Logistic regression and link spatial statistics toolbox with rasters

09-07-2011 11:17 AM
Status: Open
Occasional Contributor III

The spatial statistics toolbox could be improved by adding logistic regression and possibly other types of regression (such as autoregressive models for incorporating autocorrelation).  The OLS and logistic regression tools could also be improved if they were allowed to sample from a raster, run the regression, and then automatically generate a map algebra statement to create predictive maps using raster data.  Raster data tends to be more useful for many people in the natural resources fields, biology, and geology because natural phenomenon tend to vary continuously while people in the social sciences tend to be served by socio-economic data that is collected at the point level or summarized into discrete polygons.  The spatial statistics toolbox therefore caters to social scientists but expanding it to rasters would help accomodate the needs of a larger community.

Great idea ;o) For my master thesis I did a logistic regression to explain the variation is damage caused by the 2004 tsunami, which worked pretty okay ... but I had to do it in SPSS ... no good for street credit being a physical geographer.
 Poisson (spatial) Regression for counts/rates would be a great improvement
I am currently trying to figure out how to do logistic regression with my rasters. If it can be easy and foolproof (lol), that would be great. 
It would be great if the spatial statistics tools (e.g. regression) would directly take rasters as input!

The best tool for doing this is the Marine Geospatial Ecology Tools https://mgel.env.duke.edu/mget/download/  .  They are awesome and run all sorts of modeling methods like random forest, classification trees, mixed effects models, etc.  However, I don't think that they have migrated to Pro yet.