIDE INFORMATION SYSTEM IN GREAT BRITAIN
File
Edit
View
Supplementaw
Help Landelass Table ICensus. Internal
Current Region [1 5007] --EE-
2
I 34 5 6 7
I
'
8 9 10 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
57 2 11 4 325 847 332 253 319 418 204 6t 508 300 5035 1093 1658 877 5 1207 554 196 174 154 184 229
5700 200 1100 400 32500 84403 19099 16107 31900 41800 20173 5611 50648 29613 503500 109300 165800 87700 500 120700 55400 19600 17400 15400 18400 22900
5700 200 1100 400 32500 84312 19099 16107 31900 41800 20173 5611 50648 29613 503500 109215 165800 87700 500 120700 55400 19600 17400 15239 18400 22900
4579 178 667 282 25666 71542 16881 13194 24426 32144 16948 4683 37503 22388 444238 87874 150395 77040 473 106040 52846 18799 15356 11801 15831 18287
Fig. 1. CountrysideInformationSystem (CIS) screen showingthe land class breakdownof National Parks (right). Columns in the Landclass Table show, for 28 land classes (Lc), the total number of squares, land area (i.e., less sea), rural land area (CS Rural; i.e., less sea and non-surveyed urban land) and open countryside (Open C; land left unclassifiedon a 1:250 000 scale map); areas are in hectares.
land class data. Using the land class areas and field survey data, the CIS produces a prediction for the region. Average values for different landscape features in each land class are held in the system. These are multiplied by the number of squares in the appropriate land classes and summed to produce national or regional totals. The figures for different types of woodland, as seen on the CIS, are shown in Figure 2. The system is capable of calculating values in different ways. In this case, the Countryside Survey 1990 (CS1990) sampled only rural land and, as the screen shows, 1-km squares which were over 75% built up were omitted. The rural part of omitted squares is included in the total and the figures at the bottom report on the proportion which related to that. It would be possible to present results using the means as land area or full square area (including sea). Users have the capability to add and edit both sample and census data using the two other sections of the system, Both the CIS Sample Database and the CIS Census Database have their own independent icons. The land class means for the conifer woodland, worked through to national estimates in Figure 2, are shown in a screen from the CIS Sample Database in Figure 3. Each numbered box in the
378
DAVID C. HOWARD AND R.G.H. BUNCE
File Edit View Supplcmenta W
Help Woodlands (Sample)
Description Conifer woodlands Mixed woodland I~'ood~ayed waods Shrub Feged woodland
% Data Cover 100.00 100.00 100.00 100.O0 100 .IX)
----(ha/sq km rural land)-Mean Std Error 7.198 1.233 0.774 0.233 3.384 0.468 0.351 0.064 0.082 0.031
I~ Total 106200 11410 49940 5176 1210
(ha on rural l a n d ) - - Std Error 18190 3442 6912 943 454
Definition of Rural Land Countryside Survey results have been applied to all land in tl',e current region except the urban areas of squares with more than 75% urban cover The rural parts 01 tl',ese 'urban squares' ware not surveyed either, b~ the results have been applied to this 'tmclessified urban frin_Qe'on the assumption that its content is proba~y similar to the content of the rural squares. Relative proportions are given below: Description Area Area (he) (%) Urban Land excluded 337 0.02 Non-Urban Land (Urban Fringe) 63 0.00 Total Urban square land 400 0.03
Fig. 2. An example screen of a CIS data window showing predictions for the coverage of woodlands, using the ITE Land Classification for National Parks, as shown in Figure ]. Note the error terms qualifying the results and the additional land excluded from the survey because it was built-up.
