A-Coordination2 is concerned with the coordination of Module A and with performing research within the module. The main objectives of the coordination activity is to ensure close collaborations and linkages between different WPs within Module A, and between Module A and other modules, especially Module D, thereby assuring the immediate relevance of results obtained in Module A for the development of a final MiKlip prediction system. The main goals of A-Coordination2 research activities are:
The objectives of A-Coordination2 are subdivided into three WPs:
WP1.1: Estimating initial conditions and developing a mode-initialisation method.
WP1.2: Testing the flux correction method.
WP1.3: Developing a coupled data assimilation approach.
WP1.1: Creating initial conditions through the global GECCO approach will continue, while taking into account all available ocean and sea ice information. Testing a new climate mode initialisation method is subdivided into the following tasks: estimating climate modes for MPI-ESM using empirical orthogonal function analysis; filtering out components from the synthesis/reanalysis that do not correspond to climate modes represented by MPI-ESM which may thus lead to unbalanced states; carrying out the assimilation run to produce the initial conditions for the initialised decadal hindcasts that will be performed and compared with the hindcasts from Module D.
WP1.2: For testing the full state initialisation with a momentum and buoyancy flux correction scheme, we plan the following: construction of flux correction terms from repeating the historical run with nudging of SST, SSS to the values from reanalyses, following the strategy described by Polkova et al. (2014a,b); carrying out the assimilation run and the initialised hindcasts with employing the diagnosed flux correction terms. Finally, we plan to analyse the predictive skill and make a comparison with the skill for the Prototype hindcasts (GECCO2, full field initialisation) using MiKlip Central Evaluation System (CES).
WP1.3: We will perform assimilation experiments with the CESAM model to test the efficiency of the approach described by Abarbanel et al. (2010) to improve model parameters by assimilating data in coupled climate models. The objective is twofold; first, we will test how the improved consistency of the model climate with the data and with model-consistent initial conditions influences the forecast skill of a coupled climate model on various space and time scales. Second, we aim at producing a coupled synthesis that can provide model-consistent initial conditions for decadal predictions. Although the synthesis may be at first build on a simpler and coarser resolution climate model than what is currently state-of-the-art and used as the MiKlip prediction system, hindcasts initialised with this optimised model fields will provide insight into how model-consistent initialisation affects the forecast skill.
The following deliverables are planned for A-Coordination2:
WP1.1: Updates of GECCO2 are being performed on a regular basis and are available through https://icdc.cen.uni-hamburg.de/1/daten/reanalysis-ocean/gecco2.html
In order to minimise initialisation shocks resulting from inconsistent grids between the synthesis used for initialisation and the model used for the prediction, we are currently working on a higher resolution version of the ocean synthesis based on a configuration that will be compatible with the ocean component of the high-resolution set-up of the MPI-ESM model (MPI-HR). The configuration has been finished and control runs without data assimilation have been performed. After preparing the data for the assimilation procedure, the synthesis is now in the stage of iteratively adjusting the state to find a model solution consistent with the assimilated data. The ocean synthesis (GECCO3) will cover the years 1948-2017 and will also provide optimised mixing coefficients.
Based on the Earth System Model from the Max Planck Institute for Meteorology (MPI-ESM) we designed a climate-mode initialisation method for the decadal prediction system MiKlip. The idea of the initialisation method is in improving the ESM’s prediction skill through the initialisation of balanced components of the initial conditions and filtering out components inconsistent with the dynamics of climate model. To this end, the ocean states from the ORAS4 reanalysis are projected onto modes of variability derived from an ensemble of historical simulations performed with the MPI-ESM. The climate modes are calculated as statistical modes based on the bivariate empirical orthogonal function (EOF) analysis. The explained standard deviation in the filtered reanalysis amounts to 66%. As this value is somewhat lower than what we expected, we assume that modes of variability of the reanalysis are not exactly compatible with the modes from the prediction system or that they are not yet sufficiently sampled by the available data used to construct the EOFs. The analysis of filtered and original reanalysis anomalies shows that the signal over the whole Pacific basin is well captured and represented, while in the Southern Ocean and the North Atlantic large fraction of the signal is filtered out. The filtered reanalysis' anomalies are then added to the model’s climatology and are used as initial states for a set of retrospective decadal predictions. The climate-mode initialisation method is compared against the commonly used anomaly initialisation method, which is implemented in the MiKlip pre-operational prediction system (Preop-LR).
