Clouds and their turbulent environment Robin Hogan, Andrew Barrett, Natalie Harvey Helen Dacre, Richard Forbes (ECMWF) Department of Meteorology, University of Reading

Overview Part 1: Why can’t models simulate mixed-phase altocumulus clouds? These clouds are potentially a key negative feedback for climate Getting these clouds right requires the correct specification of turbulent mixing, radiation, microphysics and sub-grid distribution We use a 1D model and long-term cloud radar and lidar observations Part 2: Can models simulate boundary-layer type, and hence the associated mixing and clouds? Important for pollution transport and evolution of weather systems We use long-term Doppler lidar observations to evaluate the scheme in the Met Office model Abstract: This talk will tackle two aspects of evaluating and improving the representation of clouds in weather and climate models using observational data, both of which involve the challenging problem of understanding the interaction between clouds and their turbulent environment. The first part will address the problem of why virtually all models grossly underestimate the occurrence of mid-level altocumulus clouds, which represent an important uncertainty in climate prediction. These clouds are difficult to simulate because are often much thinner than the vertical resolution of the model, and it is necessary to correctly represent the processes of longwave cloud-top cooling, turbulent mixing and the exchange of water between the ice, liquid and vapour phases.   We have developed a flexible 1D model that can address this problem by running it at a cloud radar and lidar site to see which parameterizations lead to the best agreement with observations.  Two key bits of physics are needed to improve the occurrence of these clouds: firstly, a new parameterization for the sub-grid vertical distribution of thermodynamic quantities enables thin liquid water layers to be represented correctly in models with a coarse vertical resolution. Secondly, we show that an improved representation of the ice particle size distribution leading on from the work of Delanoe and Hogan (2008) provides a much better prediction of the rate at which ice crystals grow and fall out of the cloud. The second part of the talk will demonstrate how ground-based Doppler lidar can be used to classify the boundary layer into nine different types (e.g. stratus-topped stable layer, decoupled stratocumulus, cumulus-capped etc). It is demonstrated that Doppler lidar diagnostics such as vertical velocity variance and skewness are able to diagnose whether boundary-layer turbulence is being generated from surface heating or cloud-top radiative cooling, which is key to distinguishing between types such as cumulus clouds and broken stratocumulus. A two-year dataset is then used to evaluate the predictions of the Met Office model, one of many models that diagnose a boundary-layer type in order to decide which mixing scheme to use, which is important for the transport of pollution and the evolution of weather systems.

Mixed-phase altocumulus clouds Small supercooled liquid cloud droplets Low fall speed Highly reflective to sunlight Often in layers only 100-200 m thick Large ice particles High fall speed Much less reflective for a given water content

Mixed-phase cloud radiative feedback Change to cloud mixing ratio on doubling of CO2 Tsushima et al. (2006) Increase in polar boundary-layer and mid-latitude mid-level clouds Clouds more likely to be liquid phase: lower fall speed and more persistent Higher albedo -> negative feedback But this result depends on questionable model physics! Decrease in subtropical stratocumulus Lower albedo -> positive feedback on climate

Important processes in altocumulus Longwave cloud-top cooling Supercooled droplets form Cooling induces upside-down convective mixing Some droplets freeze Ice particles grow at expense of liquid by Bergeron-Findeisen mechanism Ice particles fall out of layer Many models have prognostic cloud water content, and temperature-dependent ice/liquid split, with less liquid at colder temperatures Impossible to represent altocumulus clouds properly! Newer models have separate prognostic ice and liquid mixing ratios Are they better at mixed-phase clouds?

How well do models get mixed-phase clouds? Ground-based radar and lidar (Illingworth, Hogan et al. 2007) CloudSat and Calipso (Hogan, Stein, Garcon and Delanoe, in preparation) Models typically miss a third of mid-level clouds This is cloud fraction – what about cloud water content?

Observations of long-lived liquid layer Radar reflectivity (large particles) Lidar backscatter (small particles) Radar Doppler velocity Liquid at –20C

Cloudnet processing Illingworth, Hogan et al. (BAMS 2007) Use radar, lidar and microwave radiometer to estimate ice and liquid water content on model grid

21 altocumulus days at Chilbolton Met Office models (mesoscale and global) have most sophisticated cloud scheme Separate prognostic liquid and ice But these models have the worst supercooled liquid water content and liquid cloud fraction What are we doing wrong in these schemes?

