You are here

Map your models with ModelMap

Date: September 13, 2016

We created a tool, ModelMap, that can combine the Forest Inventory and Analysis plot data with remote sensing satellite images to predict forest characteristics (such as species composition, crown cover, and forest disturbances) over any geographic extent


Working in the Forest Inventory and Analysis (FIA) Program, we have access to a valuable collection of detailed information about forests on thousands of sample plots distributed across the country. This information is used to produce summaries of forestland characteristics for a variety of geographic areas such as states or individual national forests. We wanted a simple tool to extend this sample data and make detailed maps of forest characteristics for all the land in between the study locations.

We began by writing code to create models and produce maps of forest and rangeland species in Nevada, using the R software environment for statistical computing and graphics. R is a powerful and flexible tool for statistical analysis, but can have a steep learning curve for new users. When we realized that the tools we were developing could be useful in many different contexts in our Rocky Mountain states and throughout the country, we developed an R package, ModelMap, to gather these modeling and map making tools together, complete with help files, training vignettes, and a graphical user interface. In 2009, the package was made available for download and use by researchers worldwide through the Comprehensive R Archive Network (CRAN). Since then it has been expanded and updated 11 times, adding additional model types, and the ability to predict and map categorical data such as disturbance types. The package is currently widely used throughout the world in fields ranging from forestry to oceanography and to the health sciences.

Key Findings

ModelMap allows the user to easily:

  • Construct predictive models using a suite of predictive tools including Random Forests, conditional Random Forests, Quantile Random Forests, and Stochastic Gradient Boosting

  • Validate models with a choice of independent test set, cross-validation, or Out-Of-Bag (OOB) predictions on the training data

  • Conduct detailed exploratory data analyses and create graphics of predictor relationships

  • Apply models to GIS raster files to make maps

A world map displaying the density of ModelMap downloads
A world map displaying the density of ModelMap downloads

Additional Information

Freeman, E.A., T.S. Frescino, and G.G. Moisen. 2016. ModelMap: an R Package for Model Creation and Map Production. Vignette for the ModelMap package: http://cran.r-project.org

 

Featured Publications

Frescino, Tracey ; Moisen, Gretchen ; Patterson, Paul L. ; Freeman, Elizabeth ; Menlove, James S. , 2016