Synthetic Radar by Combining Lightning and Satellite Data

Radar measurements are an essential part of accurate weather forecasting and greatly improve public safety. However, radars are expensive and have limited operational range. As a result, these types of measurements are difficult to obtain over oceans, in mountainous areas, and in developing countries. Therefore, it would be beneficial to develop cheaper and more widely available alternatives. The continued expansion of ground-based global lightning detection networks, as well as recent improvements in satellite observations, provides a unique opportunity in this regard. Lightning location data is highly correlated with the convective core of storms but poorly correlated outside these convective areas. Satellite data can image an entire storm but cannot accurately map the fast evolution of storm cores. This study aims to produce a synthetic radar product by combining lightning and satellite data through sophisticated machine learning algorithms that would leverage each dataset’s strengths. In this initial study, the field of view is limited to a region with good lightning, satellite, and radar coverage, North America, with the eventual goal of expanding the results to global coverage in the future. Lightning data comes from the Earth Networks Total Lightning Network (ENTLN) and satellite data from the Advanced Baseline Imager (ABI) aboard GOES-16. These data are pre-processed and then fed into a convolutional neural network (CNN), which are especially useful in extracting features from images, and trained using composite scans from the U.S. Next-generation Radar (NEXRAD). The high time resolution of the lightning data allows for this product to update every minute and effectively track the evolution of the storm core. Early results for the product are positive, showing that a synthetic radar approach can be a viable option when radar is not available.