Pristine Nocturnal Elevated Convective Initiation: A Preliminary Climatology and Evaluation of Predictability during PECAN

The prediction of convective initiation remains a challenge to forecasters in the Great Plains, especially for elevated events at night.  This study examines a subset of 287 nocturnal elevated convective initiation events that occurred without direct influence from surface boundaries or pre-existing convection (pristine nocturnal convective initiation, hereafter PNCI) over a four-month period during the summer of 2015 (May, June, July, and August).  Events were first classified into one of four types based on based on apparent formation mechanisms and location relative to any low-level jets.   Variations in timing and location, along with elevation of the PNCI, were examined.  Because the study period included the period when the Plains Elevated Convection At Night (PECAN) field campaign (June 1 – July 15, 2015) took place, simulations from five convection allowing models run to assist forecasters during PECAN were evaluated to study the predictability of the different types of PNCI.  In addition, four versions of a 4 km horizontal grid spacing WRF model, each using a different planetary boundary layer (PBL) scheme, were analyzed. 

The climatology revealed a dual-peak pattern in timing of PNCI, with one peak near 0400 UTC and another near 0700 UTC.  The dual peak was not due to differences in typical timing of the four different types of PNCI, and was most pronounced in the central region of the Plains where the primary PECAN study area was located.  Differences in location and elevation of the initiation for the different types of PNCI were at best subtle. 

Model simulations were more deficient with location and elevation rather than with timing during the PECAN project, with the models consistently initiating convection at a lower level than what actually happened.  The five models evaluated, the NSSL WRF, the NCAR MPAS, the OU MAPS 1 km WRF, the NCEP HRRR, and the CSU WRF, in general correctly predicted between half and two thirds of all PNCI occurrences, with hit rates ranging from 56% for the 1 km OU MAPS WRF run to 67% for the CSU WRF run.  Probabilistic forecasts made by a team of forecasters relying heavily on the output of these models were generally worse for PNCI in 6 hour windows than they were for bores in 6 hour windows, and MCS occurrence within 3 hour windows.  The forecasters were not able to distinguish between environments marginally favorable for PNCI (low probability forecasts) and those more favorable (moderate probability), with PNCI occurring within the risk areas a roughly equal percentage of the time, around 75%.  Sensitivity tests using different PBL schemes did not find substantial differences between three of the schemes (MYNN, QNSE, YSU) but did find more occurrences where the PNCI was missed when the ACM2 scheme was used.  As evidence of the challenges in obtaining accurate forecasts of PNCI, soundings from the various model runs had only very subtle differences.