Development of a neural network model to nowcast/forecast the coastal water level anomalies on the entrance to Galveston Bay, Texas

Abstract

The water level anomaly, or the difference between the observed water level and that predicted by harmonic analysis (meteorological tide), was studied in this thesis for two locations on the northwestern Gulf of Mexico. Wind and water level observations for a total of 375 days during the winter season from 1998 to 2000 were used to quantify the relative importance of the remote and local forcing in Galveston Bay and Corpus Christi Bay, Texas. For both locations, the analysis showed that the water level fluctuations are forced primarily by the remote effects which was the water level at the mouth of the estuary, consistent with earlier findings in the literature. A neural network model was optimized to forecast the remote forcing at Galveston Bay. The model was then retrained and applied to Corpus Christi Bay. This work shows conclusively that the neural network significantly improves the water level forecasts for a range of 6 to 24 hours for both systems. Improvement to the model by including additional stations for input of the wind stress is discussed.