36C3 - Climate Modelling

36C3 preroll music

Herald: ... so I'm looking forward and I
hope you are, too. I am looking forward to

be told the difference between..
laughs ... we will all be told the

difference between an input model and a
climate model, and we are going to be told

this difference by karlabyrinth.
There you go.

applause

karlabyrinth: Thank you. Hello and
welcome, everyone. I would like to see my

slides. Are the slides ...? Ah, OK.
laughs nice. So welcome, everyone, to my

talk about Climate Modeling - the science
behind climate reports. First of all, I

will shortly introduce myself and what I
do. I work at the UFZ. That's the

Helmholtz Center for Environmental
Research in Leipzig and I work for the ESM

project, which is short for 'advanced earth
system' modelling capacity. I am also a PhD

student at the University of Potsdam and I
am part of the developer team for the

middle scale hydrologic model, which is an
impact model. And I'm also a scientist for

future and an artist. So what this talk is
about? This talk is partitioned into three

sections, mainly. First is the
introduction where I will introduce some

nomenclature like what is weather, what is
climate and what we can say about

predictions. For example, why we can't
tell the weather in three years but we can

say something about the climate, and what
are climate models. Then the second part

will be the longest, the science behind
warming graphs. I will show you a graph

that's typically shown when people tell
you about climate change, and I will

explain that graph in detail and what is
behind it. The third part would be

installing an impact model to your local
PC if there is time. If there is no time I

will skip that. And in the end, there is, as
always, a summary and conclusion. So

starting with the introduction. Weather is
defined as the physical state of the

atmosphere at a given time whilst climate
is averaged weather. Most of the time a

time period of 30 years is taken for that
averaging but also other time periods

could be taken. So, while, the main
question was, while we

are not able to predict whether at a
specific date in a

decade, for example, let's say the 27th of
December in 50 years or so. Why does it

still make sense to propose general trends
for the climate? That is a question that

often arises when... and I'll answer that.
So, first of all, it is about average.

Average cloud coverage gives us
information on average reflection. And

average reflection is ... has an impact on
the warmth on the earth. And the same is

true for another scenario. For example,
average precipitation - meaning rain or snow

and temperature - has an impact on
vegetation and vegetation influences the

carbon cycle. And that again influences
the warming or cooling and that has an

influence on the ice coverage. And that,
again, on the reflection. So there are

lots of processes that are connected to
each other and if we know something about

the average of some of these physical
state of the atmosphere, we can say

something about the climate trends. So the
question is, what is a climate model? And

the AR5 defines a climate model is a
numerical representation of the climate

system. The AR5 is a source I will cite
quite often. So I have one slide

with the whole citation. It's the fifth
IPCC report. IPCC is the

Intergovernmental Panel on Climate Change
and the fifth assessment report is, yeah,

so AR5 is the abbreviation for fifth
assessment report. But coming back to a

climate model. So a climate model could,
for example, be a GCM - a general

circulation model - which is a global
climate model that usually consists of an

ocean and atmosphere circulation. An RCM
is not a GCM but it's a regional climate

model, meaning a climate model at a
limited area, and mainly it has a higher

resolution. And for it is at a limited
area, that usually means that there is

some input and output going in because
it's not a closed system. And an impact

model again has usually a higher
resolution in time and space and it's not

a climate model, but it's for simulating
extreme weather events

like floods. So if you want to build a
dam or a dike and you want to know how

high this dike or dam should be, then you
would usually run an impact model that

gives you information about water height
over decades or longer or so. And then

you would decide on the height. So this is
the use for impact models. So that's for

the introduction part. Now I come to the
main part and I will start with a

question: is it proven? Or with a climate
graph. As that, I will show you a graph, a

typical image people would show you when
they address climate change. This graph

has an x-axis with a time scale and you
see it's reaching far into the future. And

it also has three or four regions
and the first region is only in the past.

