Please join us for a UW Data Science Seminar featuring UW Atmospheric and Climate Science PhD student Miles Epstein on Thursday, February 19th from 4:30 to 5:20 p.m. PT. The seminar will be held in IEB G109.

“Predicting Biases in Day 1 Probabilistic Convective Outlooks from Time-Evolving Environmental Conditions with Convolutional Neural Networks”

Abstract: Severe thunderstorms (those producing straight-line winds at least 58 mph, hail at least one inch in diameter, and/or a tornado) pose a substantial threat to life and property in the United States. To help mitigate these risks, the National Oceanic and Atmospheric Administration/National Weather Service’s Storm Prediction Center issues daily Convective Outlooks (COs) outlining spatial risk levels for severe thunderstorms with up to eight days of lead time. Previous work (in review) details two methods for evaluating probabilistic COs; this work quantifies day-level biases in total storm coverage and in location (broken into north–south and east–west components) for Day 1 probabilistic Convective Outlooks.

We train a Convolutional Neural Network (CNN) to predict the probabilistic Sinh-arcsinh-normal (SHASH) distribution for coverage bias and both components of directional bias in Day 1 probabilistic Convective Outlooks from time-evolving, gridded environmental data (via Fifth Generation ECMWF Atmospheric Reanalysis (ERA5)). The trained model may be used operationally by inputting modeled environmental conditions (e.g., with one or two days of lead time). The distributions predicted by the model provide “anticipated feedback” and quantify expected uncertainty for Convective Outlooks associated with those conditions. Ultimately, by better understanding the environments in which severe thunderstorms develop, we can issue more accurate outlooks that will best mitigate loss of life and damage caused by those storms.

Speaker Bio: Miles Epstein is a PhD student in the Department of Atmospheric and Climate Science at the University of Washington. His research focuses on improving forecasts for severe thunderstorms and other high-impact weather events with data-driven methods, particularly machine learning.