TY - GEN
T1 - Coastal beach salinity prediction using data-driven and deterministic approaches
AU - Thoe, W.
AU - Wong, H. C.
AU - Lee, J. H.W.
PY - 2010
Y1 - 2010
N2 - Areal-time coastal beachwater quality forecast system has recently been developed for HongKong. Daily beach E. coli level is predicted as a function of different hydro-meteorological inputs using data-driven methods including the Multiple Linear Regression (MLR) and theArtificial Neural Network (ANN) models. Rainfall and salinity are found to be important parameters in the forecast models, especially on beaches fed by streams ending at the beach shoreline. However, daily measurement of salinity - which mirrors the mixing of freshwater sources with the marine water - is usually not available. This study discusses the prediction of beach salinity by both data-driven (Artificial Neural Network) and deterministic Tidal Prism (TP) methods for BigWave Bay - a beach dominated by pollution sources from a stream. The stream flow is predicted by a physically-based hydrological model (MIKE-SHE). The model parameters are first calibrated against measured stream flows at a nearby gauged catchment with similar hydro-climatic and geomorphologic characteristics. For the ANN model, the beach salinity is predicted from the measured hydro-meteorological data (rainfall in the past 3 days, wind speed, tide level and pastsalinity data) and the streamflow predicted from the rainfall data. Alternatively, the beach salinity (or freshwater concentration) can also be estimated from the tidal prism (predicted as a function of tidal range), the predicted stream flow volume, and the ambient sea water salinity.A correlation coefficient of about 0.8 is achieved between the prediction and the observation for both models. The calibrated ANN and TP models are validated against daily observations of beach salinity in June and July 2007. Both models can predict the observed salinity trends satisfactorily, particularly with higher salinity. However, both methods fail to capture accurately rapid drops in salinity brought about by heavy rain of short duration; the prediction typically has a 1-day phase lag with the observation. Nevertheless, using the predicted salinity as input to real time forecasting of beach water quality, reasonable forecasts of the compliance and exceedance of beach water quality can still be obtained.
AB - Areal-time coastal beachwater quality forecast system has recently been developed for HongKong. Daily beach E. coli level is predicted as a function of different hydro-meteorological inputs using data-driven methods including the Multiple Linear Regression (MLR) and theArtificial Neural Network (ANN) models. Rainfall and salinity are found to be important parameters in the forecast models, especially on beaches fed by streams ending at the beach shoreline. However, daily measurement of salinity - which mirrors the mixing of freshwater sources with the marine water - is usually not available. This study discusses the prediction of beach salinity by both data-driven (Artificial Neural Network) and deterministic Tidal Prism (TP) methods for BigWave Bay - a beach dominated by pollution sources from a stream. The stream flow is predicted by a physically-based hydrological model (MIKE-SHE). The model parameters are first calibrated against measured stream flows at a nearby gauged catchment with similar hydro-climatic and geomorphologic characteristics. For the ANN model, the beach salinity is predicted from the measured hydro-meteorological data (rainfall in the past 3 days, wind speed, tide level and pastsalinity data) and the streamflow predicted from the rainfall data. Alternatively, the beach salinity (or freshwater concentration) can also be estimated from the tidal prism (predicted as a function of tidal range), the predicted stream flow volume, and the ambient sea water salinity.A correlation coefficient of about 0.8 is achieved between the prediction and the observation for both models. The calibrated ANN and TP models are validated against daily observations of beach salinity in June and July 2007. Both models can predict the observed salinity trends satisfactorily, particularly with higher salinity. However, both methods fail to capture accurately rapid drops in salinity brought about by heavy rain of short duration; the prediction typically has a 1-day phase lag with the observation. Nevertheless, using the predicted salinity as input to real time forecasting of beach water quality, reasonable forecasts of the compliance and exceedance of beach water quality can still be obtained.
UR - https://www.scopus.com/pages/publications/84860238405
U2 - 10.1201/b10553-96
DO - 10.1201/b10553-96
M3 - Conference Paper published in a book
AN - SCOPUS:84860238405
SN - 9780415595452
T3 - Environmental Hydraulics - Proceedings of the 6th International Symposium on Environmental Hydraulics
SP - 595
EP - 600
BT - Environmental Hydraulics - Proceedings of the 6th International Symposium on Environmental Hydraulics
PB - Taylor and Francis - Balkema
T2 - 6th International Symposium on Environmental Hydraulics
Y2 - 23 June 2010 through 25 June 2010
ER -
