Operational predictability of monthly average maximum temperature over the Iberian Peninsula using DEMETER simulations and downscaling

Abstract

The multi-model ensemble for seasonal to interannual prediction developed in the European Union project DEMETER has been used to quantify the predictability of monthly average maximum temperature that could be achieved operationally over the Iberian Peninsula. Statistical downscaling based on canonical correlation analysis is applied to increase the spatial resolution available from the global models. The downscaling is based on empirical connections between the North Atlantic sea level pressure and monthly average maximum temperature over the Iberian Peninsula.

The maximum temperature estimated from the multi-model ensemble and the single models is compared to the observations. The statistical downscaling model skill is characterized by means of the correlation, variance fraction and the Brier skill score. The results suggest the following: the downscaling model works properly when driven by observed large-scale fields in terms of the correlation and the variance fraction scores, despite some problems owing to sample degeneracy; the predictability is almost limited to February, which is one of the initialization months of the DEMETER ensemble, and it is lost when this month is not considered as starting month. This result is supported by the fact that the areally averaged reproducibility is lower during non-initialization months. In any case, the analysis of the variance test performed reveals that the monthly average maximum temperature is scarcely predictable. Finally, the results also support the advantage of using a multi-model ensemble approach instead of single models participating in DEMETER.