LaunchCast: Deep Learning for Space Operations

Natural and triggered lightning pose significant safety risks to the general population and to rocket launches carrying high value payloads; accompanied by an increased demand for access to space due to commercialization/monetization. Climate change may increase lightning strikes over the next century by nearly 41% globally (Pérez-Invernón, Gordillo-Vázquez et al. 2023), potentially decreasing access to space given current safety considerations in the Lightning Launch Commit Criteria (Merceret 2010). The Space Coast (Florida’s central-Atlantic Coast) is home to a unique, disparate, and dense network of meteorological sensors to support space launch operations. These sensors include standard surface observations, a WSR-88D and WSR-74C weather radar, electric field mills, and the Mesoscale Eastern Range Lightning Information Network (MERLIN; Vaisala TLS200 sensors). Artificial Intelligence and Machine Learning (AI/ML) is known to provide insights into such disparate sensor networks, along with non-disparate and traditional sensing methods. In 2019, Mostajabi proved that XGBoost, a supervised decision tree ML algorithm, can hindcast lightning flashes from simple meteorological parameters like air pressure, temperature, humidity, providing 10-30 minute lead times across 12 different sensors in varying terrain (Mostajabi, Finney et al. 2019). Additionally, ProbSevere’s LightningCast product uses a convolutional neural network to quantify the threat of lightning for specific locations, correlating GOES 16 Advanced Baseline Microwave Imagery and Global Lightning Mapper Imagery (Cintineo, Pavolonis and Sieglaff 2022). Motivated by these results, this study investigates if similar meteorological parameters from 10 weather towers near the Space Coast can be used as predictors, to provide similar lead times for both cloud-to-cloud and cloud-to-ground lightning sensed on MERLIN, from 2017-2022. The training data includes the observations from 2017-2021, with the test dataset in 2022. This study is unique because the lightning detection efficiency near the Space Coast is much higher due to the dense sensor network within MERLIN. Future work may expand to include gridded radar datasets, to include dual polarization parameters, geo-stationary satellite data, and potentially electric field mills while exploring explainable AI techniques.

DISCLAIMER: “The views expressed in this article are those of the author and do not reflect the official policy or position of the United States Air Force, Department of Defense, or the U.S. Government.”

References:

Cintineo, J. L., M. J. Pavolonis and J. M. Sieglaff (2022). "ProbSevere LightningCast: A Deep-Learning Model for Satellite-Based Lightning Nowcasting." Weather and forecasting 37(7): 1239-1257.

Merceret, F. J. W., John C.; Christian, Hugh J.; Dye, James E.; Krider, Philip; Madura, John T.; O’Brien, Paul T.; Rust, David; Walterscheid, Richard J. (2010). A History of the Lightning Launch Commit Criteria and the Lightning Advisory Panel for America’s Space Program. D. o. Commerce. Kennedy Space Center, NASA.

Mostajabi, A., et al. (2019). "Nowcasting lightning occurrence from commonly available meteorological parameters using machine learning techniques." npj Climate and Atmospheric Science 2(1).

Pérez-Invernón, F. J., et al. (2023). "Variation of lightning-ignited wildfire patterns under climate change." Nature Communications 14(1): 739.