Project Description

The frequency and severity of coastal flooding and erosion can result from several factors such as storm surges, tsunamis, subsidence, high rainfall events, increasing average air temperature and sea-level. Investigating patterns in the time course of these variables can be extremely useful for understanding, forecasting, and managing flooding in coastal locations. The project focused on two directions.

Direction 1 focused on enhancing the understanding of extremal dependence between several variables related to flooding in coastal locations. The focus of Direction 2 was to understand the risk factors for flooding across roads in coastal towns and create an early alert system for residents and drivers. The goals in these two directions were achieved by state-of-the-art statistical and machine learning approaches for time series and extreme events analysis.

For Direction 1, the research team employed a combination of bivariate and trivariate extreme value analysis methods to quantify the joint probability of exceedance in these variables, focusing on stratifying by wind speed and direction, using data from Bridgeport, Connecticut. This research has demonstrated that wind speed and direction are critical factors in understanding the tail dependence between precipitation and sea-level in the context of compound flood events. Removing harmonics from sea-level data significantly improves our ability to detect and quantify joint tail dependence. Stratified EVA frameworks helped us identify regimes most vulnerable to compound floods.

In Direction 2, the research team evaluated the reliability of four statistical forecasting models for predicting hourly sea level exceedance events in New Haven using 10- and 30-day training windows. In addition, they assessed forecast reliability using various relevant metrics important for disaster management authorities.  The conclusion was that no single time series model gave the best forecasts in all situations. The best model to use depends on the forecast horizon, and considerations such as high sensitivity or low false positives. The ETS method trained on 30 days of data is most useful for generating early warnings with low false alarms, while the ARIMAX trained on 10 days is the best forecasting model for generating alerts for transportation safety, etc

Project Timeline: June 2024 - June 2025

Project Outcomes

Project Final Report