Climate variability from annual to multimillenial timescales: Insights from statistical and conceptual models

Abstract

A concept for the physical understanding of insolation-driven temperature variability on orbital timescales is developed. It relies on the observed annual cycle of temperature to estimate the climate's sensitivity to local insolation at different seasons. Based on this concept, the temperature evolution of the last 750 ky related to local insolation forcing is estimated. The seasonal templatemodel largely reproduces the Holocene temperature trends as simulated by a coupled climate model. It predicts significant temperature variability in the eccentricity and semipreccession frequency band in the tropics and indicatesthat the temperature response to local insolation is highly spatially dependent. In explaining observed climate variability, the local time-independent approach complements the global Milankovitch hypothesis (climate variations are driven by northern summer insolation) and potentially offers new insights in interpreting paleoclimate records.The interpretation of the proxy records is complicated by the seasonality inherent in many climate proxies. As this interferes with the seasonal temperature signal, the true climate response has to be isolated from the recordereffect when interpreting the proxy data. This is demonstrated on the example of Antarctic temperature records. Reconstructed Antarctic temperatures from ice cores are nearly in phase with the boreal summer insolation but out of phase with the local summer insolation. Therefore, Antarctic climate is often thought to be driven by northern summer insolation although a clearmechanism that links the two hemispheres on this timescale is missing. Here, an alternative hypothesis is proposed that key Antarctic temperature records derived from ice cores are biased towards austral winter because of a seasonal cycle in snow accumulation. Using present day estimates of this bias in the 'recorder' system, it is shown that the local insolation can explain the temperaturerecord without having to invoke a link to the Northern Hemisphere.For global climate reconstructions the relationship between the local records and the large-scale climate has to be known. Using a coupled climate model, surface temperature teleconnections are examined on pre-industrial and glacialclimate states. An increase in spatial correlations and significant changes in the teleconnection pattern during glacial times related to enhanced wave activity in the Northern Hemisphere are detected. It is concluded that present-day teleconnections cannot be used in the interpretation of glacial climate as they are used in temperature reconstruction of the last 150 years.Together with an ensemble of climate change simulations these temperature datasets are the basis for an interannual temperature prediction scheme proposed in this thesis. By calibrating the multi-model ensemble mean with the observed climate state, skilful interannual forecasts of mean temperatures can be obtained. The method is validated using extensive hindcast experiments and compares favorably with both statistical predictions and predictions fromdynamical models with assimilated initial conditions.