Tuesday, 30 January 2024
Hall E (The Baltimore Convention Center)
Andrew DeSimone, Texas A&M University-Corpus Christi, Corpus Christi, TX; and A. Beasley, A. Anand, B. Colburn, S. Dasu, P. E. Tissot, and H. M. Marrero Colominas
Cold-stunning syndrome is a condition that can affect cold blooded aquatic life when water temperatures drop. Cold stunning leaves aquatic life lethargic and eventually unable to swim. The Laguna Madre is a shallow body of water along the southern Texas coast of the Gulf of Mexico where endangered sea turtles and different fish species reside. During strong cold fronts, water temperatures can drop rapidly and leave sea turtles and fishes cold-stunned. Without interventions, many stunned turtles can grow ill or even die. In 2021, over 13,000 sea turtles experienced cold-stunning syndrome and the majority of these turtles died. During cold-stunning events, marine life often seek warmer waters such as deeper waters or special areas, such as marinas, or in our case, canals. These areas are often referred to as thermal refuge areas, providing marine life with protection against low temperatures. However, these locations can also become prime targets for fishermen who may take advantage of the accumulated marine life that are vulnerable to capture during cold-stunning events. New Texas regulations are protecting marine life in these thermal refuges.
Stakeholders within the Coastal Bend community have communicated that there is a need to verify thermal refuge locations along the southern Texas coast, in order to improve responses during cold-stunning events. This research proposes to use a machine learning model to nowcast water temperatures in the canals of the Laguna Madre (or other possible thermal refuges), where real time measurements do not exist. The proposed model takes its inputs from the Laguna Madre itself for which there are not only real time measurements but also operational AI models which forecast water temperature several days in advance. This research also tested several loss functions in order to optimize performance and found that a weighted Mean Absolute Percentage Error (MAPE) loss function helped to improve water temperature predictions below 15°C with minimal impact on the overall performance of the model. This model is capable of forecasting the water temperatures in the Laguna Madre canals by replacing measured inputs from the Texas Coastal Ocean Observation Network (TCOON) with forecasted inputs from the Conrad Blucher Institute operational AI model and the National Digital Forecast Database air temperature predictions. These models could be used to validate existing thermal refuges along the southern Texas coast as well as identify additional thermal refuge areas in other areas of the Texas coast for community stakeholders.

- Indicates paper has been withdrawn from meeting

- Indicates an Award Winner