And the y-axis is the global surface
temperature change, meaning how much

degrees in Celsius or in Kelvin, if it's
different it's the same, we will have in

future or we had already. And then, you
see several lines and different colors and

with the names RCP something. And I will
explain all the numbers and everything

about that graph because it's a pretty
important graph. So first of all, I will

tell you... no, no. I will tell you
something about the numbers and

uncertainties. The uncertainties are the
transparent colors behind the lines. I

will tell you something about the
representative concentration pathways,

which is short RCP, and so it's reflecting
the colors of the lines. I will tell you

something about the source of the graphs.
So where does this graph actually come

from? So I will tell you something about
the assessment report. And first of all, I

will answer the question: is it proven or
is there scientific evidence that we will

face that climate change? So, you will see
that graph quite often. First of all, I

took a definition for proof, for
scientific evidence, from Wikipedia. The

strength of scientific evidence is
generally based on the results of

statistical analysis and the strength of
scientific controls.

Meaning, you make an experiment over and
over again and you change basically some

influences on the experiments where you
want to know that this does not influence

the output. So you can narrow it down and
know what is the source of your results

and so you can prove a physical law or
something. Yeah, I took this comic from

xkcd because it's a nice... it's somehow
connected. So there is a person who pulls

a trigger and then gets struck by a bolt
or some something. Something bad happens,

for example climate change. And then,
yeah, there are two scenarios. For

example, a person usually would decide,
okay, I would not pull the lever again.

But scientists usually or more often would
say, okay, maybe: Does that happen every

time if I do so? Because yeah, that's
basically how you prove something. That's

experiments. But in case of climate
change, even scientists say you shouldn't.

Although it's pretty interesting for us
from a scientific perspective. But the

problem is we only have one earth. We
cannot do this experiment very often,

except we had a time machine. Then we
could go back, but we haven't so we

shouldn't do that experiment. And that's
something scientists before, long ago in

1957 said already: "Human beings are now
carrying out a large-scale geophysical

experiment of a kind that could not have
happened in the past nor be reproduced in

the future." So another question is, yeah,
if you ask this question if it is proven

or that it probably is not happening or so
to climate deniers, they usually would

tell you: Okay, maybe it's not happening.
And the other side would take the position

and ask you, okay, if you stand in front
of a road and you want to cross the road

and there's a car approaching very fast,
would you cross that road? Because it

could happen that the car stops or makes a
U-turn or something. But well, usually it

doesn't. And sadly, we know lots about
this experiment, because it's done very

often before. We know
something about traffic, and that it's

pretty dangerous. So let's change the
factors a little so that we don't know so

much about that situation. Let's say a
cube approaches us with a high velocity on

something that is maybe not a road. Would
you still cross that something? The answer

is you still probably wouldn't. And why
wouldn't you do so although you know

nothing about this situation? Well, you do
know something. You know conservation of

the momentum, which is a physical law you
know about. So you have a situation, you

know not so much about. You have never had
an experience before, but you still are

able to make some assumptions because you
know the physical laws behind it. And

that's basically the same we do with, in
fact, in the context of climate. So we

have, let's say, just an earth and the sun
and the sun has some radiation and that

comes to the earth and gets partially
reflected and the earth radiates itself

because it has some temperature. We know
something about this sun. We know the

solar insolation. And we know parts of the
light is reflected. And the factor that is

reflected is usually called albedo. And so
the reflected energy is albedo times the

solar insulation and albedo is something
about 30 percent. And we know then that

the light that is absorbed must be all the
remaining energy. So the energy of the

surface is 1 minus albedo times the solar
insolation. Then knowing Stefan-Boltzmann

law for energy emissions where the
temperature goes in to the power of four

and with the Stefan-Boltzmann constant we
can actually find find out the surface

temperature which then is derived to -19.5
degrees Celsius. Well, we know, probably

we know, that the earth is much warmer,
and that's because our model in this case,

which is maybe a climate model, is far too
simple. So we change something about that.

We add atmosphere, and atmosphere has some
interesting impact. So atmosphere has some

trace greenhouse gases,
for example CO2 but also H2O, ozone,

methane, O2 and nitrous oxide. And these
greenhouse gases reflect the radiation of

earth back to earth, partially. Meaning
the atmosphere has a transparency and this

transparency we call t is something
between 15 percent and 30 percent. So it's

not fixed. And that's another interesting
fact. The atmosphere emits energy, which

we call j atmos, and that goes out in
space and to earth and the energy that

goes into the atmosphere is 1 minus the
transparency times the energy. So we know

two equations. The first is the energy
that goes into the atmosphere also goes

out of the atmosphere. The second is that
the surface energy of the earth

is the term we had before,
1 minus albedo times the solar

insolation plus the one part of the energy
that is reflected by the atmosphere. And

