Orleans House Area Desirability:
Multi-Criteria Evaluation Using Vector and Raster Functionality

GIS has become an integral part of dealing with spatial data: acquisition, storage, and systems development. The power of GIS can help us make decisions based on intelligent maps, algorithms and systems. This is perhaps best displayed in output GIS applications. By linking information (statistical, tabular) to map data, the power of visual analyzation is added to the decision making process. We can also create maps based on assumptions or given attributes, to assess and predict poten tial phenomenon.

This paper will demonstrate raster-based analysis with multi-criteria evaluation to extract desired information. The environment will be working with a client to find best location for housing in a particular area, based on the user’s needs. This will show the power of GIS in producing quick and clever outputs based on user identified input. The software used for this study is ESRI’s ArcView 3.1, with the 3D and Spatial Analyst Extensions.

The hypothetical client profile is a professional Francophone couple with two pre-school children who are interested in moving to the Orleans, Ontario area, as shown in Figure 1. Their needs are as follows:

• Somewhat close to Catholic educational institutions
• Close proximity to bus routes
• Francophone majority
• University educated
• Higher income bracket environment

The couple would have also expressed the relative importance of each of these factors respectively into weights as percentages.

We are given the following data for the area: a demographic profile from the 1996 census, by census tract for Orleans and Blackburn Hamlet; a shape file of the census tracts; a 1997 RMOC street network; school locations, and an Ottawa Ri ver shape file for orientation. We are also given a digital image of the bus network to help analyze routes.

The first step to solve this problem was to join the demographic data to the shapefile, to connect the statistical data to the shapes they represent in 2D. The school location data was then imported into ArcView against the census data.

We then digitized a coverage consisting of the bus routes in Orleans. MapInfo was used for this operation, as its ArcView counterpart was found to be cumbersome in comparison. The resulting line theme was then analyzed for continuous distance from a bus route. The resulting output grid theme was reclassed according to the couple’s criteria (see Appendix A).

The census data was then queried to select all Catholic elementary school in Orleans, both English and French. The same function was applied to show distance from the selected schools, in accordance with the couple’s input, the reclassed relative to A ppendix A, into grid data.

At this point we have two grids based on distance, reclassed to variable or field scores. The three other variables specified by the couple were then calculated. The demographic data was given two new fields, percentFrMt and percentUnivEdu, to show t he values of university educated and French mother tongue as percentages relative to the entire sample of the area. Three more grids were created, with the field representing cell values being ppercentFrMt, percentUnivEdu, and Avinc respectively. The ou tput grids were also reclassed to reflect the field scores specified in Appendix A.

A note on calculations: the demographic data percentage calculation was evaluated against the total population of the census tract, as opposed to population over 15. The values can very for things such as income and university education. The total p opulation

was used in this study to show a true comparison of the area as a whole, and due to the fact that the couple had two younger children ("kids are people too") to consider. It is

believed that this would not ‘hide’ any values, and that results would be more concise to the application.

Now all the data desired is present in output grids, reclassed to the client’s needs. Working with Spatial Analyst’s Map Calculator, an equation was formulated to assign weights to all grids as specified in Appendix A, for all five grids. The field w eights illustrate the relative importance of each variable to the client. Figure 2 shows the algorithm.

The result was a grid, which displayed the most desirable area for the client to live, in the Orleans area, based on their input. The output grid was also output to a 3D scene for visual enhancement (Appendix A).

The final map (Appendix A) returns the most desirable area to live in Orleans for the clients. This illustrates the typical GIS process outlined in this paper: data acquisition / importing, input, analysis, output product. Theoreticall y, the GIS has done what it has been asked. It is now up to the client to decide if this suits their needs. The algorithm to

compute the final grid cell values was correct, but the data inputted may be misleading. For example, the clients placed the highest importance of house desirability on average income of people in the neighbourhood. The income values did not exceed $ 40000, thus the output grid for average income showed one class, as those under $30000 were considered as not applicable. Perhaps the output product would have benefited from further parsing down these values to better represent the data. Also, since th e weighting of the areas that applied was given a value of zero, this made no effect on the final product. If the couple had put 100% of their interest into this variable, they would get a null impression of the area. In fact, the deliverable is somewha t misleading, as the couple may interpret the most desirable area as one of higher affluence, when in reality is a product of the four other variables.

This issue is also relative to the client and methodology; should the data be presented to the client before their criteria is submitted, so as to give them a high-level overview of the area, or should the client yield their wants to the consultant suc h as in this case study?

Although the 3D display capabilities of GIS software can be very appealing, it made little difference to this case study, even with high vertical exaggeration values. The 3D display of GIS is better utilized when working with more continuous data, and at a denser rate. More adequate examples would be a dense, continuous surface DEM of height values. The values this study worked with were from 1-5, and very discrete.

Another issue raised here is that of working with grids as continuous surfaces as opposed to classification to what was specified by the client. There are many benefits and caveats to treating surfaces in two different manners. First, by scaling down the data to the various field scores and weights, we further generalize our findings to make analysis

more feasible, and computational resources less exhausted. The client receives a higher lever view and analysis of the area and criteria. However, by doing so, the upper and lower edges of each variable are reclassed into a discrete cell value and ma y result in misleading data. For example, if two areas are identical in all other variables, but the average incomes are $49900 and $51000 respectively, how will it reflect the reality of income in the area? Will they be partial to the latter area based on a difference of $2000? Will they disapprove of the former area in comparison?

Investigating the approach to treat the grids as continuous surfaces yielded similar results. The final data in this case would be more realistic in terms of actual values. While this would use more computing power to generate, the results would be v ery precise and articulate, allowing a great level of detail to be displayed / analyzed. However, how much detail is needed for a study such as house area desirability? Creating such a complex data set and display may confuse the client rather than enli ghten them. Remember, you are asked to provide an analysis of the area, to make the decision easier and feasible for the client. All analysis has some degree of generalization, assumptions to encounter.

I believe the answer lies in judgement of the client and the consultant, and especially the application. In any event, when the final data is presented, along with a report and notes, the consultant should make recommendations and expla in the context of the data to the client. While the models, calculations and algorithms are all efficient for both modes of

data analysis, it is up to the user to decipher what to use, and when to use it. For example, take into account a lab generating ground control data. While the output data points are very accurate, their use is relative to the application. If a cli ent wants to create a mosaic of Radarsat imagery, they would not have need for dead-on accurate point data. Conversely, an organization compiling DEM data for military purposes will need to use these points in a very accurate manner to verify their spot heights.

In conclusion, the data and algorithm is relevant to the application, which in turn is dependent on the needs of the client or stakeholder.

Hybrid Home