On timescales of several years to several decades, weather and climate patterns depend not only on the anthropogenic rise of greenhouse gas concentrations in the atmosphere, but also on natural climate variability. Natural climate variability is both internal – ultimately induced by weather in both atmosphere and ocean – and forced – caused by solar variability and volcanic eruptions. Owing to the chaotic nature of the internal variability, its predictability horizon is limited. But the clear signal of decadal-timescale variability in almost all climate records suggests that, if we are able to observe the phase and amplitude of the current decadal variability “event”, we can exploit the memory of the climate system to predict the further evolution of this event. Strategically, FLEXFORDEC stands out as a pivotal project within MiKlip: it is the key project for the MiKlip Module D “Synthesis”, one of the five cornerstones of MiKlip; it comprises the development of the central prediction system, one of the key objectives of MiKlip; and it includes the overall coordination of MIKlip. The scientific synthesis of the MiKlip project has both prediction research and infrastructure components. Decadal climate prediction research is an emergent field of climate science, so substantial research progress is required to improve prediction skill. Because we show by construction which elements of the climate system are predictable, FLECFORDEC build the most state-of-the-art prediction system. The prediction system will continuously be improved by incorporating the research progress across all of MiKlip, according to the development stages (DS) 1-3, where during DS1 we employ the prediction system currently being used for the WCRP CMIP5/IPCC AR5 simulations. Effective incorporation into the prediction system used in DS2 and DS3 requires an infrastructure that gives us a large degree of flexibility, so that we can test various suggestions arising from other MiKlip Modules for improving decadal climate predictions.
Based on the prediction system used for the CMIP5 set of experiments, we successively form an ensemble prediction system (EPS) for global-scale decadal climate variability. This system considers a technically complex data assimilation architecture comparable to the one provided by NWP and by seasonal climate prediction. This system will also consider scientifically complex questions such as the inclusion of relevant climate system components (e.g., land surface and sea-ice), and the generation of a sufficiently large ensemble. For these reasons we embed the EPS in a well-designed prediction system and model environment infrastructure.
During the course of MiKlip a succession of central prediction systems will be developed (one in each of the development stages DS1 to DS3), employing successively more sophisticated process representation. For each system we will perform sets of hindcasts and forecasts, control runs and 20th-century experiments with the model of highest possible resolution (for example, T63L95/TP04 during DS1). Integrating novel work into the system will naturally occur predominantly in DS2 and DS3.
An integrated system for the standardised evaluation of experiment output will be implemented in DS1 and will in DS2/DS3 be further developed. This allows comparability of forecast skills and systems performance for different configurations of the prediction system, thereby supporting its development process and allowing reproducible decisions regarding extensions and innovations to the system. Furthermore the utilization of an online post processing module and a data flow and management system as an integrated part of the forecast system will assure prompt access to model output data and efficient usage of computer resources.
Relevant simulation data from the global ensemble forecasting system will be provided to all participants in MiKlip and to potential users. Therefor an extensive documentation and inclusion of a metadata system will be applied to guarantee transparency and straightforward usage of the system. An integrated tool for analysis and visualization enables deeper analysis of model data and related forecasts.
Managing and coordinating a research project as large and complex as MiKlip is a challenge in itself, because the information flow between the Modules must be organised and monitored. In particular the flow of suggestions from Modules A and B to the central prediction system and the flow of data from the central prediction system to Modules C and D must be synchronised. FLEXFORDEC therefore establish a MiKlip office supporting the MiKlip Coordinator in the science management. The MiKlip office will take care of all overarching organisational work, such as all-hands seminars (“Statusseminare”), reporting, and collecting deliverables.
Scientifically the most important work by the MiKlip Office is the organisation of the assessment of progress in the central prediction system. This assessment must occur in conjunction with the all-hands seminar at the end of DS and then at a mid-term self-evaluation in the form of a progress report at month 30 in DS2. This timing is chosen such that a cut-off at the end of DS2 can be performed the case of insufficient progress. The final all-hands meeting and the final report, both organised by the MiKlip Office, assess whether the MiKlip system is ready for operational use. Moreover smooth data dissemination within MiKlip and potential outside partners requires a proven data storage infrastructure such as provided by the German Climate Computing Centre (DKRZ). Within MiKlip a central data server is hosted by the DKRZ to buffer and disseminate results from the central prediction system and to make the observation readily available.
This description regards the project during the first phase of MiKlip. FLEXFORDEC continues in MiKlip II, in Module D WP1 - FLEXFORDEC/INTEGRATION.
Max-Planck-Institut für Meteorologie
Prof. Dr. Jochem Marotzke
Dr. Wolfgang Müller
Dr. Holger Pohlmann
Dr. Freja Vamborg