The thunderstorm probabilities are based on the Localized Aviation MOS Program (LAMP) concept, whereby Global Forecast System (GFS) based model output statistics (MOS) thunderstorm probabilities issued four times daily are updated hourly. The forecast updates provided by LAMP thunderstorm probability regression equations arise from additional predictors specified from real time lightning strike and quality controlled radar reflectivity measurements, observed surface weather variables, and fine scale topography and lightning climatology. The lightning and radar predictors dominate in the first few forecast projections, whereas the MOS probability and various other predictors prevail thereafter.
The grid-oriented LAMP thunderstorm probability regression equations are geographically regionalized. With conventional specification of the regions and use of a fine-mesh grid, non-meteorological discontinuities in the probability patterns appeared along region boundaries. The problem was mitigated through two new procedures: (1) expanding the regions to introduce overlap among neighboring regions, and (2) performing smoothing along region boundaries where a residual discontinuity is detected. These procedures effectively mitigate the discontinuities without degrading forecast skill.
Verification of the operational thunderstorm probability and categorical forecasts shows strong skill in the 0-2 h period. Thereafter, the skill falls off rapidly with time such that in the 8 24 h projection range, it is just slightly higher than that for similar GFS-based MOS thunderstorm probabilities. The high skill in the first few projections is due to predictors derived from the latest lightning strike and radar reflectivity measurements, whereas the more modest skill thereafter results from simple advective models, supplementary GFS and climatological predictors, and model regionalization.