1. Updates of convection scheme in the 5km resolution operational system
Kengo Matsubayashi
Numerical Prediction Division, Japan Meteorological Agency
with Tabito Hara, Kohei Aranami and Kohei Kawano
November 30, 2016
4th International Workshop on Nonhydrostatic Models
Hakone, Japan

2. Current operational NWP systems at JMA
NWP systems at JMA (deterministic)
Newly developed non-hydrostatic numerical model called ASUCA has been launched into operation as the 2km LFM since 2015.
JMA even plans to introduce ASUCA into the 5km grid spacing operational model (MSM) this winter.
Parameterization schemes including convection are also updated, and many improvements are made.
Global
Meso
Local
Objectives
Short- and Medium-range forecast
Disaster reduction
Short-range forecast
Aviation forecast
Disaster prevention
NWP model
Global Spectral
Model(GSM)
Meso-Scale
Model(MSM)
Local Forecast Model (LFM)
Horizontal resolution
TL959
( deg)
5 km
(817x611)
2 km
(1581x1301)
Vertical
levels / Top
100
0.01 hPa 48
21.8 km 58
20.2 km
Forecast
Hours
(Initial time)
264 hours
39 hours
(every 3 hours)
9 hours
(every hour)
Forecast model
GSAM
JMA-NHM
ASUCA
Domain
Global
Meso
Local 2 2

3. Improvement in a precipitation forecast
Observation
Radar obs. and MSLP(analysis)
New MSM
Current MSM

4. Improvement in the precipitation forecast
Threat score
Bias score
overestimated
better
Updates of physical processes improved accuracy of precipitation forecasts measured by the threat score remarkably.
In particular the convection scheme strongly affects prediction accuracy.

5. Convection parameterization4
Parameterized processes
convection initiation
updraft
entrainment
detrainment
microphysics
condensation, evaporation
freezing, melting
sedimentation
downdraft
compensational subsidence 5 7 2 3 6 1

6. Convection scheme in MSM
Based on Kain and Fritsch(1990; KF)
Overhauling the KF scheme in detail
Just entirely changing the current scheme with another one seldom enhances the accuracy because other compensational errors often arise.
It is necessary to thoroughly analyze the current scheme.
The current scheme was carefully inspected using not only 3D model but also SCM, which exposed a lot of problems.

7. Overhauling Too high convective cloud top
Large difference of heating rate between deep and shallow convection.
Too much precipitation over windward side of mountains
Less sedimentation produced in convective updraft around freezing level.
Parcel produced by entrainment below cloud base is not identical to updraft parcel at cloud base.
Double count between resolved and parameterized vertical transport.
In the tropics, heating rate is relatively weak compared to midlatitudes.
Shallow convection is triggered even in zero CAPE condition.
KF keeps heating rate constant for convectional lifetime(~30mins).
Ice and water categories KF produces

8. Too high convectional cloud top
Cloud top height(CTH) of the parameterized convection in the MSM is often too higher than that estimated by satellite observations.
[m]
[m]
[m]
CTH estimated by satellite (Himawari 8) observation
parameterized convection CTH
parameterized - observed

9. The relation between CTH and entrainment
Entrainment dilutes buoyancy and vertical velocity of updraft.
Weaker entrainment rate leads to larger Available Buoyant Energy(ABE) and higher CTH.
Too high CTH implies entrainment is too weak.
In the KF scheme, the strength of convection highly depends on ABE (CAPE closure). 𝐾 𝑧
𝜃 𝑒𝑠 : enrivonment
𝜃 𝑒
weak entrainment
large ABE
strong entrainment
small ABE
updraft parcel

10. Too high convectional cloud top
One of the causes of excessively high parameterized CTH is related to too small entrainment rate.
Entrainment rate 𝜀 𝑢 in the KF scheme
𝑅 is a constant and the only parameter to change, but not observable.
𝜀 𝑢 ∝ (−0.03𝛿𝑝) 𝑅
𝑅:updraft radius

11. Too high convectional cloud top
Histogram of the parameterized CTH – observed CTH
changed 𝑅 such that the parameterized CTH becomes consistent with observation.
Since the original 𝑅 (weak entrainment) caused too strong convection, the modified 𝑅 weaken the strength of convection. As a result dry bias over lower troposphere is also reduced.
original 𝑅(1000m)
[m]
modified 𝑅(750m)
[m]

12. Convection Parameterization(CP) updates in MSM
The updated CP scheme significantly improves quantitative precipitation forecasts and reduces dry bias over the lower troposphere.
Refinement of entrainment rate, triggering, treatment of sedimentation were critical for this improvement.
Threat score
Bias score
Qv ME
Qv RMSE
[kg/kg]
[kg/kg]

13. The importance of CP in 5km resolution
However relatively heavy precipitation is still overestimated.
With aggressively exerted CP (weaken entrainment rate over lower layer)
predicting precipitation accuracy can be further enhanced.
however dry bias becomes more serious.
Threat score
Bias score
[kg/kg]

14. The importance of CP in 5km resolution
reference
w/ aggressively exerted CP
Observation
The precipitation predicted by reference is too strong and it occurs in few grid (grid point storm).
On the other hand, aggressive CP can stabilize before grid point storm arise.
It is estimated that grid scale vertical updraft lacks processes such as entrainment which weaken vertical updraft.

15. The importance of CP in 5km resolution
Entrainment weakens convection activity by incorporating dry air outside cumuli and dilutes moist air inside, however it is too small to resolve.
Detrainment and convective initiation are also unresolved.
If stabilization by CP is weak, poorly resolved entrainment/detrainment and convective initiation lead to frequent excessively strong updrafts and too intense precipitation.

16. Convection parameterization in high resolution
In high resolution model, convectional vertical transport is partially resolved.
Convection permitting
Scales of processes such as entrainment/detrainment, convective initiation are too small to resolve.
These processes should be still parameterized.
Grey zone

17. Summary Overhauled convection scheme in MSM.
rather than entirely changing the current scheme with another one
SCM reveals a lot of problems in the KF scheme.
Exposed problems are fixed and forecast accuracy is improved.
CP is inevitable at 5km resolution NWP model.
Even in high resolution models which can resolve vertical transport, smaller unresolved phenomena around convection should be still parameterized.