MotivationsWhat is Superparameterization (SP)?Superparameterized GCM is a type of GCM that the conventional cloud parameterizations are replaced by cloud resolving models (CRMs) in order to reduce uncertainties from those statistical subgrid schemes.

How do GCM and CRM communicate in SP?The current SP scaffold allows GCM and CRM to communicate only at each GCM time step. Sale coupling frequency between GCM and CRM increases with decreasing GCM time step.

Scale coupling frequency (fscale) is an important parameter that controls GCM − CRM communication frequency, but its effect is hardly knownBetter understanding on fscale would tell us its potential as a tuning parameter and provide useful insight for future SP model development.

So, we explored the effect of fscale in a SPGCM! Our major findings include:

1. fscale monotonically impacts climate. With a higher fscale,• Shortwave and longwave cloud forcing biases lessen.• Tropical rainfall extreme becomes more frequent.

2. Convective organization changes with fscale, and it seems to be the main cause of the climate sensitivities.

Convection becomes bottom-heavy ➣ fscale appears to affect the net

updraft mass flux profile of convection rather than simply boosting mass flux at all level, promoting more bottom-heavy convection with a high fscale.

Shortwave and Longwave cloud forcing biases decreases

Effects of Scale Coupling Frequency on Convective Organization in the Superparameterized CAM 3.0Sungduk Yu (sungduk@uci.edu) and Mike Pritchard, Department of Earth System Science, University of California, Irvine, CA

Cloud physics isolated within a GCM box for 𝜏 < 1/fscale

Multiple CRMs feeding back with GCM dynamics for 𝜏 > 1/fscale

CRM grids embedded withina single GCM grid box

GCM grids

[Illustration: UCAR]

MethodsSPCAM Version: 3.0 CRM setup: 1x8 columns with 4km horizontal grid size Control simulation: dtime (GCM timestep, ~ 1/fscale) = 1800 [s]Experiment simulation: dtime = 600, 900, 3600 [s]Simulation length: 10 years with 4 months of spin-upBoundary conditions: prescribed monthly SST

Poster#A51F-0133

1. fscale monotonically impacts simulated climate

deg N-60 -45 -30 -15 0 15 30 45 60

W/m

2

-100

-50

0

50

dtime600dtime900dtime1800dtime3600CERES-EBAFShortwave cloud forcing

Longwave cloud forcing

Net

(Direction of arrow is from low to high fscale)

ii) dtime600 - control

iii) dtime900 - control

iv) dtime3600 - control

-47.2 -23.6 0 23.6 47.2

i) control (dtime1800)

0 87 174 261.1 348.1

b) Liquid water path [g/m2]

ii) dtime600 - control

iii) dtime900 - control

iv) dtime3600 - control

-20.5 -10.3 0 10.3 20.5

i) control (dtime1800)

-172.9 -129.7 -86.5 -43.3 -0

a) Shortwave cloud forcing [W/m2]

ii) dtime600 - control

iii) dtime900 - control

iv) dtime3600 - control

-47.2 -23.6 0 23.6 47.2

i) control (dtime1800)

0 87 174 261.1 348.1

b) Liquid water path [g/m2]

ii) dtime600 - control

iii) dtime900 - control

iv) dtime3600 - control

-20.5 -10.3 0 10.3 20.5

i) control (dtime1800)

-172.9 -129.7 -86.5 -43.3 -0

a) Shortwave cloud forcing [W/m2]

ii) dtime600 - control

iii) dtime900 - control

iv) dtime3600 - control

-10 -5 0 5 10

i) control (dtime1800)

-2.1 16.8 35.7 54.6 73.5

a) Longwave cloud forcing [W/m2]

ii) dtime600 - control

iii) dtime900 - control

iv) dtime3600 - control

-23.6 -11.8 0 11.8 23.6

i) control (dtime1800)

2 35.6 69.2 102.8 136.5

b) Ice water path [g/m2]

ii) dtime600 - control

iii) dtime900 - control

iv) dtime3600 - control

-10 -5 0 5 10

i) control (dtime1800)

-2.1 16.8 35.7 54.6 73.5

a) Longwave cloud forcing [W/m2]

ii) dtime600 - control

iii) dtime900 - control

iv) dtime3600 - control

-23.6 -11.8 0 11.8 23.6

i) control (dtime1800)

2 35.6 69.2 102.8 136.5

b) Ice water path [g/m2]SWCF [W/m2] Cloud water [g/m2] LWCF [W/m2] Cloud ice [g/m2]

Zonal-Mean mean cloud forcing

➣ Liquid clouds systematically become less dense and less bright as fscale increases.

