Prediction of urban stormwater quality using artificial neural networks

Estimated human health risks from recreational exposures to stormwater runoff containing animal faecal material

“…Application of animal manure on land is a common practice in the United States and many other countries. Modelling of runoff and stormwater contamination is a welldocumented research activity (e.g., (Bhattarai et al, 2011;Burian et al, 2001;Kara et al, 2012;Liu, 1994;L opez-Vicente et al, 2014;Luna et al, 2006;May and Sivakumar, 2009;Vezzaro and Mikkelsen, 2012;Vezzaro et al, 2014;Whelan et al, 2014)) Prior studies of pathogen and indicator mobilisation via overland flow from land applied manures have explored the influence that numerous factors have on mobilisation (Cardoso et al, 2012;Ferguson et al, 2007;Muirhead et al, 2006;Stout et al, 2005). Those factors include manure type and method of land application (e.g., Hodgson et al, 2009;Miller and Beasley, 2008;Ramirez et al, 2009;Saini et al, 2003;Thurston-Enriquez et al, 2005), slope and ground cover (e.g., Cardoso et al, 2012;Davies et al, 2004;Ferguson et al, 2007;Hodgson et al, 2009;Miller and Beasley, 2008;Stout et al, 2005;Thurston-Enriquez et al, 2005;Trask et al, 2004;Winkworth et al, 2008;Yeghiazarian et al, 2004), rainfall intensity and antecedent soil moisture (Bradford and Schijven, 2002;Davies et al, 2004;Ramirez et al, 2009;Saini et al, 2003;Schijven et al, 2004;Sistani et al, 2009;Yeghiaz...…”

Artificial neural network modeling of the water quality index for Kinta River (Malaysia) using water quality variables as predictors

“…Kovalishyn et al (1998) accentuated that the over-fitting problem does not have any effect on the predictive capability of the ANN when overtraining is precluded by the cross-validation technique. Training according to this measure is stopped if the error on the cross-validation subset stops changing or begins increasing (Atiya and Ji, 1997;May and Sivakumar, 2009;Tiron and Gosav, 2010); (ii) a mean-squared error (MSE) value on the training set of 0.01; (iii) a minimum improvement in error of 0.0000001; and (iv) a maximum of 10,000 iterations. On the other hand, each model was retrained 10 times, following the recommendation of the ANN software developer who specifically suggested retraining each network architecture within the range of three to, preferably, 10 times with weight randomization and initial weight adjustment (Alyuda Research Company, 2003).…”

Estimating monthly total nitrogen concentration in streams by using artificial neural network

“…These variable are selected as the basic input for the ANN model and they are combined with other variables for the selection of the final ANN model. Sensitivity analyses may be used to check the redundancy amongst the input variables (May and Sivakumar, 2009). The variables were retained if improving the model performance when added to the ANN model.…”

“…However, the essential information for calibration may not be readily accessible (Loke et al, 1999), which makes very expensive and time-consuming (Suen and Eheart, 2003) or results in large inaccuracy if parameterized without using data from the site of interest (May and Sivakumar, 2009). There is a great need for statistical models capable of predicting water quality at unmonitored sites (May and Sivakumar, 2009).…”