The goal of MiKlip Module C is to, for the first time, systematically regionalise decadal predictions. The downscaling is expected to support the supply of climate-information for relevant regions in a resolution that is required by many commercial users. Eight projects were involved in Module C during the first phase of MiKlip. The downscaling was achieved by decadal simulations with regional climate models (RCMs), which were driven with data from the global ensemble of MiKlip climate predictions of the MPI-ESM at the boundaries of the model region.
To evaluate the expected quality of a future prediction, simulations of past periods (here 1960-2010), so called “hindcasts”, were performed. The regionalisation of ensembles represents with an additional computational effort, which makes it essential to show improvement or at least preservation in skill through the application of regionalisation. The processes in the climate system, which contribute to decadal predictability, run at a global scale (approximately ocean currents or planetary waves in the atmosphere). Thus their origin is more often beyond the limited model space of the RCMs, and the information has to be transferred from the driving global-model across the regional model edges. However, smaller-scale processes plays an important role for regional climate and RCMs can better reproduce characteristics such as regional topography or state and classification of soil in the region. In Module C there are three main focus areas in which different regional and thematic aspects were investigated:
Examination of decadal climate variability has shown that the north Atlantic, compared to the rest of the globe has very high predictability on this time scale. Since the climate in Europe is strongly influenced by North-Atlantic, it should as a consequence be possible to find predictability over land.
Regionalization with dynamical RCMs
To use a regional prediction system efficiently, several aspects were considered: It can be pointed out, that RCMs in contrast to the global model have generally perceived and partially increased predictive skill in Europe and in many cases even better reproduced the climate parameters after regionalisation. In Figure 1, summer mean temperatures for the first five years are shown as example, of how well the observed climate variability in the period of 1960-2010 is represented by regional(left) and global(right) ensembles. The correlation of the two ensembles (top) with observations provides comparable results. Positive correlation can be found in most parts of Europe. The lower half shows “reliability”, a number for the confidence of the ensemble. The Metric expresses if the distribution of probabilities of a climate parameter is in balanced relation to the model error. Best values would therefore be zero, contrary to the correlation where higher values represent better results. Through regionalisation reliability is enhanced in many regions.
There were also studies on the impact of model resolution, the composition of ensembles with only one or various RCM as also the importance of ensemble-size. Potential added value of regionalization appears also in the prediction of user relevant climate-parameters such as heat waves, heavy precipitations or wind blasts.
Regionalisation using a statistical dynamic approach
Besides the direct dynamic downscaling, module C was also testing an approach for statistical dynamic downscaling (SDD), which, if optimised for certain climate parameters, enables a regionalisation with little effort. With SDD a certain predictability in regards to frequency of storms and the wind energy potential on a scale of a few years in advance can be reached (Reyers et al., 2015). Figure 2 demonstrates the prediction skill for the wind energy potential in central Europe for the period 1979 to 2010, using the second MiKlip ensemble generation (baseline 1). The prediction skill is defined via the ratio of mean squared error of the ensemble average (MSESS) of the hindcasts and non-initialised climate projections, which do not contain information of the observed starting condition. The hindcasts provide an additional value if the MSESS is positive (red colour), no added value is given with negative values (blue colours). Positive results can be found predominantly close to the coast lines of North and Baltic Sea for the 4 year average (fig 2 left). For longer prediction periods the results become worse (fig. 2 middle and right).
The Project DEPARTURE explores the decadal predictability of the West African Monsoon precipitation and the Atlantic hurricane activity. For a variety of decades in the hindcast cycle global simulations of the coupled climate model MPI-ESM based on the three regional climate models REMO, CCLM and WRF with 50 km resolution were regionalized. Furthermore REMO was coupled with a global ocean model and CCLM was driven with improved boundary conditions(aerosols, sea surface temperatures (SSTs), vegetation, land use) and soil initialisation for selected decades, to make advantage of the potential skill of ocean, atmosphere and land.