➣ High clouds reduce with fscale but this response is weaker and more complex.

2. Convective organization changes seem to cause the fscale sensitivity

kg/m 2 /s #10 -40 5 10 15

hPa

100

200

300

400

500

600

700

800

900

1000

a) This study

dtime600dtime900dtime1800dtime3600

kg/m 2 /s #10 -30 1 2 3 4

hPa

100

200

300

400

500

600

700

800

900

1000

b) PBD14

32x116x18x1

GMS decreases in active convection areas

➣ GMS (following Raymond et al. 2009, JAMES) decreases with fscale in convective regions, as expect from more bottom heavy convection.

➣ Reduced GMS enhances net precipitation efficiency to a given diabetic forcing and may link to the rain distribution shift toward extreme and possibly to cloud water and ice.

dtime (s)600

9001800

3600

GM

S

0.4

0.6

0.8e) Regional mean GMS

AB

deg

N

-20

0

20

AB

a) control (dtime1800)

-2 -1 0 1 2

deg

N

-20

0

20

AB

b) dtime600 - control

deg

N

-20

0

20

AB

c) dtime900 - control

deg E0 100 200 300

deg

N

-20

0

20

AB

d) dtime3600 - control

-0.2 -0.1 0 0.1 0.2dtime (s)

600900

18003600

GM

S

0.4

0.6

0.8e) Regional mean GMS

AB

deg

N

-20

0

20

AB

a) control (dtime1800)

-2 -1 0 1 2

deg

N

-20

0

20

AB

b) dtime600 - control

deg

N

-20

0

20

AB

c) dtime900 - control

deg E0 100 200 300

deg

N

-20

0

20

AB

d) dtime3600 - control

-0.2 -0.1 0 0.1 0.2

Convective organization hypothesis

It seems possible to explain a broad set of simulated climate responses to fscale as the result of an overarching change in convective organization favoring more bottom-heavy convection, reduced gross moist stability, and ultimately enhanced precipitation efficiency at a high fscale. In addition, the systematic responses of cloud water and ice—accordingly, SWCF and LWCF—could be viewed as stemming from changes in precipitation efficiency.

3. Weak-temperature-gradient conforms better with a high fscale

Better WTG conformity with high fscale

➣ Increasing fscale reduces T300′ such that SPCAM3’s behavior becomes more WTG-like.

Frequent dynamical adjustment toT anomalies helps WTG conformity

➣ Local convection confined in an embedded CRM can build up to produce larger thermal anomalies.

➣ These local heat anomalies (e.g., T′>0) are then spread to adjacent GCM grid columns via dynamical adjustment.

➣ In SPGCMs, dynamical adjustment is limited by fscale, indicating a high fscale can reduce T′.

a) dtime600

-0.4-0.3-0.2-0.100.1

hPa

100

300

500

700

900

b) dtime900

-0.4-0.3-0.2-0.100.1

hPa

100

300

500

700

900

c) dtime1800

-0.4-0.3-0.2-0.100.1

hPa

100

300

500

700

900

d) dtime3600

!500 (Pa/s)-0.4-0.3-0.2-0.100.1

hPa

100

300

500

700

900

K-1.2 -0.8 -0.4 0 0.4 0.8 1.2

[Left] Daily horizontal-mean anomalies of temperature from its horizontal field at 300 hPa (T300′) across vertical velocity at 500 hPa (ω500) in equatorial region (5°S–5°N). [Right] Relative frequency of vertical velocity at 500 hPa.

Large waves dominates T anomalies

➣ Planetary-scale disturbances, i.e., small zonal wavenumbers, dominate the sensitivity to fscale of the covariance between T′ and ω.

zonal wavenumber, k1 4 8 12 16

T·P

a/s/

k

-0.12

-0.1

-0.08

-0.06

-0.04

-0.02

0

dtime600dtime900dtime1800dtime3600

Vertically resolved profiles of temperature anomalies from its horizontal mean, binned by vertical velocity at 500 hPa (ω500) in equatorial region (5°S–5°N).

Cospectrum of daily T anomalies at 300 hPa and ω at 500 hPa from their horizontal field in equatorial region (5°S–5°N)

Take-home Points1. fscale impacts simulated climate monotonically. With a high fscale,• Both shortwave and longwave cloud forcings decrease. • Tropical precipitation tail amplifies.

2. fscale also impacts the organization of tropical convection. With a high fscale,• Convection becomes more bottom-heavy, and accordingly GMS decreases. • WTG conforms better due to dynamical adjustment to thermal anomalies.

3. fscale can be a useful tuning parameter for SP models and provide some insights for future SP model development.

