Degree of Certainty

A major task of climate modelers is to estimate the degree of certainty within a model. This lets us know what the range of possible outcomes will be, if a model predicts a rise of 1°C in a certain scenario, for example. The following methods are ways to estimate the degree of certainty in models, so that we can be more certain what climate conditions to expect in the future.

Ensembles:    An ensemble consists of many simulations of the same region over the same time by the same model. By starting the simulations with initial conditions that are very slightly different, one may sometimes obtain a result that differs quite a bit from “most of the pack.” However, most of the results will cluster within a range of values. Scientists have more confidence that the real-world value will fall within this range.
This ensemble approach, by the way, is now a preferred tool of weather forecasters.

Multi-model ensembles:     Since models differ in their handling of climate feedbacks, one might obtain more confidence by collecting results of many different models for the same region over the same time period. Each model contributes its own ensemble of simulations.  Again, one looks for a range of values in which most of the predictions are found; and to be sure, these predictions should come from many different models, not just a few.

When model results are verified against reality, the mean value of many simulations from many models consistently performs better than the values from any one model.

Perturbed physics:     Observations of some property of the environment (such as sea-surface temperature) do not always agree, because of errors in measuring the property. Therefore, it is reasonable to assume that the parameters of a model should not have a fixed value, but should vary within a reasonable range.

In this approach, we change some of the physical parameters slightly (not just at the initial time), and run the model to obtain an ensemble of results. Again, we place more confidence in the values that are predicted more often.

A group of experts.  This approach is not favored by scientists, who seek to quantify their level of certainty.

Precautions:    Although the methods above allow us to reject outlying projections and accept something approaching a consensus, there are certain barriers to obtaining more certainty about projections of future climate:

• the future forcing of climate may not be knowable;   forcing factors that today are not important, may become important.
• Different climate models today may share common errors. These errors may not show up in simulations of present or past climates (in which modeling results can be verified against reality). These errors are real, but not yet known. Some of these errors will lead to erroneous predictions.