Limitations of Equilibrium Or: What if τ LS τ adj?

Limitations of Equilibrium Or: What if τ LS τ adj? Bob Plant, Laura Davies Department of Meteorology, University of Reading, UK With thanks to: Steve Derbyshire, Alan Grant, Steve Woolnough and Jeff Chagnon

Outline Deliberately depart from a key quasi-equilibrium assumption The timescale separation issue A simple model with memory CRM results with variable forcing timescale Memory and its origins Conclusions / implications What if τ LS τ adj? p.1/34

Timescale Separation What if τ LS τ adj? p.2/34

Adjustment Timescale Consider an adjustment timescale τ adj Describes rate at which a pre-existing convective instability would be removed by convective activity in the absence of forcing Key assumption of quasi-equilibrium thinking is that τ LS τ adj Parameterizations often include some such adjustment timescale in computing the closure Their behaviour is sensitive to the timescale used What if τ LS τ adj? p.3/34

Adjustment Timescale May be related to cumulus lifetime? Or related to time taken for gravity waves to travel between clouds? i.e., to communicate local temperature perurbations from clouds throughout environment For a step change in forcing, adjustment of mass flux is proportional to cloud spacing in unadjusted state (Cohen and Craig 2004) Implies that it depends on magnitude of forcing What if τ LS τ adj? p.4/34

Timescale Separation Arakawa and Schubert (1974), p691 Usually the large-scale forcing is changing in time and therefore the cumulus ensemble will not reach an exact equilibrium. The properties of the cumulus ensemble... depend on the past history of the large-scale forcing, but this dependency should be significant only within the time scale of the adjustment time. If τ LS τ adj, past history of forcing is effectively encoded in the current state of the atmosphere. What if τ LS τ adj? p.5/34

Timescale Separation Arakawa and Schubert, continued... We assume that this is the case for the cumulus ensembles we wish to parameterize. We call this assumption the quasi-equilibrium assumption. It is also an assumption on parameterizability, if by parameterization we mean a relation between the properties of the cumulus ensemble and the large-scale variables at some instant. What if τ LS τ adj? p.6/34

Separation Issues But... What forcing timescale can be considered large enough? What happens as quasi-equilbrium starts to break down? What if τ LS τ adj? p.7/34

Simple Model for Convective Memory What if τ LS τ adj? p.8/34

Simple Model Two-layer model of atmosphere and surface (1) dt dt = COOL+Q 1 where Q 1 adjusts to a rate R with a memory timescale t mem, (2) dq 1 dt = R Q 1 t mem and the rate R is such as to produce a neutral temperature with a closure or adjustment timescale t close ( ) Tsurf T (3) R = MAX,0 t close What if τ LS τ adj? p.9/34

Simple Model Forcing Surface temperature prescribed, and is characterized by a timescale τ Half of a sine wave during day, constant at night What if τ LS τ adj? p.10/34

Regime Diagram Qualitative diagram for t close = 1hr 24 Memory timescale 12 A B C 5 2 12 D E Forcing timescale 24 Choose τ = 24hr and look at regimes for increasing t mem What if τ LS τ adj? p.11/34

Regime Type E Type E: t mem << τ, repetitive response, dominated by evolution of forcing 0.8 0.7 0.6 Q 1 (K/hours) 0.5 0.4 0.3 0.2 0.1 Q 1 timeseries for t mem = 1hr 0 0 5 10 15 Time (days) What if τ LS τ adj? p.12/34

Regime Type D Type D: a transitional regime, with variability and skipped cycles 0.7 0.6 0.5 Q 1 (K/hour) 0.4 0.3 0.2 0.1 Q 1 timeseries for t mem = 5hr 0 0 5 10 15 20 25 Time (forcing cycles) What if τ LS τ adj? p.13/34

Regime Type C Type C: the heating over each forcing cycle is variable 0.45 0.4 0.35 0.3 Q 1 (K/hours) 0.25 0.2 0.15 0.1 0.05 0 0 5 10 15 20 25 30 Time (days) Q 1 timeseries for t mem = 12hr What if τ LS τ adj? p.14/34

Regime Types B and A Type B: heating oscillates about a mean response 0.35 0.3 0.25 Q 1 (K/hours) 0.2 0.15 0.1 0.05 0 0 5 10 15 Time (days) Q 1 timeseries for t mem = 24hr Type A: t mem >> τ, very slow response, not adjusted after 20+ cycles What if τ LS τ adj? p.15/34

In summary As t mem increases to τ the response is not directly related to current forcing Implies feedback, with current heating dependant on the time-history of the convection Are there such regimes for atmospheric convection? Hard to vary t mem (!), but we can vary τ... What if τ LS τ adj? p.16/34