West African Monsoon
The Results of several RCMs with improved initial and boundary conditions regarding absolute bias and decadal predictability of the west African Monsoon precipitation can be summarised as follows: The Bias of the RCMs simulated west African monsoon precipitation shows in contrast to MPI-ESM an added value in west and central Sahel, but a clear gain of the warm-bias on the coast of guinea, which is based on too warm SSTs in Southeast-Atlantic. The ocean coupling of REMO can sharply improve this SST Bias and the precipitation bias over the guinea coast (Figure 3, left and middle). On top of that precipitation bias can be reduced in each region through different boundary conditions (Guinea-Coast: SSTs and land cover, central Sahel: vegetation, west Sahel: aerosols).
The decadal predictability of the West African monsoon precipitation over a whole decade commonly shows an added value in single RCMSs, but the positive correlation is not significant and variations between individual decades. The intradecadal predictability in the analysed decades and the interdecadal predictability between those decades reveal statistically more robust results and a clear added value for at least one RCM in each region.
Atlantic tropical storms
Regarding the Atlantic tropical storms and hurricanes all uncoupled RCMs display a positive Bias in the simulated quantity and intensity, which can be also considerably reduced through ocean coupling with REMO (figure 3, right). The decadal predictability of both variables exhibits some improvements through ocean coupling but rarely reaches statistical relevance and fluctuates heavily between the decades.
The seasonal to interanual development of weather in Europe is shaped by the long-term-variability and large-scale dynamic in general and in the north Atlantic in particular. The large-scale dynamic in the north Atlantic is not only determined through the large-scale interaction between earth-system components such as ocean and atmosphere and the global phenomena in different regions of the earth but also through the quasi steady interaction with small-scale phenomena (so called mesoscale, in scopes of kilometres to several 100 kilometres). For the weather in Europe upwind interactions have particular importance.
In climate models the frequency and duration of heat waves in summer and cold periods in winter is underestimated. The reason lies presumably in the underestimation of frequency and intensity of, in lower latitudes arising and in the north Atlantic propagating, cyclones and anticyclones, in the earth-system-models. The formation of evolution of cyclones and anticyclones is partially mesoscale and can therefore not be captured correctly by climate models, because simulation is not explicitly realized on that scale. Successful improvements in the statistic of large-scale weather conditions could be a critical step for improvements on interannual to decadal prediction.
As a result of restricted computational resources it will not be possible in the foreseeable future to simulate mesoscales explicitly in global climate simulations. However it is possible, to calculate these scales in a spatially confined area explicitly, e.g. the north Atlantic. In the context of MiKlip the MiKlip models which have been tested multiple times will be coupled for their respective scale, so that mesoscale dynamics in the coupled area can react upon the large scale dynamic. Therefore the regional climate model COSMO-CLM for the region Central America to north Atlantic (CANA, see figure 4) will be configured and with the help of the recently developed parallelised coupler OASIS3-MCT coupled to the global earth system model MPI-ESM.
Mesoscale Feedback and large-scale dynamics in the north Atlantic
Figure 4 shows a clear underestimation of cyclone-storm tracks-density through the MPI-ESM(left) and the corresponding raise in cyclone frequency through TWC (right). A promising second result is shown in figure 5. The MPI-ESM underestimates the blocking-rate about 25% in contrast to ERAInterim in the mid latitudes at 20° West to 20° East. This underestimation is not present with T
It can be demonstrated, that mesoscale simulations of cyclone-development in the Atlantic can correct a number of known deficits in the large scale dynamic of the north Atlantic. This can be accomplished by slight increase in resources, which is equivalent to an increase of the global model by the factor of 1.2. The model concept can basically be applied on other relevant regions-
Mieruch, S., Feldmann, H., Schädler, G., Lenz, C.-J., Kothe, S., and Kottmeier, C.: The regional MiKlip decadal forecast ensemble for Europe: the added value of downscaling, Geosci. Model Dev., 7, 2983-2999
Reyers, M., Pinto, J. G. and Moemken, J. 2015. Statistical-dynamical downscaling for wind energy potentials: Evaluation and applications to decadal hindcasts and climate change projections. Int. J. Climatol. 35, 666, 229 - 244.
This description summaries the achievements of Module C during the first phase. Module C continues in MiKlip II, read more here.