Abstract
Decadal Climate Predictions (DCP) have gained considerable attention for their potential utility in promoting optimised plans of adaptation to climate change and variability. Their effective applicability to a targeted problem is nevertheless conditional on a detailed evaluation of their ability to simulate the near-term climate evolution under specific conditions. Here we explore the performance of the IPSL-CM5A-LR DCP system in predicting air temperature over Europe, by proposing a systematic assessessment of the prediction skill for different time windows (periods of the calendar time, forecast years and months/seasons). In this framework, we also compare raw and de-biased hindcasts, in which the temperature outputs have been corrected using a quantile matching method. The systematic analysis allows to discern certain conditions conferring larger predictability, which we find to be intermittent in time. The predictions appear more skilful around the 1960s and after the 1980s, in coincidence with large shifts of the Atlantic Multidecadal Variability, which are well reproduced in the hindcasts. Averages on longer forecast periods also generally imply better prediction skill, while the best predicted months appear to be mainly those between late spring and early autumn. Moreover, we find an overall added value due to initialisation, while de-biased predictions significantly outperform raw predictions only for a few specific time windows. Finally, we discuss the potential implications of the proposed systematic exploration of skill opportunities in DCPs for integrated applications in climate sensitive sectors.
Highlights
One of the biggest scientific challenge for the 21st century concerns the capacity of simulating the future climate evolution through numerical models (Dutton, 2002)
To establish a reference point in our systematic analysis, we evaluated the skill of the IPSL-CM5A-Low Resolution (LR) model in predicting air temperature over Europe for the reference context
Non-initialised historical simulations (Figs. 1a,1d) exhibit only limited skill concentrated over the Mediterranean sector, the anomaly correlation coefficient (ACC) is not statistically significant at the 95% level, as for the rest of Europe, which is characterised by both low ACC and relatively high root mean square error (RMSE)
Summary
One of the biggest scientific challenge for the 21st century concerns the capacity of simulating the future climate evolution through numerical models (Dutton, 2002). Let us consider the tourism sector: a climate service for ski resorts in the Alpine regions should demonstrate skilful predictions of snow fall during the winter in mountains areas, while a climate service for seaside activities in the Mediterranean region should rather demonstrate skilful prediction of temperature and precipitation in coastal areas during summer This implies that forecasts based on the same DCP system may show a wide range of confidence when applied to these very different scopes, as the prediction skill considerably depends on a multitude of contingent factors like the relevant climatic variable, the specific season, the particular period, and the region under investigation. The Copernicus Climate Change Services (https://climate.copernicus.eu/) will soon include operational decadal predictions implemented by different Institutes, which will possibly allow for an optimal selection of model experiments as a function of the specific study
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
