C3-WP3 - Ensemble Generation

Project aims

The goal of work package 3 (WP3), “Ensemble Generation” of objective C3 “Optimising regional ensembles” is to produce a comprehensive set of dynamically downscaled regional decadal hindcast simulations from which recommendations for the optimum operational ensemble can be derived. For this purpose, the already existing ensemble from the first phase of MiKlip (baseline1 CCLM4.8, 0.44° resolution) was completed covering all annual starting years and all members. Additional MiKlip II hindcast generations (preop, dcppA) were generated with an improved model version and a higher resolution (0.22°). Furthermore, simulations with a different land surface scheme and an perturbed physics approach (Breil and Schädler, 2017) were tested. The simulations were performed with the regional climate model (RCM) COSMO-CLM (CCLM). They focus on the EURO CORDEX region (see Figure 1) and cover up to now the staring-years 1960 – 2018. The full MiKlip hindcast period from 1960 was considered in order to account for inter-decadal variability and to assess the robustness of the analyses. The skill and added value of the regional hindcasts have been assessed in Mieruch et al. (2014), Reyers et al. (2017) and Feldmann et al. (2018).

Project structure

The IMK-TRO/KIT and the German Weather Service ( DWD) share the work.

Tasks of the project

Task C3-3.1: Generation of the basic regional hindcast ensemble

Extend the existing decadal ensembles a) by downscaling the MPI-ESM decadal hindcasts for an extended ensemble at a horizontal grid spacing of 0.22° (initial conditions perturbation) to analyse the impact of inter-decadal variability on the ensemble metrics, and b) by using advanced perturbed physics methods based on those developed in MiKlip I, in order to achieve a balanced spread.

Task C3-3.2: Provide ensemble members at high resolution
For selected decades and realisations high-resolution simulations at 0.0625° will be performed, followed by an analysis of the added value of the higher resolution and the impact on the ensemble metrics, especially for high/low quantiles.

Task C3-3.3: Provide ensemble members using a different SVAT and/or coupled ocean
In cooperation with the work package C1-WP2, “Soil vegetation processes”, simulations using the soil-vegetation-atmosphere transfer scheme (SVAT) Veg3d instead of the standard scheme TERRA will be performed for selected decades and realisations. Furthermore, coupled ocean simulations provided by C1-WP1, “Regional ocean coupling”, will be integrated in the ensemble. The impact of the additional members on the ensemble metrics, especially for high/low quantiles, will be analysed.

Task C3-3.4: Very high-resolution simulations

Depending on the overall progress of this work package and the available computing power, very high-resolution simulations (0.0025°) will be carried out for a selected subregion of Central Europe, and their added value will be assessed.

Task C3-3.5: Overall ensemble assessment, analysis and optimisation

This is the synthesis task of work package C3-WP3. The various ensemble compositions are analysed over several decades to assess the inter-decadal and intra-ensemble variability in terms of ensemble size and quality criteria and. The aim is to derive recommendations for an optimum ensemble size and composition taking into account also technical and computing efficiency issues. This work will be done in cooperation with work packages C3-WP1, “Post-processing”, and C3-WP2, “Optimized ensemble characteristics”.
 

Deliverables

  • comprehensive decadal ensemble matrix at a spatial resolution of 0.22°
  • provision of ensemble members for selected decades at high (0.0625°) and very high (0.0025°) spatial resolutions
  • provision of ensemble members for selected decades based on alternative SAVT and/or coupled ocean models
  • synthesis of tasks 3.1 to 3.4 in order to derive recommendations for an optimum ensemble size and composition

Progress so far

  • The efficiency of the ensemble generation has been improved to allow for the production of full hindcast sets of decadal predictions(about 60 starting years. 10-years/simulation ~6000 simulation years)

  • Three generations of regional decadal hindcasts were provided: baseline1 (completion of the MiKlip I ensemble, 0.44° resolution), and two using the higher resolved MiKlip II setup – using MPI-ES_HR forcing and the CCLM5 version with 0.22° resolution – namely: preop and the dcppA-hindcasts (CMIP6).

  • It could be shown that the higher resolution and updated model version reduces the model bias and provides an added value w.r.t predictive skill.

  • Robust estimates of the skill and added value are given

  • A recommendation is given for an operational use of the regional hindcast system, which balances the skill and the computational efforts. This recommended version uses the MiKlip II setup.

  • The application of post-processing methods (Module E CALIBRATION or Module C C2-WP2) increases the skill of decadal hindcasts further, allowing the application of the results for user oriented variables (C2-WP1) and extremes (C2-WP2).

  • The regional forecasts are now part of the MiKlip decadal forecast web-page.

C3-WP3 Fig1.
Fig. 1: Model domain for the decadal CCLM simulations within MiKlip II and its orography. Red boxes represent the location of 8 sub-regions (PRUDENCE) generally been used for standard analyses.

Figure 2 shows the temperature correlation of the regional preop hindcasts with observations. The skill is significant over the full domain. It is generally higher in Southern and Western Europe than in parts of Eastern and Northern Europe. This is a robust feature of all MiKlip hindcast generations.

Fig. 2: Correlation between near-surface temperature for the regional preop hindcasts (5 members) and the CRU TS4.01 observations over Europe for the period 1967 – 2016. Dots denote significant skill scores.

An added value of regional downscaling was found compared to the forcing data from the global model as well as an added value of initialised hindcasts compared to un-initialised simulations (Figure 3). Positive values of MSESS indicate a skill; a perfect agreement would have a skill score of 1. All three ensembles show mostly positive values. Compared to the un-initialised simulations (green bars) the MiKlip hindcasts (red, blue) are shifted towards a higher skill. This is especially true for the regional hindcasts (blue) which show a significant higher percentage of high skill scores.

Fig. 3: Distribution of the Mean Square Error Skill Score (MSESS) for near-surface temperature over Europe for the period 1967 – 2016 with the climatology as reference. Green: Ensemble of 10 un-initialised simulations with MPI-ESM-LR; red: MPI-ESM-LR baseline1 hindcasts (10 member, lead-years 2-5); blue: regional CCLM baseline1 hindcasts (10 member, lead-years 2-5). Observations: E-OBS

Contact

Karlsruher Institut für Technologie (KIT), Institut für Meteorologie und Klimaforschung – Bereich Troposphäre (IMK-TRO)
Hendrik Feldmann
hendrik.feldmann(at)nospamkit.edu
+49 (0)721 608 22802

Deutscher Wetterdienst - Climate and Environment Consultancy
Dr. Sascha Brand
sascha.brand(at)nospamdwd.de

Deutscher Wetterdienst - Klima- und Umweltberatung
Dr. Barbara Früh
Barbara.Frueh(at)nospamdwd.de
+49 (0)69 8062-2968