As I have noted from time to time, these interactions are frequently used to
tune hindcasts so that a better representation of the past response of the Earth's climate systems is obtained.
I understand the argument that past projections are based on estimated future forcings which can change, but this amounts to the same things as
tuning hindcasts and declaring matching a hindcast to observations as a validation of your model.
Not exact matches
Just because models have been
tuned to
hindcast, can we assume they can forecast as well?
Let's say that we are given the results of calculations using the original Lorenz 1963 system and that these represent the data that we're going to use to
tune up our model of the data in a
hindcast exercise.
In a more general manner, if we take
tuning through
hindcasting to be some kind of (weak) form of proper parameter estimation, what is the theoretical basis for parameter estimation within the framework of chaotic response; especially spatio - temporal chaos?
If the models show a lack of skill and need
tuning with respect to predicting (in
hindcast) even the current climate statistics on multi-decadal time scales (much less than CHANGES in climate statistics), they are not ready to be used as robust projection tools for the coming decades.
This is surely what is happening in
hindcasts where aerosol forcing appears to me to be fine
tuned to fit the data.
The models are the weakest point of climate science, and until they are validated this estimate of temperature rise is not possible (and using
hindcasting with
tuned models doesn't have any credibility in my book)
If you intend to use your model to attribute observed warming to human forcing, then it is inappropriate and unethical to
tune your model so that observed warming equals
hindcast warming.
Those parameter sets are fine
tuned by
hindcasting.
Truth be told, most warming in
hindcast is
tuned in because the models can not reproduce the warming with just model physics, therefor future warming in model output is
tuned in
These parameters are
tuned to best
hindcast; for CMIP5 explicity from YE2005 back to 1975.
e.g., take HALF the date to
tune the model, then see how well it forecasts /
hindcasts on the other half of the data.
Unavoidable Parameterization
tuned to best
hindcast, which introduces the general attribution problem lurking in AR4 WG1 SPM figure 8.2.
Since the data are historical, the analysis here is essentially that of a
hindcast, and since some of these data may have been used during model construction and
tuning, it is debatable to what extent they can be considered to provide validation of the models.