Assessment of arsenic concentration in stream water using neuro fuzzy networks with factor analysis.

We propose a systematical approach to assessing arsenic concentration in a river through: important factor extraction by a nonlinear factor analysis; arsenic concentration estimation by the neuro-fuzzy network; and impact assessment of important factors on arsenic concentration by the membership degrees of the constructed neuro-fuzzy network. The arsenic-contaminated Huang Gang Creek in northern Taiwan is used as a study case. Results indicate that rainfall, nitrite nitrogen and temperature are important factors and the proposed estimation model (ANFIS(GT)) is superior to the two comparative models, in which 50% and 52% improvements in RMSE are made over ANFIS(CC) and ANFIS(all), respectively. Results reveal that arsenic concentration reaches the highest in an environment of lower temperature, higher nitrite nitrogen concentration and larger one-month antecedent rainfall; while it reaches the lowest in an environment of higher temperature, lower nitrite nitrogen concentration and smaller one-month antecedent rainfall. It is noted that these three selected factors are easy-to-collect. We demonstrate that the proposed methodology is a useful and effective methodology, which can be adapted to other similar settings to reliably model water quality based on parameters of interest and/or study areas of interest for universal usage. The proposed methodology gives a quick and reliable way to estimate arsenic concentration, which makes good contribution to water environment management.