24AI Ensemble Uncertainty Visualization Development for Effective Communication of Cold-Stunning Event Threats

Cold fronts along the southern Texas coastline can cause hypothermic stunning of ecologically and economically valuable marine life. Major cold-stunning events occur when water temperatures fall below 8°C for about 24 hours, leaving endangered green sea turtles lethargic and vulnerable to cold-related illnesses and death if not rescued (Shaver et al., 2018). Using a combination of the Conrad Blucher Institute’s Coastal Dynamics Lab’s (CDL) Cold-Stunning model (120-hr lead time deterministic water temperature predictions) and The Weather Company’s (TWC) 100-member ensemble air temperature forecasts, advisory guidance is provided to coastal stakeholders to help them prepare for cold-stunning event risk and the threat they pose to marine life. The TWC’s ensemble model’s provision of air temperature prediction ranges (i.e., 90% confidence intervals) is critical for informing the timing of cold-stunning events.

CDL cold-stunning advisory experts are responsible for not only providing water temperature predictions along southern Texas but also communicating and disseminating water and air temperature information leading up to and during cold-stunning events to coastal stakeholders. While TWC’s model provides valuable predictions, its utility is limited without further analysis and real-time visualizations that enhance interpretation, communication, and decision-making processes for effective cold-stunning advisory guidance.

The main goal of this project is to develop a graphical user interface (GUI) designed to provide real-time visualizations and analysis of TWC’s air temperature ensemble model’s skill and spread at various forecast and initialization times to improve the decision-making process for experts and stakeholders alike.

For this project, extra attention was given to various areas such as data storage, visualization construction, and co-refinement to ensure the resulting tools provide useful information to experts, and the tools are refined and ready to be implemented into the GUI. To retrieve certain portions of data from TWC’s large repository of air temperature predictions, various functions were created to search for and retrieve data segments. TWC utilizes a proprietary Global High-Resolution Atmospheric Forecasting (GRAF) model that is based on physics and utilizes machine learning and atmospheric observation data to predict weather activity across the globe (The Weather Company, 2024). GRAF outputs predictions hourly due to the utilization of a GPU-accelerated supercomputer which also causes the model to run at a rate five to six times faster than other models (The Weather Company, 2024). Developing new visualization tools will help experts and stakeholders view data in new ways. These tools include time series graph visuals such as spaghetti, fan, and boxplots. Other visualizations include spread-skill plots, dprog/dt plots, MAE plots, and probability plots. These plots were created using Python and Plotly, a visualization package supporting interactive plots. After considering expert input on the design and functionality of the created visualizations, various improvements were implemented during a refinement period to ensure multidimensional context was examined for decision-making during cold-stunning events. We will use the Python package Tkinter, and the refined visualizations will be integrated into the GUI, which will run locally on a user’s machine.

We plan to develop an effective GUI to host various visualization tools needed for assessing historical bias in TWC’s air temperature predictions. After integrating visualizations into the GUI tool, expert communicators can be equipped to effectively convey to coastal stakeholders the uncertainty of air temperature prediction. Increasing the effectiveness and efficiency of this information dissemination will help experts communicate the approaching risks posed by cold-stunning events.

In future work, we plan to investigate whether TWC’s air temperature ensemble member predictions can be incorporated into machine learning models to assist with quantifying water temperature uncertainty. We also will expand upon the visualizations by examining water temperature uncertainty