One of the products demonstrated at the Hazardous Weather Testbed (HWT) in Norman in May and June this year was GREMLIN (GOES Radar Estimation via Machine Learning to Inform NWP) a tool created with Machine Learning (by Kyle Hilburn at CIRA) to use ABI and GLM data to create synthetic radar reflectivity fields. (This article describes the product fully). GREMLIN was trained on GOES-16 CONUS data, but in these images it is displayed over American Samoa (the product has been incorporated into the AWIPS machine at WSO Pago Pago), and it’s being compared to hourly estimates of rainfall from CMORPH-2. The images below show 6 fields (at 10-minute intervals) of simulated radar and an hourly CMORPH-2 accumulation derived independently and displayed in RealEarth.

Consider the 15 different animations shown below, that step through an event with rain over the Samoan Islands. In general, the CMORPH-2 hourly accumulations on the right agree in general with where the radar estimates derived from GREMLIN suggest rain might be falling.

What should a user do when GREMLIN does or does not show rain where CMORPH-2 is or is not showing rain accumulation. It should not make a user think “Oh, this is wrong”; rather, it means a user should investigate those regions using other satellite imagery to determine whether or not rains are occurring.

—————
Free Secure Email – Transcom Sigma
Transcom Hosting
Transcom Premium Domains