screen holds an average area for one land class. There is a similar screen for loading and editing standard errors, as well as screens for adding textual descriptions of the feature and its accuracy. Both sample and census data can also be imported via ASCII files. The CIS Sample Database also allows the composition of the data tables viewed within the full CIS to be determined and manipulated. Apart from calculating the area of different features for different regions, the CIS can also map predicted distributions and allow simple overlay. The predicted distribution of coniferous woodland across GB, as seen on the CIS, is shown in Figure 4. It is possible to alter the number of categories and the break points of each category. The distribution map is produced by ranking, and then colouring, different land classes; care must be taken not to over-interpret this type of map. The land classes do match an intuitive breakdown of GB, but they are generalisations and are merely used in this instance to predict the probability of finding woodland. The CIS is also designed to manipulate and present census data (i.e., where each 1 km square has an individual data record). CS1990 used two approaches to measure the stock of GB's landscape features. The first, as described above, used the ITE Land Classification to target field survey of a restricted sample; the second used satellite imagery to produce a LCM (Fuller et al., 1993). The CIS can be used to make comparisons between different datasets and an example is given for broad-leaved woodlands (field survey) and deciduous woodlands (LCM). A region was stored which contained the upper quartile of the squares in GB when ranked on the area of deciduous woodland in the LCM. This was overlayed onto the upper quartile of broad-leaved woodlands as predicted from field survey. The result, shown in Figure 5, shows the size and distribution of the four groups; the distributions are somewhat different, but with a broad
"Internal census dataset 0 Square area line sea on coastal sq] 0 Land area ~) ITE Countryside Survey area 0 Open Countryside area 0 ITE Hedgerow Reporl area
Full Title I I Conifer wo~ I Means will be applied to an internal census datasct modified by an optional secondary sample dalaset
per sq km
Units:
Data Umits'
Title: Conifer woodlands "Data by Land Class
-Data Values with special meanings]
Fig. 3. The input screens in the CIS Sample Database utility showing the opening screen (left) and land class averages for conifer woodland surveyed in 1990 (right).
Dense bracken 1 Dense heath Drier northern bog Dune Felled woodland Field beans Hard areas Secondary sample dataset Hard coast l [N o n el Horticulture Selected dataset is used in the I Confidence Limits I "Means representfollowing Tables: 0 Values ~) Asymmetric AJl data 121 0 Symmetric Densities
Short Title Coniler woodlands Datasets in this file: Agrlc. buildings Barley Barry-bush heath Broadlcavcd woods
,.,J
r~ ,-1 m
Z
0
.-]
380
_File
DAVID C. HOWARD AND R.G.H. B U N C E
Edit
View
Help
_Supplementary
egs- 7s4
I:~.|
Analyse Dataset Table Key to Analysis of SAM: Conifer woodlands Range O1o [0.1925] 0.1925 to [1.4155] 1.4155 to [2.1585] 2.1585 to [5.8115] A ~ 5.8115to [8.3975] 8.3975to 29.641 I t No DatafOutof Range Total Squares in range
Squares 21668 32703 50014
55275 26321 41672 12569 227653
Figures in square brackets are not included in the range.
Fig. 4. A distribution map of conifer woodlands in 1990 based upon the national field survey, and presented in 7 categories (grey-tones).
area of overlap. The value of such a comparison is not to reject either estimate (they will both differ from the truth for different reasons), but to help a user to take several estimates and make a value judgement. In the example, deciduous woodland appears to be most common in southern and central England. Differences between datasets do have many causes; some are purely methodological (e.g., survey versus census, or manual versus automatic recording), others are temporal (e.g., recorded over different time-scales or in different years) and yet others reflect the presentation. The terms used to describe features may often suggest recording similar attributes, but these always need to be checked. In the example here, deciduous and broad-leaved woodlands were considered comparable. Although in terms of tree species that may be almost true (exceptions being species such as larch, Larix decidua Mill.), the terms are general and widely used offering no sharp definition. Within the dataset details held on the system are descriptions, covering features such as the relationship to mixed woodland, scrub and orchards, the minimum parcel size recorded, and the area measured (e.g., canopy or base area). Another part of the CIS called Land Use Classification Information and Documentation (LUCID) offers assistance in making comparisons between published datasets (Wyatt et al., 1994). The utility, separately commissioned by DOE, holds
381
THE COUNTRYSIDE INFORMATION SYSTEM IN GREAT BRITAIN
_File _Edit _View Supplementary, Current R e g i o n [ O v e r l a y ]
Help Overlay Analysis Table
I~
S e l e c t D a t a s e t to be a n a l y s e d f o r t h i s o v e r l a y :
iiiiiiiill
ICEN:deciduous woodland Region
Density ha/sq km 1.985 17.34 15.16 1 225
I~ % Data Cover 100.0 100.0 100.O 100.0
Region k,ey Bose Region only - Field survey woodland Overlap (Base 8, Overlay) Overlay Region only - LCM deciduous woodland I : ; 1Neither Region
Total ha
92510 5061 O0 471 go0 163300 Area in sq km 46616 29186 31064 133356
Fig. 5. Areaspredicted to have most deciduous woodland in 1990 (in both cases, the upper quartile of the distribution). The data window shows the predictions for deciduous woodland in each of the four areas (field survey alone, field survey and LCM overlapped, LCM alone, and neither field survey nor LCM). definitions of categories from 17 different surveys; these include both the LCM and the CS 1990 field survey. Apart from giving a text definition there is also a matrix which identifies how a category from one classification would be presented in another. For our example, if deciduous woodland from the LCM is matched with the CS1990 field survey, four categories appear. Broad-leaved woodlands is obvious, but the LCM may also identify cropland with perennial crops (the code including orchards in the field survey), shrub and mixed woodlands as broad-leaved woodlands.
