Premium
Stochastic simulation and spatial estimation with multiple data types using artificial neural networks
Author(s) -
Besaw Lance E.,
Rizzo Donna M.
Publication year - 2007
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1029/2006wr005509
Subject(s) - categorical variable , artificial neural network , cluster analysis , covariance , covariance function , autocorrelation , computer science , spatial analysis , data mining , algorithm , statistics , machine learning , mathematics
A novel data‐driven artificial neural network (ANN) that quantitatively combines large numbers of multiple types of soft data is presented for performing stochastic simulation and/or spatial estimation. A counterpropagation ANN is extended with a radial basis function to estimate parameter fields that reproduce the spatial structure exhibited in autocorrelated parameters. Applications involve using three geophysical properties measured on a slab of Berea sandstone and the delineation of landfill leachate at a site in the Netherlands using electrical formation conductivity as our primary variable and six types of secondary data (e.g., hydrochemistry, archaea, and bacteria). The ANN estimation fields are statistically similar to geostatistical methods (indicator simulation and cokriging) and reference fields (when available). The method is a nonparametric clustering/classification algorithm that can assimilate significant amounts of disparate data types with both continuous and categorical responses without the computational burden associated with the construction of positive definite covariance and cross‐covariance matrices. The combination of simplicity and computational speed makes the method ideally suited for environmental subsurface characterization and other Earth science applications with spatially autocorrelated variables.