Hi class,
This week, in addition to talking about your projects, we will look at the NEON program at NSF. NEON stands for the National Ecological Observatory Network. As part of the network, they are building planes with an imaging spectrophotometer, a LiDAR and a multi-spectral imager on board (one has been delivered, two more planes on the way). A page describing these planes (including a video) may be hound here: http://www.neoninc.org/science/aop
In addition, some of you are planning on doing regression analysis with some of your map data. There are some things you need to worry about when doing spatial regression. A good introduction is presented in the free training course on ESRI.com called ‘Regression Analysis Basics in ArcGIS 9.3’ (http://training.esri.com/gateway/index.cfm?fa=catalog.webCourseDetail&courseid=1640). This is a video lecture and demonstration that has good advice for any regression, especially spatial. The tools haven’t changed very much in 10.0, but there are a few new ones in 10.1, so it is worth looking at the talk if you are going to do regression. To view it, you will have to register, but it is free.
In class today (Thursday), I’ll go through a regression tutorial that you can download from the ArcGIS.com gallery (or you can download my copy for ArcGIS 10.0: https://udrive.oit.umass.edu/xythoswfs/webui/_xy-12027721_1-t_eO3ed61U).
The first tool to try is called ‘Ordinary Least Squares Regression’. This tool is primitive compared to R or SAS. For example, it will not automatically use categorical data! To make it do categories, you have to construct dummy (0/1) variables (just like R does) by yourself. I’ll show you how this works in class today. To make it work you need a set of points or polygons with attributes for each one. One of the attributes will be the response variable you are trying to predict, and one or more will be predictor variables. You must know the response for some of your points, so that you can construct your statistical models. You then specify the variables to use in the OLS tool and you will get out a regression equation and a bunch of statistics. You can use the equation to predict the response of other places on the map that you didn’t measure!
You need to look at whether their is spatial autocorrelation in the data that isn’t accounted for yet. Calculating Moran’s I on the residuals will tell you that. If there is autocorrelation, then you should use the Geographically Weighted Regression (GWR) tool, which takes into account spatial autocorrelation and computes a set of coefficients for each point in the dataset (and nearby points as well). This can be very time consuming, so be careful. If you need to do this, I’ll show you how.
Jack