Lightning Prediction for Space Launch Using Machine Learning Based on Electric Field Mills and Lighting Detection and Ranging Data

The United States Air Force is looking at its busiest year in space launches down in Cape Canaveral, Florida. For the year 2019, USAF has 32 projected launches, 66 percent of the way to the goal of 48 and double of 24 launches, the previous standard. However, this goal most likely will not be reached as lightning continues to be the leading factor in delayed and cancelled launch missions.

According to the Annual Lightning Report from the National Lightning Detection Network, Florida has the most concentrated area of lightning strikes. This is a challenge that the 45^th Weather Squadron (45^th WS), based out of Patrick Air Force Base, has dealt with since forming in 1991. The 45^th Weather Squadron works with the Kennedy Space Center (KSC) and Cape Canaveral Air Force Station (CCAFS), performing weather assessments to ensure that any safety requirements involving weather during pre-launch and the actual launch are met. Part of the way the 45^th WS does this is by analyzing electric field mill (EFM) data, which measures the strength of the electric fields in the atmosphere and is then used to forecast when a strike might occur, from the 31 EFMs around Cape Canaveral.

Past studies have investigated EFM data, looking for patterns to forecast lightning strikes. Results has yielded decent results and this study hopes to build upon previous findings while taking a different approach. Using four years’ worth of lightning detection and range (LDAR) data logged by the KSC and corresponding EFM data, several Long Short-term Memory (LSTM) Neural Networks. Separate LSTM models are built for each EFM and trained from adjusting parameters to forecast lighting 30 minutes out in the surrounding area of each mill.