the need for data/model integration
DESCRIPTION
THE NEED FOR DATA/MODEL INTEGRATION. Retrodiction/prediction is meaningless without meaningful error bars or probability distribution of model results Major challenge for climate change assessments - PowerPoint PPT PresentationTRANSCRIPT
THE NEED FOR DATA/MODEL INTEGRATION
Retrodiction/prediction is meaningless without meaningful error bars or probability distribution of model results
Major challenge for climate change assessments Are PMIP/observation discrepancies due to faulty boundary
conditions or to problems in the models? even dynamical process modelling needs constraints on
boundary conditions for under-constrained systems need data/physics
integration -> calibrate model against observational data
Criteria for calibration methodology
Complicated under-constrained non-linear system with threshold behavior
effectively large number of poorly constrained model parameters
Large set of diverse noisy constraint data Data and & model limitations -> need a fundamentally
probabilistic approach bumpy phase and likelihood spaces (shown below) further
rule out gradient-based approaches such as adjoint (eg. 4D var) methods
-> stochastic methodology accurate propagation of data uncertainties -> Bayesian
approach => Markov Chain Monte Carlo
Bayesian calibration
Sample over posterior probability distribution for the ensemble parameters given fits to observational data using Markov Chain Monte Carlo (MCMC) methods
Other constraints: Minimize margin forcing LGM ice volume bounds Hudson Bay glaciated at -
25 kyr Post MCMC scoring:
Marine Limits Strandlines
Large ensemble Bayesian calibration
Bayesian neural network integrates over weight space
Self-regularized Can handle local minima
It works
Bumpy likelihood space
Another data/model link: guide future data collection
North American Climate and meltwater phasing ->
meltwater/iceberg discharge
is a critical link between cryosphere and climate
system
The meltwater link Need more marine
observations to corroborate/refute results
What happens to a meltwater plume in the Arctic Ocean?
Mixing dynamics in the GIN Seas?
A couple of other interim results
The calibration tends to favour an ice volume for North America that is too low to meet global LGM eustatic constraints
-> can be addressed by strong H2/H1/mwp1-a events -> also starting to give consideration to larger marine
components (especially given recent HOTRAX data)
Where to:
Completion of interim global calibrated deglacial ice/meltwater chronology
EMIC/GCM recursion to get climatological self-consistency Calibration of glacial inception ice & climate with the
glacial systems model coupled to a reduced AOGCM Ditto for deglaciation -> forward in time: P(future cryospheric evolution) eventual calibration of full glacial cycle ice & climate