5. Application As described earlier, the user-centred development has lead to the system being widely bench-tested before its development was complete. For example, changes in hedgerow length were analyzed and presented using the CIS during the drafting of legislation to protect hedgerows and encourage better management (Barr et al., 1991). As Steering Group members worked in different government departments,
382
DAVID C. HOWARD AND R.G.H. BUNCE
information could be passed in CIS format. Ministries, departments and agencies could load their own data confidentially to compare with the CS1990. The sharing and availability of data are starting to fulfil one of DOE's aims for the CIS, namely to provide a level playing field of information, on which policies can be discussed and shaped. In part due to the success of the CIS, some government research contracts are now being let with the proviso that information will be suitable for manipulation and presentation in the CIS. The further goal is to define and set standards within a framework, so that information can be compared and integrated with confidence.
6. Discussion Although CIS is now in a form that can be sold as commercial software, there are areas which need further development. The value of the system will grow as more data are available in an appropriate format and structure. DOE has let a contract to identify, acquire and install different datasets onto the CIS; again this is user-driven, and the Steering Group has identified the topics and areas of highest priority. The other aspect of development involves functionality. Some sections of the system can be improved (for example, the control of colour in map production) and as a wider group of users comes into contact with the package, more will, no doubt, be identified. Other developments may take place outside the package. The open architecture that is encouraged in Windows programs means that utilities can be written to pass information back and forth from within the system. A simple model buffering points has already been written which will define a CIS region as the squares within a specified radius of a given point. The datasets within CIS can then be used to map and characterize the region. Other work, which is structured around the ITE Land Classification, can also be loaded and presented on the system. So, for example, the Land Use Allocation Model (Harvey et al., 1986), which models changes in agriculture through agricultural economics, can use the CIS to estimate the environmental consequences of predicted change. The breadth and diversity of application make the system difficult to categorise; it has been compared with expert systems (ES), information systems (IS), decision support systems (DSS) and GIS (Wadsworth, 1992; Stott, 1994). Unfortunately, the categories themselves overlap considerably adding to the confusion. GIS is a utility which can load, edit, manipulate, analyze and present spatial data; CIS can perform those functions to a limited extent. However, CIS is not a GIS in that its limitations in each function make a true GIS a valuable tool to use in tandem. ES and IS differ in the type of information held (knowledge and data) and the approaches to processing (heuristic and algorithmic respectively). CIS, arguably, holds both types of information, although the processing is predominantly algorithmic. DSS falls somewhere between the other categories, being designed to assist decision
THECOUNTRYSIDEINFORMATIONSYSTEMIN GREATBRITAIN
383
makers confront poorly-structured problems by interacting directly with data and analysis models (Sprague and Watson, 1993). For a decision maker to make use of CIS as a DSS requires not only answers to problems, but also information about the quality and reliability of the data. The presentation of error terms may offer some measure of the data quality, but it still requires expertise to interpret. CIS is designed to also carry further qualifiers in the form of alternate datasets (either spatially disaggregated or as simple tables in a supplementary data catalogue) and text descriptions. The latter can be updated by users so that their experience with the system becomes available to them and their colleagues. The production of an information system to carry data, including samples collected using an objective stratification, should increase interest in sampling efficiency. The ITE Land Classification has been used since the 1970s and has acquired a good reputation for providing a means to target ecological and environmental surveys. The CIS will, hopefully, increase familiarity and maintain its reputation.