A comparison of the climate-mode initialisation with Preop-LR indicates an improved surface temperature skill over the tropical Pacific Ocean at seasonal-to-interannual time-scales in terms of accuracy and correlation with the observations. This result shows a potential for improving seasonal forecasts of the El-Nino Southern Oscillation. For the 2-5 lead years averages, the climate-mode initialisation method outperforms the skill of the anomaly initialisation for the surface temperature as well as the upper ocean heat content over the central and northern Pacific Ocean. For the North Atlantic sub-polar gyre region, the climate-mode initialisation experiments rather resemble a trend of the historical simulations than that of ORAS4 or the observations. Also, they show smaller amplitudes of variability as compared to the non-filtered initialisation. This suggests the need to further improve the design of the climate-mode initialisation method attempting to capture better the variability modes in the North Atlantic in a larger EOF-basis or/and using different weighting in combination with regional modes to better determine variability structures in regions of interest.
This and the other initialisation and ensemble generation methods developed within the MiKlip-II project have been compared in a coordinated inter-comparison study (contributions encompass results from A-Coordination2 and AODA-PENG2 from Module A, ATMOS-MODINI from Module B and FLEXFORDEC from Module D). The main results of this inter-comparison led to a manuscript by Polkova et al. (2018, submitted to JAMES). The inter-comparison shows that the tested methods provide an added value for the prediction skill as compared to Preop-LR in that they improve prediction skill over the eastern and central Pacific and different regions in the North Atlantic Ocean. In this respect, the Ensemble Kalman Filter (AODA-PENG2 from Module A) and the climate-mode initialisation show the most distinct improvements over Preop-LR for surface temperatures and upper ocean heat content, followed by the other methods.
WP1.3:
In cooperation with colleagues from the University of Hamburg, the research for this work-package was carried out in the two following directions:
(i) The research on ad-joint sensitivities with CESAM resulted in the study by Stammer et al., 2017. The study compares modelled and observed fields for sea level pressure, wind, temperature and radiative fluxes and suggests that the intermediate complexity CESAM is comparable in quality to the CMIP-type coupled models. The ad-joint sensitivities are computed to provide patterns of ocean anomalies that cause a maximum air temperature response in northern Europe for the time-scale of one month. The study analysed climate ad-joint sensitivities to establish a reference against which longer-term climate experiments can be compared.
(ii) Parameter estimation and extension of the assimilation window: Producing a coupled data assimilation product for applications in the MiKlip system appears to be challenging due to the limitation related to the chaotic nature of the atmospheric component of the climate models. However, concerning retrieving parameters within an assimilation window over climatological relevant time scales, progress has a better perspective. Tuning climate models via data assimilation over longer assimilation windows became possible with the implementation of the regularisation scheme for the 4D-VAR method (synchronisation with observations; Abarbanel et al., 2010). The regularisation scheme has been implemented in the atmospheric component of CESAM, which equips the model with the restoring terms toward observations to suppress growth rates of atmospheric instabilities. The study by Lyu et al., 2018, describes in detail experiments based on assimilating model-observations (twin experiments) and ERA-Interim reanalysis. The advantage of this method is that it allows extending the assimilation window with minimal additional computational cost.
Institut für Meereskunde, Universität Hamburg
Prof. Dr. Detlef Stammer
Polkova, I. | Köhl, A., Stammer, D.
Köhl, A. | Vlasenko, A.
Polkova, I | Brune, S., Kadow, C., Romanova, V., Gollan, G., Baehr, J., Glowienka-Hense, R., Greatbatch, R.J., Hense, A., Illing, S., Köhl, A., Kröger, J., Müller, W.A., Pankatz, K., Stammer, D.
D. Stammer | A. Köhl, A. Vlasenko, I. Matei, F. Lunkeit, and S. Schubert
Guokun Lyu | Armin Köhl, Ion Matei, Detlef Stammer