Author(s)

Patrick A. Sullivan, Donna M. Rizzo, and David E. Dougherty

Abstract

Hierarchical Artificial Neural Network for Regionalized Cokriging Patrick A. Sullivan, Donna M. Rizzo, and David E. Dougherty University of Vermont, Department of Civil and Environmental Email: {patrick.sullivan, donna. rizzo, david.dougherty}@ uvm.edu Abstract A new method of estimating spatially dependent parameters (e.g., log conductivity) is being developed using artificial neural networks (ANNs). The method involves designing and training an ANN to estimate parameter structure and the variance of estimation error based on multiple correlated data types, like in cokriging. The research is motivated by current data collection approaches, such as are being used at the National Environmental Technology Test Site at Dover Air Force Base, Delaware. The neural network method allows for the use of more data without the hindrance of large co variance matrices. In addition to incorporating expanded data types or quantity, the ANN method yields a measure of error uncertainty. 1 Int

Keywords