current PROJECTS

My dissertation explores how flooding from atmospheric river storms negatively impacts California communities.

TEMPORALLY COMPOUNDING FLOOD RISK

2022—2023

Temporally compound (back-to-back) atmospheric river storms have historically caused the most negative hydrologic and economic impacts in California; recent examples include the Oroville Dam crisis in 2017 and the incredible precipitation seen in the first few months of 2023. I and my coauthors have presented a definition of atmospheric river “sequences,” which are periods of time when hydrologic risk is elevated due to temporal compounding, and we have found that being in a sequence can triple the expected losses from an atmospheric river event. This work represents one of the first attempts to (a) identify when temporal compounding is occurring and (b) quantify the contribution of temporal compounding to overall flood damage and loss.

MEDIA COVERAGE:

“California could face more atmospheric river ‘super-sequences,’ study finds” (San Francisco Chronicle, 1/18/24)
“Clusters of atmospheric rivers amp up California storm damages” (Stanford News, 1/19/24)
“What is an atmospheric river storm? New study gives us an understanding” (NBC Bay Area, 1/19/24)
“Study projects sequences of atmospheric rivers will increase. Here’s what that means for Sonoma, Napa counties” (The Press Democrat, 1/30/24)
“What is an atmospheric river? A hydrologist explains the good, the bad, and how they’re changing” (PBS News Hour, 2/5/24)
“Why record rain hasn’t washed away California’s water woes” (Washington Post, 2/6/24)

atmospheric river damage drivers

2020—2023

Flood damage from atmospheric rivers depends on the strength and severity of the event (hazard), but also on the number of people and buildings (exposure) and their degree of preparedness for the event (vulnerability). We collected metrics of hazard, exposure, and vulnerability in California communities to fit a random forest model that predicts the probability of a specific atmospheric river causing flood damage. We used interpretable machine learning techniques to identify the most important damage drivers, which are the variables contributing the most to the model’s prediction, and determined the values and locations where those damage drivers caused the biggest increase in flood risk.

PERFORMANCE-BASED ATMOSPHERIC RIVER RISK ANALYSIS (PARRA)

2019—2022

If we know how big an atmospheric river will be, can we predict how much damage it will cause? To answer this question, I created a framework to integrate existing hydrological and meteorological research about how atmospheric rivers drive flooding with a more engineering-focused perspective on how flooding leads to property damage. Linking the chain of events from atmospheric river occurrence all the way to economic damage and incorporating uncertainty at every step can help to produce cost-benefit analyses of different flood mitigation strategies in northern California.

previous work

Photograph of cars in an apartment complex parking lot, all under several feet of water.

VEHICLE FLOOD LOSS IN SAN RAFAEL, CA
2020

I estimated the average annual loss for passenger vehicles in San Rafael, California under present and future flood scenarios. I created geospatial location tags for over 40,000 vehicles based on block group-level reports of both the number of housing units per building and the number of vehicles per household, and calculated how many locations would experience nonzero inundation under a range of return periods and sea level rise projections. I applied vehicle depth-damage fragility curves from the US Army Corps of Engineers and produced an average annual loss estimate for passenger vehicles within the study area. Results showed that vehicle-related flood loss in San Rafael is expected to increase almost linearly from 2020 to 2050, but that the additional losses are not distributed equally in space across the city and could exacerbate existing economic inequalities.

Map of northern California, with locations of fires highlighted in orange.

CALIFORNIA WILDFIRE RISK
2019

I conducted exploratory data analysis to investigate trends in the frequency, spatial distribution, and seasonality of California wildfires. I matched fire occurrence and severity to spatial and temporal weather information and looked at the relationships between the two for a few case study locations. I then created a predictive classification model to label each day in the time series as fire/no fire for the area surrounding Oroville, CA. The model was trained with data from 2000-2012 and tested with data from 2013-2015, and was able to achieve 70% prediction accuracy on the unseen test data.

TORNADO HAZARD MODEL
2018

I worked with two other people to create a custom tornado hazard model for portfolio risk assessment at BHSI. We developed a brand-new methodology for correcting biases in the NOAA tornado record based on population distribution, which I was responsible for implementing because of my geospatial experience. I helped write the model code in R, ran beta tests to troubleshoot the model, and wrote a script to visualize tornado paths and intensity across a single-site insurance portfolio.

Photograph of an industrial facility, with smokestacks giving off plumes of smoke.

INDUSTRIAL FACILITY STORM SURGE RISK
2017

At BHSI I worked on projects to generate views for unusual and client-specific risks. One of these projects involved generating custom damage functions for heavy industrial facilities exposed to hurricane wind and storm surge. Because the site had such unique equipment, typical portfolio risk assessment methods were insufficient, so I helped to create new component-level fragility curves for a wide variety of machinery and equipment. After spending over a month digging into these industrial facility damage functions, I spearheaded the analysis on a strategic account and created an analysis tool for other team members to use.

Map of downtown Boston, Massachusetts, with flood inundation extents colored by the intensity of the driving hurricane. Cat 1 hurricanes cause flooding in the light green area, Cat 2 cause the dark green, Cat 3 cause the yellow, and Cat 4 cause the …

HURRICANE EMERGENCY RESPONSE IN BOSTON
2017-2018

As a senior at Northeastern I completed a year-long undergraduate Honors in the Discipline thesis with Dr. Jerome Hajjar predicting hurricane risk and resilience in the Boston area. I aggregated geospatial exposure and fragility information regarding Boston’s building stock and transportation infrastructure, as well as its water, communication, and power networks, for the purpose of predicting hurricane risk outcomes. I was able to improve upon existing regional damage simulations, and I was recognized for my work at Northeastern’s annual research expo, where my poster came in second out of all Undergraduate Engineering and Technology submissions.