HR: 0800h
AN: H31D-0428    [Abstracts]
TI: Artificial neural network application to the solute transport through unsaturated zone
AU: * Yoon, H
EM: oolahee1@snu.ac.kr
AF: Seoul National University, School of Earth and Environmental Science, Seoul, 151-747 Korea, Republic of
AU: Lee, K
EM: kklee@snu.ac.kr
AF: Seoul National University, School of Earth and Environmental Science, Seoul, 151-747 Korea, Republic of
AU: Hyun, Y
EM: yjhyun@snu.ac.kr
AF: Seoul National University, School of Earth and Environmental Science, Seoul, 151-747 Korea, Republic of
AU: Bae, G
EM: gokbae@snu.ac.kr
AF: Seoul National University, School of Earth and Environmental Science, Seoul, 151-747 Korea, Republic of
AB: For effective groundwater management, the amount of solutes transported from surface to groundwater table through the unsaturated zone should be estimated. Recently, artificial neural network (ANN) receives much attention as a tool for analyzing solute transport through the unsaturated zone. The ANN is considered to be a versatile tool for approximating complex functions without taking into account the complicated physical mechanisms of the unsaturated zone. In this study we develop an ANN model to evaluate the amount of solute reaching to the groundwater table when a certain amount of surface input is given. Using the developed model we investigate the applicability of the ANN to solute transport through the unsaturated zone using numerical and laboratory experiments. For numerical tests, 2624 training data are obtained using HYDRUS2D. We group the data into two training sets : one for arrival time of solutes and the other for solute mass with respect to time. For the latter, the data are trained in the form of normalized mass, cumulative mass, and logarithmic mass, respectively to examine the influence of data form on training. The optimal neural network architecture is determined through a numerical case study with the number of nodes in the hidden layer, momentum, learning rate and the range of initial weights. Then we estimate the solute transport for various inputs. Results show that ANN model can estimate the solute transport efficiently when sufficient training sets are available. Also, the choice of network parameters and formation of training set patterns are important for estimation of solute transport. For validation, the developed ANN model is applied to soil column tests data. The results are compared with the estimation using conventional convection-dispersion equation (CDE).
DE: 1899 General or miscellaneous
DE: 1818 Evapotranspiration
DE: 1866 Soil moisture
DE: 1875 Unsaturated zone
DE: 1894 Instruments and techniques
SC: Hydrology [H]
MN: 2004 AGU Fall Meeting