Jeffrey Smith
Dr. Jeffrey C. Smith began his academic passion in the field of Accelerator Physics. After building a cyclotron, a small particle accelerator, as an undergraduate at Knox College, Jeff matriculated at Cornell University furthering his passion for high energy particle accelerators and uncovering the inner workings of fundamental particles & the universe. His Ph.D. thesis was on the design of the International Linear Collider (ILC), a 22 mile-long electron-positron accelerator that will complement the discoveries being made at the Large Hadron Collider (LHC) at CERN in Geneva, Switzerland. After Cornell, Jeff joined the SLAC National Accelerator Laboratory at Stanford University to continue his work on the ILC and also to develop upgrade hardware for the LHC. After a successful career looking into the tiniest of inner-spaces Jeff decided to look up to the stars. Dr. Smith switched fields and began developing data processing and planet detection algorithms for the Kepler and TESS Missions. These missions combined have discovered thousands of extrasolar planets. With the Kepler and TESS planet detection pipelines now quite mature, Dr. Smith has diversified his research interests. Among other projects, he is centrally involved in a project funded by NASA's Planetary Defense Coordination Office to develop an automated pipeline to identify bolides (exploding meteors) in weather satellite data. The goal is to create a rich data set to inform the planetary defense community of the risks associated with large asteroidal impacts.
Session
Weather satellite data contains a wealth of information well beyond its application to meteorology. The GOES weather satellite lightning mapper instruments detect millions of lightning strikes per day. Within these haystacks are a handful of bolides (exploding meteors). Through a combination of hard manual work, advanced machine learning techniques, statistical analysis and supercomputers, our multi-disciplinary team has succeeded in creating an efficient pipeline to identify the bolides. Our algorithms are also sensitive to other interesting phenomena in the data. Funded by NASA's Planetary Defense Coordination Office (PDCO), our goal is to create a rich, calibrated, and statistically consistent data set of bolide light curves to inform the planetary defense community of the risks associated with large asteroidal impacts. We utilize a three-stage detection pipeline, with successively more computationally expensive algorithms: 1) simple Hierarchical Clustering, 2) Random Forests and then 3) Convolutional Neural Networks. Detections are promptly published on a NASA hosted publicly available website, https://neo-bolide.ndc.nasa.gov. We present the evolution of our pipeline, the ML techniques utilized and how we continue to incorporate new information to improve detection performance.