Effect of Scale Coupling Frequency! on Simulated Climatology! in the Uncoupled SPCAM 3.0

Transcription

1 CMMAP Winter 215 Team Meeting Effect of Scale Coupling Frequency on Simulated Climatology in the Uncoupled SPCAM 3. Sungduk Yu and Mike Pritchard UC Irvine (Special thanks to Gabe Kooperman and Brian Mapes)

2 Most versions of SP models have not been optimally tuned DJF SWCF RMSE DJF LWCF RMSE (Pritchard, CMMAP Team Meeting, 212 August)

3 Most versions of SP models have not been optimally tuned DJF SWCF RMSE DJF LWCF RMSE (Pritchard, CMMAP Team Meeting, 212 August) SPCAM3. is surprisingly well-tuned A newer model a better tuned model

4 Tuning is increasingly important for existing and emerging SP models SP models are getting more useable: - several promising versions already exist - computational resources keep expanding - acceleration techniques are being suggested e.g. reduced CRM set up (Pritchard, Bretherton, and DeMott, 214) time scale separation (Jones, Bretherton, and Pritchard, 214) - UP (ultra-parameterization) models are under development It s unclear what are tuning knobs in SP models

5 Scale coupling frequency as a tuning parameter in SP models? Scale coupling frequency (fscale) ~ 1/(timestep of GCM) Cloud physics isolated within a GCM box for τ < 1/fscale CRM grids embedded within a single GCM grid box GCM grids Multiple CRMs feeding back with GCM dynamics for τ > 1/fscale

6 Model timestep known to be important in GCMs ω6 (15-225, 15S-15N) Precip (15-225, 15S-15N) DT = 5 min DT = 5 min (Williamson, 213) mm/day grid point storms CAM4 T34 resolution dtime: 5, 2, 4 min

7 Model timestep known to be important in GCMs ω6 (15-225, 15S-15N) Precip (15-225, 15S-15N) DT = 5 min DT = 5 min (Williamson, 213) mm/day grid point storms GCM Δt >> τ of deep/shallow convective parameterizations (5min) (6min and 3min) ==> excessive precipitation (e.g. grid point storms)

8 Model timestep known to be important in GCMs ω6 (15-225, 15S-15N) Precip (15-225, 15S-15N) DT = 5 min DT = 5 min What about in SP models? (Williamson, 213) mm/day grid point storms GCM Δt >> τ of deep/shallow convective parameterizations (5min) (6min and 3min) ==> excessive precipitation (e.g. grid point storms)

9 Simulation setup Model: SPCAM3. CRM setup: Micro-CRM (4km x 8) Control simulation: Experiment simulation: Simulation length: SST: dtime = 18 [s] dtime = 6, 9, 36 [s] 1 years prescribed

10 Striking quasi-linear thermal and SWCF responses to increased scale coupling frequency 2S-2N 5E-2E temperature bias T dtime6 ERAI dtime9 ERAI dtime18 ERAI dtime36 ERAI Mid-troposphere warming K Zonal mean cloud forcing low fscale (dtime36) high fscale (dtime6) LWCF W/m 2 net SWCF weakening LWCF weakening 5 1 SWCF dtime6 dtime9 dtime18 dtime36 CERES EBAF Latitude (degn)

11 Reversing key biases introduced by reduced CRM domain CRM setup [Pritchard, Bretherton, and DeMott (214)] Smaller CRM (acceleration) 32x1 16x1 8x1 W/m 2 LWCF net? SWCF Latitude (degn)

12 Reversing key biases introduced by reduced CRM domain 2S-2N 5E-2E temperature bias T dtime6 ERAI dtime9 ERAI dtime18 ERAI dtime36 ERAI [Pritchard, Bretherton, and DeMott (214)] K 5 Zonal mean cloud forcing LWCF LWCF? W/m 2 net W/m 2 net 5 1 SWCF dtime6 dtime9 dtime18 dtime36 CERES EBAF Latitude (degn) SWCF Latitude (degn)

13 Liquid clouds systematically become less dense and less bright as scale coupling frequency increases SWCF [W/m 2 ] cloud water [g/m 2 ] dtime18 (control) dtime6 - control dtime9 - control dtime36 - control

14 Liquid clouds systematically become less dense and less bright as scale coupling frequency increases SWCF [W/m 2 ] cloud water [g/m 2 ] dtime18 (control) dtime6 - control dtime9 - control dtime36 - control geographically robust, quasi-linear response strong response along deep convective zones > convective mixing efficiency affected

