Forecasts of Southeast Pacific Stratocumulus with the NCAR, GFDL and ECMWF models.

1. Forecasts of Southeast Pacific Stratocumulus with the NCAR, GFDL and ECMWF models. Cécile Hannay(1), Dave Williamson(1), Jim Hack(1), Jeff Kiehl (1), Jerry Olson(1), Chris Bretherton (2), Steve Klein (3) and Martin Köhler (4). (1): National Center for Atmospheric Research, Boulder, Colorado(2): Department of Atmospheric Science, University of Washington, Seattle, Washington.(3): Program for Climate Model Diagnosis and Intercomparison, Lawrence Livermore National Laboratory, California (4): European Centre for Medium-Range Weather Forecast, Reading, England 2. The Eastern Pacific Investigation of Climate (EPIC) column Motivation We illustrate the way the NCAR, GFDL and ECMWF models represent regions of persistent stratocumulus with global forecast simulations examined at a column in the South Eastern Pacific (20S-85W). Stratocumulus clouds play an important role in the seasonal cycle of the Eastern Pacific and the global climate by exerting a strong cooling effect on the surface. These clouds are very complex to parameterize in GCMs because :- they are only a few hundred meters thick.Therefore, they are difficult to represent with the current climate model vertical resolution. - they are maintained by a complex set of interactions between the cloud layer and its environment, which are not always well understood (Figure 1). • This location (Figure 2) has been chosen because of the availability of observational datasets and accurate analyses. • the WHOI buoy provides a long-term time-series of surface meteorological variables. • the 2001 EPIC cruise provides a comprehensive dataset of remote sensing and surface measurements for Oct 16-21, 2001. • the MK ECMWF analysis (which uses a combined Mass-flux/K-diffusion scheme) provides a realistic state of the EPIC column except the boundary layer height is too shallow (Figures 3-4). EPIC * Potential temperature subsidence Inversion jump LW cooling Entrainement of dry air Stratocumulus Figure 2: The EPIC column (20S-85W) is located in a region of persistent stratocumulus off the coast of South America. Annual low-level clouds from ISCCP. Buoyancy flux BL height sfc Figure 1: Some processes controlling stratocumulus. Figure 3: Time-height cross-section of potential temperature at the EPIC column during the 2001 EPIC cruise: Radiosondes versus MK ECMWF analysis. Observations show a very stable PBL under a sharp inversion. One of the problem of the MK ECMWF analysis is that it underestimates the height of the boundary layer and the strength of the inversion. Figure 4: Liquid water path and water vapor mixing ratio during the 2001 EPIC cruise: ECMWF short-term forecasts versus observations. The MK ECMWF forecasts (moist PBL) and an ECWMF forecasts with a dry PBL are compared to observations. The MK ECMWF forecasts represents the liquid water path and its diurnal cycle fairly well . Specific humidity (g/kg) Vertical model grid Figure 10: Modeled specific humidity versus observations. The modeled specific humidity is averaged over the entire 5-day forecasts and then, averaged over all forecasts (October 11-22). The observations are averaged over the period October 16-22. The NCAR and NCAR-UW produces more shallow boundary layers than the ECMWF and GFDL models. The GFDL model inversion is less pronounced than the ECMWF one. Note: doubling the vertical resolution in the NCAR model does not change the moisture profile • 5. Mechanisms controlling the PBL height. • A key problem common to the 4 models is that the forecasted PBL height is too low compared to observations (Figure 10). In the section 4, we presented the results of a single 5-day forecast starting on October 16. However, the ensemble of forecasts (starting on October 11-22) reveals various features. • The ECMWF model shows a steady PBL with no significant decrease or increase of the inversion height. • The NCAR and NCAR-UW forecasts show 2 typical behaviors: either the PBL is maintained or it collapses, while the observations during the same period do not show any shallowing of the PBL. Climate runs with the NCAR model also show an inability to maintain the proper PBL depth. • The GFDL forecasts have periods during which the boundary layer becomes shallower. The collapse is less dramatic than in the NCAR forecasts but it is more pronounced than in the ECMWF ones. Role of omega in the collapse of the PBL SCM simulations with the NCAR model shows that the vertical pressure velocity is one of the determinant factors in the maintaining/shallowing of the PBL height (Figure 11). Vertical pressure velocity Specific humidity Figure 11: - Left panel shows the mean vertical velocity for the ECMWF and NCAR models (5-day forecast starting on Oct 12). The NCAR omega is about twice as large as the ECMWF omega below 900 mb. The mean value is 15 mb/day (r. 30mb/day) for the ECMWF (r. NCAR) model. SCM experiments with the NCAR model show that the PBL is maintained when using the ECMWF omega and it collapses when using the NCAR omega . - Middle panel shows the evolution of the specific humidity (g/kg) in a SCM experiment using ECMWF omega. - Right panel shows the evolution of the specific humidity (g/kg) in a SCM experiment using NCAR omega. Correlation between PBLH and LHFLX The NCAR model decently represents the latent heat flux. However, the correlation between latent heat flux and boundary layer height in the NCAR model is stronger than in the observations (Figure 12). Figure 12: Left panel: Latent heat flux averaged over the forecasts. Right panel: Scatter plots of the PBL height versus the latent heat flux. The correlation between the PBL height is stronger in the NCAR model than in observations.

