The Laguna Madre is the longest hypersaline lagoon in the United States and extends southward for over 200 miles from Corpus Christi, Texas to the United States-Mexican border. The Laguna Madre is home to fragile young finfish, shrimp, and shellfish as well as redfish, spotted sea trout, a host of birds, and endangered sea turtles. On occasion the passage of strong cold fronts dramatically lowers air temperatures by more than 19 C in less than 20 hours and leads to a considerable decrease in water temperature. Over the past 20 years a few of these cold water events resulted in massive fish kills. In 1997, more than 94,000 fish died in the Lower Laguna Madre and over 48,000 in the Upper Laguna Madre. To mitigate the impact of these cold events, local agencies and stakeholders are considering interrupting activities such as fishing and boating during these events. To help manage such interruptions accurate predictions of occurrences and length of events are critical.

Artificial Neural Network (ANN) models were trained and optimized to predict water temperatures based on previous water temperatures and previous and forecasted air temperatures. Other inputs considered include along shore and across shore wind squared, water level history, and tidal forecasts. As the performance of the ANN varies depending on the length of the forecast, both a short-term model (3 and 12 hour forecasts) and a long-term model (24 and 48 hour forecasts) were developed and optimized. The inputs that maximize performance in both models include previous water temperatures and previous and forecasted air temperatures. Adding a time stamp reduces the error in the short-term model but does not significantly affect the long-term model. Previous water temperatures from a nearby location decrease the error for long-term forecasts but do not have a similar effect on short-term forecasts. Different ANN topologies are tested and compared. The optimized ANN design is then compared to the conclusions of multivariate statistical analysis in an attempt to determine whether statistical analysis aids in the development and optimization of the ANN. Finally, the performance and operational applicability of the model are discussed.