Tropical Cyclone Intensity via Deep Learning

The Dvorak Technique had a great run. Seldom does an application come along that provides reasonable data for such a long period of time. In fact, there's a special session at this meeting celebrating 50 years since its introduction. Well done. Over the years, the Dvorak Technique has been tweaked, objectified, and automated. It remains the standard for determining tropical cyclone intensity to this day. But it's time is over. No longer do we need to stare at gray-scale enhanced IR images that repeat shades because...uh, color hasn't been invented yet? No longer is it required to read through 126 replies on a tropical storms thread about why a storm should have a CI of 5.0 vs. 5.5. Done are the days of conflicting global best track data resulting from subjective applications of the technique. Put to rest the frustration of complicated decision trees, cloud patterns that could be shear or embedded center, questioning your neighbor's banding feature assessment, and numerous caveats and special rules that may or may not be violated depending on the mood of the analyst, day of the week, or what's for dinner later. Joe Montana was a great quarterback, but there's a reason why he is no longer playing. It's time to retire the Dvorak Technique and celebrate its valuable contributions to tropical meteorology. Create the bronze bust and induct Dvorak into the Hall of Fame. But at the same time, let's recognize that artificial intelligence is here and is already better than humans in image pattern recognition. Specifically, images of tropical cyclones. And best of all, it never complains nor asks to be treated to dinner. In this presentation, we will demonstrate how a simple convolutional neural network, trained on several years of infrared and visible geostationary satellite imagery, can produce robust estimates of tropical cyclone intensity quickly and accurately, freeing us from hours of frustration trying to answer the same question over and over again - "are those pixels dark gray or medium gray?"