Methodology

1. Procedures
2. Cartographic Model

1  Procedures

The set of procedures taken to reach the final selection of suitable site locations include:
  1) querying the datatable and then analyzing the displays to see any interesting trends
  2) specifying degrees of suitability using FUZZY module
  3) determining relative importance of the criterias and assigning weights to the criteria
  4) finally using MCE (multi-criteria evaluation) to aggregate the criterias and selected constraints       into outputing a final map showing potential locations for the site

1) In this intial step I ran single and multi-attribute queries on the datatable to see the immediate results on the generated display. For instance I did a query (or filtering) to find all census tracts with high densities, or all census tracts which have a medium/high median income and has high population densities. The categories of high, medium and low for each demographic field were chosen subjectively by looking at the range of values for each field in the datatable and then choosing arbitrary break points. The results of each query on the datatable were saved by exporting them to an AVL file so that I can generate a raster image later using EDIT and ASSIGN.

The importance of querying the database in this initial step was to see if it was necessary to maintain certain criterias set from the beginning. For instance, from the querying I found that there is not one census tract which isn't adjacent or nearby another census tract where there were medium/high median household income or where there was medium/high proportion of families with children. Therefore the results suggest that these criterias (median household income and proportion of families with children) will not be necessary to maintain because they don't factor much into the evaluation of a location for the restaurant site.

2) I used the FUZZY module to specify a more continuous suitability range for each criteria. The simple Boolean constraints were not used because they didn't provide enough flexibility required for the analyses. For instance the image comfuzz.rst shows a continuous suitability range as a distance decay function from the commercial centers. This reflects a better real life simulation than the Boolean method.

3) To have some understanding of what criterias are most important, I decided to use restaurant.rst to see how other competing fast food restaurants have decided to locate within the region. My query will try to find a relationship between the location of the competitor restaurants with factors such as distance to major roads, distance to commercial areas, distance to high populated regions, and distance to bus routes.. Idrisi32 provides the QUERY module which generates an image for analysis (but not for display). The module takes an input file from which a query is drawn and a mask file which specifies the pixels to be included in the query. After running QUERY, I ran HISTO which displays the frequency of distances to [commercial area, high density areas, etc.] among the restaurants. As it can be seen, most of the restaurants are within commercial areas, many are within 100m of a major road, and many are close to bus routes (most within 200m), etc.. From these findings I can give a relative weight to the criterias. However the final decision on assigning weight values is still somehwhat arbitrary although it now reflects careful consideration of the histogram results. One area that is somewhat problematic and will be discussed in the section Problem Issues is that this analysis relies heavily on the accuracy of the absolute locations of my digitized restaurant points.



Query1: This shows the frequency of competitor restaurants distance to major roads in Burnaby.




Query2: Shows frequency of restaurant distances to bus routes.




Query3: Shows frequency of restaurant distances to commercial areas.




Query4: Shows frequency of restaurant distances to medium and high density census tracts.

The results of looking at the histograms allowed me to make a more objective judgement of relative weights. However, contrary to the results of the histograms I gave a slight nod for major roads to be more important than population density due to the realization that there might be more potential for error in the histogram for major roads than for population density. I also decided that more customers will use cars than bus. I then ran WEIGHT which allowed me to input the relative importance of the criteria into a pairwise comparison matrix. It also automatically computed the weights so that they total 1. I decided to keep these weights since they showed a high consistency.

The eigenvector of weights is :
mjroadsfuzz : 0.2349
comfuzz : 0.4520
burnbusfuzz : 0.0966
popdenfuzz : 0.2166
Consistency ratio = 0.04 Consistency is acceptable.

4) I Ran MCE and Weighted Linear Combination(WCE) using the bestweight.dsf (parameter file) created from WEIGHT in the previous step. Then I chose the four factor images listed in the table above as well as choosing two constraints, burnbool and combuff, where burnbool masked out areas not in Burnaby and combuff masked out all areas not designated commercial or within 100 meters of a commercial zone.

Last thing to consider is to specify a reclass of the suitability values to reflect either a one or a zero. The reclass gave all pixel values with suitability values of greater than 200 a value of one and the rest zero. The result is shown in the next section Spatial Analysis.


2  Cartographic Model

• The main cartographic model   >>
• The cartographic model for the spatial query   >>