RE:  Spatial Logistic Regression (again)

Discussion created by doncatanzaro on Dec 27, 2012
Hi All,

I have a few questions regarding spatial logistic regression.

First off, I need to make sure that this is the correct model to create.  Here is what I want to do: 

[INDENT]  I have a binary variable that is polygon based that I want to predict (Flood/Not Flood). I have a soil polygon map that has data associated with it (e.g. percent soil type, landform etc., there may be other data that could be helpful but let us wait on that).  So the soil map is by counties and some counties in a state are mapped, others are not and same thing with my flood data.  I want to predict flooding in the areas where I have no flood data.

The soil data has a very strong spatial component to it (some states do not have certain landforms, counties are soil mapped differently etc.).  I have tried a simple classification model (e.g. all polygons where landform ='XXX' are flooded) but the performance is not that great at the state level. 

So I figured I could create a model that is spatially aware and varies through geographic space.  To me, this means I need some sort of spatial logistic regression model to predict the probability of flooding based on a certain amount of variables. 

Q1:  Is Spatial Logistic Regression the 'best' type of model to perform flood predictions ?  I think there are others (e.g. Spatial Bayesian predictions). 

[INDENT] I've seen some chatter in the forums over the last couple of years regarding software to allow users to do this and want to check on the status of software tools to help you accomplish this. [/INDENT]

Q2:  What kind of software can help me with this?  I am very familiar with R so I could certainly do that.  I've not tried PySAL yet so that is an option as well.

Q3:  Has this kind of thing been done with polygons ?  I wanted to check before I went to far.  How does the polygonal nature of the data change how the model behaves (e.g. are the equations weighted by polygon area and geographically?)