Sandia National Laboratories - Spatial Statstics

		Spatial Statistics Projects U.S. Department of Energy and Russian Academy of Sciences, Nuclear Safety Institute

The U.S. Department of Energy and the Russian Academy of Sciences, Nuclear Safety Institute (IBRAE) have entered into a joint agreement to develop more efficient algorithms for the characterization of environmental variables. This project focuses on the development of advanced algorithms for the characterization of soil contaminants and the estimation of variables that vary both spatially and temporally (e.g., the long-term variation in a groundwater contaminant plume). Recent developments by Professor Mikhail Kanevski, at IBRAE, in the coupling of neural networks with geostatistical simulation algorithms to estimate spatially variable properties are being coupled with cost-risk approaches to characterization as developed under the SmartSampling program at Sandia.

We hope to extend the neural network approaches to estimating longer-term trends in time series data that can then be coupled with spatial simulation techniques to estimate and predict data that vary both spatially and temporally. As an example application, such an algorithm could be used to design long-term monitoring networks for DOE facilities. Additional work in this project is an examination of the feasibility of using Support Vector Machines (SVM) and Support Vector Regression (SVR) techniques to classify spatial data.

The principals in this project met at the 2001 DOE Containment and Remediation Technology Conference in Orlando, FL in June, 2001.

Papers:

Kanevski M., A. Pozdnukhov, S. McKenna, Ch. Murray and M. Maignan, (in review), Statistical Learning Theory for Spatial Data, submitted to GeoENV 2002, April, 2002.
Kanevski, M., A. Pozdnukhov, and S. A. McKenna, (in press), Transductive Decision-Oriented Mapping of Environmental Data, accepted for presentation at: IAMG Annual Meeting, Berlin, Germany, September, 2001.
McKenna, S.A., 2001, Comparing Approaches to Locating Boreholes in Spatially Heterogeneous Aquifers (extended abstract), in proceedings of: DOE 2001 International Containment & Remediation Technology Conference, June, 10-13, 2001, Orlando, Florida.
M.F. Kanevski, L.A. Bolshov, V.V. Demyanov, E.A. Savelieva, V.A. Timonin, S. Chernov, 2001, Spatial Data Analysis and Modeling of Radioactively-Contaminated Territories: Lessons Learned from Chernobyl, in proceedings of: DOE 2001 International Containment & Remediation Technology Conference, June, 10-13, 2001, Orlando, Florida.