Simulation of field injection experiments in heterogeneous unsaturated media using cokriging and artificial neural network

Ming Ye, Raziuddin Khaleel, Marcel G. Schaap, Jianting Zhu

Research output: Contribution to journalArticle

22 Scopus citations

Abstract

Simulations of moisture flow in heterogeneous soils are often hampered by lack of measurements of soil hydraulic parameters, making it necessary to rely on other sources of information. In this paper, we develop a methodology to integrate data that can be easily obtained (for example, initial moisture content, θi, bulk density, and soil texture) with data on soil hydraulic properties via cokriging and Artificial Neural Network (ANN)-based pedotransfer functions. The method is applied to generate heterogeneous soil hydraulic parameters at a field injection site in southeastern Washington State. Stratigraphy at the site consists of imperfectly stratified layers with irregular layer boundaries. Cokriging is first used to generate three-dimensional heterogeneous fields of bulk density and soil texture using an extensive data set of field-measured θi, which carry signature about site heterogeneity and stratigraphy. Soil texture and bulk density are subsequently input into an ANN-based site-specific pedotransfer function to generate three-dimensional heterogeneous soil hydraulic parameter fields. The stratigraphy at the site is well represented by the estimated pedotransfer variables and soil hydraulic parameters. The parameter estimates are then used to simulate a field injection experiment at the site. A relatively good agreement is obtained between the simulated and observed moisture contents. The spatial distribution pattern of observed moisture content as well as the southeastward moisture movement is captured well in the simulations. In contrast to earlier work using an effective parameter approach (Yeh et al., 2005), we are able to reproduce the observed splitting of the moisture plume in a coarse sand unit that is sandwiched between two fine-textured units. The simple method of combining cokriging and ANN for site characterization provides unbiased prediction of the observed moisture plume and is flexible so that additional measurements of various types can be included as they become available.

Original languageEnglish (US)
Article numberW07413
JournalWater Resources Research
Volume43
Issue number7
DOIs
StatePublished - Jul 1 2007

ASJC Scopus subject areas

  • Water Science and Technology

Fingerprint Dive into the research topics of 'Simulation of field injection experiments in heterogeneous unsaturated media using cokriging and artificial neural network'. Together they form a unique fingerprint.

  • Cite this