Discipline: Ecology Environmental and Earth Sciences
Subcategory: Civil/Mechanical/Manufacturing Engineering
Session: 3
Room: Exhibit Hall
Stephen Rosario - Hunter College (CUNY)
Data Science and Machine Learning have proven to be powerful tools recently, in flood frequency analysis. This research aims to determine the spatial concordance of flood events, and their relationship to large-scale climate and atmospheric teleconnections. It also focuses on classifying such spatially concordant events into various types of risk based on their impact in terms of drainage area and population and assets exposed. The at-site risk, i.e., the probability of any site (region) under a spatially concordance event at any given time is also estimated. Daily streamflow data from the United States Geological Survey’s (USGS) HCDN stream gage stations over a period of 68 years (1950-2015) was used in this analysis. These stations are minimally impacted by anthropogenic influence. Using the 95th percentile of the daily streamflow data as a threshold for extreme flood events, a Bernoulli event matrix and Poisson Counts vector were created, detailing the days and stations with extreme flood events and the daily total number of stations simultaneously flooding. The drainage areas for the extreme flood events were then mapped into the Bernoulli event matrix and Poisson Counts vector, which resulted in the computation of the total area exposed to extreme floods. Risk typology is established based on the bivariate distribution of Poisson counts and the total area exposed. Further, spatial clustering techniques are employed to separate regions at high risk from region at low risk. Various non-linear dependence techniques are employed to detect the predictability of this risk conditional on climate and atmospheric teleconnections, which then allows us to study the similarities in the simultaneous floods and the causes and effects of these floods. This risk data and analyses will be crucial in creating forecast models to predict future flood events and the percentage of drainage area this might affect. Our current findings show a considerable number of stations flooding simultaneously across the United States. The highest number of simultaneous floods observed reached 142 stations (out of 318) on a single day, with a total area exposure of close to 100,000 square kilometers.
Funder Acknowledgement(s): National Science Foundation
Faculty Advisor: Naresh Devineni, ndevineni@ccny.cuny.edu
Role: I conducted all parts of this research with the guidance of my mentor and my faculty advisor.