WP3b: EXTREMES

Results of WP3b: Extremes

Contents

  1. Validation of simulated extreme indices for present climate
  2. Projected changes of extremes
  3. Influence of the error correction (WP2) on projected changes of extremes

1. Validation of simulated extreme indices for present climate

Before assessing the projected change of extreme events over Central and Eastern Europe from the climate change simulations, the observed present-day spatial and temporal distribution of the extremes has to be assessed and the model's ability to correctly simulate these extreme events has to be demonstrated.
To do this, indices for extreme events have been derived from all three CLAVIER ERA40 simulations (the so called "validation simulations"). For these simulations, the models have been driven and initialized by the ERA40 Reanalysis data provided by the European Centre for Medium-Range Weather Forecasts (ECMWF).
The extreme indices have been defined in WP2 and are based on the STARDEX definitions and on specific requirements of the CLAVIER partners. In addition, the same quantities (where possible) have been derived from an observational dataset. This has been done using the gridded observational dataset compiled in the framework of the ENSEMBLES project (ECA Observations, Haylock et al., 2007).
For this comparison, the choice of the indices has been restricted by the availability of observational data.

Temperature-based indices:

  • tnn20 - mc [days] (tropical nights): number of days where the minimum temperature (tas-dn) exceeds 20°C per month
  • tnfd - dc [days] (number of frost days): number of days with minimum temperature (tas-dn) below 0 °C per month

Precipitation-based indices:

  • px1d - mx [mm] (greatest 1-day rainfall): maximum precipitation sum in one day per month
  • px3d - mx [mm] (greatest 3-day rainfall): maximum precipitation sum in 3 consecutive days per month
  • px5d - mx [mm] (greatest 5-day rainfall): maximum precipitation sum in 5 consecutive days per month
  • pxcdd - mx [days] (consecutive dry days): maximum number of consecutive dry days per month.
  • pint - mm [mm/day] (precipitation intensity): mean daily precipitation sum on rainy days (days where pr-24hc exceeds 1 mm/day) per month

Figure 1: DJF 1961 to 1990: Number of frost days [days/month]Figure 1: DJF 1961 to 1990: Number of frost days [days/month]
Figure 1 shows the mean number of frost days [days/month] in winter (DJF, December, Janury, February) for the period 1961 to 1990 for the three different model simulations (REMO-MPI, REMO-HMS, and LMDZ) and for the ECA Observations. From figure 1, you can see that the spatial distribution of frost days in the Central end Eastern European region is mainly dominated by orographic effects, leading to more frost days along the mountain ridges and less in the flat areas. This is well reproduced by all models. However, Figure 1 shows that especially REMO-MPI has some problems with the temperatures in the Danube basin along the border between Romania and Bulgaria, in northern Serbia and western Hungary, where it drastically underestimates the number of frost days.

Figure 2: JJA: maximum 5-day precipitation [mm/5days] 1961 to 1990Figure 2: JJA: maximum 5-day precipitation [mm/5days] 1961 to 1990
Figure 2 shows the mean number of tropical nights [nights/month] in summer (JJA, June, July, August) for the period 1961 to 1990 for the three different model simulations (REMO-MPI, REMO-HMS, and LMDZ) and for the ECA Observations. Figure 2 shows that all three models simulate too many tropical nights in summer. However, this overestimation is much stronger in the REMO models than for LMDZ and again stronger for REMO-HMS than for REMO-MPI. When looking at the spatial distribution of the tropical nights, the models overestimate the minimum daily summer temperatures especially over the flat areas of e.g. the Danube basin in the region of the Bulgarian/Romanian border and in the low elevated regions of Northern Serbia and South-Eastern Hungary. The overestimation of summer temperatures by the models is part of the summer drying problem, which is a model bias known for several regional models, resulting in a drying and warming of the Central and Eastern European region. The reasons for the summer drying problem are a major part of the investigations in the CLAVIER WP 1.

Figure 3: Mean annual cycle of precipitation intensity [mm/day] for Hungary 1961 to 1990Figure 3: Mean annual cycle of precipitation intensity [mm/day] for Hungary 1961 to 1990
Figure 3 shows the mean annual cycle of precipitation intensity for Hungary for the period of 1961 to 1990, again for the three models and for the ECA Observations. For all models, the annual cycle is well captured, both in shape and magnitude. Some small discrepancies show up in autumn, where the models have problems with the decreasing precipitation intensity.

