The Machine Learning Foretold the Future

While monitoring convective initiation in a sea of Altocumulus castellanus clouds as they moved into a more convectively unstable environment, I noticed that the GREMLIN Emulated reflectivity in Briscoe County was higher than the MRMS (GIF 1).  The GREMLIN emulation was producing a DbZ of 50+ while the MRMS was showing a 30 DbZ. To me, this difference was operationally significant as I would pay closer attention to the developing 50 DbZ feature than I would the 30 Dbz feature. So far in our experiment, this was the opposite of what I had experienced and I initially thought the emulated radar was wrong. I jumped over to the OCTANE products to get a sense of how rapidly the cell was building (GIF 2). My assessment of the cell seemed to confirmed my initial opinion that it was not growing as rapidly as I would have expected given that there was not a strong gradient in the speed or direction products.

Gif One: GREMLIN Emulated Radar on the top and MRMS Composite Reflectivity on the bottom.

 

Gif Two: OCTANE Products (Speed Upper-left, Direction Upper-Right, Divergence Bottom-Left) and Day cloud Phase RGB (bottom-right)
My opinion changed 10 minutes later.  Seen in image one, the MRMS with the next update showed a similar storm intensity to the GREMLIN emulation. This is impressive when one considers the latency of GREMLIN is greater (about 10 min) than the latency of MRMS (about 2 min). The GREMLIN product actually delivered a more operationally useful product sooner, despite having a greater lag. The GREMLIN product continued to show this ability for two more storms developing further back in the line!
Image One: GREMLIN Emulated Radar on the top and MRMS Composite Reflectivity on the bottom.

-Kilometers

Tags: None

Leave a Comment