Global models

Content

Principles of Earth System Models and global models Global aerosol models as part of Earth System Models

Model input Computation

Spatial discretization

Parameterizations, look-up tables

Output Evaluating model results

Postprosessing

1-D models

• Representative of surrounding area

• Timestep: seconds

• Vertical levels: even 100

• Timescale: usually days

3-D global models

• Grid box represents ~100 km x 100 km

• Timestep: >10 minutes

• Vertical levels: few tens

• Timescale: years to centuries

Parameterizations

CirculationAerosolsClouds

CirculationBiogeochemistryHeat transport

VegetationLand useSoil moisture

Aerosol emissionsGaseous emissionsDeposition

Heat transferMomentum fluxAerosol emissions

Earth System (Model)

Earth System Model: choice of components

Choice of ESM components is based on timescale of the experiment: years, decades or millenia

variables of interest: air quality, climate change, process study

availability of computational resources

Mixed layer ocean

Ocean circulation model

Dynamic vegetation model

Prescribed sea surface temperatures and sea ice

Prescribed meteorology

Model of everything related to Earth

ComplexityComputational expenseModel noise

Cloud microdynamics

Prescribed vegetation (type, leaf area index)

Population model

Earth System Model: black box modeling

ESM can easily have >200 000 lines of code A single researcher usually contributes only to a single

module Rest of the model is considered black box (“need to know” basis)

Not a significant problem with ESM users, but developers do not

always know all of the

consequences their code

has on the overall model

performanceAerosol module

Global aerosol models

Global aerosol model has to describe all possible

combinations of atmospheric aerosol composition and

size Dust, seasalt, black carbon, organic carbon, sulfate, ...

Atmospherically relevant aerosol processes Nucleation, condensation, coagulation, deposition, ...

Model must be easily coupled with the host-model Emissions

Parameters for radiative effects

Formation of cloud droplets

Still, the model has to be computationally efficient

Transport of gases

Aerosol microphysics

Transport of aerosols

SOx, NO

x

Organicaerosolchemistry

Direct effect

Indirect effect

Inorganic aerosol chemistry

Development of global aerosol models

Increased primary sulfateActivation nucleationPrimary emissions

Global aerosol models

Fixed aerosol climatologies Monthly/yearly average radiative properties of aerosol

Based on simulations and satellite observations

Aerosol mass-only models No aerosol microphysical processes

Modal size-resolved aerosol microphysics models Aerosol distribution is represented with superposition of

several log-normal modes

Sectional size-resolved aerosol microphysics models Better representation of aerosol processes

Example model setup: ECHAM5-HAM

ECHAM5 is an atmospheric General Circulation Model

developed from ECMWF (global weather forecast model)

HAM module describes aerosol population with seven log-

normal distributions and solves related microphysics

(condensation, coagulation, wet deposition, etc.)

INSOLUBLE

NUCLEATION

SOLUBLE

AITKEN ACCUMULATION COARSE

SU = sulfate

BC = black carbon

OC = organic carbon

SS = sea salt

DU = mineral dust

SU

SUSU SU

BC BC BC

BC

OC OC OC

OC

SSSS DU

DU

DU

DU

Modularisationof a global aerosol model

Emissions and fields

Dust, sea salt, DMS

Water, aerosols, SO4

Online

Fossil-fuel, SO2

chemical fields: OH, H2O2, NO2, ozone

Offline

EmissionsFields

Dust

Black carbon

Emission inventories usually contain static monthly or yearly average

emission fields Online emissions use meteorological conditions and surface properties to

calculate emission of e.g. dust and sea salt

Examples of online/offline variables in a global model

Vertical discretisation

Sigma coordinates Hybrid coordinates

Pressure/height coordinate is not a good choice for a

vertical coordinate Typically 20-30 hybrid levels are used Choice of model vertical extent:

troposphere+lower stratosphere

+stratosphere + lower mesosphere

+ mesosphere + lower thermosphere

Dense, terrain-following near surface

Sparse, flat pressure-levels at top of atmosphere

Horizontal discretisation

Linear terms of temperature, divergence, vorticity and surface

pressure are usually presented in spectral space using

spherical functions with a certain truncation (21, 42, 63, ...) Other terms (humidity, concentrations) are calculated in

gridspace

T runcat ion No. of Longitudes No. of Lat itudes Cell size at equatorT21 64 32 ~630x630 kmT42 128 64 ~310x310 kmT63 192 96 ~210x210 kmT106 320 160 ~130x130 km

