| 11.520: A Workshop on Geographic Information Systems |
| 11.188: Urban Planning and Social Science Laboratory |
| Distributed: | Monday, November 5th, 2007 |
| Due: | Task 1: Friday, November 16th, 2007, 5 PM (via Stellar) |
NOTE 1: If you need help, and you think that your question might be of
interest to the whole class, send e-mail to 11.520@mit.edu.
If you would prefer to ask just the class staff, send e-mail to 11.520staff@mit.edu.
Regardless, please don't be shy about asking for help if you are struggling
with understanding the assignment or having trouble with a particular aspect
of ArcMap. We've heard many stories of students struggling for many hours over
minor issues that had easy solutions. We'd like to avoid these misadventures,
so please contact us sooner rather than later if you get stuck.
NOTE 2: This assignment builds directly on the endpoint of Homework 2. To avoid problems resulting from the use of different endpoints from homework #2, we provide a shapefile, finalresult.shp, in the hm2 sub-directory of the class data locker. This shapefile identifies areas that are acceptable locations based on the accessibility, health, land use, and demographic criteria from HW#2. Please note, however, that this shapefile differs somewhat from the actual answer to HW#2 - it was developed using slightly different cutoffs for the various criteria. (So, do not worry if the shapefile looks different from the shapefile you developed to answer HW#2!)
In Homework 2, you identified those places that met the accessibility, health, land use, and demographic criteria. You present your first set of results to the non-profit group in charge. Now you are asked to add another factor -- the density of poor senior citizens. After some discussion, you agree that it will be desirable to have the center close to as many poor seniors as possible. Consequently, you decide to augment your analysis by adding another criteria related to the density of poor seniors.
Please use the block group layer M:\data\eastma2000bg_stateplane.shp for the calculation of the poor senior population density. (Otherwise you would have to project the Library's blockgroup layer to Mass stateplane, and then recompute area, before you could get an accurate measure of area in square meters so that you could compute the density of poor seniors.) .
The Homework 2 screening exercise identified more than one site that met all the initial criteria. Adding in this new 'density of poor seniors' consideration can help you further narrow down the identification of a preferred site. Among those feasible sites already identified, you will prefer those that fall in the higher density-of-poor seniors locations. You know that the census data provide counts of poor seniors (aged 65+) for each block group so you can use those data (and the area_sqm field) to compute a density-of-poor seniors measure. A thematic map of this measure (by block groups) will show you where the density of poor seniors is high. But that isn't quite what you want. You would like a measure of how many poor seniors have easy access to the center -- and those could include poor seniors from neighboring block groups as well as any from the block group in which the center is located. So, you decide to use ArcMap's Spatial Analyst tools to develop a better measure. First, you'll create a rasterized (grid) version of the block groups (with the block group poor senior citizen density as the cell value). Then you can use the 'neighborhood statistics' tools to create a new grid layer that has, for its cell value, the average of all the cells within some buffer distance (a circle of 1000 meter radius is what we suggest below).
To follow through on this idea, create a grid coverage that rasterizes the block groups into grid cells that are 100 meters on edge. Be sure to mask off all but the 5-town area when you create the grid coverage. Set the value in each cell to be your estimate of the density of poor seniors in the block group at the center of the grid cell. Create a thematic map showing the density of poor senior citizens across your Cambridge grid cells.
Now, use the Spatial Analyst > Neighborhood Statistics option to generate a new grid layer where the value of each grid cell is the average (mean) of the density of poor seniors in the grid cells. Define the neighborhood to be a circle (of grid cells) of 1000-meter radius centered on each grid cell. Notice how the new layer looks different from the original rasterization of your block group densities. (Do you understand why?)
Now, intersect this smoothed density-of-poor seniors layer with finalresult.shp, in the hm2 sub-directory of the class data locker - the shapefile identifying the locations that we will consider to be acceptable based on criteria similar to those stated in Homework 2.
One way to select a specific location for the poor senior center would be to pick the (1 hectare) site among the acceptable locations from Homework 2 that had the highest density of nearby poor seniors. Add another view to the layout of the thematic map you just created which shows this density of nearby poor seniors. Annotate this layout to clarify which map shows the densities: before and after doing the neighborhood averaging.
Summary of Task 1 Requirements:
Turn in one map layout with two views -- one showing the block group density-of-poor seniors before the neighborhood averaging and one showing the density-of-poor seniors after the neighborhood averaging. Adjust and annotate the map so it is readable and add a few sentences explaining what each map is measuring and commenting on any shift in pattern that you observe. As in Homework 2, you will be graded on the quality of both your analysis and your presentation of your results.
What to Make of All This:
Task 1 asks for one map with two views (of the block group poor senior density before
and after neighborhood averaging). Make sure that the neighborhood average view
includes some visual representation (i.e. visual overlay) of the HW #2
acceptable sites so the reader can see which high-density cells fall within
the HW #2-acceptable locations. Doing this is more a matter of attention to
cartography and visual display rather than computing a new combined index.
