Machine Learning Inland Compound Flooding Predictions for Corpus Christi, Texas

Oso Creek, located near Corpus Christi, Texas, within the Oso Creek Watershed, is prone to inland compound flooding, which results from the combined effects of heavy precipitation and tidal influences from Corpus Christi Bay. This flooding poses significant risks to infrastructure and residents. Traditional flood prediction models often fail to capture the complex interactions between these flood drivers, necessitating innovative solutions. This study introduces a Multilayer Perceptron (MLP), a type of artificial neural network, to improve flood prediction accuracy for Oso Creek. The MLP is implemented using the TensorFlow library with the architecture tuned using the KerasTuner framework. The model is then compiled and optimized using Adam Optimizer as part of the Keras library.

By integrating diverse data sources, including hourly rainfall rate data from NOAA’s Multi-Radar Multi-Sensor (MRMS) system, Texas Coastal Ocean Observation Network (TCOON) primary water level data, and Oso Creek air gap measurements from a Conrad Blucher Institute (CBI) sensor, the model addresses the nowcasting challenges in predicting water levels at Oso Creek. The combined data spans the period of February 2015 through March 2021. The model’s effectiveness in nowcasting—achieved without using air gap measurements as an input feature, but only as a target variable—enabled the application of Permutation Feature Importance. This approach allowed for the investigation of the relative importance of precipitation in different parts of the watershed compared to coastal water levels.

The performance and relative importance of each feature of the nowcasting model will be presented, highlighting its role in laying the groundwork for a future forecasting model. Outcomes of this model include enhanced prediction accuracy, comprehensive data integration, and improved flood preparedness for the local community. This approach not only benefits Oso Creek and the surrounding area, but also offers a framework that should be adaptable to other areas prone to compound flooding.