C2-WP2 - Prediction of large regional climate anomalies

Project aims

The general goal of this work package is to quantify the capability of the regional climate model COSMO-CLM to provide reliable decadal forecasts for Europe, and to determine how far and why the forecasts vary on decadal time scales. The aim is thus to uncover the processes underlying this predictability. We analyse the predictive skill of the MiKlip hindcasts (in collaboration with C3-WP3) and identify the physical processes that lead to predictive skill over Europe.

For this task, we follow two different routes: the first one concentrates on the statistics of the regional predictive skill depending on the large-scale teleconnection pattern. The second one focuses on large climate anomalies related to an enhanced occurrence of extreme events like heat waves, droughts, floods and windstorms in the target region because of their high impact and importance to the society. Both approaches complement each other.

Project structure

Two groups Deutscher Wetterdienst (DWD) and Karlsruhe Institute of Technology (KIT) cooperate in work package C2-WP2.

Tasks of the project

T1 Conditional evaluation of the hindcasts

The regional decadal hindcasts are stratified according to the phase of the relevant large-scale teleconnection pattern (e.g. the North-Atlantic Oscillation (NAO, etc.) and their impact on the predictive skill analysed (contribution of DWD).

T2 Predictability of large climate anomalies

Within this topic three types of extremes are analysed: Heat waves, heavy precipitation and wind storms. The decadal variability and frequency of these extremes as well as their relation to the teleconnection pattern are analysed to estimate the predictability of such climate anomalies (contribution of KIT).


D1a Methods for investigating the stratified hindcasts
D1b Results of the stratified evaluations
D2a Method for investigating the large climate anomalies
D2b Results of the large climate anomaly analysis

Progress so far

  • A method to assess the stratified skill of the decadal hindcasts (c.f. C3-WP3) w.r.t teleconnection pattern has been developed

  • The (multi-)decadal variability and its relation to teleconnection pattern has been analysed for different types of extremes (in collaboration with C2-WP1 and C2-WP3)

    • Heat waves

    • Heavy precipitation potentially causing floods

    • Wind storms

Stratified Evaluation

For the stratified evaluation, data classes based on terciles of various large scale circulation indices are generated. This means, that e.g. the global time-series of the NAO index is used in order to determine negative, neutral and positive NAO phases and data of all variables to be analysed (temperature, precipitation, wind etc.) are split into the corresponding three classes of equal size regarding to their NAO phase. Since the circulation indices obtained from the modeled data and the circulation index calculated from reanalysis would only be equal for a perfect forecast, there are still two choices within each class: Cases, where the indices do not agree (different phase), and cases, where they do (same phase).

When structured regarding to the NAO, the latter cases (same phase) generally show a clearly reduced Root Mean Square Error (RMSE) and a higher prediction skill (Mean Square Error Skill Score, MSESS) for all lead years considered. The improvements, however, differ temporally, spatially and depending on the variable. While, e.g., there is an appropriate surface temperature signal for the first lead year largely within the Scandinavian region, this signal shifts towards the British Isles for the following lead years (2-5). Thus, the improvements of the prediction for the NAO refer mainly to characteristic centers of action. There is neither a monotone (spatially equal) nor a constant (temporally equal) enhancement in case of a correctly predicted circulation phase.

Heat Waves

A percentile based definition of heat waves (HW) is used to calculate HW days using daily maximum temperatures (cf. Fig. 1 left). Temporal and spatial characteristics of the HWs across Europe were analysed to determine their duration, location and intensity.

Fig.1 left: maximum HW temperatures across Europe for E-OBS 1960-2010, right: Correlation of HW days between the CCLM b1 Hindcasts lead-years 2-5 and E-OBS 1960-2010 recalibrated via the CALIBRaTION tool (Module E, CALIBRSTION) and evaluated with the MiKlip CES plug-in MURCSS (Module D, INTEGRATION)

A large year-to-year variability were found for the spatial distribution of the HW, which often occurred in cohesive patterns across wide areas, indicating a dependence on circulation anomalies (CA). Figure 1 right shows the correlation of the 4-year mean values of the HW-days between the recalibrated regional baseline1 (b1) hindcasts and the E-OBS observations. For Central and Southern Europe high correlations between the hindcast and observations were found.

To analyse the connection between the HW index and the CAs monthly sums of HW days are used to calculate correlations between them and the monthly indices of the teleconnection patterns (defined after Barnston and Livezey, 1987) e.g. the North Atlantic Oscillation (NAO), the Pacific Transition (PT), the East Atlantic (EA), the Polar (POL) or the Scandinavian pattern (SCAND) for Europe and different PRUDENCE regions (Fig. 2; cf. Fig. 1 work package C3-WP3 for the location of the PRUDENCE regions).

To capture the different phases of the CAs, the indices were separated into positive and negative phases. Correlations of up to r ≤ ±0.5 for certain regions and phases have been found with large variations for regions and index phase. Partly vice versa signals occurred between In the different phases of the indices the signals partly switch their sign, but sometimes the correlations do not react symmetric in opposite phases.

