Constraining a climate model with ocean reanalyses for the initialisation of decadal predictions

Climate prediction at interannual to decadal time scales is mainly an initial value problem. In addition, the memory of the climate system lies mostly in the ocean. Therefore, the initialisation of coupled models is a crucial point for the quality of such predictions, which are of high interest both for the society and for climate change assessment, as shown by the IPCC/CMIP5 requirements for the so-called u201cshort term projectionu201d experiments. One characteristic of coupled models is that their climatological behaviour is substantially different, in a statistical sense, from the real world climate. One clear consequence is that their reaction to an initialisation procedure at a close-to-the-real-climate situation will be to drift towards a state compatible with their own climate attractor. The time scale adjustment involved here is of the order of one to several years, of the order of the lead time looked for. Hence, alternatives to that method, which is extensively used for seasonal prediction, and which has been shown to be efficient despite a posterior systematic error correction, must be explored. One reason is that it is not insured that the systematic error is stationary enough to be properly estimated over the limited sample we are able to verify against with the current observed and reanalysed period (e.g. over ERA40) at our disposal, covering at most five decades as of now. In this poster, we show a series of sensitivity experiments with the APREGE-NEMO CM5 coupled model, developed at CNRM and CERFACS. These experiments involve a full field initialisation method, and an anomaly initialisation method, in coupled mode, and several tests in both coupled and forced mode, used to carefully check the influence of constraining the tropical oceans. A particular focus will be on the influence of the time varying ocean observation system. The reference observed/reanalysed ocean state is the most recent ECMWF reanalysis using the NEMOVAR system. Observation estimates of upper ocean heat content as well as long term sea level tide gauge and satellite measurements and reconstructions are the test bench used in these sensitivity experiments.

up arrow