A Probabilistic Multispectral Pattern Recognition Method for Detection of Overshooting Cloud Tops Using Passive Satellite Imager Observations

Deep convective updrafts often penetrate through the surrounding cirrus anvil and into the lower stratosphere. Cross-tropopause transport of ice, water vapor, and chemicals occurs within these overshooting tops'' (OTs) along with a variety of hazardous weather conditions. OTs are readily apparent in satellite imagery, and, given the importance of OTs for weather and climate, a number of automated satellite-based detection methods have been developed. Some of these methods have proven to be relatively reliable, and their products are used in diverse Earth science applications. Nevertheless, analysis of these methods and feedback from product users indicate that use of fixed infrared temperature-based detection criteria often induces biases that can limit their utility for weather and climate analysis. This paper describes a new multispectral OT detection approach that improves upon those previously developed by minimizing use of fixed criteria and incorporating pattern recognition analyses to arrive at an OT probability product. The product is developed and validated using OT and non-OT anvil regions identified by a human within MODIS imagery. The product offered high skill for discriminating between OTs and anvils and matched 69% of human OT identifications for a particular probability threshold with a false-detection rate of 18%, outperforming previously existing methods. The false-detection rate drops to 1% when OT-induced texture detected within visible imagery is used to constrain the IR-based OT probability product. The OT probability product is also shown to improve severe-storm detection over the United States by 20% relative to the best existing method.