Machine Learning Water Level Predictions for an Intermediate Location Using Connected Bodies of Water

Water level predictions, whether coastal or inland, often rely on large amounts of data obtained from the target prediction location. These predictions can help keep communities safe, allowing time for preparation and evacuation in the case of a flooding event. In the absence of sufficient data, an alternate approach in which data from nearby locations relative to the target location was explored. The target location for this research, Magnolia Beach, is located in the town of Indianola, Texas. The town, formerly the county seat for Calhoun County, has a tragic history in relation to coastal inundations. Indianola was once a thriving community, home to one of the largest ports in Texas, second only to the port located in Galveston. However, in September of 1875, a storm surge following a hurricane hit the coast and nearly wiped the town away. After a second storm hit the town in 1886, Indianola was all but abandoned. While the town has never returned to its former glory, it is still home to some residents. During weather events which include strong winds or approaching tropical storms, some of the town’s roadways are known to flood, leaving some residents inaccessible to emergency services. A sensor measuring water levels was installed in a pond near one of the roads prone to flooding. The pond is the source of the flooding waters and is also connected to Matagorda Bay where one finds long term tide gauges. Because sufficient water level data from the Magnolia Beach area is not yet available, a method was tested in which metocean data from surrounding tide gauges were used. These gauges include measurements from the Texas Coastal Ocean Observation Network (TCOON) of Port Lavaca and Port O’Connor, and were used to create synthetic water levels for Lavaca Bay near the target location. A machine learning model was then trained to create a synthetic data set of predictions of present and future water levels at that location. Finally a simple model was calibrated to transfer the water level predictions from Matagorda Bay to the new sensor location near the roads at risk of flooding. The model is evaluated for several months of data starting on June 13, 2023. Results show the promise of the method to create a flood warning system but also that more data is needed to calibrate the model relating predictions for Matagorda Bay and actual measurements in the nearby pond.