15 Liquid clouds systematically become less dense and less bright as scale coupling frequency increases dtime18 (control) dtime6 - control dtime9 - control dtime36 - control SWCF [W/m 2 ] cloud water [g/m 2 ] precipitation-binned cloud water relative frequency 2 4 Cloud water from CRM (g/kg): dtime18 6 Cloud water from CRM (g/kg): dtime Cloud water from CRM (g/kg): dtime dtime6 control dtime6 control dtime6 control dtime6 control dtime9 control dtime9 control dtime9 control dtime9 control dtime36 control dtime36 control dtime36 control dtime36 control

16 High clouds reduce with scale coupling frequency but this response is weaker and more complex LWCF [W/m 2 ] cloud ice [g/m 2 ] dtime18 (control) dtime6 - control dtime9 - control systematic more complex, not know why yet 1/2 weaker than SWCF response dtime36 - control

17 High clouds reduce with scale coupling frequency but this response is weaker and more complex LWCF [W/m 2 ] cloud ice [g/m 2 ] precipitation-binned cloud ice relative frequency dtime18 (control) dtime6 - control dtime9 - control dtime36 - control Cloud ice from CRM (g/kg): dtime dtime6 control Cloud ice from CRM (g/kg): dtime18 dtime6 control dtime6 control dtime6 control dtime9 control dtime9 control dtime9 control dtime36 control dtime36 control dtime36 control

18 SWCF response seems consistent with reduced CRM throttling Pritchard, Bretherton, and DeMott (214) s hypothesis: Artificially throttled deep convection by trapped subsidence Reduced CRM domain -> stronger subsidence -> preventing ventilation -> too much liquid cloud -> too strong SWCF

19 SWCF response seems consistent with reduced CRM throttling Pritchard, Bretherton, and DeMott (214) s hypothesis: Artificially throttled deep convection by trapped subsidence CRM is not a closed system This artifact is corrected by GCM s large scale dynamics More frequent scale coupling more ventilation less liquid cloud

20 LWCF response is new and mysterious LWCF response 5 scale coupling frequency experiment Zonal mean cloud forcing CRM domain size experiment LWCF LWCF? W/m 2 net W/m 2 net 5 1 SWCF dtime6 dtime9 dtime18 dtime36 CERES EBAF Student Version of MATLAB Latitude (degn) SWCF Latitude (degn)

21 Similar to GCMs, rain intensity tail amplifies Amount distribution of precipitation 2S 2N 3N 9N mm day GPCP dtime6 dtime9 dtime18 dtime36 mm day GPCP dtime6 dtime9 dtime18 dtime mm day mm day 1 low fscale (dtime36) high fscale (dtime6)

22 Similar to GCMs, rain intensity tail amplifies Amount distribution of precipitation 2S 2N 3N 9N mm day GPCP dtime6 dtime9 dtime18 dtime36 mm day GPCP dtime6 dtime9 dtime18 dtime mm day mm day 1 Rain intensity tail amplifies as fscale increases. (not expected with the trapped subsidence hypothesis) Rain intensity tail amplification is mostly from the tropics

23 Similar to GCMs, rain intensity tail amplifies Precip (15-225, 15S-15N) 2S 2N DT = 5 min mm day GPCP dtime6 dtime9 dtime18 dtime mm day 1 Student Version of MATLAB mm/day SPCAM3 CAM4 (Williamson, 213)

24 No single mode of tropical organization dominates the rain intensity change mean power shifted to higher frequency

25 No single mode of tropical organization dominates the rain intensity change mean power shifted to higher frequency moist Kelvin wave strengthen

26 Scale coupling frequency is an interesting tuning parameter for superparameterized models (with caution) 5 Zonal mean cloud forcing SWCF and LWCF biases decrease LWCF as fscale increases W/m 2 net 5 1 SWCF dtime6 dtime9 dtime18 dtime36 CERES EBAF Latitude (degn) Tropical Precipitation (2S-2N) Tropical precipitation tail boosts as fscale increase mm day GPCP dtime6 dtime9 dtime18 dtime mm day 1 Student Version of MATLAB

27 Appendix A: SPCAM3 vs. CAM3 SPCAM3., T42, Δt=1, 15, 3, 6 min 5 Zonal mean cloud forcing CAM3, T63, Δt=6, 2, 5 min (Mishra and Sahany, 211) LWCF W/m 2 net 5 1 SWCF dtime6 dtime9 dtime18 dtime36 CERES EBAF Latitude (degn) Higher fscale -> weaker SWCF -> weaker LWCF Higher fscale (lower Δt) -> stronger SWCF -> stronger SWCF opposite sensitivity

28 Appendix B: CRM mass flux statistics (Scale coupling frequency experiment) (CRM domain experiment by Pritchard et al, 214) Something more than throttled deep convection

29 Appendix C: Precipitation std dev