Applying machine learning methods to detect convection using Geostationary Operational Environmental Satellite-16 (GOES-16) advanced baseline imager (ABI) data

摘要: Abstract. An ability to accurately detect convective regions is essential for initializing models short-term precipitation forecasts. Radar data are commonly used convection, but radars that provide high-temporal-resolution mostly available over land, and the quality of tends degrade mountainous regions. On other hand, geostationary satellite nearly anywhere in near-real time. Current operational geostationary satellites, Geostationary Operational Environmental Satellite-16 (GOES-16) Satellite-17, provide high-spatial- high-temporal-resolution only of cloud top properties; 1 min data, however, allow us observe convection from visible and infrared even without vertical information system. Existing detection algorithms using infrared look for static features clouds such as overshooting or lumpy cloud top surface growth occurs periods 30 min an hour. This study represents a proof concept artificial intelligence (AI) is able, when given from GOES-16, learn physical properties automate the detection process. A neural network model with convolutional layers proposed identify convection high-temporal resolution GOES-16 data. The takes five temporal images channel 2 (0.65  µ m) 14 (11.2  as inputs produces map In order provide products comparable radar products, it trained against Multi-Radar Multi-Sensor (MRMS), which radar-based product uses rather sophisticated method classify types. Two channels each related optical depth (channel 2) top height (channel 14), expected best represent convective clouds: high reflectance, surface, low top temperature. has correctly learned those convective clouds resulted reasonably false alarm ratio (FAR) high probability (POD). However, FAR POD can vary depending on the threshold, proper threshold needs be chosen based on the purpose.