Faculty Interaction

Bryan Pijanowski

Department of Forestry and Natural Resources
Purdue University

RCAC staff: M. Sayeed

Brief Project Description

The LTM group at Purdue uses SNNS (Stuttgart Neural Network Simulator) to identify land use patterns from GIS data by training the simulator and use it to forecast future land use patterns. The model uses a set of spatial interaction rules and machine learning, through neural net technology, to determine the nature of spatial interactions of drivers, such as transportation, urban infrastructure and proximity to lakes and rivers, that have historically contributed toward land use change in the past. This information is then used to conduct forecasting studies.

The future land use data is used by RAMS (Regional Atmospheric Modeling Simulator) to predict impact on regional climate change. Currently our group uses both RAMS and SNNS in serial mode and thus seriously limiting research possibilities. Initial idea was to develop a parallel version of SNNS or to obtain the parallel SNNS (for research only) from the developers (provided only to associated research groups). Figure below shows the urban expansion for the Detroit metropolitan area.

An alternative to this approach has been proposed and involves running SNNS using Condor (hundreds of pattern matching runs for training and prediction) throughput computing environment. Also a new parallel version of RAMS v6.0 package has been ported to Linux clusters. The group is very exited as a major challenge to their research has been overcome.

Sayeed from Performance Engineering group proposed and tested SNNS using Condor and ported and tested RAMS v6.0 package on Linux clusters.

 

Last updated 03/09/08