HR: 0830h
AN: H11F-0904    [PDF]
TI: Use of artificial neural networks in prediction of subsurface hydrological processes
AU: Warrick, A W
EM: aww@ag.arizona.edu
AF: Soil, Water, and Environmental Sciences, Univ. of Arizona, Tucson, AZ 85721 United States
AU: * Furman, A
EM: alex@hwr.arizona.edu
AF: Hydrology and Water Resources, Univ. of Arizona, Tucson, AZ 85721 United States
AU: Zerihun, D
AF: Soil, Water, and Environmental Sciences, Univ. of Arizona, Tucson, AZ 85721 United States
AU: Sanchez, C A
AF: Spoil, Water, and Environmental Sciences, Yuma Agricultural Center, Univ. of Arizona, Yuma, AZ 85364 United States
AB: Modeling of surface runoff for hydrological or agricultural purposes typically makes use of empirical or semi-empirical infiltration functions such as those by Horton or Kostiakov. The alternative of detailed numerical solution of the Richards equation is generally limited to research purposes because of greater complexity than the empirical forms and the need for a greater effort in order to obtain results. A major drawback of the use of empirical functions with regard to the subsurface is that boundary and initial conditions are not explicitly taken into consideration and the details of water distribution are not calculated. This limits the accuracy of the surface models to be site and event specific. In addition, it limits the applicability of surface models for studying recharge. We present here a new alternative to the detailed numerical solution of the Richards equation which is based on artificial neural networks (ANN). The methodology relies on extensive numerical solutions for various soil properties and geometries, building a large training set for the ANN. A dimensionless form of the numerical increases the efficiency of the database generation. At a second stage, the ANN in its recall mode can replace the numerical solution to obtain a dramatically faster solution, while retaining acceptable accuracy for both infiltration rates and water content distributions. The methodology is presented for several case studies with an emphasis on infiltration from trenches and furrows.
DE: 1842 Irrigation
DE: 1866 Soil moisture
DE: 1875 Unsaturated zone
DE: 3210 Modeling
DE: 3230 Numerical solutions
SC: Hydrology [H]
MN: 2003 Fall Meeting