CRM Results for Variable τ LS What if τ LS τ adj? p.17/34

The CRM Experiments Using Met Office LEM 1km horizontal resolution on 64x64km 2 domain Prescribed radiative cooling of troposphere, constant with height No wind shear imposed, f = 0 Run to radiative-convective equilibrium with prescribed sensible and latent heat fluxes Then vary surface fluxes: half sine wave during day and switched off at night Not a diurnal cycle simulation: e.g., no shallow convection phase What if τ LS τ adj? p.18/34

Equilibrium: Weak Definition If τ is long (e.g., for constant forcing), equilibrium state easily defined... 0.18 0.16 Cloud base mass flux (kg/m/s) 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0 50 100 150 Times (hours) What if τ LS τ adj? p.19/34

Equilibrium: Weak Definition If τ is finite... A strict definition is to compare the instantaneous convective heating to the instantaneous forcing Not satisfied by convection due to adjustment Does not respond instantaneously: eg, when forcing first switches on Here we consider a weak definition of an equilibrium: compare the total convective heating to the total forcing, integrated over the forcing cycle A balance between these is a necessary condition for AS s quasi-equilibrium What if τ LS τ adj? p.20/34

Timeseries of Mass Flux For τ = 24hr, with similarities to regime E Type E: repetitive response, dominated by evolution of forcing What if τ LS τ adj? p.21/34

Timeseries of Mass Flux For τ = 3hr, with similarities to regime C Type C: the heating over each forcing cycle is variable What if τ LS τ adj? p.22/34

Timeseries of Mass Flux For τ = 1hr, with similarities to regime B Type B: heating oscillates about a mean response What if τ LS τ adj? p.23/34

Variations in Convection Per Cycle As measured by cycle-integrated cloud base mass flux, 0.024 Mean integrated CB mas flux per cycle (kg m 2 ) 0.023 0.022 0.021 0.02 0.019 0.018 0.017 0.016 0.015 0 5 10 15 20 25 30 35 40 Forcing timescale (hr) Enhanced variability at τ < 12hr What if τ LS τ adj? p.24/34

Variations in Convection Per Cycle As measured by cycle-integrated precipitation, Enhanced variability at τ < 12hr What if τ LS τ adj? p.25/34

Feedback from one cycle to the next? For τ =36h (pink), 24h (dark blue), 18h (light blue) 2 ) 0.028 Total CB mass flux per cycle, following cycle m (kg 0.026 0.024 0.022 0.02 0.018 0.016 0.014 0.012 0.012 0.014 0.016 0.018 0.02 0.022 0.024 0.026 0.028 Total CB mass flux per cycle, 1st cycle m(kg 2 ) No correlation between convection on successive cycles for τ > 18hr What if τ LS τ adj? p.26/34

Feedback from one cycle to the next? For τ =12h (green), 6h (red), 3h (black) 2 ) 0.028 Total CB mass flux per cycle, following cycle m (kg 0.026 0.024 0.022 0.02 0.018 0.016 0.014 0.012 0.012 0.014 0.016 0.018 0.02 0.022 0.024 0.026 0.028 Total CB mass flux per cycle, 1st cycle m(kg 2 ) R 2 0.5 for 3 < τ < 12hr What if τ LS τ adj? p.27/34

Physical Origins of Memory What if τ LS τ adj? p.28/34

Mean State Is the cycle-to-cycle variability explained by differences in domain-mean profiles? Domain-mean profiles of θ and q v at start of forcing conditioned on the convection occurring in the following cycle: integrated mass flux exceeds mean (strong cycles) integrated mass flux exceeds mean (weak cycles) What if τ LS τ adj? p.29/34

Mean State Solid line and dark shading for strong cycles; dashed line and light shading for weak cycles. Strong/weak difference < variations between strong cycles What if τ LS τ adj? p.30/34

Spatial Water Vapour Anomaly Snapshots during active convection, for τ = 3h (left) and = 24h (right) Convection develops coherent structures on scales of 10 20km What if τ LS τ adj? p.31/34

Spatial Water Vapour Anomaly Snapshots just before onset of convection, for τ = 3h (left) and = 24h (right) Note different scales used. What if τ LS τ adj? p.32/34

Power Spectrum of RH 0.3 0.25 Max convection Min convection (3 hr) Min convection (24 hr) Normalised power 0.2 0.15 0.1 0.05 0 10 20 30 40 50 60 Wavelength (km) For longer τ, the power at intermediate scales is lost during the break in convective activity What if τ LS τ adj? p.33/34

Conclusions Memory appears to be carried by spatial structures in the moisture field, not in the domain-mean state Memory effects found for timescales < 12hr For u 10ms 1, this translates to a spatial scale of 500km Suggests that for forcing mechanisms with scales < 500km, convection may not be in quasi-equilibrium Convection in upstream grid boxes may be relevant The perfect parameterization might have non-local aspects, and not be purely 1D? What if τ LS τ adj? p.34/34