Use this URL to cite or link to this record in EThOS:
Title: Using artificial neural networks to predict storm surge in the North Sea and the Thames Estuary
Author: Prouty, Daniel Bruce
ISNI:       0000 0001 3502 4537
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2007
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
An artificial neural network (ANN), was developed to predict storm surge magnitudes and arrival times at selected locations in the North Sea. The model predicts storm surges based solely on past measured water level residuals at one or more tidal stations. The research focuses on the performance of the model at the Sheerness tide station near the entrance of the River Thames in the UK. To take advantage of the specificity of surge propagation in the North Sea, the ANN uses input from both the target station and an additional station located where the peak of the storm surge has just passed. The ANN is trained to relate surge at the primary station from measured surge at a secondary station. The optimal secondary location is correlated to the forecast interval and the storm surge's propagation time between the secondary and primary station. ANN performance is analyzed on an annual basis and on a 72-hour window centred on individual storm events which focuses the evaluation on a time when it is most critical. Performances are also compared at the times of both maximum surge and maximum water elevation during the passage of individual storm events. The simplest ANNs developed uses data from Sheerness only and predict surges with an absolute average error of 0.11 m for 3-hour predictions when analyzed on an annual basis. Models were systematically made more complex in an attempt to increase model performance by changing the both the size of the models, and the number of inputs used to train the ANN. A new ANN modelling method using input from several possible secondary stations was developed, decreasing the error to 0.08 m. This ANN model was compared to the continental shelf model (CS3) for 1, 3, and 5-hour predictions. The ANN model performed better than the CS3 model on an annual basis, but results were mixed when evaluating performance over the shorter 72-hour storm intervals. This research further explores new forecasting methods using ANN ensembles to reduce variance and minimize error. The ensemble forecasting method averages results from multiple ANN models trained based on different model initializations. The use of ensemble forecasting with ANNs was found to significantly reduce variance when analyzed over a 72-hour storm window, but not model accuracy. The average absolute error for an ensemble ANN using 5 repetitions had 50% of the variance of a single ANN model. An ensemble model using 50 repetitions had 5% of the variance of a single ANN model. A significant result of this research is the ANN's ability to accurately predict maximum water elevations. A single ANN model had a 4-hour forecast error of 0.017 m, while a simple [1,1] ensemble model using 20 repetitions performed better with an average 4-hour forecast error of 0.008 m. When over-training is included to reduce the model bias, the error is further reduced to 0.004 m. ANN ensemble model performances for predicting maximum storm surge were however less impressive. Best results were obtained for ensembles of [30,1] models with an average 4-hour forecast error of 0.68 m.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available