Climatological observations seem to indicate that at middle and high latitudes there exist climate oscillations with periods of several decades. The origin of these oscillations may lie in the internal dynamics of the climate sytem, especially in the coupled interacion between the atmosphere and the ocean. However, the available observations do not allow for a clear identification of the physical mechanism that may give rise to those oscillations. The analysis of long-integrations of coupled atmosphere-ocean models may give clues about the real mechanism that are operating in nature. Since the amount of data generated in this integration is enormous, sophisticated statistical techniques to identify oscillations such data sets have been used, such as Singular Spectrum Analysis and Principal Oscillation Patterns.

Possible climate change is currently estimated with numerical models that simulate the present and the future earth climate. Their skillful resolution is nowadays of the order 2000 Km and therefore they cannot properly the local and regional features that are important to study the impact of climate change on the environment. One way to overcome this scale mismatch is statistical downscaling techniques: A statistical transfer function is estimated by analyzing observations of the large-scale circulation and the regional climate. This transfer function can be used to translate the changes of the atmospheric circulation simulated by a climate model to changes of the regional climate. The transfer functions may be designed by means of multivariate linear thecniques, such as canonical correlation, or by nonlinear techniques, such as analogue methods, classification methods or neural networks.