About Me

I am a Research Associate at the Harvard T.H. Chan School of Public Health and Harvard University’s Data Science Initiative, HDSI, in Cambridge, MA. I completed a postdoctoral fellowship at Princeton University, advised by Prof Michael Oppenheimer. I obtained my DPhil (PhD) in Climate Science with a focus on climate attribution and climate econometrics from the University of Oxford, and obtained a MS in Sustainability Management from Columbia University.

My research approach is strongly interdisciplinary and collaborative, integrating numerical models and data science methods from probabilistic event attribution and econometrics to answer research questions related to estimating socio-economic impacts of extreme weather and climate change, and, more recently, the rapid build-out of AI infrastructure. The main questions that underlie my ongoing research activities include:
1) Whether and to what extent does anthropogenic climate change impact the frequency and intensity of extreme weather events and resulting human activities and economic outcomes?
2) What are the human responses and health-related sensitivities to climate change?
3) What disaster risk reduction mechanisms prepare populations and infrastructure before natural hazards turn into disasters?
4) How is the rapid growth of AI-driven electricity demand reshaping electricity markets, grid reliability, and the distribution of environmental and health costs and benefits across communities?
5) Can we predict the energy and health impacts of a proposed AI data center before it is built, to give communities, regulators, utilities, and developers a scientific, rigorous, and unbiased basis for deciding where facilities should — and should not — be sited? (See this Harvard Chan School write-up of our Frontiers of Innovation for Societal Impact-funded project with Francesca Dominici and Le Xie.)

At the Dominici Lab, I also serve as Guest Editor, with Francesca Dominici and Shaolei Ren (UC Riverside), of a collection in the Nature Portfolio journal npj Climate Action on the health and environmental impacts of AI data centers.

Latest Research

Environmental and health impacts of the AI data center boom, working paper with Kevin Butler and Francesca Dominici

The build-out of AI and hyperscale data centers is one of the fastest sustained increases in US electricity demand in decades, and it raises questions that go beyond price: facility siting is already tied to air pollution from fossil-fuel generation and backup diesel generators, heavy water use for cooling, and noise, with these burdens falling unevenly across host communities. In our working paper, we assemble a newly linked utility-level panel (2021–2025) connecting verified hyperscale campus locations and capacity to US electricity price and sales data. Industrial rates rise after a data center opens in every specification we run — by roughly $13 to $51 per megawatt-hour — though with only 12 treated utilities in our panel, this effect can’t yet be called statistically reliable. Our clearest result, drawn from a much larger sample of 1,199 utilities, is that a utility’s own residential price rises when its neighboring utilities host more data center capacity, a pattern that lines up with an active complaint Maryland’s utility regulator has filed with federal energy regulators alleging that grid-upgrade costs are being spread across an entire region rather than charged to the data centers driving them. A companion analysis asks whether host counties see offsetting local benefits in jobs or income, and a geographic crosswalk shows that the counties bearing these costs and the counties capturing the benefits frequently are not the same ones. We read our results alongside related work finding that how a campus sources its power materially shifts who bears the resulting air-quality burden, and a state-level accounting exercise finding that a single state’s projected air-pollution health costs from data-center growth could exceed its projected electricity-price costs several times over. For a plain-language walkthrough of the data, methods, and findings, see our summary memo.

Socio-economic dimensions of climate change, published in Nature Communications

We underscore the economic dimensions of involuntary immobility—a critical but often ignored consequence of climate change, where individuals and communities are unable to migrate away from high-risk areas due to financial, structural, and systemic barriers. While climate change intensifies natural hazards, economic constraints—such as poverty, lack of access to capital, and limited livelihood alternatives—prevent vulnerable populations from relocating, even when staying poses severe risks. In our study, we highlight how economic disparities exacerbate climate vulnerability: those with fewer resources face greater exposure to climate impacts and fewer options to escape them. Addressing this challenge requires policies that not only reduce economic barriers to migration but also invest in local adaptation, ensuring that climate action is equitable and supports both mobility and resilience for those who cannot or choose not to move. Without targeted economic interventions, climate change will deepen inequality and leave the most vulnerable populations further behind.

Bayesian modeling assessing short-term conflict-displacement dynamics, published in Annals of Applied Statistics

Together with Gregor Zens, I investigate methodological challenges when modeling imperfect data collected in conflict zones. Employing a Bayesian panel regression, we aim to understand why and under what conditions higher temperatures and meteorological drought lead to conflict-related migration. The results suggest a rapid and non-linear migration response post-conflict, with significant heterogeneity in effects dependent on the nature of conflict events. Our model outperforms standard benchmarks in a forecasting exercise, underscoring its potential for informing decision-makers in crisis scenarios.

Advancing flood displacement modelling, published in Earth’s Future

The 2022 summer flood in Pakistan displaced about 7 million people in the Sindh province alone. Up to one-third of the country’s area was flooded. Using the open-source impact assessment platform CLIMADA, we study the extent to which flood-related hazards can be used to quantify displacement outcomes in a data-limited region. Using flood depths, exposed population, and impact functions, we link flood vulnerability to displaced people. This allows us to estimate internal displacement resulting from the flood event and assess how displacement varies across different areas. We find that a flood depth threshold of 0.67m (CI 0.35 - 1.10) provides a best fit to all data from Sindh province. We find a negative correlation between displacement and the degree of urbanisation. By testing our model’s performance in explaining differences in displacement estimates across Pakistan, we demonstrate the limitations of existing impact assessment frameworks.

AI for Early Warning Systems and reducing climate change impacts, published in iScience

Artificial intelligence (AI) is increasingly being used to analyze natural hazards and their impacts, which is crucial for improving Early Warning Systems (EWS). These systems help protect people and communities by providing timely, actionable alerts about risks such as floods, storms, and earthquakes. However, AI’s potential in EWS remains underutilized, especially in risk assessment, forecasting, communication, and preparedness. This study reviews how AI is currently used in EWS, identifies challenges and opportunities, and emphasizes the need for responsible AI frameworks. The goal is to ensure that AI-powered warning systems are inclusive, accessible, and focused on people’s needs, supporting global efforts to make early warnings available to everyone by 2027.