Upload
ioan-muntean
View
98
Download
0
Tags:
Embed Size (px)
Citation preview
1
FROM MODELS TO MECHANISMS. FEEDBACK AND OPTIMIZATION IN CMIP5
IOAN MUNTEAN
THE REILLY CENTER FOR SCIENCE, TECHNOLOGY, AND VALUES
UNIVERSITY OF NOTRE DAME
2
PREVIEW
Main issue: A transition from models to mechanisms in climate change
Argument for a mechanistic view in climate change
Feedback
Optimality
Control/ manipulation
Understanding
Arguments against mechanisms in climate science
Holism
Failed mechanisms in physical sciences
So what?
Not yet
Will never happen
Not needed
3
WHAT IS UNDER SCRUTINY HERE?
The internal structure of climate models
Feedback in climate models
Mapping models to mechanism (M2M)
Many to one?
Many to many?
Optimality and plurality of models
Communicating results, metadata and mechanisms
4
SOME TOPICS OF INTEREST IN CLIMATE SCIENCE
Social values (viz. Epistemic values) in creating models (Winsberg 2012)
Complexity of models and “analytic understanding” (Parker 2014)
Multiplicity/plurality of climate models, (Parker 2010a)
Uncertainty of models
Stability, reliability of models
Explanatory power and understanding of climate models
Modularity
Adapted from (Knutti and Sedláček 2012)
5
ROLES OF MODELS
Climate change is mainly about building and assessing models
Climate models are mainly:
predicting tools
generate other models or hypotheses
Quantification of theories of climate change
“hybrid”: predict and explain
Do climate models really explain? How?
Do we have an IBE with climate models?
6
OVERLAPPING MODELS IN MECHANISMS
Differenct communities focus on different parts
They do not necessarily look at the “coupled model”
Climate scientists are specialized
7
SOME VIEWS ABOUT MODELS AND MECHANISMS
Models are not related to mechanisms
mathematical models exist in physics, without being related to any mechanism
some models summarize data (phenomenal models)
some other models predict (are phenomenally adequate) but do not explain
Models represent mechanisms
One task of model building is to represent the dynamics of mechanisms (Bechtel and Abrahamsen 2011)
Models needs mechanisms to be explanatory
Models are explanatory when they describe a mechanism (Craver 2006)
Models map to mechanisms (M2M)
Let us call these models “mechanistic models”
8
MODEL ASSESSMENT IN CLIMATE SCIENCE
Confirmation of “the truth” of existing models (Lloyd 2010)
Adequacy-for-a-purpose: (Parker 2009)
Realism: accurate description of the actual climate system
Bayesian view
Possibilism (Katzav 2014)
Present focus: mapping models to a mechanism
How does model X map on the mechanism Y?
9
THREE EXTREME PREDICTIONS
A. Where do I need to look in the sky to find the moon in London ON at 16:30 on 25.10.2044?
B. What will be Ioan’s state of health on 4:30 on 25.10.2044, given this and this constraints on the world and what we know of his diet, genes, etc.?
C. How far can I drive a Honda Civic car from London ON with a full tank of gas, in this direction, in the weather conditions, all things being equal?
A= one theory, simple simulation, simple data, perfect prediction
B= no theory no model, some mechanisms
C= one mechanism, no theory, some initial conditions, no need of models
10
QUESTIONS 1 Are some (all?) climate models mechanistic?
2 Why explanation?
3 Can climate models explain without being “mechanistic”?
What advantages does a mechanistic view bring to climate science?
4 So what? Why do we need mechanistic explanation anyway?
1 yes, those in which feedback plays a role
2 We want to understand the “causal story” of the climate system. The understanding of why a phenomenon occurs (Parker 2014).
Question to Parker: is a mechanistic explanation better than causal explanation in improving our understanding of a phenomenon and of its question “why?”
3 yes, they can, but still mechanistic explanations can do better
4 Because with explanation, control, understanding and manipulation come!
4’ we can hope for the optimal model
11
EXTRAPOLATING MECHANISMS
Universality: Model-building occurs anywhere in science
Neuroscience/cognitive science (empirical data and laws/equations)
Biology (empirical data)
Physics (laws, symmetries)
Life science, medicine
Scientific revolution can be read, charitably as a process building models, mechanisms, unifying, eliminating models, creating theories etc.