so we have two formulas, two equations
with two unknowns and with the Stefan-

Boltzmann law from before we can derive
the surface temperature, which is 15

degrees of Celsius. And that actually is
not so far from what it actually is. In

2000 it was measured that the surface
temperature is 14.5 degrees. So, I did

this for a specific t which is 22.5
percent but when we change that t a little

to, for example, 20 percent, so we add
more CO2 because, for example, we would

add a factory that would do carbon
emissions. Then the transparency goes down

and the temperature rises to, for example,
16.6 degrees in case of 20 percent. This

is also a very old knowledge. So this is
maybe a little much on a slide but it is

still very interesting because it is
copied directly from a paper that was

published from Svante Arrhenius in 1896
already. And it's on the influence of

carbon acid in the air upon the
temperature of the ground. And carbon acid

is the old term for carbon dioxide. So if
we have a look to the percentage... So he

investigated: What if we change
carbon dioxide? So what is the impact of

our behavior? Let's say carbon dioxide
in our atmosphere would

double, so would increase by a factor of
2, then the average temperature rise in

Leipzig in December, so I choose the
region for Leipzig, would be 6.1 degrees.

Well, that's probably a little high, but
what we can't see is already that

Arrhenius back then already knew that
there is a seasonal impact on

climate... that climate change is seasonal
and also spatial. So it is not just one...

not the average temperature is the only
interesting knowledge we get. So

Arrhenius said something like the
temperature in case of carbon acid doubled

would be around four to six degrees. And
the current models predict something like

an increase of two to four degrees for
that scenario. So there is maybe overlap

already with that simple model from back
then. So, then I come to the question...

a climate model represents physical laws.
That's what we learned. Where do the

uncertainties come from? So if we know all
the physics laws and we would just

calculate everything with this physics
laws, why are there even uncertainties?

And there are some reasons for that. For
example, the initial conditions is one

main source of uncertainties, meaning how
is the current state of the climate system

now? How fast does something move?
Where are the clouds exactly?

And so on. We don't
know these precise initial

conditions and therefore errors
occure. Second, would be the

resolution of a model. So the
temporal and spatial step length, meaning

we can't... always represent our
climate system with differential equations

and we approximate everything. We have not
the movement of every molecule but we have

some average on cells. And if we increase
the resolution then usually the

uncertainties go down. But sometimes they
even don't for some question, for some

questions it's better to have a lower
resolution. But mostly it's better to have

a higher. Then, truncation, so we have

computational limits. And lack of
understanding, for example, clouds. Clouds

are not understood pretty well. And when I
read the fifth assessment report, I found

a sentence a little amusing:
Climate model... Clouds in climate models

usually tend to rain too early. Yeah, so
but if you know all these sources of

uncertainty, why is there no such thing as
the one best climate model? Meaning, why

can't we go to the highest resolution and
to the best... the best computer we get

and do everything just in the best way and
then we would have our best climate model?

And there are some reasons for that. For
example, the so-called dynamic core,

including the method for differential
equations or something like grids.

For example, if we have a triangular
grid or a rectangular grid. On a

rectangular grid we usually can
calculate faster but on a triangular grid

we could, for example, increase the
resolution locally. That might be

differences. And both have advantages and
disadvantages. Also, the parametrization:

parameters in our last slide were for
example the t and the albedo which will

probably be not the final parameters
because they are derived from other

parameters, but physical laws or something
are often represented by rules with

parameters, and these parameters can be
estimated. And they can be calibrated with

different error measures.
And there this is another reason for

uncertainties and differences in climate
models and then their schemes. For

example, there are different formulations
of physical processes, for example, that

again, clouds. And last the truncation,
again, we can also decide how we limit due

to our lack of computational power. So,
yeah, what do we do? We investigate all

the models we have. So there are different
climate models that are representing our

climate and we take all the models that
match certain conditions - I come to that

later - and we average the output and then
we

get a climate prediction and also that
uncertainty band you see. So what climate

models do we investigate? They are so-
called coordinated GCMS. So climate models

are compared in so-called coupled model
intercomparison projects in different

phases. These coupled model
intercomparison projects are called CMIP

4, 5 and 6. So there might have been four
earlier ones. But currently, for the AR6,