1D “EMPIRE” model Single column model High vertical resolution Default: Dz = 50m Five prognostic variables u, v, θl, qt and qi Default: follows Met Office model Wilson & Ballard microphysics Local and non-local mixing Explicit cloud-top entrainment Frequent radiation updates (Edwards & Slingo scheme) Advective forcing using ERA-Interim Flexible: very easy to try different parameterization schemes Coded in matlab Each configuration compared to set of 21 Chilbolton altocumulus days Variables conserved under moist adiabatic processes: Total water (vapour plus liquid): Liquid water potential temperature

EMPIRE model simulations

Evaluation of EMPIRE control model More supercooled liquid than Met Office but still seriously underestimated

Effect of turbulent mixing scheme Quite a small effect!

Effect of vertical resolution Take EMPIRE and change physical processes within bounds of parameterized uncertainty Assess change in simulated mixed-phase clouds Significantly less liquid at 500-m resolution Explains poorer performance of Met Office model Thin liquid layers cannot be resolved

Effect of ice growth rate Liquid water distribution improves in response to any change that reduces the ice growth rate in the cloud Change could be: reduced ice number concentration, increased ice fall speed, reduced ice capacitance But which change is physically justifiable?

Summary of sensitivity tests Main model sensitivities appear to be: Ice cloud fraction In most models this is a function of ice mixing ratio and temperature We have found from Cloudnet observations that the temperature dependence is unnecessary, and that this significantly improves the ice cloud fraction in clouds warmer than –30C (not shown) Vertical resolution Can we parameterize the sub-grid vertical distribution to get the same result in the high and low resolution models? Ice growth rate Is there something wrong with the size distribution assumed in models that causes too high an ice growth rate when the ice water content is small?

Resolution dependence: idealised simulation Liquid Ice

Resolution dependence Best NWP resolution Typical NWP resolution

Effect 1: thin clouds can be missed Consider a 500-m model level at the top of an altocumulus cloud Consider prognostic variables ql and qt that lead to ql = 0 θl qt ql T P1 But layer is well mixed which means that even though prognostic variables are constant with height, T decreases significantly in layer Therefore a liquid cloud may still be present at the top of the layer Gridbox-mean liquid can be parameterized P2

Effect 2: Ice growth too high at cloud top Diffusional growth: qi = ice mixing ratio, ice diameter RHi = relative humidity with respect to ice qi zero at cloud top: growth too high dqi RHi qi dt P1 Assume linear qi profile to enable gridbox-mean growth rate to be estimated: significantly lower than before P2 100%

Parameterization at work Liquid Ice Liquid Ice

Parameterization at work New parameterization works well over full range of model resolutions Typically applied only at cloud top, which can be identified objectively

Standard ice particle size distribution log(N) Inverse exponential fit used in all situations Simply adjust slope to match ice water content Wilson and Ballard scheme used by Met Office Similar schemes in many other models N0 = 2x106 Increasing ice water content D But how does calculated growth rate versus ice water content compare to calculations from aircraft spectra?

Parameterized growth rates log(N) Ice growth rate D Ratio of parameterization to aircraft spectra N0 = constant Ice clouds with low water content: Ice growth rate too high Fall speed too low Liquid clouds depleted too quickly! Fall speed Ice water content

Adjusted growth rates log(N) New ice size distribution leads to better agreement in liquid water content Ice growth rate D Ratio of parameterization to aircraft spectra N0 ~ IWC3/4 Delanoe and Hogan (2008) result suggests N0 smaller for low water content Much better agreement for growth rate and fall speed Fall speed Ice water content

Mixed-phase clouds: summary Mixed-phase clouds drastically underestimated in climate models, particularly those that have the most sophisticated physics! Very difficult to simulate persistent supercooled layers Experiments with a 1D model evaluated against observations show: Strong resolution dependence near cloud top; can be parameterized to allow liquid layers that only partially fill the layer vertically More realistic ice size distribution has fewer, larger crystals at cloud top: lower ice growth and faster fall speeds so liquid depleted more slowly Many other experiments have examined importance of radiation, turbulence, fall speed etc. Next step: apply new parameterizations in a climate model What is the new estimate of the cloud radiative feedback?

Part 2 Boundary layer type from Doppler lidar Turbulent mixing in the boundary layer transports: Pollutants away from surface: important for health Water: important for cloud formation, and hence climate and weather forecasting Heat and momentum: important for evolution of weather systems Mixing represented in four ways in models: Local mixing (shear-driven mixing) Non-local mixing (buoyancy-driven with strong capping inversion) Convection (buoyancy-driven without strong capping inversion) Entrainment (exchange across tops of stratocumulus clouds) Models must diagnose boundary-layer type to decide scheme to use Getting the clouds right is a key part of this diagnosis Doppler lidar can measure many important boundary layer properties Can we objectively diagnose boundary-layer type?

