Premium
Forecasting of Turbid Floods in a Coastal, Chalk Karstic Drain Using an Artificial Neural Network
Author(s) -
Beaudeau P.,
Leboulanger T.,
Lacroix M.,
Hanneton S.,
Wang H.Q.
Publication year - 2001
Publication title -
groundwater
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.84
H-Index - 94
eISSN - 1745-6584
pISSN - 0017-467X
DOI - 10.1111/j.1745-6584.2001.tb00356.x
Subject(s) - turbidity , surface runoff , environmental science , precipitation , hydrology (agriculture) , aquifer , drainage basin , karst , groundwater , geology , meteorology , ecology , geotechnical engineering , geography , paleontology , oceanography , cartography , biology
Water collected at the Yport (eastern Normandy, France) Drinking Water Supply well, situated on a karst cavity, is affected by surface runoff‐related turbidity spikes that occur mainly in winter. In order to forecast turbidity, precipitation was measured at the center of the catchment basin over two years, while water level and turbidity were monitored at the well site. Application of the approach of Box and Jenkins (1976) leads to a linear model that can accurately predict major floods about eight hours in advance, providing an estimate of turbidity variation on the basis of precipitation and water level variation over the previous 24 hours. However, this model is intrinsically unable to deal with (1) nonstationary changes in the time process caused by seasonal variations of in ground surface characteristics or tidal influence within the downstream part of the aquifer, and (2) nonlinear phenomena such as the threshold for the onset of runoff. This results in many false‐positive signals of turbidity in summer. Here we present an alternative composite model combining a conceptual runoff submodel with a feedforward artificial neural network (ANN). This composite model allows us to deal with meaningful variables, the actioneffect of which on turbidity is complex, nonlinear, temporally variable and often poorly described. Predictions are markedly improved, i.e., the variance of the target variable explained by 12‐hour forward predictions increases from 28% to 74% and summer inaccuracies are considerably lowered. The ANN can adjust itself to new hydrological conditions, provided that on‐line learning is maintained.