so for the sixth assessment report CMIP6
investigated. And I showed you on the

map the research centers which took part
in CMIP6, so which take part in the 6th

assessment report. These research centers
are mainly specialized research centers,

university and metereological offices, but
generally it's open for any institution to

participate, as long as they follow a
protocol for their contribution, where

there are some rules so you cannot just do
anything. These institutions

need to produce variables for a set of
defined experiments and a historical

simulation from 1850 to present. This blue
part is a link, so if you go to my slides

afterwards, you can see these variables
you need to reproduce and then you can do

something like this. So we have a graph
here again. On the x axis, we see again a

timescale that reaches from 1850 to today.
And on the y axis we again see the

temperature anomaly or the temperature
difference between.. so, exactly the

temperature difference, so how much the
earth has warmed up. We see CMIP3 and CMIP5

compared, which were the models that were
investigated for the AR5. So we see a

band. This uncertainty with the yellow and
bluish and the background and then we see

these two lines, the blue and the red one
from CMIP3 and CMIP5 and then we see the

black one. And that is what actually was
observed. And we see that this differs

quite a lot. And that's due to there was
only investigated the natural forces,

meaning excluded what the human did. And
if we

also put the human forces into it, then
it's quite matching. And that is the best

kind of proof we can get. And again, I
said we investigate the physical laws and

the physical laws were actually results of
scientific experiments. And so, yeah,

there's this kind of proof. And yeah. So
maybe a little addition. There are also

other coordinated model intercomparisons
projects that are outside of the IPCC and

so the ones that are inside the IPCC
where the scientific focus is on a

subtopic, on something like land surface
for example (and that's what I do). And

they're also published work
outside from IPCC. So back to the graph.

We talked about the part: Is it proven?
And I hope I convinced you that it is. And

now I will talk about the sources of the
graph. So I talked a lot about the IPCC.

The IPCC, the Intergovernmental Panel on
Climate Change, publishes reports. So for

example, the 5th assessment report and
what you see here is part of the cover.

But there have been 4 ones before, as the
name 5th suggests, the first assessment

report FAR was published in 1990. The
second SAR in 1995. Then there was the TAR

and then for the 4th assessment report
they changed the name scheme for some

reason to AR4 and then there was AR5,
which I'm talking about. The IPCC consists

of several working groups, including
Working Group 1 to 3, providing the

assessment reports and I mainly focus on
the assessment report from a working group

1, which investigates the scientific
aspects of the climate system and climate

change. But there is also a working group
investigating on vulnerability and

economic impact. And the third one on the
options of limiting greenhouse gas

emissions and others. So I totally show
you a history of the climate models. In

something like the 70s, climate models
were investigated where there was just an

atmosphere, the sun, rain - clouds were
missing - and CO2 emissions. And

I hope you believe that the sun is behind
the atmosphere and not in this atmosphere.

In the mid 80s there was prescribed
ice added and already clouds and land

surfaces and yeah, you see a nice mountain.
But actually in that time the resolution

was so low that the Alps only had one or
two grid cells, meaning that was not so

much about land surface, but it was added.
And for the first assessment report there

was a swamp ocean added, meaning an ocean
was added, but it was had no depth. For

the second assessment report, the ocean
got some depth. So it was a normal ocean

with surface circulation and there was
added volcano activity and sulphates. For

the third assessment report, there was
added... So this is all about which kind of

processes were there in the climate models
that were investigated in these assessment

reports. Meaning there were climate models
before that already had those processes

included, but they were not investigated
in the assessment reports. So this is a

history of which climate models or which
processes and climate models were

investigated in assessment reports. And
the third assessment report, there was

another circulation edit for the ocean,
the overturning circulations. And there

were rivers added, which is interesting
because I do something with rivers and

there were aerosols added and a carbon
cycle, meaning that the carbon that goes

into the atmosphere also goes out. But
yeah, not everything, or that half-time is

not so good. For the AR5, er 4, there was
chemistry added in the atmosphere and

interactive vegetation, and for the AR5
there was ozone added and biomass burning

emissions. And there is a history of
processes, but there is also a history of

computer modeling that might be really
interesting. It started more or less in

1904 with Vilhelm Bjerknes, who found
equations that could be solved to obtain

future states of the atmosphere. And he
thought about that maybe these equations

are really hard to
solve and that task should be split and

distributed to many people. So he
basically mentioned a human computer and

then Lewis Fry Richardson came in 1922 and
did actually calculate all this.