Computational demand

If memory use ~ (number of vertical levels) x

(number of latitudes) x

(number of longitudes) x

(number of tracers) Common resolution with simple aerosol model:

- 19 x 64 x 128 x 20 x 8 bytes = 25 Megabytes

Slightly better resolution and a sectional aerosol model:

- 31 x 128 x 192 x 50 x 8 bytes = 305 Megabytes

Arithmetic operations (105 / timestep / gridbox) ~ 1015 operations per simulation year

Computational demand:what is being calculated?

Atmospheric circulation is calculated with primitive

equations:

( )

( )

( )( )

( )

x

y

turb cond rad

p p

q

uA u fv F

t xv

A v fu Ft y

Q Q QT RTA T

t c p c

qA q S

t

( ) N

NA N S

t

( ) M

MA M S

t

…adve

cted

trac

ers

Model dynamics: advection, Coriolis force

Physical processes: all subgrid-scale non-adiabatic effects (friction, turbulence, phase change of water)

Computational demand:parameterizations and look-uptables

To decrease computation time, included submodels

are usually parameterized Parameterization is not as accurate as original model,

and cannot be used outside parameterization limits

Parameterizations are also needed to include subgrid-

scale processes, such as Convection scheme

Cloud structure

Aerosol processes

Look-up tables are used to store frequently needed

data for fast access

Evaluation of results

Results of global models can be evaluated against

field observations Flight observations

Long-term and campaign in situ observations

Satellite observations

Inter-model comparison Global models have differences in representations of

atmospheric physics

Running experiments with several models (e.g. IPCC)

Model output

Status of the climate every 30 minutes Direct (predicted) variables

Temperature, winds, humidity, aerosol concentrations

Derived variables AOD, aerosol forcing

Due to model noise, a single

datapoint is unimportant Statistical tools have to be

used to get useful information

from results

More complexity more noise more averaging needed

Optical thickness at one gridpoint near Finland

Model output: averaging

Selection of averaging dimension: Time, latitude, longitude, vertical

Global averaging (both latitude and longitude)

decreases noise significantly Shows the effect on global climate

Averaging over few (tens) of years makes it possible

to investigate local changes Averaging dimension depends also on variable of

interest Comparing AOD to satellite observation

Studying effect on global 2-meter temperature

Model output: length of simulation

When planning the duration of the model run,

response time of different model components must be

taken into account With an ocean model included, it might take a few

decades for the temperature to reach a new stable

state Response time of mixed-layer ocean model is much

shorter due to lower mass of water

Simulat ion t ime Ocean

1-5 years Prescribed sea-surface temperatures and sea ice5-20 years Mixed-layer ocean model

-20 Fully coupled ocean model

Model tuning

Why do climate models produce so “good” results? Partly because they are tuned to do so

Climate system includes several variables whose

values are poorly known For e.g. cloud-related variables (convective cloud

systems)

Values can be taken “from a hat”, or used in tuning

Usually modeled Top-of-Atmosphere radiation flux is

matched to observed This makes the overall climate (temperatures etc.) look

close to observed

Almost all models are tuned with different variables and

different tuning criteria

What are global modelsgood for?

Importance of individual processes in the Earth system add/remove/modify a single process

e.g. role of new particle formation in climate system

Predicting the future e.g. climate change in 100 years

need to construct scenarios for emissions/conditions

validity of parameterizations in new conditions?