It is generally recognized that a correct initialization can improve climate predictions up to a few years ahead. However, the systematic error of the models makes the initialization process challenging. When the observed information is transferred to the model at the initialization time, the discrepancy between the observed and model mean climate causes the model to drift away from the real-world attractor towards its own biased attractor. To account for such a bias, a-posteriori bias correction needs to be applied. This, although introduces additional errors in the forecast, is unavoidable for the forecasts to be usable. On top of the correction of the bias, decadal predictions also have the additional challenge of disentangling the smaller magnitude of climate signal to be predicted, from the initial drift to be removed. Previous studies have shown the comparison of different initialization strategies designed to reduce the initial drift. The results have shown improvements limited to specific regions, that could even be model dependent. In this study, we have designed an innovative initialization technique that aims at reducing the initial drift by scaling the observed amplitude of the initial data with the model state distribution. This method consists of performing a quantile matching between the observed state at the initialization time step and the model state distribution. Multi-annual climate predictions with the EC-Earth v3.3.1 global model have been initialized with this innovative methodology. We will show the impact of the quantile matching initialization technique on the prediction skill at multi-annual time scale. These results will be compared with the non-initialized historical simulations from CMIP6 and to the Decadal Climate Prediction Project experiments initialized with standard techniques, carried out with the same model version.
An innovative initialization technique for decadal climate predictions
Volpi Danila;Meccia Virna;Corti Susanna
2019
Abstract
It is generally recognized that a correct initialization can improve climate predictions up to a few years ahead. However, the systematic error of the models makes the initialization process challenging. When the observed information is transferred to the model at the initialization time, the discrepancy between the observed and model mean climate causes the model to drift away from the real-world attractor towards its own biased attractor. To account for such a bias, a-posteriori bias correction needs to be applied. This, although introduces additional errors in the forecast, is unavoidable for the forecasts to be usable. On top of the correction of the bias, decadal predictions also have the additional challenge of disentangling the smaller magnitude of climate signal to be predicted, from the initial drift to be removed. Previous studies have shown the comparison of different initialization strategies designed to reduce the initial drift. The results have shown improvements limited to specific regions, that could even be model dependent. In this study, we have designed an innovative initialization technique that aims at reducing the initial drift by scaling the observed amplitude of the initial data with the model state distribution. This method consists of performing a quantile matching between the observed state at the initialization time step and the model state distribution. Multi-annual climate predictions with the EC-Earth v3.3.1 global model have been initialized with this innovative methodology. We will show the impact of the quantile matching initialization technique on the prediction skill at multi-annual time scale. These results will be compared with the non-initialized historical simulations from CMIP6 and to the Decadal Climate Prediction Project experiments initialized with standard techniques, carried out with the same model version.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