This did a six hour forecast solving the
equations by hand, alone. And 42 days user

time, meaning he himself calculated 42
days on it. But those 42 days were

distributed over two years in total. So he
was a little behind the weather, only to

find out that it didn't give the correct
answer. audience laughs That was long

forgotten. But people said, yea, that's
not quite practical. We cannot do that.

But then computers came. In 1950, the
first successful weather model was run on

a computer called ENIAC, and in 1950,
weather predictions were run twice a day

on an IBM 701. Nowadays we use
supercomputers much larger and there's a

whole list and rank and I will shortly
introduce JEWELS to you: the Jülich Wizard

for European Leadership Science. That's a
supercomputer in Jülich and I would have

shown you a picture, but you are not
allowed, you are not simply allowed to

take pictures on that campus. But since
super computers are fancy shiny cupboards

anyway, I thought this is OK. So we have
these cupboards that look at shiny covers

and then this covers their blades and each
blade is called a standard node and

consists of, in case of JEWELS, 2 times 24
cores with 2.7 GHz and it's hyper-

threaded, meaning you can actually run 96
threads or processes on one of these

nodes. And these notes have 12 times 8 GB
of memory. And that's not quite much if

you want to run a climate model but
I'll come to that a little later. And in

fact, in case of JEWELS, you have like
three rows of five of these cupboards or

something. And so there are in total 2271
standard nodes, 240 large memory nodes and

56 accelerated nodes having something like
GPUs. And I tell you about JEWELS, not

because it's the
fastest, actually, it's maybe the 30th,

not even because it's the fastest in
Germany - it was when it was built but

that's a while ago - but I told you about
that because JEWELS provides actually

computing budget for the ESM project, the
Advanced Earth System Modelling Capacity.

And so there are actually earth system
models run on that machine. So what I told

you before, there is not so much memory on
each node.

So what you need to do is you need to cut
down your problem and distribute them over

the nodes. And then there needs to be some
communication. So usually if the task

is so simple, you can cut down your grid
and put a number of grid cells to each

node. And then there's communication
between the nodes on the boundaries to

solve the differential equations. Talking
about grids, I would talk about the

resolution. Also, again, a history of
resolution of the climate models. For the

1st assessment report, the region
resolution was 500 kilometers times 500

kilometers. And as I said before, you see
these two yellow yellowish cells in the

middle that are the Alps. For the second
assessment report, the resolution already

doubled or halved, depends on how you want
to phrase it. For the TAR, it was 180

kilometers for AR4, it
was 120 kilometers. For the AR5, it's a

little bit a different section I show you.
And also, I show you two resolutions.

There's a resolution for the higher
models, which is 87, for example, 87.5

kilometers. And for the very high
resolution, 30 kilometers. And that's

because climate models are not just one
model, but they are different kinds of

models that are coupled. And each model
has its own resolution. So it's more or

less like something like this. So we have
a model for ice, we have a model for

atmosphere, for ocean and for terrestrial.
And this is coupled. So they all sent

their data to a coupler or something. And
that set was there as an input to the

other model. So this is more or
less like how climate models look.

And each of the models, again, has several
layers. For example, the terrestrial layer

has a ground water part and the
atmosphere. And so there's some input from

the atmosphere to the soil and plant
system. And then there's some water that

is sinking into the groundwater and then
coming out to the rivers. And yeah. So

then we have the runoff. So meaning rivers
get water. And then if you have a look to

rivers and want to parallelize rivers,
then it's not so easy because we have a

source somewhere and the water has to go
from the source or something that happens

at the source has an impact to the sink,
meaning this has to communicate all the

way along to the sink. And that's where I
come in. I actually do. So I show you the

Danube, which you probably know better
with the name Donau. At a resolution of

five kilometers and basically cut down the
Danube into sub river domains. And we

need. If we parallelize these we need to
calculate the subriver domains that are

farther away from the sink first and it
uses that in the first graph a little. So

the grayish areas are calculated first and
then it goes down farther to the sink. So

just to tell you about what I do. So now
we come back to the main question. So we

answered where the sources of the graphs
come from. Now we answer the questions:

What is a representative concentration
pathway? Meaning what we all did before

was more or less telling how we get to
that black line in the first section. And

now we concentrate on the colored part
where we have more graphs than one. So the

working group 1 of the IPCC generally
tests the selection of coupled models, that

is what I told you before, matching
specific conditions and investigates the

output assuming different emission
scenarios. Meaning we have a couple of

climate models that are somehow different,
for example in their grid. And then we

have input data. The input scenarios would
be, for example, the first one where we

just do business as
usual and don't reduce carbon emissions.

