Assessing the uncertainties in regional climate predictions of the 20th and 21th century

1
Assessing the uncertainties in regional climate
predictions of the 20th and 21th century

• Andreas Hense
• Meteorologisches Institut
• Universität Bonn

2
Overview

• The problem
• Climate system and climate models as random
  systems
• The consequences of randomness
• Estimation of randomness at various levels
• Predictability of forced climate variations
• Comparison of simulations with observations
• The conclusions

3
The problem the climate system as a random system
4
The problem the climate system as a random system

• Due to the high dimensionality 10 32 degrees
  of freedom statistical physics
• Due to the nonlinearities in the atmosphere,
  ocean and the interactions dynamical systems
  theory

5
The problem continued climate models as random
systems

• Due to high dimensionality 10 8 degrees of
  freedom
• Due to nonlinearities in the model atmospheres,
  oceans and interactions
• Due to parametrized subgrid scale processes
  (clouds, rain, convection etc..)
• Due to model errors

6
The consequences Estimation of randomness

• From the real climate system
• one observation / realisation available
• randomness has to be modelled
• e.g. assuming ergodicity, probabilities by
  counting, frequentists approach
• bayesian approach, modelling by probability
  densities
• ... more at the end

7
The consequences Estimation of randomness

• In models by Monte Carlo simulations, sampling
  the uncertainties in initial conditions,
  parameters, models

Initial conditions
8
The consequences estimation of randomness
Sampling models
9
The consequences predictability of forced
climate variations

• Forced variations Greenhouse gases, solar
  forcing, volcanoes
• overlaid by random variations
• in models
• in reality
• Forced variations gt random variations ?
• Predictability of the 2nd kind
• In models Analysis-of-Variance
• on specified space and time scales

10
ECHAM3/LSG HadCM2
11
ECHAM3/LSG HadCM2
12
(No Transcript)
13
The Bayes Theorem
14
The consequences comparison of simulations with
observations, Bayesian Classification
(Attribution)
15
A Bayesian attribution experiment

• ECHAM3/LSG 1880-1979 Control
• ECHAM3/LSG in 2000 Scenario
• NCEP Reanalysis Data 1958-1999 Observations
• Northern hemisphere area averages
• near surface (2m) Temperature
• 70 hPa Temperature
• joint work with Seung-Ki Min, Heiko Paeth and
  Won-Tae Kwon

16
(No Transcript)
17
Conclusions

• Inherent uncertainty in the climate system
• due to the chaotic nature
• strong dependance on space and time scales and
  type of variable
• annual temperature on a regional scale 70
  predictable
• annual sum of precipitation on a regional scale
  20
• decadal sum of precipiation 70

18
Conclusion

• Uncertainty introduced by model errors are large
  on the regional scale
• Uncertainty introduced by randomized
  parametrizations not yet explored
• Despite of all uncertainties climate change
  signals on the global / hemispheric scale can be
  detected
• Uncertainty has to be quantified as additional
  input for impact studies, meta-information
• scales in space, time and variable have to be
  selected from the discipline