BA 302 - Managing Space - Site Selection - GIS

• Problem: Does 'space' or 'location' matter in business operations? Is geography really "history"?
• Learning outcomes:
• Recognize spatial/locational aspects of operations.
• Recognize how to use spatial information to make spatial decisions.
• Recognize the role of Geographic Information Systems (GIS):
• Know the basic difference between vector and raster GIS.
• Realize that 'flat' maps are projections that imply distortion.
• Realize that different projections imply different distortions.
• Realize that GIS ≠ mapping.

• Regression as a way to find a pattern in data (not at all limited to spatial data!!!!).
• Know how to ask tough questions about your textbook.

• Metters et al. (p. 328): single or infrequent vs. multiple or frequent site selection:
•  Single/infrequent:
  
• Car manufacturer wants one or two production plants in North America.
• Fujitsu needs a new call center.
• Where does the new nuclear power plant go?
• Where will I put my Indian-/US tax filing business?
• Where do I  establish my  one-person furniture  making business? My restaurant? My bed-and-breakfast?
  

• Multiple/frequent:
  

• Where do the next 10 Starbucks' go?
• Where to locate mail (drop off) boxes?
• How do I lay out bus lines through the city and where to put in stops?
• If I shuttle people to/from the airport and back, where do I have (fixed) stops?

• Recurring spatial decisions:
  
• If I shuttle people to/from the airport who do I drop of first, second, etc.?
• Which routes to follow when delivering pizza?
• Which taxi gets dispatched to which address?

• Overcoming 'friction of distance' implies:
• using time ==> cost.
• transportation ==> cost.
• uncertainty (space is not always identically configured; e.g., empty vs. crowded roads).

• Metters et al. offer some thoughts on site selection:
• Problem: Ideas on how to select a site?
• Metters et al. (p. 330-331) factor-rating; multicriterion evaluation; utility model:
1. Determine choice dimensions: Metters et al. example of restaurant:
• Neighborhood income.
• Proximity to shopping center.
• Accessibility(?)
• Visibility(?)
• Traffic(?)

2. Determine the dimensions' weights.
3. Determine how the dimensions and their weights combine into an overall score (determine the combination rule); e.g.,
   
• Additive rule (f):  U_i = Σ_d(i_dw_d).
• Multiplicative rule (f): U_i = Π_d(i_dw_d).

4. Score the alternatives (the choices) on the dimensions.
5. Compute the alternatives' utilities.
6. Pick the one with the highest utility.

• Problem: Have we seen this model/approach before?
  
• Problem: Do Metters et al. like this factor rating? Do they like the model previously?

• Metters et al. (p. 332-335): regression:
1. Collect two types of data:
• Performance data on similar businesses; e.g., existing Starbucks stores, fast food restaurants, hotel franchises, bank branch offices, etc.
• Spatial/locational information for each of these businesses; e.g.,
  • Neighborhood income.
  • Proximity to working population.
  • Pedestrian traffic.
  • Car traffic.
  • Etc.

2. Model the performance of the businesses as a function of the locational characteristics:
   
• Performance = f(Neighborhood income, Proximity to working population, Pedestrian traffic, Car traffic, ...).

• Performance = α + β₁(Neighborhood income) + β₂(Proximity to working population) + ... + β_n(Variable_n).
• Performance is the dependent variable.
• Neighborhood income, Proximity, etc. are the independent variables.
• β₁, β₂, ...,  β_n are called the regression coefficients.
• Each regression coefficient indicates how much the dependent variable will change if the associated independent variable changes one unit.
• How to find β₁, β₂, ...,  β_n? ==> least squares criterion; i.e., find values for β₁, β₂, ...,  β_nsuch that the squared differences between the observed values and the predicted ones are minimized.
  

3. Use the found β_n values and the spatial/locational information of a new site to predict the performance of the business at that site.

• Example: La Quinta Inn case study pp. 348-355.
  
• An observation:
  • Regression model is mathematically equivalent to the factor rating model.
  • Metters et al.'s strongest critique on the factor rating model is that the weights are determined arbitrarily.
  • However, it seems that the regression model can be used to estimate these weights.
    

• Problem: can you see some potential problems with this approach?
• Model assumes that the dependent variable is a linear function of the independent variables (guess what will happen if we put in an exponential data set? e.g., Figure 16.3 (p. 334).
• Model makes strong assumptions about how data are distributed.
• Every site is special; the model might not be valid for a specific site.
• How to get the data with which to fit the model?
• Have lots of other similar business establishments (Starbucks, Banks, Hotels, etc.).
• Use trade organization.
• Use specialized consulting company.
• Collect your own using a Geographic Information System and available spatial data.

• Geographic Information Systems (GIS):
• Metters et al. (p. 336): "software that links location to (nonspatial RR) information."
• "A system for capturing, storing, checking, integrating, manipulating, analyzing and displaying data which are spatially referenced to the Earth." (UK Dept. of the Environment, 1987).
• Problem: what do we mean by 'spatially referenced?' ==> things have position: longitude, latitude and optionally elevation.
  
• Absolute position.
• Relative position:
• Albany is downstream of Corvallis on the Willamette River.
• Albany is north of Corvallis.
• Madagascar is an island in the Indian ocean off the southeast coast of Africa.
• How many people between the ages of 35 and 40 live in Corvallis?

• The problem of projections:
• On the earth's surface things have longitude/latitude coordinates; Can get with techniques such as GPS.
• However, expensive to collect if we must map many objects; e.g., homes and land, sewer lines, etc.
• ==> digitize locational information from maps.

  

• Problem: earth is a sphere; map is flat ==> map is not accurate; has distortion:

  

• Some map projection families:
• Cylindrical:
• e.g. Mercator projection

  

• Maps entire sphere except poles.
• Maps entire sphere except poles.
• Good for equatorial areas but distances and hence, areas sharply exaggerated towards the poles.

• Azimuthal:
• Maps half or less of the sphere.
• Distortion strong along edges of the map.
• Distance mostly preserved.

• Conic:
• Area distorted; shape mostly preserved.
• Distance strongly distorted at bottom of the map.

• Therefore, when combining several map layers, must always transform into a single, common coordinate system.

• Vector or raster?
  • Vector: All things spatial are either points, lines or polygons.

    
    

    • Points have no size, just position.
    • Lines are points connected; have no width.
    • Polygons are lines connected in sequence.
    • Infinite precision.
    • Great for building complex spatial structures such as route and sewer networks.
    • Low date volume requirements.
    • Very good for topological computations (relative position).
    • Complex computing.

  • Raster: Space is 'quantized' in discrete, square chunks:

    

    • Everything has size; even points and lines.
    • Precision is limited to the size of the raster cell (pixel).
    • Need more detail? --> make cells smaller --> more data.
    • Easy computing.
    • Goes well with satellite imagery.

      

  • Vector and raster can be converted into each other but:
  • Vector --> raster implies loss of information.
  • Raster --> vector implies making assumptions.

  • Some vendors:
    • ESRI: ARCGIS
    • Clark labs: IDRISI
    • US Corps of Engineers: GRASS
    • Intergraph
    • MapInfo

  • Some cases:
  • Metters et al. (p. 340, 341): site selection for 'delivered' services; e.g., ambulance, pizza, meals-on-wheels, FedEx/UPS/DHL, etc.
  • Problem: What are some of  the spatial considerations that factor into this decision?
  • Problem: Would you favor a vector or raster GIS for this problem?

  • As the US Bureau of Land Management (BLM) you worry about nonnative grass infestations across the Western US. You rely on satellite imagery to inventory populations of these species.
  • Problem: Would you favor a vector or raster GIS for this problem?