On June 15th 2022, a dynamic setup was unfolding across Iowa, Minnesota, and Wisconsin with multiple hazards that NWS forecasters would have had to message and warn. For this case, we were on watch for a DSS event representing the Cranberry Blossom Festival in Wisconsin Rapids in the GRB CWA. The main concerns with the event were lightning and any severe storms, both of which seemed certain for this case and the name of the game was timing the oncoming convection.
LightningCast uses machine learning with numerous satellite inputs that yields the probability of lightning occurring at a location within the next hour. This product immediately jumps to the front of a forecaster’s mind to apply for decision support services (DSS) or assessing lightning probability for airport forecasting. Below is a table showing the probabilities from LightningCast versus the “time of arrival” tool that estimated storm timing based on the movement of storms:
First vicinity lightning strikes (within 10 mi of event): 4:21 PM
Arrival of storms (within 10 mi of event): 4:30 PM
Immediately the usability of this product is fantastic. It shows the probabilities of lightning occurring in a contour format, making it a great pairing with satellite imagery, lightning data, and radar. WIth this case being a very well forced event the main evaluation was the percentages and how they did with the advection of the storms. The LightningCast seemed to ebb and flow with the eastward acceleration and deceleration of the storms between 3 and 3:30 PM, while the next 15 minutes showed accelerating storms, giving a 61% chance of lightning within the next hour at 3:45 PM. With the acceleration of the storms, it was good to see the model adjust, with the 50% threshold being crossed before 3:45 PM. The 50% threshold is very important for forecasters, as values above that are typically used in several products and gauges of confidence. The LightningCast model giving upwards of 40 minutes of lead time for advecting storms gives me a lot of confidence in the product, leaving me wishing it was already available within our datasets for immediate use.
– aerobeaver