The talk will present an overview of the National Invasive Species Forecasting System project developed jointly by the US Geological Survey and NASA.

Current statistical modeling procedure include the combined use of field data, GIS layers (such as soils information and elevation), and satellite imagery into stepwise regression modeling. To account for spatial dependency in the residuals, current modeling uses geostatistical techniques to create an "error surface" that is added to the model resulting from the stepwise procedure. Initial work at NASA has been done to conduct such modeling in a parallel computing environment - reducing the computing time by more than an order of magnitude.

The talk will go on to describe the future direction of the statistical modeling within the Invasive Species Forecasting System. This will include the development of generalized least square modeling in a parallel computing environment. This would account for spatial dependence in the residuals in the initial modeling. However, questions remain on how to create parallel code for this modeling and how to screen through all possible variables in a parallel, generalized least square, modeling environment.

In addition to the generalized least squares work, current work is being done to build a suite of time-series satellite data products specifically tailored to invasive species management. The effort will produce image layers that summarize time-series data. These new summary layers will be used as additional explanatory variables within the Invasive Species Forecasting System.