CLIMATE CHALLENGES IN HIGHPERFORMANCE COMPUTING

1
CLIMATE CHALLENGES IN HIGH-PERFORMANCE COMPUTING

• Jean-Claude ANDRE
• CERFACS, Toulouse (France)
• andre_at_cerfacs.fr
• http//www.cerfacs.fr

2
The 2 directions for model evolution
Resolution
Complexity
3
for answering a series of societal and
scientific questions

• The cloud problem and the climate sensitivity
• Extreme events and tropical cyclones
• Oceanic intake of CO2
• Climate change and ozone depletion
• Earth-system modelling
• Probabilistic aspects
• Climate surprises
• . . . .

4
Toward more realism and higher accuracy1. The
cloud problem

• What is the sensitivity of climate to an increase
  in GHG, with particular concern to feedbacks
  involving the hydrological cycle?
• How is it possible to reduce the factor 3
  uncertainty which presently affects predictions ?
  

5
Toward more realism and higher accuracy1. The
cloud problem
Oklahoma July 27-31, 1997
Errors for 1d parameterized models
Water vapour (g/kg)
Errors for cloud-resolving models
Mean errors
Rms errors
EUROCS project
6
Toward more realism and higher accuracy1. The
cloud problem
100 km
30km
15 km
7
Toward more realism and higher accuracy1. The
cloud problem
8
Toward more realism and higher accuracy1. The
cloud problem

• Toward solving the cloud problem, i.e. non
  hydrostatic with a 3 km horizontal grid size
• ? 10 Petaflops (sustained)

9
Toward more realism and higher accuracy2. The
future of extreme events?

• What type of evolution for dangerous extreme
  events such as heavy precipitation, wind storms,
  tropical cyclones, ... ?

10
B2 Scenario
Present (1970-99)
Future (2090-99)
120 trajectories
109 trajectories
10-year simulation Grid 0.5x0.5 (50km) B2
Scenario SST CNRM
11
A2 Scenario
30-year simulation Stretched Grid A2 Scenario SST
Hadley Centre
Future (2070-99)
285 trajectories ? 95 tr/10 year
Present (1970-99)
377 trajectories ? 125 tr/10 year
12
Precipitation (mm/day)
ARPEGE-Climat 50 km
Down to ?x20km
13
Toward more realism and higher accuracy2. The
future of extreme events?

• ? 10 Teraflops (sustained)

14
Linking climate evolution to other environmental
issues 3. Oceanic intake of CO2

• Will ocean continue to absorb the same fraction
  of the released CO2 ?
• ? Accounting for the full carbon cycle, and
  consequently for oceanic biogeochemistry

15
Linking climate evolution to other environmental
issues 3. Oceanic intake of CO2
Vorticity
3d fine-scale structures
Vertical entrainment
Phytoplankton
Planktonic bloom
Absorption of atmospheric CO2
OPA model (6 km) Levy Klein 2004
16
Linking climate evolution to other environmental
issues 3. Oceanic intake of CO2

• Relative vorticity field near the surface
• (ES30002000200)?
• Filaments with large
• vorticity values
• Associated with horizontal
• density gradients.

17
Linking climate evolution to other environmental
issues 3. Oceanic intake of CO2
18
Linking climate evolution to other environmental
issues 3. Oceanic intake of CO2

• ? Requires resolution of the sub-meso-scale 2 km
  (to be consistent with observed budgets)
• ? 100 Teraflops (sustained)

19
Linking climate evolution to other environmental
issues 4. Climate and stratospheric O3

• What are the links between climate change and the
  future of ozone depletion  ?

20
Linking climate evolution to other environmental
issues 4. Climate and stratospheric O3
SIMULATION
OBSERVATION
Total 3 column in the southern hemisphere during
an event where coupling between chemistry and
dynamics is important. Simulation at low
horizontal and vertical resolution seems quite
realistic, but fails to reproduce all the
smaller-scale features and consequently does not
allow for an acurate quantification of ozone
destruction and corresponding impacts on middle
latitudes.
21
Linking climate evolution to other environmental
issues 4. Climate and stratospheric O3

• ? Requires a high-resolution (both vertically and
  horizontally) of the upper atmosphere)
• ? 100 Teraflops (sustained)

22
Linking climate evolution to other environmental
issues 5. Earth-system modelling

• What are the physical (sub-)systems, other than
  atmosphere and ocean, which influence the future
  of the planet ?

23
Linking climate evolution to other environmental
issues 5. Earth-system modelling
24
Linking climate evolution to other environmental
issues 5. Earth-system modelling

• ? 10 Petaflops (sustained)

25
Quality assessment and probabilities6.
Probabilistic aspects

• What are the probabilities linked with the
  various uncertainties ?

