Regional estimation of groundwater arsenic concentrations through systematical dynamic-neural modeling
Journal
Journal of Hydrology
Journal Volume
499
Pages
265-274
Date Issued
2013
Author(s)
Abstract
Arsenic (As) is an odorless semi-metal that occurs naturally in rock and soil, and As contamination in groundwater resources has become a serious threat to human health. Thus, assessing the spatial and temporal variability of As concentration is highly desirable, particularly in heavily As-contaminated areas. However, various difficulties may be encountered in the regional estimation of As concentration such as cost-intensive field monitoring, scarcity of field data, identification of important factors affecting As, over-fitting or poor estimation accuracy. This study develops a novel systematical dynamic-neural modeling (SDM) for effectively estimating regional As-contaminated water quality by using easily-measured water quality variables. To tackle the difficulties commonly encountered in regional estimation, the SDM comprises of a neural network and four statistical techniques: the Nonlinear Autoregressive with eXogenous input (NARX) network, Gamma test, cross-validation, Bayesian regularization method and indicator kriging (IK). For practical application, this study investigated a heavily As-contaminated area in Taiwan. The backpropagation neural network (BPNN) is adopted for comparison purpose. The results demonstrate that the NARX network (Root mean square error (RMSE): 95.11μgl-1 for training; 106.13μgl-1 for validation) outperforms the BPNN (RMSE: 121.54μgl-1 for training; 143.37μgl-1 for validation). The constructed SDM can provide reliable estimation (R20.89) of As concentration at ungauged sites based merely on three easily-measured water quality variables (Alk, Ca2+ and pH). In addition, risk maps under the threshold of the WHO drinking water standard (10μgl-1) are derived by the IK to visually display the spatial and temporal variation of the As concentration in the whole study area at different time spans. The proposed SDM can be practically applied with satisfaction to the regional estimation in study areas of interest and the estimation of missing, hazardous or costly data to facilitate water resources management. ? 2013 Elsevier B.V.
Subjects
Arsenic (As)
Bayesian regularization
Gamma test
Groundwater quality
Indicator kriging
NARX network
Other Subjects
Bayesian regularization; Gamma test; Ground-water qualities; Indicator kriging; NARX network; Arsenic; Estimation; Groundwater; Health risks; Interpolation; Maps; Mean square error; Neural networks; Space division multiple access; Water pollution; Water quality; Water resources; Water supply; Information management; arsenic; artificial neural network; Bayesian analysis; concentration (composition); drinking water; estimation method; groundwater pollution; groundwater resource; kriging; numerical model; public health; resource management; spatial variation; temporal variation; water quality; World Health Organization; Taiwan
Type
journal article