The effect of the grid-scale scheme doing the work of the convective scheme depends on the scale of the model involved and the intensity of the grid-scale convection. In a model such as the 12-km Eta, grid-scale convection effects are limited to a local area and do not result in significant degradation to the forecast when they occur (See vignette.).
Typically, the environment in which the grid-scale convection develops is favorable to the formation of mesoscale convective systems (MCSs). There are some elements of a GFS forecast containing precipitation bombs which are reminiscent of these mesoscale systems, particularly the precipitation rates and (to a small extent) the scale and propagation of the resulting disturbance. However, MCSs cannot be accurately reproduced in the GFS because the model is hydrostatic (vertical accelerations from buoyancy are ignored), and MCSs are decidedly not. Even if the GFS were non-hydrostatic, the motions at the scale of an MCS cannot be resolved at the GFS equivalent grid-scale resolution of 80-km.
In the 12 UTC 20 August 2002 GFS forecast run, we saw that the effects are significant in all three occurrences over the Midwest and Great Plains, with:
These effects rendered the forecast essentially useless, without adjustments to account for the effects of the grid-scale convection by the forecaster. For example, dry air advection in the wake of the overly intense wave resulted in what would have been a forecast for sunny, dry weather over the lower Great Lakes on 22 August. In fact, showers and thunderstorms were observed in this area.
Grid-scale convection occurs in the GFS at T254L64 resolution as well, as can be seen below in two forecasts for 24-hour precipitation valid at 12 UTC 23 August 2002 (during the time of this case). On the left is the parallel T254 which was placed into operations on 29 October 2002, and on the right is the T170 with 42 levels. Note that in the T254 forecast, there is a small scale precipitation maximum of about 125-150 mm (5-6") over Iowa. The precipitation bomb in the T254 is in the same general area as it was in the T170, but is forecast for 24 hours later. We can conclude, at least in this case, that the occurrence of grid-scale convection is sensitive to the initial conditions (the parallel run of the T254 had its own data assimilation system) and to the resolution of the model.
Additionally, the GFS is run at T170L42 after 84 hours; there is no evidence that the high resolution used up to 84 hours will remove the conditions conducive to the development of grid-scale convection later in the forecast.
The effects are most readily seen in situations with relatively weak synoptic-scale forcing. However, the effects of excessive low-to-middle tropospheric heating resulting from excessive conditional instability can be seen in the cold season as well. An example from a winter case appears below (from December 2001), with the 24-hour grid-scale precipitation forecast on the left and the low positions for the forecast (black) versus verification (red) on the right. The verifications comes from the GFS data assimilation system (GDAS) analyses.
Note that the developing low was forecast at 1003-hPa near Pittsburgh; verification was 1012-hPa near Lynchburg, VA. Associated with the deeper low than verification is heavy grid-scale precipitation. Heavy convective precipitation did actually occur, to the south of the forecast grid-scale precipitation axis shown above (not shown). Later, the pressure difference between forecast and verification reached 10-hPa (996 versus 1006-hPa) as the low moved offshore southeast of MA, with a significant forecast displacement to the north and west of verification. This storm was associated with the only significant December 2001 snowfall over interior NY and New England.