The interpretation and use of biases in decadal climate 1 predictions

5 Decadal climate predictions exhibit large biases, which are often subtracted and forgotten. 6 However, understanding the causes of bias is essential to guide efforts to improve prediction 7 systems, and may offer additional benefits. Here we investigate the origins of biases in decadal 8 predictions, and whether analysis of these biases might provide useful information. We focus 9 especially on the lead time dependent bias tendency. We initially develop a ’toy’ model of 10 a prediction system and use it to show that there are several distinct contributions to bias 11 tendency. Contributions from sampling of internal variability and a start-time dependent 12 forcing bias can be estimated and removed to obtain a much improved estimate of the true 13 bias tendency, which can provide information about errors in the underlying model and/or 14 errors in the specification of forcings. We argue that it is the true bias tendency, not the 15 total bias tendency, that should be used to adjust decadal forecasts. 16 We apply the methods developed to decadal hindcasts of global mean temperature made 17 using the HadCM3 climate model, and find that this model exhibits a small positive bias 18 tendency in the ensemble mean. When considering different model versions we show that the 19 true bias tendency is very highly correlated with both the Transient Climate Response (TCR) 20 and non-greenhouse gas forcing trends, and can therefore be used to obtain observationally 21 constrained estimates of these relevant physical quantities. 22