1. Characteristics of precipitating convection in the UM at Δx≈200m-2km
Bob Plant1
and
Emilie Carter2, Peter Clark1, Heather Guy2, Carol Haliwell2, Kirsty Hanley2, Robin Hogan1,3,Humphrey Lean2, Thorwald Stein1,Mark Weeks2
1University of Reading 2 Met Office 3ECMWF
RMetS NCAS Conference, 6-8 July2016,
Challenges in using high resolution numerical weather prediction to forecast high impact weather in the atmospheric boundary layer

2. 1:3.33 aspect ratio, 25/08/12
Robin Hogan

3. Example of US forecast Hazardous weather testbed
00 UTC 17th May 2015
Hazardous weather testbed
Despite biases in representation of convective elements, good forecasting tool
Both qualitatively and improved skill according to FSS
Mark Weeks

4. Example of US forecast 2km UM Radar Hazardous weather testbed
00 UTC 17th May 2015
Hazardous weather testbed
Despite biases in representation of convective elements, good forecasting tool
Both qualitatively and improved skill according to FSS
Mark Weeks

5. Explicit to Grey Zone to Fully Parameterized
Grey zone for deep convection when Δx≈ cell size
Model can produce cells on near-grid scales
But don’t expect them to be well represented
Probably needs some form of parameterization, but often better without one than with a conventional parameterization
Grey zone for boundary layer turbulence for Δx≈ h
Model can produce BL overturning structures on near-grid scales
Needs a turbulence parameterization, but neither a 1D NWP nor a 3D LES parameterization is well suited
Resolutions in this talk range from explicit to grey for deep convection and from grey to fully parameterized for boundary layer turbulence

6. Convection-permitting models (e. g
Convection-permitting models (e.g. UKV) struggle with timing and characteristics of convective storms
Example, 07/08/11
Rainfall radar (Nimrod)
1.5 km forecast model (UKV)
By eye, UKV does not have enough small storms in this case
Too much heavy rain, not enough light

7. Average over strong cases Average over shower cases
Cell sizes
Average over strong cases
Average over shower cases
200-m model best
Storms are identified using an area threshold of 10 km2 and a rain rate threshold of 4 mm/hr.
40 days of convective weather in over southern England.
1.5km model is expected to be poorly resolved for the small storms!

8. Radar effective diameter (km) UKV effective diameter (km)
Average cell size in the UKV
Threshold (mm/hr)
Radar effective diameter (km)
UKV effective diameter (km)
0.125
8.8
18.0
0.5
7.0
14.9
1.0
6.2
13.2
2.0
5.0
11.2
4.0
3.7
9.0
8.0
2.9
6.7
16.0
2.4
3.8
UKV around twice the size of radar cells

9. Rain rate distributions
Average over strong cases
UKV struggles to get the largest rain rates
Improves for stronger cases at higher resolutions
In general, model cells seem to be not variable enough

10. Treatment of turbulence
The UM boundary-layer parameterization is based on semi-empirical vertical mixing profiles in a range of boundary layer types, with horizontal mixing based on Smagorinsky-Lilly
The 500m and 200m models use the full 3D Smagorinsky-Lilly scheme
The subgrid eddy-viscosity,
is a mixing length, Δ is the horizontal gridlength, and cs is a constant
Here we vary l such that it is equivalent between the different resolution models

11. Mixing length effects, 500m model
Example shower case
Example strong case, 25/08/12
500m-model run with a mixing length of 300m, 100m (standard) and 40m.
The mixing length plays a key role in determining the number of small storms.

12. Mixing length effects, 500m model
Example shower case
Example strong case, 25/08/12
By altering Smagorinsky mixing length, we can produce storm morphologies similar to a higher resolution simulation with the equivalent mixing length
i.e., the turbulent mixing formulation is a (the?) leading control on convective cell characteristics
500m-model run with a mixing length of 300m, 100m (standard) and 40m.
The mixing length plays a key role in determining the number of small storms.

13. Convective cell life cycle
Cumulative fraction of total rainfall
UKV does not get enough rainfall from short-lived storms.
200m and 100m simulations lack long-lived events (more break-ups and mergers).
200m with increased mixing length behaves like a lower resolution run 13

14. Conclusions
The highest resolutions of current NWP run at resolutions for which the dynamics produces convective clouds, but not with many grid points
UKV has not enough light rain, too much heavier rain, not enough heaviest.
UKV has not enough small cells (under-resolved) and not enough large ones
Consistent across different models in different regions
Improved to some degree by resolution
Higher resolutions give more small cells and possibly too many, so appearing fragmented
All of the above is sensitive to mixing formulation!

15. East Africa Again too few large cells and too many small.
1.5km better than 4.4km but now too fragmented (like the higher res ones in UK) cases.
Chloe Eagle

16. 21/07/10, Δx=500m better sea breeze convergence line
Rob Warren