Finding Fireballs in Lightning: A Daily Pipeline to Find Meteors in Weather Satellite Data

Abstract

Weather satellite data contains a wealth of information well beyond its application to meteorology. The GOES weather satellite Geostationary Lightning Mapper instruments detect millions of lightning strikes per day. Within these “haystacks” of lightning are a handful of “needles” of bolides (aka bright fireballs, or exploding meteors). We have developed an efficient pipeline to identify these bolides. Our goal is to create a rich, calibrated, and consistently processed data set of bolide light curves that can be used by the planetary defense community in assessing the risks associated with large asteroid impacts. We utilize a multi‐stage detection pipeline, with successively more computationally expensive machine learning algorithms. Detections are published on a NASA hosted publicly available website, https://neo‐bolide.ndc.nasa.gov , with a typical latency of less than 2 weeks from the event occurrence. Virtually all aspects of the pipeline have been automated and we have deployed a “selective classification” system where most bolide candidates can be automatically published with no human intervention. The remaining candidates, where the machine classifier abstains, are human vetted using an efficient web‐based GUI interface. The work presented here is an improvement on a previously deployed pipeline, and a key new component to the automation is a neural network‐based auto‐validator which minimizes the manual effort required by human subject matter experts. Improvements to other parts of the pipeline are also presented, including a unified database and data pipeline for efficient detection, assessment and publication. User tools are available to aid in the exploration of our data set.