LightningCast Use for Wildfire Situational Awareness

LightningCast is a convolutional neural network that predicts the probability of lightning within the next 60 minutes using GOES-R ABI data as inputs. Core inputs to the model are the 0.64-µm reflectance (red portion of the visible), the 1.6-µm reflectance (shortwave infrared), the 10.3-µm brightness temperature (longwave infrared), and the 12.3-µm brightness temperature (longwave infrared), with the truth dataset being the gridded Geostationary Lightning Mapper (GLM) flash extent density (FED) product. This research combines LightningCast probabilistic lightning predictions with fire detections from a prototype version of the Next Generation Fire System (NGFS) to examine how LightningCast, now operational within the National Weather Service, can be used by incident meteorologists (IMETs) for situational awareness to increase firefighter personnel safety. Using a retrospective NGFS dataset for the western U.S. spanning June-October 2020 (~1800 fires), the analysis shows that the Critical Success Index for GOES-17 LightningCast maximizes around prediction thresholds of 50-70%. It has a regional dependence, with the highest skill found in northern and eastern regions of the domain. A diurnal signal is observed, with skill maximized during the day. Results are also compared to predictions from GOES-16 for the same 5-month period. Case studies examining lead time provided by LightningCast for select 2024 fires are also presented.