3rd Conference on Artificial Intelligence Applications to the Environmental Science


Optimization and Performance of a Neural Network Model Forecasting Water Levels for the Corpus Christi, Texas, Estuary

Philippe E. Tissot, Texas A&M University, Corpus Christi, TX; and P. R. Michaud and D. T. Cox

Accurate water level forecasts are of vital importance along the coasts of the Gulf of Mexico (GOM) as its waterways play a critical economic role for a number of industries including shipping, oil and gas, tourism, and fisheries. For example, GOM ports account for more than half the total annual U.S. tonnage. While astronomical forcing (tides) is well tabulated, water level changes along the GOM coasts are frequently dominated by meteorological factors, which are unaccounted for in present forecasts. The impact of these meteorological factors is often greater than the tidal range. Wind forcing has been identified as the principal input unaccounted for by present models [1]. Artificial Neural Network (ANN) models are being developed to predict short-term water levels, up to 48 hours, based on previously observed water levels, wind speeds, wind directions, and forecasted tidal and meteorological data [1,2]. The observational data is collected as part of the Texas Coastal Ocean Observation Network (TCOON), which has platforms covering the coast of Texas from the Mexican border to Louisiana. The present work focuses on the Corpus Christi (CC) estuary. The performance of a set of ANN models predicting water levels is evaluated for six stations within CC Bay, on the coast, and at the inlet of the CC ship channel. The importance of the respective input components is evaluated, and the topologies of the ANN models are optimized. The performances of the models are measured by a variety of skill assessment variables. ANN models provide significant improvements over harmonic forecasts for all skill variables with improvements often better than 50% for the absolute average error and more than a factor of 4 or 5 for skills such as the positive and negative outlier frequencies and the maximum duration of positive and negative outlier frequencies. Model performances for stations deep inside the bay are found to be better than for stations at the entrance of the bay. For these locations including input data from a station on the open coast improves significantly the accuracy of the forecasts. While wind hindcasts are used in this work the excellent agreement between the National Center for Environmental Predictions (NCEP) Eta-12 forecasts and measured winds, once a consistent bias is included, should lead to similar model performances for the real-time model. A database of Eta-12 historical wind forecasts is presently accumulated to confirm the performances and applicability of the model. An operational model is in development and starting during the spring of 2003 ANN water level forecasts will be published on the World Wide Web to provide better predictions for mariners and other coastal users.

[1] D.T. Cox, P.E. Tissot, and P. Michaud, ?Water Level Observations and Short-Term Predictions Including Meteorological Events for Entrance of Galveston Bay, Texas?, Journal of Waterway, Port, Coastal and Ocean Engineering, 128-1 (2002) 21-29. [2] P.E. Tissot, D.T. Cox, P. Michaud, ?Neural Network Forecasting of Storm Surges along the Gulf of Mexico?, Proceedings of the Fourth International Symposium on Ocean Wave Measurement and Analysis (Waves `01), ASCE, 1535-1544, 2002.

extended abstract  Extended Abstract (480K)

Supplementary URL: http://dnr.cbi.tamucc.edu/projects/NN

Session 3, Neural Networks
Tuesday, 11 February 2003, 8:30 AM-12:15 PM

Previous paper  Next paper

Browse or search entire meeting

AMS Home Page