While our formulation of the problem does narrow down the number of possible site locations, many of the criteria were rather arbitrary -- why within 300 meters of a major road instead of 200 or 500 meters? In general, you will want to do some reasoning and make some judgments about how to balance (or adjust!) the various criteria that you have incorporated into your site selection screening process. You will also want to cast your analysis within a framework within which is it as easy as possible to adjust or add criteria, or reweight the analysis to take into account different priorities. Within stock ArcGIS, the best way to do this is to create a ModelBuilder model of your suitability analysis using the "weighted overlay" tool to "rate and weight" the final criteria. Fortunately, having worked through much of this in the prior homework, you should already understand the data and their limitations.
Revise the analysis done in homework 2, creating a ModelBuilder suitability analysis model. In this model, the primary criteria should be as continuous as possible, rather than discrete. For example, proximity to major roads and distance from TRI facilities should be treated as continuous distance surfaces rather than specific buffer distances. Recall that the desired land use characteristics are both discrete (buildable land), and semi-continuous. In practice, various categories of existing land use would be easier or more difficult to build, so even that factor could be made continuous by reclassifying land use into a "buildability" index. In the vector analysis, we used the criterion "abuts, but is not actually in, a residential area." In the raster analysis, you could also require adjaceny to residential on a grid cell basis, but a more flexible formulation might be "as near as possible to residential areas." Similarly, the raster formulation would lead you to express the density of poor seniors as continuous. With all this in mind, we have four criteria to consider:
For the purposes of this exercise, we will ignore the criterion of a 1 hectare mimum site size in the original formulation of the problem. This concept was easily expressed in vector by measuring the size of the polygon overlay fragments which met all criteria. In raster, this same concept is more difficult to express because we lack an appropriate bounding geometry. By facilitating the use of continuous measures, raster analysis tends to produce broad gradient patterns. If you were given a vector parcels layer, you could use "zonalStatistics" to summarize suitability at the parcel level. ("ZonalStatistics" is one of a few GIS operators which summarizes rasters using vector data, the other being "Extract Values to Points"). However, you are not given a parcels database, so we don't include that in this homework.
The final component in this model should be a "weighted overlay" in which criteria are explicitly weighted. Start with an equal weighting of factors, but experiment with at least a couple of alternate weightings. Do you see a significant difference in pattern? What happens when you "restrict" or exclude extremely bad areas from particular criteria?
Write a short report (maximum of 2 double-spaced pages, 10-point font, 1-inch margins) interpreting the results of your analysis and explaining any comments you may have about the allowable locations. Should the density criterion for seniors in poverty be relative or absolute? Would you prefer to relax one of the other criteria and shift the site elsewhere? Would you suggest some tradeoffs among the criteria? Do the criteria restrict the sites a lot or a little? Do they appear to capture the individual criteria reasonably?
There are not 'correct' answers to these questions -- and we do not expect you to research the many real-life characteristics of Cambridge that could influence your decision. The intent is for you to back away from ArcMap and GIS and spend some time reflecting on the problem of locating a poor senior center and thinking about how your specific variables, cutoffs, and visualization tools may have biases, artificial limits, or have otherwise colored or overlooked relevant factors. Why do you think you have (or have not) zeroed in on one or more sensible suggestions for where to put the center?
Recognize in Task 2 that we are not trying to get you to pick a 'best' location. Rather we want you to use the data and GIS tools to boil down the many criteria and possible sites to a reasonable and informed discussion of the pluses and minuses of several specific sites and the primary criteria and tradeoffs that are likely to be involved in making the selection. You'll want to be able to refer to your Task-1 map in your discussion and it will be easier to make your points if your maps help the reader visualize the spatial patterns (in terms of buffers, land use density, etc.) that you've identified for the various criteria. There will be times in your professional career when you will be asked to conduct an analysis and recommend a single "best" site based entirely on your own analysis. However, there will likely be many more occassions in which you will work with a group of people (experts or general public) to identify criteria, and to weight and rate them. Doing this fairly, using a transparent process, can be difficult. For example, what if two experts or two citizens disagree on what the weights should be? Part of the point in tackling the same problem with different tools is for you to start to gain an appreciation of how the technical tools choosen can influence the decision - sometimes in subtle ways.
Finally, notice that we've simplified the site selection task to ignore the cost of acquiring the property. Write a few sentences explaining how such cost considerations might steer the site selection toward one or another of the better locations that your analysis identified and what further analysis you might do (using GIS tools and data of the kinds we've been using) to sort through this question. You do not actually have to perform this analysis -- just indicate how you would approach it. You are welcome to reference your answer from HW # 2, but you should amplify on how the additional work in HW # 3 (this assignment) augmented your analysis.
Homework 2 and Homework 3 developed
by Kamal Azar.
Modified by Joe Ferreira on 7 March 1999.
Revised by Anne Kinsella Thompson on November 5, 1999.
Revised for Fall 2000/01 by Thomas H. Grayson
Revised for Fall 2002-4 by Myounggu Kang and Jinhua Zhao.
Revised for Fall 2006 by Michael Flaxman.
Last modified: November 5, 2007 by Joe Ferreira.
Back to the 11.520
Home Page.
Back to the CRN Home Page.