In class Exercise: Week 8

Predictive (Locational) Models

As discussed in class, predictive models range in complexity from a simple collection of landform criteria for planning efficient survey strategies, to complex logistic regression functions where the contribution of each variable is made explicit. As described in the readings, these various models can be categorized by three major elements: the types input data, types of output data (form of results), and goals of the model.

A simple Rule-based survey model can be constructed by assembling various criteria and progressively delimiting the terrain included in the survey. For example, one might survey all terrain in a river valley that is less than 500m from the highest river terrace, and slopes under 15 degree slopes. The construction of such a model in Arcmap would involve several Spatial Analyst operations.

First, all the land below the high terrace is delimited with a polygon.

Next, the distance from that polygon can be calculated using

Toolbox > Spatial Analyst Tools > Distance > Euclidean Distance...

for a more complex approach consider using a "cost distance" model such as Tobler's hiking function based on travel time that can be implemented in

Toolbox > Spatial Analyst Tools > Distance > Cost Distance...

The resulting distance GRID can be classified into two regions: one within (with 0 values) and one outside of (1 values) the specified travel distance (or time) using Spatial Analyst > Reclassify...

Next, the land under 15 degree slope can be selected by reclassifying the output from a Spatial Analyst Slope calculation. Again if two GRIDs are specified with 0 and 1 values.

Finally, these output GRIDs can be combined in raster calculator (GRID1 + GRID2) and all the places with "2" in the output are in the overlap zone between the two GRID files.

 The output might be converted to Vector (polygon) and mobile GIS can be used to guide the survey to remain within the bounds specified with the model.

Consider do a swath of "100%" coverage in order to test the assumptions of your model. What kinds of sites were found with the 100% survey that were missed in the model survey?

Keep in mind that you are not modeling human behavior -- you are modeling the areas with high probability of encountering sites in modern contexts.

Rule based model using ModelBuilder in ArcGIS

This week we will make use of a lab developed by UCSB Geography that uses a simple Rule-based predictive model based on Anabel Ford's research on the border of Belize and Guatelamala. We will take the results of this model and evaluate it in JMP.

Begin by going to the UC Santa Barbara Geography 176b lab 4 page.

Go through Exercise II of Lab 4 "Developing a Predictive Model".
Note: choose a larger sample. We want at least 100 sites in the output.

At the end of the Geography lab when you have arrived at the point with a file entitled Arch_Sites_ID_Freq return to this page

Now export this data to a comma-delimited text file.

• Right click the Arch_Sites_ID_Freq table and choose Data > Export…
  Click the Folder and save it locally, but change the file type to TXT

Analysis in JMP

• Now open the program JMP 5.1 from the Statistics folder on your desktop.
• Open the data table
• Chi-Squared and other analyses functions are found under

Analyze > Fit Y by X (Use the SITE field for Y)

• Logistic Regression is found under

Under Analyze > Fit Model

• Click SITE and Click the Y button
• Select the other variables (except ObjectID and Frequency) and click the Construct Model Effects: Add button

• Click Run Model

See the discussion on this webpage for interpreting the results
We will further discuss analysis in class.