Start
February 12, 2021 - 3:30 pm
End
February 12, 2021 - 4:30 pm
Categories
Convective Meteorology (Mesoscale Dynamics)Convective Meteorology (Mesoscale Dynamics) Seminar
Preliminary Analysis of Hail Size Prediction using Deep Learning Models
Amanda Burke
Friday, February 12th
3:30pm
Join Google Meet
https://meet.google.com/ksh-txvg-kni
Using Deep learning (DL) models to predict different atmospheric phenomena has rapidly grown in the past few years. The ability to learn both spatial and temporal data patterns, instead of distributions, highlights one reason why DL is becoming popular in the weather community. To test the capability of DL models to skillfully predict hail size over the contiguous United States (CONUS), where previous traditional machine learning (ML) models have shown success, ten environmental variables from the High-Resolution Ensemble Forecast system version 2 (HREFv2) are input to a UNET. Gridded maximum estimated size of hail (MESH) values are used as observations, where direct MESH values can be input within the DL framework instead of distributions of values. In depth details of the UNET framework will be presented as well as preliminary results of the DL forecasts, which show spatially continuous forecasts that can learn relevant areas of hail threat with a smaller number of variables than the traditional ML framework.