Thus, only for certain regions and patterns clear dependences between LCAs and HW days are found.

Fig. 2: Correlationof monthly HW day sums to positive phase of the CA indices for Europe and the PRUDENCE regions for E-OBS from 1961-2010
Heavy Precipitation

In this section, heavy precipitation events, as an example for large climate anomalies, will be discussed concerning the link to teleconnection patterns and their predictability.

First of all we analysed how the baseline1 hindcast ensemble represents different teleconnection patterns. Focussing on Europe, we choose the North Atlantic Oscillation (NAO), the East Atlantic (EA), the East Atlantic/Western Russia (EAWR) and the Scandinavian Pattern (SCAND) as defined by Barnston and Livezey (1987). We found that in the baseline1 hindcasts the first lead-years provide positive scores concerning time-correlation, mean square errors and probabilistic forecast-skill. Concerning large climate anomalies, we defined a percentile based index, which captures heavy precipitation events potentially causing floods. This index consists of field-means over the main European river catchments (from west to east: Rhine, Elbe, Oder, Vistula and Danube, c.f. Fig. 3) and 7 day running means of daily precipitation data.

Fig. 3: E-OBS winter precipitation. European main river catchments depicted in colours: Rhine (yellow), Elbe (purple), Oder (orange), Vistula (green), and Danube (red).

Combining this index with the chosen teleconnection patterns, we found a clear relation between heavy precipitation events in winter and teleconnection patterns which changes over the catchments respectively over longitudes (c.f. Fig. 4).

The index is positively correlated to the three Atlantic patterns (EA, EAWR, NAO) for the western river catchments, but anti-correlated for the eastern. For the Scandinavian pattern, which is basically a southward shifted NAO pattern, the correlation trend between east and west switches its sign. From this we conclude that in the positive phase of the three Atlantic patterns, the potential for heavy precipitation events is increased in the Western catchments. In the Eastern catchments it is higher for the positive phase of the Scandinavian pattern. For the intermediate catchment of the Oder no clear tendency was found.

Fig. 4: Time-correlation between precipitation index (based on 95th percentile and per catchment) for winter-months DJF and depicted teleconnection patterns.

The long-term variability of extreme indices and their teleconnection to the Atlantic Multi-decadal Oscillation index (AMO) has been studied using the centennial downscaling experiments from work package C2-WP3, which cover the period since 1900. Several precipitation indices indicate a higher intensity in AMO+ phases. These findings still have to be evaluated carefully, since few reliable observational data are available before the mid of the 20th century.

Additionally, the regionalised and recalibrated (method developed within MiKlip Module E RECALIBRATION by Pasternack et. al 2017) baseline1 hindcast ensemble was analysed in terms of high percentiles of 7 day running mean precipitation. We found large areas of positive skill respectively correlation over Europe, again for the early lead-years (c.f. Fig. 5).

Overall we found that the teleconnection pattern provide an indicator for heavy precipitation events potentially causing floods, which changes from West to East. Moreover the high percentiles of precipitation shows positive skill for the early lead-years.

Fig. 5: Time-correlation between baseline1 precipitation lead-years 2-5 (7 day running mean, 95th percentile) and observation E-OBS
Wind storms

Wind storms and phases with increased storm activity for several European regions (PRUDENCE) are analysed using a meteorological storm severity index (MI). This index depends on the exceedance of the 98th percentile of daily maximum surface wind speeds and the affected area. Within this work package, we focus on the number of MI days per extended winter season (ONDJFM) and its correlation with different teleconnection patterns. For CCLM-ERA, results show that the MI (and therefore the storm activity over Europe) is related to the occurrence of certain teleconnection patterns (Fig. 6).

Fig. 6: Correlation between number of MI-days per ONDJFM and different teleconnection patterns for the individual PRUDENCE regions for CCLM-ERA


Deutscher Wetterdienst - Climate and Environment Consultancy
Dr. Barbara Früh
+49 (0)69 8062-2968

Deutscher Wetterdienst - Climate and Environment Consultancy
Dr. Sascha Brand
+49 (0)69 8062-2967

Institute for Meteorology and Climate Research (IMK-TRO) Karlsruhe Institute of Technology (KIT)
Hendrik Feldmann
+49 (0)721 608-24942

Institute for Meteorology and Climate Research (IMK-TRO) Karlsruhe Institute of Technology (KIT)
Natalie Laube
+49 (0)721 608-22844

Institute for Meteorology and Climate Research (IMK-TRO) Karlsruhe Institute of Technology (KIT)
Julia Mömken
+49 (0)721 608-22805

Institute for Meteorology and Climate Research (IMK-TRO) Karlsruhe Institute of Technology (KIT)
Prof. Dr. Joaquim Pinto
+49 (0)721 608-28467

Institute for Meteorology and Climate Research (IMK-TRO) Karlsruhe Institute of Technology (KIT)
Benjamin Buldmann
+49 (0)721 608-24942