This section ivolves the specifics behind the analysis.

 DATA: 
     For the type of analysis being performed (Ordered Weighted Averaging), all factors must be standardized to a continuous scale.  This allows the comparison of factors.  For example, without standardizing the data, it would be impossible to tell the relative value of factors.  Is a site 412 meters from a highway at an elevation of 100 meters better than a site next to a highway but at 3 meters elevation?

      This standardization is accomplished by making the data of every layer be in the range 0 - 255.  For example, the enumeration areas with the highest population values through standardization now had a value of 255, while the least populated areas had a value of 0.  See the Data section for details on standardization for each factor.

 ANALYSIS: 
     Once the data was gathered converted to IDRISI grid format, and standardized, the actual spatial analysis was performed.  There were now 8 layers: 2 constraints, 6 constraints. The factors were all standardized to a continuous scale.

     The factors were now weighted using the WEIGHT function.  In this function, the user creates a wieght of importance for each factor by to a pairwise comparison. Each factor is compared to each other factor and given a value.  Once all factors are compaired, a weight is developed for each factor.  All weights add up to 1.  To see the factor weights developed and the reasons behind them, click here.  These weights make a large difference to the final analysis, as can be seen by these images. 

      Once the weights were decided on, the next step is to look at the way in which the weights established in the previous step tradeoff with each other.  This can be called controling risk and tradoff by establishing Order Weights.  The reason for this step is to answer questions similar to the following: 

"If a candidate is very near an exising site, should the candidate be instantly ruled out, or will the fact that it is on the optimal land use (Resource and Industrial) make it a strong candidate?". 

     Factor weights alone express the relative importance of each criteria for the overall objective.  The use of Order weights allows us to control how our factor weights influence the final suitability map.
    I divided my factors into 3 classes.  Those with no tradeoff, some tradeoff and full tradeoff.  For those factors with no tradeoff, there is very low risk.  For example if proximity to existing cell sites is given an Order Weight of "no tradeoff", then the answer to our question above would be:

"Yes, the candidate would be instantly ruled out, there is no room to trade proximity to exisiting cell sites for any of the other factors." 

If proximity to existing cell sites is given an Order Weight of  "some tradeoff", then the answer to our question above would be:

"No, the candidate would not be instantly ruled out, maybe it IS still a strong candidate."

     To see the Order Weights chosen, and the reasons behind them, click here.

     Once the factor weights were established, the next step was the actual processing to create a map showing suitability. This image shows a continuous surface of the study area ranging from 0 - 255.  The most suitable areas would have a pixel value of 255.  This is accomplished through the IDRISI function MCE (stands for Multi-Criteria Evaluation).  Using the Ordered Weighted Averaging option, the image is created.  In this function, each factor layer is examined. Every pixel is altered using the weights established above.  The factors are then added together to create the final suitability image.

SOURCES OF ERROR AND POSSIBLE IMPROVEMENTS:

     In gathering and cleaning the data, many problems were encountered.  The DEM was in rough shape.  It was digitized by following along contour lines which makes it visually not very appealing.  There were also "NO DATA" points littered equally spaced throughout the layer.  Errors in the DEM were of course felt in the SLOPE coverage as well.  If time allowed, this layer would have been filtered and cleaned up a little better.
     It was also found that Idrisi was not very stable when importing raster layers, or converting vector layers to raster.  One of the shortcomings of FME 2002 is its inability to handle raster conversions.  This lead to my reliance on ArcInfo for most of the conversions.
     In the actual methodolgy, there are many shortcomings.  This is a simple analysis that does not take into account much of the relevant factors influencing RF signal propogation.  For example, just because a PCS antenna candidate location falls on Industrial land, this does not garuantee that it would be an adequate location.  What if the candidate is surrounded by dense forest on all sides?
     The analysis also does not take into account actual cell phone usage patterns.  The analysis is based on population values collected by Census Canada.  These numbers represent residence population.  Most cell phone usage occurs in the day time, when people are not in their homes.  Some kind of day time population figures should have been included for a more accurate analysis. For information on how day time populations can be predicted, click here.
     The analysis also simplifies the value of individual land use types.  This is in part because of the land use coverage.  By classifying Burnaby into only 7 land uses, it limits the scope of analysis.  Although it is said that residential is not a favoured land use, dense residential is in fact a favoured location for antenna placement.  This however is not mentioned.
     Also, nothing is mentioned about building height.  This is very important when assessing antenna placement.  The city of Burnaby is very restrictive about building towers for antennae.  This leaves carriers reliant on attaching antennae to exisitng buildings.  This factor was not included in the analysis.

     Furthermore, to finish off the analysis, it would have been nice to re-run the dB Planner propgation including the new chosen PCS antenna sites.