26
Quality assessment and probabilities6.
Probabilistic aspects
Same model, different parameters for some
parameterizations
Unperturbed model Sensitivity 3.4 K
Low sensitivity 2.5 K
High sensitivity 10.5 K
From Stainforth et al., 2005 (Nature)
27
Quality assessment and probabilities6.
Probabilistic aspects

• Detection signal-to-noise ratio problem,
  anthropogenic effects (signal) vs. natural
  variability (noise)
• The later requires control simulations of a few
  thousands of years (observations too short and
  possibly influenced by anthropogenic signal)
• Attribution detecting changes for all
  combinations of possible forcings, and estimating
  the linear responses
• Three main sources for uncertainty
• GHG emission scenarios
• Natural variability
• Representation of the climate system by
  numerical models

28
Quality assessment and probabilities6.
Probabilistic aspects

• Need for a probabilistic approach
• ? Will allow for progress in climate change
  detection and prediction
• ? 10 Teraflops (sustained)

29
Requirements for climate issues
0zone layer
Climate surprise
Earth system modelling
Oceanic CO2
Extreme events
Climate sensitivity Cloud problem
Land-use change
Tropical systems
Probability aspects
10T
100T
1P
10P
100P
1T
100G
Sustained perfomance
30
Requirements for climate issues
31
Computing power is not the only condition

• Storage

32
Storage requirements

• Short term (on site)
• - Substantial online storage to insure
  efficient utilisation of the system
• - Local off line storage for handling outputs and
  inputs ( at least 6 to 12 months)
• - Local storage 1 to 10 Pb

33
Storage requirements

• Long term (possibly off site)
• Need easy access to a common distributed archive
  (fast network requirement)
• Database system based on data-grid technology
• Storage for climate could be gt 100 Pb
• National institutions use data-grid for
  visualisation, analyses of the data

34
Computing power is not the only condition

• Storage
• Parallelism

35
Parallelism

• Parallelism for each (sub-)system
• Parallel assembling of parallel (sub-)systems
• Various levels for parallelism
• Parallelism for shared memory (OpenMP)
• Parallelism for distributed memory (MPI)
• Processor distribution
• Code coupling

36
Computing power is not the only condition

• Storage
• Parallelism
• Programming

37
Programming

• Embedding regional models within global models
• Revisiting interfaces between sub-systems
• Benchmarking and running demonstration
  simulations

38
Thanks

• Acknowledged help and contributions from
• Pascale BRACONNOT, Daniel CARIOLLE, Marie-Alice
  FOUJOLS, Patrice KLEIN,
• Jean-François ROYER, Sylvie JOUSSAUME, Laurent
  TERRAY,
• and a few others

39
(No Transcript)
40
Thank you
41
Spare
42
Uncertainties in Climate Change Predictions2.
Precipitation
43
(No Transcript)
44
(No Transcript)
45
(No Transcript)
46
(No Transcript)
47
Uncertainties in Climate Change Predictions3.
Carbon cycle
48
(No Transcript)
49
Climate modelsThe three-dimensional coupled
approach
50
Toward more realism and higher accuracy3.
Climate surprise ?

• Is it possible to quantify the risk of "climatic
  surprises" linked to oceanic instabilities (e.g.
  a sudden weakening of the Gulf Stream) ?
• ? Longer simulations (a few centuries), with high
  spatial resolution (oceanic convection), repeated
  a very large number of times (probabilistic
  aspects)
• ? 1-10 Petaflops (sustained)

51
Linking climate evolution to other environmental
issues 8. Land-cover change

• How to evaluate the climatic consequences of
  land-use changes and of urban development  ?

52
Linking climate evolution to other environmental
issues 8. Land-cover change
Land use
Significant changes in surface physical
properties and vegetation cycles

• ? Population growth
• ? Global evolution of farming practices

Significant regional and global impacts on
climate, especially on hydrological cycles ?
? Relevant climate simulations must be performed
with various land-use scenarios and
sophisticated SVATs
53
Linking climate evolution to other environmental
issues 8. Land-cover change
Urban areas
Rapid development of megacities with ? drastic
change of surface properties ?emission of
pollutants
Severe impacts on atmospheric flow and
composition at local scales
Significant regional and global impacts on
climateand environment ?
Mesoscale atmospheric flow simulated around Paris
on a summer anticyclonic day
54
Linking climate evolution to other environmental
issues 8. Land-cover change

• ? Relevant climate simulations must be performed
  at horizontal resolution finer than 5 km
• ? 1-10 Petaflops (sustained)

55
Quality assessment and probabilities9. Quality
control of climate models

• Data assimilation, as for numerical weather
  prediction (and ocean modelling)

56
Quality assessment and probabilities9. Quality
control of climate models
t
5C
-5C
57
Quality assessment and probabilities9. Quality
control of climate models
The MERCATOR programme
58
Quality assessment and probabilities10.
Probabilistic aspects
Climate sensitivity at 2xCO2
Red 127 30-year simulations (IPCC-type), Met
Office
Black 2579 15-year simulations
(climateprediction.net)
59
A way forward
2008
2013
2018
ENES Europe
2005
2020
France
2003
2018
2023
0zone layer
Japon, USA
Climate surprise
Earth system modelling
Oceanic CO2
Extreme events
Climate sensitivity Cloud problem
Tropical systems
Land-use change
Probability aspects
10T
100T
1P
10P
100P
1T
100G
Puissance efficace
60
European computing needs for climate simulations
61
Societal issues and scientific questions
62
Where are we in Europe ?
The Earth Simulator Inter-connected 640-node NEC
SX-6s (1node 8 vector processors)
63
Where are we in Europe ?

• France
• IDRIS 30 of 320/100 Gflops
• CEA 35 of 350/110 Gflops
• Météo-France 30 of 1190/380 Gflops
• (S France 0,2 TFe)
• Great Britain 50 of 1920/610 Glops (0,3 TFe)
• Germany 100 of 1530/490 Gflops (0,5 TFe)
• S Europe 1 TFe
• Japan 50 of 40960/13116 Gflops (4 TFe)
• USA amounting to 4-5 TFe

64
Where are we in Europe ?
65
Where are we in Europe ?