--> ABSTRACT: Methodology for Enhancing and Evaluating Geologic Effects of Time Series Models: A Case of Ground Response in Santa Clara Valley, California, by Samuel-Ojo, Olusola; Olfman, Lorne; Reinen, Linda; Flenner, Arjuna; Oglesby, David; Gareth, Funning; #90155 (2012)

Datapages, Inc.Print this page

Methodology for Enhancing and Evaluating Geologic Effects of Time Series Models: A Case of Ground Response in Santa Clara Valley, California

Samuel-Ojo, Olusola¹; Olfman, Lorne¹; Reinen, Linda³; Flenner, Arjuna²; Oglesby, David4; Gareth, Funning4
¹School of Information Systems and Technology, Claremont Graduate University, Claremont, CA.
²School of Mathematical Sciences, Claremont Graduate University, Claremont, CA.
³Geology Department, Pomona College, Claremont, CA.
4Department of Earth Sciences, University of California, Riverside, Riverside, CA.

Unlike cross-sectional data models of geologic properties where datasets of regionalized (random) variables take levels (values) that are ‘frozen' in time, time series data models pose unique characteristics and require special enhancement in order to get the most out of them. Using the groundwater and remote sensing field data between 1992 and 2000 of Santa Clara Valley, USA, we investigate liquefaction deformation, a groundwater-induced ground response. We examine the uniqueness of hydrogeologic (hydraulic heads) and geodetic (PS-InSAR) time series data, and fashion a methodology that enables local and regional geologic features to be more pronounced so that they are better identified and interpreted.

A time series three-phase methodology consisting of disaggregation, application, and aggregation phases is proposed. In the disaggregation phase, observational data are partitioned into inliers and outliers. The inliers are de-noised and decomposed into four main components: levels, trends, cycles, and residuals. In the application phase: parametric and non-parametric quantitative methods are applied, training and validating datasets, local (shallow) and regional (deep), level and residual models are built. These models constitute threads, a computer executable unit that can concurrently be processed by server clusters to achieve high performance results. Levels and residuals are recombined in the aggregation phase in order to make subsurface prediction. The predictive performances of different smoothing techniques are compared by performing a ten-fold cross-validation experiment.

Using the methodology, we produced sixteen models of interaction between groundwater and surface deformation. We also generated from the models, enhanced regional hydraulic head, local direction, velocity, path of groundwater flow isoline maps, describing the hydrogeology variation and ground response in the study area. In a reservoir media consisting of sandstones (solid phase), empty pores (air phase), and filled pores (liquid phase) such as an aquifer, we found that enhanced feature-based time series models produce better plausible results incorporating environmental uncertainties. Resource managers, geoscientists and engineers are likely to use the maps to monitor groundwater volumetric flow for water budget and predict hydraulic conductivity of the aquifer storage for land use. They may use the methodology to monitor other reserves including ores and petroleum as well.


AAPG Search and Discovery Article #90155©2012 AAPG International Conference & Exhibition, Singapore, 16-19 September 2012