I think it makes sense to talk about:
“mechanisms & models (together) in climate science”
12
A SIMPLIFIED VIEW
I. Convergence from a plurality of models to a limited number of models
Culling models
Coupling submodels
Constraint models
II. Mapping from a limited number of models to a limited number of mechanisms
III. Convergence of mechanisms to a theory (unification of mechanisms and models)
I think II deserves attention in the light of CMIP5
I am quietist about III. And I is already discussed in the literature
13
ADVANTAGES OF MECHANISMS
Introduce new explanations
Integrate causal stories
Introduce levels
Facilitate communication between submodels and between subroutines
Can map elements of models to mechanisms and give them materiality
Cluster of different models into mechanisms based on the M2M
Move from statistical explanations/arguments close to what the layman wants to hear (not probabilities, but conditionals)
14
“BOTTOMING OUT” MECHANISMS
Ignore the fundamental and fundamentality (deep physics)
Work at scales
Relative to a scale (time space energy)
Multiscale modeling
15
FROM MODELS TO MECHANISMS
Why do we need mechanisms? A Kantian innuendo:
“Dynamical models without mechanistic grounding are empty, while mechanistic models without complex dynamics are blind.” (Bechtel and Abrahamsen 2011)
This suggests a relation among models and mechanisms.
Normatively: models and mechanism should be mapped one onto the other.
16
DO MODELS EXPLAIN?
The Craver-Kaplan hypothesis (Kaplan and Craver 2011):
Models explain only when there is a model-to-mechanism mapping. M2M
Models needs to be modular in order to explain (Weber 2008)
17
THE MECHANISM-MODEL MAPPING
Biologists discover mechanisms
Models resemble the mechanism
Some models are better, some are worse, in representing the mechanism
18
MECHANISMS IN MODELS’ CLOTHING
Are mechanisms already in the climate science?
Try to identify in CMIP-5 the mechanistic mindset (not language)
Unveil their explanatory role
Explain the M2M mapping.
19
SYNTHETIC MODELING
Mechanism complements the computational modeling
It is not a question of reinterpretation of what climate scientists already do
It is more or less a reconstruction based on M2M
It does bring in a clearly stated language of levels
Cycles of amplification are called amplifying mechanisms
20
MECHANISTIC OPERATIONS IN THE MODELS
Decomposition is a procedure that happens in mechanisms
Switch on and off various components: Inhibition
Stimulation
Recomposition of the operation of the mechanisms (Bechtel, 2011)
21
CLIMATE FEEDBACK 101
Feedback is never linear
Apply a forcing (CO2)
Temperature raise
Feedback changes
Look for mechanisms that are not switched off al low temperature
Once these processes go on, there is amplification or reducing of the temperature
22
FEEDBACK
Feedback can be positive or negative
The net feedback from the combined effect of changes in water vapor, and differences between atmospheric and surface warming is almost surely positive.
The net radiative feedback due to all cloud types combined is positive.
23
CMIP-5: A LOLLAPALOOZA OF FEEDBACK
AOGCM are not enough!
Earth System Models
Earth System Models of Intermediate Complexity
Includes cycles
Since AR4, the understanding of mechanisms and feedbacks of extreme in temperature improved
25
FEEDBACK: M2M
Feedback can be captured by:
Non-linear equations
A cycle in a mechanism
Simple mechanisms are serial: start to finish. They contain only linear causal chains
Feedback loops complicate mechanisms.
They are non-sequential
Introduce timescale
Synchronization of feedback (makes them positive or negative, depending on phase factor)
27
CARBON CYCLE IN CMIP5 AND FEEDBACK
Increased atmospheric CO2 increases land and ocean uptake
Limitations on plant growth imposed by nitrogen availability
28
VAPOUR-CO2-CLIMATE
Vapour is a feedback not a Forcing of climate change
It is a fast and strong feedback (see Ch 8)
29
HOW DO WE REACH OPTIMALITY?
Optimality does not belong to a model
Through mechanisms (Machamer, Darden, and Craver 2000)
Optimal mappings between models and mechanisms
Reduce uncertainty
30
TIMESCALE MATTERS!
The effect of feedbacks is clear for longer timespans
Some feedbacks are delayed by centuries or millennia
31
Lifetime (years) GWP20 GWP100 GTP20 GTP100
CH b 412.4a No cc fb 84 28 67 4
With cc fb 86 34 70 11
HFC-134a 13.4 No cc fb 3710 1300 3050 201
With cc fb 3790 1550 3170 530
CFC-11 45.0 No cc fb 6900 4660 6890 2340
With cc fb 7020 5350 7080 3490
N2O 121.0a No cc fb 264 265 277 234
With cc fb 268 298 284 297
CF4 50,000.0 No cc fb 4880 6630 5270 8040
With cc fb 4950 7350 5400 9560
32
ARGUMENTS AGAINST M2M IN CLIMATE SCIENCE Climate science is a physical science in which mechanisms do
not play the same role as in neuroscience/life science/ Some “disastrous” examples of mechanism thought in physics (ether,
phlogiston, Cartesian physics)
Climate models are mathematical models, unlike models in neuroscience
Climate science is holistic, in pursue of complexity, not reductionist. Emergence looms large
Climate science is more about statistical reasoning, not about discovering reality/mechanisms.
Climate modelers are partially blackboxing, or probably grey-boxing their object of study