Potential of Synthetic Weather Radar for Squall Identification and Prediction

Squall event impacts are a long-standing problem offshore, especially in regions where radar imagery and reliable nowcasts are unavailable. Existing methodologies for monitoring and advising of impacts have been near-stagnant for decades. In this paper, initial results of using machine learning tools paired with advancements in weather satellite imagery processing are presented. The approach is based on a novel method of processing satellite, lightning, radar, and numerical weather model datasets trained against observed weather radar as truth to create gridded synthetic radar and short-term forecast. The capability has demonstrated to be an effective system in simulating and predicting the high precipitation rates that are associated with squall activity in real-time. The resulting output provides precipitation rates among other attributes at 1-km resolution, updated every five minutes, and gridded extrapolative nowcasts produced to four hours ahead. Initial results over multiple geographic domains of the system have performed exceptionally well at identifying and tracking strong thunderstorm activity, with and without ground radar, including detection rates over 90% and false alarm rates near 20%. As the technique is improved and deployed more broadly on a global scale, the objective is to provide a consistent, high-fidelity dataset that enables squall risk identification and advisories within a minimum two-hour planning horizon. The primary visualization for situational awareness is a commonly used format: weather radar. New levels of productivity and safety are possible with the global expansion and application of this system. This work was completed as a collaboration between Solcast, WeatherZone, and DTN.