Presentation on theme: "Past and Future Climate Simulation Lecture 3 – GCMs: parameterisations (1) From last time – discretising the advection equation (2) Parameterisations:"— Presentation transcript:
Past and Future Climate Simulation Lecture 3 – GCMs: parameterisations (1) From last time – discretising the advection equation (2) Parameterisations: clouds/precip, land surface, dust, the oceans. (3) Implementation: boundary conditions, initial conditions. (4) Model output and model-data comparison (5) Experimental Design (6) Model tuning
2 main parts to atmospheric GCM: 1)Adiabatic (no heat exchanged) – e.g. advection, surface friction. 2) Diabatic (heat exchanged) – e.g. radiation, boundary layer, clouds Adiabatic advection of a tracer. E.g. a volcanic ash cloud moving around the equator, in a wind of constant speed, u: 180E 180W u Example of numerics – atmospheric tracer
nameABCDEFGHIJ longitude036E72E108E144E180E216E252E288E324E Initial Concentration U=0.1 A 0 =0, B 0 =1, C 0 =0,…… A 1 =0, B 1 =B , C 1 =C ,D 1 =0,…… nameABCDEFGHIJ longitude036E72E108 E 144E180E216E252E288E324E Initial Concentration Concentration after 1 timestep Concentration after 2 timesteps Excel demonstration
2 main parts to atmospheric GCM: 1) Adiabatic (momentum equation, last lecture) 2) Diabatic (heat exchanged) – e.g. convection, radiation (including clouds, greenhouse gases, aerosols), precipitation, surface energy balance. All parameterisations. e.g. precipitation: If (relative humidity > 85%) then precipitation = (relative humidity - 85%)*constant relative humidity = 85% e.g. convection: If (temperature gradient > 10 o C/km) then clouds = 1 temperature gradient = 10 o C/km precipitation (2) Parameterisations
e.g. land surface and turbulence:
(1) Potential dust source regions e.g. aerosols (here, dust):
Simulate just the uppermost approx 50m of the ocean (homogeneous slab of water). Typically, atmosphere calculates the surface energy fluxes for each gridbox (net-solar, net-infrared, sensible, latent heats). The sum will not be zero; this is the net energy flux at the surface. If it is positive, the ocean absorbs this and warms up appropriately. If it is negative the ocean will cool down. Need to parameterise ocean heat transport! Therefore no good for time periods/climates very different from modern. 1)2) e.g. oceans:
(3) Configuring Models – boundary conditions/initial conditions Boundary Conditions: Prescribed (by the user) fields. e.g. land- sea mask. The model can not change these. May be time-varying (e.g.SST). Initial Conditions:Fields used for initialising the model. After first timestep, model calculates. e.g. surface temperature
Land-sea mask Boundary conditions
Surface albedo (for models not predicting vegetation)
Sea surface temperatures (for models without an ocean)
Incoming solar radiation
Greenhouse gases, aerosols
Initial conditions Surface Temperature Pressure in mid-atmosphere Cloud coverSoil moisture + for ocean: temp,salinity,u,v,seaice
(4) Model output and model-data comparison
Produce a climatology
Surface Temperature: observations Surface Temperature: HadCM3 How good are GCMs? (1) temperature
Precipitation: observations Precipitation: HadCM3 Seaice: observations vs models How good are GCMs? (2) Precip and seaice
How good are GCMs? (3) El Nino
(5) Experimental Design Key concept: Testing hypotheses. Typically, a control + a number of sensitivity studies Modify a boundary condition… If everyone painted their roofs white, could this mitigate against global warming? Modify an internal parameter… Can the fact that all models predict too-cold poles in deep-time palaeoclimates be due to the lack of anthropogenic aerosols? Modify an initial condition… Was the Sahara bistable in the mid-Holcoene, 6,000 years ago? Change a parameterisation… Does poor representation of clouds in models result in poor ENSO simulation? Change the whole model… Which is the best model to use for future climate prediction?
(6) Model Tuning We know that internal model parameters affect the control climate produced by a model…often these are not well constrained by data. Therefore we can legitimately tune the model towards observations of modern climate by tweaking these parameters… For a small number of parameters, we can cover parameter space well…but….N=A x, where N is number of simulations, A is how well we sample the parameter space, and x is the number of parameters….soon become unmanageable. So, various approaches, including random sampling…..
And latin hypercube sampling….. Skill score generated, and then experiments ranked...
Tuned model outperforms original model….. observationstuned modeloriginal model