Quantifying the Uncertainties in an Ensemble of Decadal Climate Predictions

Abstract

Meaningful climate predictions must be accompanied by their corresponding range of uncertainty. Quantifying the uncertainties is nontrivial, and different methods have been suggested and used in the past. Here we propose a method that does not rely on any assumptions regarding the distribution of the ensemble member predictions. The method is tested using the Coupled Model Intercomparison Project Phase 5 1981–2010 decadal predictions and is shown to perform better than two other methods considered here. The improved estimate of the uncertainties is of great importance both for practical use and for better assessing the significance of the effects seen in theoretical studies.

Learning algorithms allow for improved reliability and accuracy of global mean surface temperature projections

“…members 7,8 and the relation between the ensemble spread and the uncertainties associated with their predictions 9 . Using an ensemble of CMIP5 long-term climate projections that was weighted according to a sequential learning algorithm and whose spread was linked to the range of past measurements, we found considerably reduced uncertainty ranges for the projected Global Mean Surface Temperature (GMST).…”

“…Recently, a new method for the quantification of the uncertainties associated with ensemble predictions was suggested 9 . The method is based on studying the relation between the spread of the ensemble member predictions (quantified by the ensemble standard deviation (STD)) and the ensemble root mean squared error (RMSE).…”

“…Obviously, this approach requires simulations of past conditions, which allow the calculation of the RMSE. The most general method of those suggested 9 is the Asymmetric Range (AR) method, which relies only on the assumption that the relation between the ensemble spread and the error does not change significantly with time (i.e., the relation found during the learning period remains the same during the prediction/projection period). The prediction of an ensemble is its weighted average (or the simple mean of the ensemble if equally weighted).…”

“…The AR method has the advantage of estimating independently the range of likely conditions above the mean and the range of conditions below the mean (in this sense, it is asymmetric). The AR method was shown to provide much more reliable predictions (relative to the standard method) for the ensemble of CMIP5 decadal predictions of surface temperature and surface zonal wind 9 . The improvement of the reliability demonstrated the validity of the assumption that the relation between the ensemble spread and its error does not vary considerably for decadal predictions.…”

“…The weighting affects not only the ensemble mean (the projection of the EGA forecaster) but also the ensemble's STD, which is often used to quantify its spread and the uncertainties associated with the projection. In the second stage of the analysis, the relation between the weighted STD and the projection errors was established using the AR method 9 in order to estimate the range of likely GMST values at different periods. The AR method calculates a pair of time-independent correction factors for each desired confidence level (confidence level c implies that the extreme higher and lower tails of the probability distribution, each with a probability (1 − c)/2, are excluded).…”