Net updraft mass flux

Spectral power shifts to higher frequencies

➣ No single mode of equatorial wave variability dominates rain intensity change.

➣ Daily mean power is shifted to higher frequencies at all zonal wavelengths.

zonal wavenumber-10 0 10

frequ

ency

(cpd

)

0

0.1

0.2

0.3

0.4

0.5a) dtime600

zonal wavenumber-10 0 10

0

0.1

0.2

0.3

0.4

0.5b) dtime900

zonal wavenumber-10 0 10

0

0.1

0.2

0.3

0.4

0.5c) dtime1800 (ctrl)

zonal wavenumber-10 0 10

0

0.1

0.2

0.3

0.4

0.5d) dtime3600

-18.1

-17.9

-17.7

-17.5

-17.2

-17

-16.8

-16.6

-16.3

-16.1

e) dtime600/ctrl

zonal wavenumber-10 0 10

frequ

ency

(cpd

)

0

0.1

0.2

0.3

0.4

0.5f) dtime900/ctrl

zonal wavenumber-10 0 10

0

0.1

0.2

0.3

0.4

0.5g) dtime3600/ctrl

zonal wavenumber-10 0 10

0

0.1

0.2

0.3

0.4

0.5

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

5S-5N precipitation wavenumber-frequency spectrum (sym)

(a–d) Raw log power spectra and (e–g) the ratio of log power of experimental simulation to control simulation. Gray contour lines in a, b, and d are from control simulation

The profiles of the CRM updraft mass flux (solid lines) and their saturated moist components (dashed lines)

(a) Control simulation, (b–d) experimental simulation anomalies against control simulation, and (e) horizontally averaged GMS in two subregion A and B. Magenta line shows the contour of GMS of 0.1.

Normalized GMS

Tropical rainfall extremes increase

➣ Rain intensity tail amplifies as fscale increases.

➣ Rain intensity tail response is mostly from the tropics.

mm day -10.1 1 10 100

mm

day

-1

0

0.4

0.8

1.2

1.6

a) Precipitation amount (20S-20N)

dtime600dtime900dtime1800dtime3600GPCPTRMM

mm day -10.1 1 10 100

%

10 - 3

10 - 2

10 - 1

10 0

10 1

c) Precipitation frequency (20S-20N)

mm day -10.1 1 10 100

mm

day

-1

0

0.2

0.4

0.6

0.8

b) Precipitation amount (30N-90N)

dtime600dtime900dtime1800dtime3600GPCP

mm day -10.1 1 10 100

%

10 - 3

10 - 2

10 - 1

10 0

10 1

d) Precipitation frequency (30N-90N)

Daily precipitation distribution

(a, b) Amount and (c, d) frequency distributions of daily mean precipitation

-0.4-0.3-0.2-0.100.1

K

-0.4

0

0.4

0.8

1.2

1.6a) T300'

!500 (Pa/s)-0.4-0.3-0.2-0.100.1

0

0.05

0.1

0.15

b) Daily !500 relative frequency

dtime600dtime900dtime1800dtime3600

-0.4-0.3-0.2-0.100.1

K

-0.4

0

0.4

0.8

1.2

1.6a) T300'

!500 (Pa/s)-0.4-0.3-0.2-0.100.1

0

0.05

0.1

0.15

b) Daily !500 relative frequency

dtime600dtime900dtime1800dtime3600

-0.4-0.3-0.2-0.100.1

K

-0.4

0

0.4

0.8

1.2

1.6a) T300'

!500 (Pa/s)-0.4-0.3-0.2-0.100.1

0

0.05

0.1

0.15

b) Daily !500 relative frequency

dtime600dtime900dtime1800dtime3600

T300′ vs. ω500 ω500 frequency

T300′ − ω500 cospec

T′ vs. ω500

Just published!

For further details, refer to: Yu and Pritchard (2015), JAMES. (doi: 10.1002/2015MS000493)

Acknowledgement: We would like to thank Stefan Tulich, David Randall, and Robert Pincus for helpful comments. We would especially like to thank Brian Mapes for his original framing idea and Chris Bretherton for much insightful feedback. Funding for this study was provided by the Department of Energy under DE-SC0012152 and DE-SC0012548 and by the National Science Foundation (NSF) under AGS-1419518. Computational resources were provided by the Extreme Science and Engineering Discovery Environment, which is supported by NSF grant OCI- 1053575, under allocation TG-ATM120034.

[Top] (i) Control simulation and (ii–iv) experiment simulation anomalies against control simulation.[Left] Zonal mean annual mean cloud forcing in SPCAM3 (thin lines) and observation (thick lines).