[cosc-grad-students-list] Thesis Defense Announcement - Beto Estrada - Machine Learning Based Groundwater Level Predictions as a Proxy for Compound Flooding Events in Miami, Florida

[Eulenfeld, Menda]

Name: Beto Estrada



Title: Machine Learning Based Groundwater Level Predictions as a Proxy for Compound Flooding Events in Miami, Florida

Abstract: Miami, Florida is prone to compound flooding (when multiple flood drivers occur concurrently or within a similar time period) which impacts property and lives. This is in large part due to anthropogenic changes to the region, relative sea level rise, and an increase in precipitation intensity. Better predictive modeling of these events is important to prepare for and mitigate their impact and improve the long-term resiliency of the region and its community. Miami, Florida is home to a karst aquifer that is vulnerable to saltwater intrusion. To combat this, water is pumped into canals throughout the city which helps to recharge the groundwater supply. Water can also be pumped into these canals during extreme precipitation events to decrease the chances or severity of flooding by diverting some of the excess water entering the water conservation areas (WCAs). The important role that groundwater plays in the flooding dynamic of Miami's urban areas suggests that groundwater may be used as a potential indicator for compound flooding. A 7-year data set (Oct. 2017 - Dec. 2024) including groundwater level, precipitation, ocean water level, and parameters from a nearby spillway serves as the basis for the study. The study confirms that the highest groundwater levels are correlated with flooding events in Miami-Dade County, Florida. Using a data-driven approach, multiple machine learning Multi-Layer Perceptron (MLP) models were trained that accurately predict groundwater levels at lead times 3, 6, 12, and 24 hours at a target location in Miami-Dade County, Florida. These groundwater level predictions were then used as a proxy for compound flooding events at the target location. Different hyperparameter tuning and regularization methods were applied to boost model performance and avoid overfitting. Several experiments were conducted to test the performance of the models using different combinations of inputs at each lead time. The performance of the models during confirmed flooding events was used as a case study. Results show that models that used a combination of past measurements for each feature as well as ocean water level, rainfall, and gate opening perfect prognosis predictions as inputs performed the best both statistically and during flooding events. This study models the different drivers of compound flooding in Miami, Florida, such as groundwater levels, surge conditions, and rainfall, using data-driven machine learning methods.


Date: November 7, 2025



Time: 9:00 AM - 11:00 AM CST



Location: NRC 2200



Meeting Link: https://teams.microsoft.com/l/meetup-join/19%3ameeting_ODcyY2RjOTYtYzhjMC00YTEzLTk5NWYtYTZjZmM4MGVlNTkx%40thread.v2/0?context=%7b%22Tid%22%3a%2234cbfaf1-67a6-4781-a9ca-514eb2550b66%22%2c%22Oid%22%3a%22518ef2f5-945b-4ead-9e6a-3b6501037d69%22%7d<https://nam12.safelinks.protection.outlook.com/ap/t-59584e83/?url=https%3A%2F%2Fteams.microsoft.com%2Fl%2Fmeetup-join%2F19%253ameeting_ODcyY2RjOTYtYzhjMC00YTEzLTk5NWYtYTZjZmM4MGVlNTkx%2540thread.v2%2F0%3Fcontext%3D%257b%2522Tid%2522%253a%252234cbfaf1-67a6-4781-a9ca-514eb2550b66%2522%252c%2522Oid%2522%253a%2522518ef2f5-945b-4ead-9e6a-3b6501037d69%2522%257d&data=05%7C02%7Ccosc-grad-students-list%40listserv.tamucc.edu%7C63b1781fbe3b4e7c97ff08de17d8359f%7C34cbfaf167a64781a9ca514eb2550b66%7C0%7C0%7C638974414536582549%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=5hD1G64OS0%2ByWrxRQ7pOqPe8rP93o41gHOvQyaFp1cE%3D&reserved=0>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.tamucc.edu/pipermail/cosc-grad-students-list/attachments/20251030/5252bbdc/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: text/calendar
Size: 9196 bytes
Desc: not available
URL: <http://listserv.tamucc.edu/pipermail/cosc-grad-students-list/attachments/20251030/5252bbdc/attachment-0001.ics>