Hi Jane,

I’m sorry for my slow response; I have been out of the office a lot recently.

I think you could probably use MGET to make a map of predicted feeding sites. MGET’s statistical modeling tools are just wrappers around statistical functions in R that make it convenient to call those functions from ArcGIS. So if you were able to conceptualize how to model feeding sites using one of types of models MGET supports—e.g. GLM, GAM, CART, or Random Forest—then you could implement that in MGET.

The main advantages MGET provides over R for this scenario are that 1) MGET can implement models of modest complexity from ArcGIS without having to do any programming, and 2) MGET can handle tedious data processing chores, such as extracting values of time series of rasters at points, or reprojecting and clipping rasters representing covariates so they all have the same coordinate system, cell size, and extent, so that they may be stacked and predicted through a model. Eventually, if you develop complex models and need control over certain details, or you just prefer to write code, you might want to use R directly.

For the project you were envisioning, you probably have point observations of manatees in which they were feeding and not feeding. These points probably have spatial coordinates (latitude and longitude) and dates associated with them. You could approach the problem by creating a binary classification model (e.g. a binomial GLM or GAM, or a classification tree or Random Forest) where the behavior (feeding or not feeding) is the response variable (e.g. coded as 1 or 0) and some environmental parameters (e.g. water temperature, water depth) extracted at those points are the covariates / predictor variables. This is similar to what is often called a “habitat suitability model” or more generically a “species distribution model” but instead of predicting “habitat” or species presence you’d predict behavior.

All the best,

Jason