Some Lessons on the Predictability of Convective Systems over a 36 h Timeframe

During the past six springs and summers, 36 hour realtime forecasts were conducted

daily with WRF-ARW using 3 to 4 km horizontal grid resolution and explicit convection

over the central and eastern US. Initial and boundary conditions have been supplied

by the larger-scale operational models such as the NAM and GFS. These forecasts

have been surprisingly successful on many days for forecasting larger convective

systems, such as squall lines, bow echoes, and mesoscale convective systems. However,

significant errors were noted on about 15% of the days where larger convective systems

were mis-forecast by over 4 h and/or 400 km, or were sometimes missed completely.

For such cases, sensitivity testing with model physics (e.g., microphysics, PBL, etc.)

and resolution generally failed to remedy the forecast error. For many of these cases,

however, significant forecast improvement was obtained by varying the initial conditions

(e.g., changing to or from NAM, GFS or RUC analyses, etc). This process has produced

a series of good-bad forecast pairs that can be analyzed to determine the nature of the

initial condition errors/improvements that lead to such dramatic forecast variability.

To date, two cases have been analyzed in detail: 1) A severe squall line system

on 19 June, 2007 that develops in the afternoon in western Kansas and propagates

southeastward over Oklahoma, and 2) A bow echo/derecho event on 23 August 2007 that

moves across north-central Iowa and Illinois, producing much wind damage, especially

in the Chicago area. Both of these cases show surprisingly large differences between the

12 UTC NAM and GFS analyses, especially in the representation of mid-tropospheric

water vapor. For each of these cases, the GFS analysis produced the better forecast.

The nature of the analysis differences in these and other similar cases would seem to

pose a significant challenge for current convective forecasting systems, both in terms of

obtaining the observations needed to sufficiently characterize the existing atmospheric

state, as well as for representing the true range of analysis uncertainty in ensemble

systems, as may be necessary to produce an appropriate range of potential forecast

outcomes.