Acknowledgements The CIS was commissioned by the Department of the Environment (UK), and throughout its development, members of the Department have been of great assistance. In particular, John Peters, Dr. Terry Parr and Dr. Andrew Stott should be thanked for their ideas, support and hard work. Members of the Steering Group should also be thanked for their active participation and constructive criticism, which helped shape the system.
References Barr, C.J., Howard, D.C., Bunce, R.G.H., Gillespie, M.K. and Hallam, C.J.: 1991, 'Changes in hedgerows in Britain between 1984 and 1991', Institute of Terrestrial Ecology, Grange-overSands, UK, 13 pp. Barr, C.J., Bunce, R.G.H., Clarke, R.T., Fuller, R.M., Furse, M. T., Gillespie, M.K., Groom, G.B., Hallam, C.J., Hornung,M., Howard,D.C. and Ness, M.J.: 1993, 'CountrysideSurvey 1990 Main Report', Countryside 1990 Series, VolumeII, Dep. Environ.,London,UK, 174 pp. Bunce,R.G.H.,Barr,C.J. and Whittaker,H.A.: 1983, 'A stratificationsystemfor ecologicalsampling', In: R. M. Fuller(ed.), EcologicalMappingfrom Ground, Air and Space, ITE SymposiumNo. 10, Grange-over-Sands, UK, pp. 34-46. Bunce, R.G.H., Howard, D.C., Clarke, R.T. and Lane, A.M.J.: 1991, 'The ITE land classification: classification of all 1 km squares in GB', Institute of Terrestrial Ecology,Grange-over-Sands, UK, 74 pp. Bunce, R.G.H., Howard, D.C., Hallam,C.J., Barr, C.J. and Benefield,C.B.: 1993, 'Ecologicalconsequences of land use change', Countryside1990 Series, VolumeI, Dep. Environ.,London,UK, 129 pp. Fuller, R.M., Brown, N.J., Ullyett,J.M., Sanders, M.E., Groom,G.B., Howard, D.C. and Gillespie, M.K.: 1993, 'CountrysideSurvey 1990 - mappingthe land cover of Great Britain using Landsat imagery: a demonstrator project in remote sensing, final report on pattern analysis and GIS', Institute of Terrestrial Ecology,Abbots Ripton, UK, 29 pp.
384
DAVIDC. HOWARDAND R.G.H. BUNCE
Harvey, D.R., Barr, C.J., Bell, M., Bunce, R.G.H., Edwards, D., Errington, A.J., Jollans, J.L., McClintock, J.H., Thompson, A.M.M. and Tranter, R.B.: 1986, 'Countryside implications for England and Wales of possible changes in CAP', Centre for Agricultural Strategy, Reading UK, 313 pp. Hill, M.O., Bunce, R.G.H. and Shaw, M.W.: 1975, 'Indicator species analysis, a divisive polythetic method of classification, and its application to a survey of native pinewoods in Scotland', Journal of Ecology 63, 597-613. Howard, D.C., Bunce, R.G.H., Jones, M. and Haines Young, R.: 1994, 'The development of the Countryside Information System', Countryside 1990 Series, Volume IV, Dep. Environ., London, UK, 55 pp. Stott, A.: 1994, 'The Countryside Information System', In: Countryside Recreation Network News, Vol. I1, University of Cardiff, UK, pp. 6-9. Sprague, R. and Watson, H.J.: 1993, Decision Support Systems, Putting Theory into Practice, Prentice Hall, Englewood-Cliffs, NJ, 437 pp. Wadsworth, R.M.: 1992, 'Software implementation of a decision support system for land use planning', In: M.C. Whitby (ed.), Land Use Change: Causes and Consequences, ITE Symposium No. 27, London, UK. Wyatt, B.K., Greatorex Davies, N., Bunce, R.G.H., Fuller, R.M. and Hill, M.O.: 1994, 'Comparison of land cover definitions', Countryside 1990 Series, Volume III, Dep. Environ., London, UK, 131 pp.
