Project aims

The aim of the project OPERATIONS is an operating system with an extensive degree of automatisation of the decadal prediction based on the MPI-ESM model on the high-performance computer of the German Meteorological Service (DWD). During the first phase of Miklip simulations were performed on the high-performance computing facility of the German Climate Computing Center (DKRZ). The plan of MiKlip II is to compute the decadal predictions at DWD in the final stage of the second phase of MiKlip.

Progress made by other modules and projects will be incorporated into the operational system and the optimal model configuration for the operational system will be selected in close co-operation with the Max Planck Institute in Hamburg (MPI-M). The model system MPI-ESM and the evaluation software FREVA, developed in the first phase of MiKlip, will be updated at the system of the DWD in co-operation with the Freie Universität Berlin (FUB).

Following the example of the operational weather forecast system, a workflow environment for the decadal prediction system will be established. This workflow environment is provided by the use of the software “ecFlow”; it facilitates routine decadal predictions and the evaluation of the model output. “ecFlow” is a task scheduling software, which is developed and used by the European Centre for Medium Range Weather Forecasts (ECMWF).

Project Structure

Three groups (Deutscher Wetterdienst (DWD), Max Planck Institute for Meteorology (MPI-M) and Freie Universität Berlin (FUB)) co-operate in Workpackage D2-WP2.1. The topic of this work package is the transfer of the global decadal prediction system und of the evaluation system to the systems of the DWD.

Tasks of the project

The tasks of the project can be outlined in two topics. The first half of the project duration (DS4) is dedicated to the transfer of the decadal forecast system to the infrastructure of the DWD. The workflow, the computational and storage demand are evaluated by setting up a “toy”-workflow with the global model. During the second half of the project duration the “toy” workflow is replaced by the agreed coupled climate model setup. This system will be capable of producing initial conditions for the global model in real time, in order to provide decadal predictions starting every year. The evaluation system will be finalised and the results will be supplied.

Summary Goals

• Porting of the model system MPI-ESM and updating the evaluation tool FREVA to the infrastructure of the DWD
• Establish a highly efficient workflow environment following the example set by the operational weather forecast system with the scheduling software ecFlow, developed and used at the ECMWF
• The work-flow environment carries out the different tasks, such as providing initial conditions (via data assimilation) and boundary conditions, the model integration, archiving procedures and evaluation.
• Optimisation of the prediction ensemble
• Visualisation and presentation of the output data
• Investigating the dependence of the prediction skill on the specific  time of the initialisation of the decadal predictions

All tasks require a close collaboration with other work packages in module D.

Progress so far

The successful porting and testing of the model MPI-ESM to the High Performance Computer of the DWD is completed for both high and low (HR and LR) resolutions. Benchmark tests have been successfully repeated with the higher resolved model showing a similar behaviour in decomposition efficiency, as compared to the lower resolution model. The operational work-flow has been established and tested in the low resolution (preop-LR) re-forecast experiment. The results of the preop-LR experiment have been shared with our project partners, and they were used as the base for the 2018-2027 decadal prediction, as published on the forecast website. Currently, the next generation of decadal simulations, in high resolution with CMIP6 forcing (preop-HR-CMIP6), is under-way using the same work-flow. The ten member ensemble will be calculated one half each at MPI-M in Hamburg and DWD.

Predictive skill on decadal time scales has been researched in the preop experiments by looking at one specific process in the coupled atmospheric-oceanic system, the Pacific Decadal Oscillation (PDO). Both preop-LR and preop-HR-CMIP5 show high correlations and therefore good skill in predicting the state of the PDO after one year. But, in both experiments the skill deteriorates quickly with increasing lead time. After lead year three, the correlation is statistically indistinguishable from zero. This result is in accordance with other studies within the decadal prediction community.