One of my colleagues has long wanted me to create an algorithm to find overshooting tops in satellite imagery. There were a couple of things holding me back -- one administrative and the other technical.
The first was a lack of funding to pursue that work. But all of a sudden, in the last six months or so, I have gotten funding (from NASA, NESDIS and HPCC) to work on satellite data. In the past few years, none of these guys would fund us. This year, they all did -- when it rains, it pours. Anyway, we have the funding, so I have the time. That's how science works. Administratively at least.
The second reason was techical. The first signature I tried to find on satellite data was v-shaped notches on infrared imagery which was at 4km resolution and came every half hour. The temporal and spatial resolution was too poor to do anything worthwhile (as a comparison, radar data is at 1-km resolution every 5 minutes). But recently, my colleague got his hands on 1-km visible data combined from the GOES east and west satellites at 1 minute intervals and suggested that I try to find overshooting tops which are indicative of strong updrafts which themselves are indicative of severe weather.
To find overshooting tops, I wrote an algorithm to look for areas with high spatial variance (taking to care to avoid using parts of the image that are unlikely to be clouds in the variance computation) and that was it! As it turned out, given 1km visible imagery, you don't even need a sequence for quality control (In many algorithms, one way to avoid lots of spurious detections is to correlate things across time).
In the image to the left, the red contours are drawn by the automated algorithm that results. Since the 1-minute temporal resolution is not required, and because 1km visible imagery is available during daylight hours in realtime, we should be able to run an overshooting tops algorithm in real-time. As the newpapers say, watch this space.