The second would be we start with our way
we do it today, but we would slowly change

to renewable energy. And the third one
would be a scenario where we do it

spontaneously now or so. And that is an
input scenario that we put into the

systems and then we get out a model output,
that says something about the future. So

there is a black line that says, OK, this
was our history until today. And from that

on, we have three scenarios and they are
represented upper to lower. So the upper,

upper and right line represents the way
where we do nothing or so. So this is

basically what we do with scenarios. And
the RCPs, Representative Concentration

Pathways are scenarios that include time
series of emissions and concentrations of

the full suite of greenhouse gases and
aerosols and chemical active gases as well

as land use and land cover. So that is
another graph from the AR5 and it shows

again in the X-axis the years, it's the
same timescale as before, but on the Y-

axis we now have the rate of forcing, that
is basically having this impact on our

climate. And so each of the RCP scenarios
has some kind of equivalent in

radiative forcing.
Yeah. So we have a 4 of these scenarios.

The data for the RCP scenarios is
coordinated by again the input4MIPS: input

datasets for model and intercomparison
projects that I told you before. And most

of it is freely available and I gave you
the link. So if you want to run your own

climate model and test it with these
input, you can find it there. And now I

will explain the last part. The numbers
and uncertainties.

So first of all, again, to the graph from
before the numbers behind the RCP refer to

the radiative forcing at the end of the
modeling period of 2100. Meaning if you

follow one of these lines, for example,
the red one to where it crosses the 2100

line, then the number there is 8.5. So
RCP8.5 is the name for this RCP scenario.

But then the numbers of these different
sections are the numbers of models used

for this scenario in this time period.
Yes. So as I said, there are lots of

models intercompared and we even have
different models for the different time

periods. So until 2100 there are 39 models
for the RCP8.5. And of all the all the

rest, there are 12. And you see this
little gap, this line break at 2100. And

that is caused by the change of numbers of
models that took took part in this

project. And another interesting thing
that we see here, and maybe the most

important, is we have quite huge model
uncertainties. So if we compare all the

models, there's a huge band where we can't
exactly say, OK, it's like this or that.

But this band is still... About human
uncertainties are more important, than this

model uncertainties. We see tiny overlap,
but mainly we can say how will the human

behave derives our future. And that there
will be this climate change we are talking

about. So that was the main part about
this three parts. And it's also it is also

the most important part. Now, I could
probably show you how you can install an

impact model to your local PC, but
probably I will have maybe something like

three minutes left. So we'll switch to the
conclusion. And yeah, maybe if it's

arising as a question, I can do it. So
what have we learned? Weather is the

physical state of the atmosphere at a
given time, while climate is average weather

over 30 years. A climate model as a
numerical representation of the climate

system. And we learned that the main
uncertainty is the way we solve a

differential equations. I would probably
have told you what a differential equation

is in particular, but that would have
taken maybe another lecture. Climate

change is not proven throughout repeating
one real experiment over and over again.

So there is only one earth it is said. But
models simulate our

past climate pretty, well based
on physical laws that were proven in real

experiments. And then maybe the most
important message. Human behavior

is the primary source of climate change.
Therefore, we talk about projections and

not predictions. Meaning if we wanted to
predict the climate, then we needed to

simulate all human minds. And what we will
decide on future. But we don't. That would

be another talk again. We take what humans
will decide in future as an input

scenario, and with these input scenarios
we create different output scenarios. So

with different inputs scenarios, we get
these different output scenarios. Where we

can tell, OK, when we behave like that,
this is the output. And human behavior

scenarios dominate model uncertainties,
meaning the question is what do we want?

And if you go to a demonstration, the
answer is usually climate justice. And I

think that's a good answer. Thank you.

postroll music

subtitles created by c3subtitles.de
in the year 2020. Join, and help us!