SPRUCE aim is to attempt to improve our capacity to predict climate variations six months ahead, by combining our best present tools and data with a high performance computation capacity which is not yet available in our production machines but is offered by PRACE Tier0 platforms. The two modelling aspects we aim to improve in SPRUCE are thus horizontal resolution and ensemble size.

Climate models, which simulate the joint evolution of the global atmosphere and ocean beyond the limit of deterministic atmospheric predictability (one to two weeks), have been developed since the 1990's, driven for a large part by increasing offer of the numerical computation hardware. They have been used to predict the statistical properties of the atmosphere few months ahead. Although substantial progress has been made in the past, the current performance of climate models at seasonal to decadal scale is still not sufficient to meet the expectations and needs of the various stakeholders at European, regional and local levels. Nevertheless, reliable seasonal-to-decadal climate predictions are of strong potential value, since society and key economic sectors (energy, agriculture, ...) have to base their short and medium term planning and decisions on robust climate information and the associated environmental and socio-economic impacts.

Horizontal resolution has always been one of the major limiting factor in climate modelling. At coarser resolution than 0.5°, the mountain pattern is unrealistic and lower atmosphere winds may have a wrong direction on average. In the ocean, high resolution is required to represent the eddies which transport heat from equator to poles. Many climate studies have shown the benefits of increasing horizontal resolution on the mean simulated climate and its variability. The standard CNRM-CMIP5 Météo-France model uses a 1.6° resolution for the atmosphere and 1° for the ocean. SPRUCE proposes to increase this resolution to respectively 0.5° and 0.25°, to bring a significant jump in the seasonal predictability.

The second aspect of the improvement in SPRUCE is the ensemble size. Due to the chaotic nature of the atmosphere at monthly to seasonal scale, a seasonal prediction is necessarily probabilistic. A single realization of the forthcoming months has little chance, even on time average, to resemble the observed behaviour. In the mid latitude, very recent results on northern mid latitude winter predictability suggest that increasing the ensemble size to 60 leads to a significant improvement. A size of 120 is necessary at SPRUCE model resolution.

The predictability evaluation is based on a series of re-forecasts, or hindcasts, covering the past years, starting from the 0.25° ocean reanalysis GLORYS, during the 1993-2009 period. The exploitation of the results will concern first the winter mid-latitude regimes (e.g. NAO) and local predictability of temperature over Europe. We will then examine the predictability of summer heat waves over Europe and North America, with a focus on 2003 summer. Our results will contribute to define a strategy for operational seasonal forecasting in Europe.