Global Synthetic Weather Radar – teaching the machine to read Published Dec. 2, 2020 By David R. Hopper 557th Weather Wing OFFUTT AIR FORCE BASE, Neb. -- This is the second in a three-part series covering the innovative work by 557th Weather Wing Airmen for the ongoing development efforts into machine-learning for a weather radar depiction across the globe, designated the Global Synthetic Weather Radar (GSWR). Early civilizations began using meteorological and astrological events to attempt to predict the change of seasons. The Babylonians tried to predict weather using visual cues like cloud formations and Chinese astronomers divided the year into 24 festivals to associate with different types of weather. It wasn’t until the 1400s that an instrument was developed to assist with weather prediction. First the Hygrometer, invented by Nicolas Cusa, was able to measure humidity in the air. After that, Galileo Galilei invented the thermometer and by 1643, Evangelista Torricelli invented the barometer. While more modern instruments and techniques are now available to assist with weather prediction, it still takes a skilled technician to utilize and interpret the vast amount of data and information available to assess current conditions and make a forecast. Today’s technological leap forward is in Machine Learning or Artificial Intelligence. For the last several months the 557th Weather Wing has been assisting Massachusetts Institute of Technology / Lincoln Labs by evaluating the operational feasibility for a potentially groundbreaking ML capability, GSWR. Forecasters from the 1st Weather Group are assessing GSWR to determine how well the system depicts and predicts meteorological patterns across the entire globe and, perhaps in the future, whether the military can use it for operations on a worldwide scale. “GSWR is designed with two roles in mind,” said Jeffrey Fries, 1st Weather Group chief of operations, standards and tactics. “Primarily, GSWR is a technical solution to fill a capabilities gap in sensing the terrestrial environment in data sparse regions. The application was originally designed to fill gaps in RADAR coverage over the Gulf of Mexico to assist the Federal Aviation Administration in routing aircraft. That has now been expanded to a global capability by taking advantage of improved ability to remotely sense lightning strikes via a global sensing network comprised of terrestrial and space-borne sensing systems.” GSWR utilizes known relationships between satellite imagery, lightning, rainfall rates and radar reflectivity to portray thunderstorm activity. It also uses current meteorological model data to provide environmental insight to precipitation occurring outside of thunderstorm activity. Most of the data is being driven by 557 WW Airmen at the 2d Systems Operations Squadron. How does machine learning work in GSWR? Using data from multiple seasons, the machine learning algorithms discover relationships between the variables over different geographic domains to produce an analysis of simulated RADAR images. Exploiting the meteorological model data from the 557th WWs Global Air Land Weather Exploitation Model allows for a complete product with lead-times out to 12 hours for the weather operators. Airmen from the 557th WW must then leverage their training in satellite interpretation and other data sources to validate GSWR output and assist MIT/LL technicians with “teaching” the machine to read and predict weather patterns. “The 1st Weather Group weapons and tactics members, working in concert with forecasters in the Operational Weather Squadrons, observed GSWR behavior during four tropical cyclones (hurricanes and typhoons) and provide subjective analysis of its ability to accurately sense and depict key features of the tropical cyclones while over water and outside RADAR sensing networks,” Fries said. “In addition, we evaluated GSWR output over mountainous regions on three continents to determine whether the application handled difficult and variable terrain influences on thunderstorm formation, lifecycle, and propagation.” We were encouraged by the results and look forward to more detailed testing as the capability matures and is made ready for operations.” At this point, 557th WW Airmen have logged hundreds of hours during the evaluation period. They have not only been actively feeding data to MIT/LL that will be used by the GSWR in order to teach the machine, but also evaluating the results of GSWR forecasting for accuracy. If the AI misses the mark, the real-time weather will be applied so the AI can learn from its mistakes. “The 557th Weather Wing is willing to assist MIT/LL with this project because we see the potential in enhancing our abilities to conduct global operations, even in vast, data sparse regions,” said Col. Patrick Williams, 557th Weather Wing commander. “If the GSWR becomes a delivered capability, it could end up being a valuable asset for our highly-skilled 557 WW Airmen. Accurate prediction of weather is a skill that can take decades to master, the amount of time that the GSWR ML/AI will need for this particular facet of weather is still unknown.