Home

Trevor Harris

I am an Assistant Professor of Statistics at the University of Connecticut. I work on uncertainty quantification, statistical learning, and methods for spatial and functional data, with applications in climate science and epidemiology.

Much of my research asks how we can assess the reliability and uncertainty of a model's predictions and use those predictions to learn about system we're modeling. Recent works include conformal prediction for functions and spatial events, the evaluation and calibration of climate models, uncovering latent dynamics in the climate system, and functional inverse modeling.

Before joining UConn in 2024, I was an Assistant Professor at Texas A&M University. I received my PhD in Statistics from the University of Illinois Urbana-Champaign.

Research

Map of temperature-distribution differences between GISS-E2-2-G and ERA5, with larger differences near the Arctic — SCWD reveals where a climate model differs from reanalysis.

Climate model evaluation, integration, and emulation

My research develops statistical methods for comparing climate simulations with observations and combining information across models and spatial resolutions. This includes optimal transport metrics for distributional bias and extremes, and operator models for estimating local climate distributions. Current work is on developing lightweight latent emulators of the climate’s forced response, with the aim of extending weather models to novel climate conditions.

Blue samples approach a red conformal boundary along the arrows of a nonconformity flow — A nonconformity flow moves samples toward a prediction-set boundary.

Uncertainty quantification and calibration

More broadly I also develop conformal methods for quantifying uncertainty in operators models for functions, fields, and trajectories. Recent work uses flows to represent conformal prediction sets and construct predictive distributions. I am extending this approach to calibrate probabilistic models through constrained optimal transport which find the nearest distribution that satisfies specified statistical and physical constraints.

Global map of wintertime conditional tail dependence in precipitation extremes relative to a boxed equatorial Pacific region — Wintertime dependence between precipitation extremes and the boxed Pacific region.

Dependence, extremes, and inverse problems

My work also studies how interactions across space and time shape environmental variability and risk. My work includes neural methods for detecting Granger-causal influences and estimating conditional dependence between precipitation extremes. In a newer direction, I am also developing simulation based inference (SBI) and confidence distributions for functional inverse problems. Applications include reconstructing unobserved climate processes from indirect observations.

I also develop graph models for West Nile virus forecasting and optimize mosquito-control intervention policies with reinforcement learning in collaboration with epidemiologists.

Selected papers

Flow-Based Conformal Predictive Distributions (2026). NeurIPS, accepted. Code
Decision-Focused Calibration with Adaptive Conformal Predictive Distributions (2026). ICLR, under review.
Manifold Constrained Conformal Prediction for Spatial Events (2026). ICLR, under review.
Locally Adaptive Conformal Inference for Operator Models (2025). Transactions on Machine Learning Research, under review. Code
Statistics for Climate Science: Differentiation and Integration (2026). Annual Review of Statistics and Its Application, advance online publication.
Quantifying Uncertainty in Climate Projections with Conformal Ensembles (2024). Annals of Applied Statistics, accepted. Code
Validating Climate Models with Spherical Convolutional Wasserstein Distance (2024). NeurIPS, spotlight. Code
Forecasting West Nile Virus with Deep Graph Encoders (2025). arXiv preprint.

See the full publication list for all papers and their current status.