Amanda Burke - February 12

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

Start

February 12, 2021 - 3:30 pm

End

February 12, 2021 - 4:30 pm

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.