2. Specific humidity (g/kg) Cloud fraction Figure 6: Time-height cross-section of the specific humidity. The observations show a very stable boundary layer with strong diurnal cycle of the PBL depth. All the models underestimate the height of the PBL and overestimate the moisture inside the boundary layer. The NCAR and NCAR-UW PBL tend to create a more shallow PBL than the ECMWF and GFDL models (see also Figure 10). The NCAR-UW better represents the diurnal cycle of the PBL due to the explicit entrainment of dry air at the top of the PBL. Figure 7: Time-height cross-section of cloud fraction. The observed cloud fraction is deduced from the cloud top (radar) and the cloud base (ceilometer). The observed cloud fraction shows a strong diurnal cycle with a minimum around 12 UTC (i.e. 18 local time). - The ECMWF model produces a thick and constant layer of clouds. - The NCAR model produces an unrealistically thick layer of clouds that sometimes extends to the surface. Note that it produces some ‘empty’ clouds (clouds with very low or no liquid water content). - The NCAR-UW and GFDL clouds are more realistic in the sense they form on a single level and show a diurnal cycle. Strategy If the model is initialized realistically, we assume the error comes from the parameterizations deficiencies. Advantages Full feedback <=> SCM Limitations Accuracy of the atmospheric state ? Initialize realistically Operational MK ECMWF analysis MODEL 5-day forecast Starting daily at 12 UT EPIC 2001 cruise WHOI buoy Figure 5: Forecast runs framework 4. Forecasts for the EPIC Period We initialize the models (NCAR, NCAR-UW, ECMWF and GFDL) with the ECMWF MK analyses for every day of the period October 11-22, 2001. For each initialization, we run the model for 5 days obtaining an ensemble of forecasts with various features. As an illustration, we present the forecasts starting on October 16 at 12 UTC and we compare the 5-day forecasts with observations (Figure 6-9). 3. Models and forecast framework • ModelsWe use the following models: • the ECWMF model with the MK PBL scheme (a combined Mass-flux/K-diffusion scheme) • the GFDL model • the NCAR model: which uses Holtslag-Boville (1993) for the boundary layer and Hack (1994) for the shallow convection. • the NCAR-UW model: similar to the NCAR model but with different parameterizations of PBL and shallow cumulus. It uses the turbulence scheme of Grenier-Bretherton (2001) which includes explicit entrainment at the top of the PBL coupled with a shallow cumulus scheme which includes the determination of cloud-base mass flux based on surface layer turbulent kinetic energy (TKE) and convective inhibition near the cloud base. Forecast frameworkIn the CAPT protocol, we realistically initialize the model with analyses and we then run the model in forecast mode to determine the drift from the analyses and/or available field data. This method allows us to diagnose model parameterization deficiencies (Figure 5). • Figure 8: 5-day forecasts of the liquid water path, latent and sensible heat fluxes and static stability. - The liquid water and its diurnal cycle are represented very well in the ECWMF forecasts. In particular, the daytime LWP is well captured while the other models unrealistically collapse the LWP during daytime, especially the GFDL and the NCAR-UW model. This affects the solar flux at the surface. • The models tend to overestimate the amplitude diurnal cycle of the surface fluxes. Notice that the NCAR latent heat flux decreases in the 2 last days of the forecast as the NCAR PBL becomes more shallow (See also Figure 12). • The static stability of the atmosphere is defined as Theta800-Theta1000. It is well represented in the models. Figure 9: Vertical velocity (mb/day).