How is the boundary layer modelled? Met Office model has explicit boundary-layer types (Lock et al. 2000) Unstable profile: Non-local scheme Buoyancy-driven mixing: diffusivity profile determined by parcel ascents/descents Stable profile: Local mixing scheme Shear-driven mixing only: diffusivity K is a function of local Richardson number Ri Shallow cumulus scheme If cumulus present, mixing determined by mass-flux scheme Entrainment scheme If stratocumulus is present, entrainment velocity is parameterized explicitly

Turbulence from Doppler lidar Input of sensible heat “grows” a new cumulus-capped boundary layer during the day (small amount of stratocumulus in early morning) Convection is “switched off” when sensible heat flux goes negative at 1800 Surface heating leads to convectively generated turbulence Turbulence from Doppler lidar Hogan et al. (QJRMS 2009)

Skewness Can diagnose the source of turbulence Stratocumulus cloud Height Potential temperature Longwave cooling Shortwave heating Cloud Negatively buoyant plumes generated at cloud top: upside-down convection and negative skewness Positively buoyant plumes generated at surface: normal convection and positive skewness Skewness Can diagnose the source of turbulence

Boundary-layer types from observations Lock type I qv Lock type III

Probabilistic decision tree Test surface sensible heat flux Test skewness Test skewness & velocity variance Use lidar backscatter Stable cloudless Clear well mixed Forced Cu under Sc Cloudy well mixed Decoupled Sc Decoupled Sc over Cu Cumulus Stable stratus Decoupled Sc over stable

Example day: 18 October 2009 Most probable boundary-layer type II: Stratocu over stable surface layer IIIb: Stratocumulus-topped mixed layer Ib: Stratus Usually the most probable type has a probability greater than 0.9 Now apply to two years of data and evaluate the type in the Met Office model Harvey, Hogan and Dacre (2012)

Comparison to Met Office model Winter Spring Model has: Too little stable Too little well-mixed Too much cumulus Note: Model cumulus needs to be >400 m thick Use radar to apply this criterion to obs Harvey, Hogan and Dacre (2012) Summer Autumn

Comparison with Met Office versus season and time of day Obs Winter Spring Summer Autumn Model

Forecast skill 6x6 contingency table is difficult to analyse Most skill scores operate on a 2x2 table: a (hits), b (false alarms), c (misses), d (correct negatives) Instead consider each decision separately Use symmetric extremal dependence index (SEDI) of Ferro & Stephenson (2011): many desirable properties (equitable, robust for rare events etc) Where hit rate H = a/(a+c) and false alarm rate F = b/(b+d)

Forecast skill: stability Surface layer stable? Model very skilful (but basically predicting day versus night) Better than persistence (predicting yesterday’s observations) d c random

Forecast skill: cumulus b a Cumulus present (given the surface layer is unstable)? Much less skilful than in predicting stability Significantly better than persistence d c random

Forecast skill: decoupled b a Decoupled (as opposed to well-mixed)? Not significantly more skilful than a persistence forecast d c random

Forecast skill: multiple cloud layers? b a Forecast skill: multiple cloud layers? d c Cumulus under statocumulus as opposed to cumulus alone? Not significantly more skilful than a random forecast Much poorer than cloud occurrence skill (SEDI 0.5-0.7) random

Forecast skill: Nocturnal stratocu Stratocumulus present (given a stable surface layer)? Marginally more skilful than a persistence forecast Much poorer than cloud occurrence skill (SEDI 0.5-0.7) b a d c random

Summary and future work Doppler lidar opens a new possibility to evaluate boundary layer schemes Model rather poor at predicting boundary layer type In addition to boundary-layer type, can we evaluate the diagnosed diffusivity profile – this is what matters for evolution of weather? How do models perform over oceans or urban areas? How can boundary layer schemes be improved? Combination of radar-lidar retrievals and 1D modelling demonstrated that shortcomings of altocumulus models could be identified and fixed The same strategy could be applied to the boundary layer

Model evaluation using CloudSat and Calipso Use DARDAR cloud occurrence Hogan, Stein, Garcon and Delanoe (in preparation)

Radiative properties Using Edwards and Slingo (1996) radiation code Water content in different phase can have different radiative impact

Modelling mixed-phase clouds - GCMs Until recently: most models diagnostic split More recently: improved computer power and desire for ‘physicality’  prognostic ice (Met Office, ECMWF, DWD) 1 -23 C 0 C Temperature Liquid fraction All liquid All ice Mixed phase

Ice cloud fraction parameterisation

Ice particle size distribution Large ice crystals are more massive and grow faster than smaller crystals Small crystals have largest impact on growth rate

Skewness Skewness defined as Positive in convective daytime boundary layers Agrees with aircraft observations of LeMone (1990) when plotted versus the fraction of distance into the boundary layer Useful for diagnosing source of turbulence