Figure 4: Mean annual cycle of maximum 3-day precipitation [mm/3-days] for Bulgaria 1961 to 1990Figure 4: Mean annual cycle of maximum 3-day precipitation [mm/3-days] for Bulgaria 1961 to 1990
Figure 4 shows the mean annual cycle of maximum 3-day precipitation for Bulgaria for the period of 1961 to 1990, again for the three models and for the ECA Observations. The maximum 3-day precipitation amount averaged for Bulgaria shows an overestimation of precipitation by the LMDZ models for all seasons except of winter. Both REMO models overestimate winter and spring 3-day precipitation amounts. This figure shows the typical model behaviour, also for Hungary and Romania and also for the maximum 5-day precipitation.
Figure 5: JJA: maximum 5-day precipitation [mm/5days] 1961 to 1990Figure 5: JJA: maximum 5-day precipitation [mm/5days] 1961 to 1990
In summer (Figure 5), the REMO-MPI simulation clearly shows an underestimation of maximum 5-day precipitation over Hungary and parts of Romania and Bulgaria, which is also part of the Summer drying problem mentionned above.

back to top

2. Projected changes of extremes

For this study, Regional Climate Model (RCM) simulations have been used which were driven and initialized by different Global Climate Models (GCMs) under different emission scenarios. Table 1: Overview of the CLAVIER regional climate change simulations (period: 1951 to 2050)Table 1: Overview of the CLAVIER regional climate change simulations (period: 1951 to 2050)
Table 1 summarizes the regional simulations which have been available for the assessment. “EC5” abbreviates the ECHAM5/MPIOM global coupled climate model from MPI-M and "IPSL" the global coupled climate model developed at the Institut Pierre Simon Laplace in Paris. In the following, the regional model simulations are named by a combination of the regional model name, the driving global model name and the IPCC scenario used, e.g. REMO-MPI-EC5-A1B for a REMO-MPI simulation driven and initialized by an A1B scenario of the GCM Echam5/MPIOM. All simulations are available for the period 1951 to 2050. For each simulation, the climate change signal for the different extreme indices has been derived by subtracting the values for the target period from 2021 to 2050 from those of the climate reference period from 1961 to 1990.

For all models (see table 1 above) and all indices indicated in section 1, the changes in the annual mean cycle (2021 to 2050 in relation to the climate reference period of 1961 to 1990) have been calculated separately for each of the three CLAVIER countries. In addition, seasonal mean changes (again 2021 to 2050 minus 1961 to 1990) were plotted for the Central and Eastern European Region to show the spatial distribution of the calculated changes. For the annual cycles, the maximum spread between the 5 model simulations is highlighted by the gray shaded background. This spread gives the uncertainty of the simulated signal inserted by the models and the GCM/RCM configuration setup. In this section, the original model results were used. How this spread is reduced / modified by the error correction applied in WP2 will be shown in section 3.

Figure 6: Mean annual cycle of projected changes for the number of frost days [days/monht] for HungaryFigure 6: Mean annual cycle of projected changes for the number of frost days [days/monht] for Hungary
Figure 6 shows the change of the annual mean cycle of the number of frost days (tnfd) between 2021-2050 and 1961-1990 for REMO-MPI-EC5-A1B, REMO-HMS-EC5-A1B, LMDZ-IPSL-A1B, LMDZ-EC5-A1B, and LMDZ-EC5-B1 for Hungary. All models agree in decreasing numbers of frost days for Hungary, with LMDZ-IPSL-A1B having the highest values of up to -6 days/month for late winter and early spring.

Figure 7: MAM: projected change in the number of frost daysFigure 7: MAM: projected change in the number of frost days
This also reflects in Figure 7, showing the MAM (March, April, May) mean changes of the number of frost days between 2021 to 2050 and 1961 to 1990 for the whole region. The two REMO simulations show similar features. The LMDZ simulations seem to be dominated by the driving global model rather than by the underlying emission scenario. This can be concluded from the fact that the LMDZ-EC5-A1B and LMDZ-EC5-B1 simulations show very similar changes of tnfd, although being based on different emission scenarios. In contrast, the LMDZ-EC5-A1B and LMDZ-IPSL-A1B simulations show drastic differences in the projected changes of tnfd. The large differences between these simulations results from large differences in the underlying global model simulations.

Figure 8: JJA: change in the number of tropical nights [nights/month]Figure 8: JJA: change in the number of tropical nights [nights/month]
Figure 8 shows the projected changes of the number of tropical nights (tnfd) for JJA between 2021-2050 and 1961-1990 for REMO-MPI-EC5-A1B, REMO-HMS-EC5-A1B, LMDZ-IPSL-A1B, LMDZ-EC5-A1B, and LMDZ-EC5-B1. All models agree on increasing numbers of tropical nights for the Central and Eastern European region. The strongest changes can be seen for all models in those regions which were mainly affected by the summer drying problem (see Figure 2 above). The projected changes are stronly affected by the summer drying problem of the models, therefore the bias correction drastically reduces the projected increase in tnn20 (see section 3).

Figure 9: projected changes in the mean annual cycle of maximum 1-day precipitation for HungaryFigure 9: projected changes in the mean annual cycle of maximum 1-day precipitation for Hungary
The projected changes in precipitation extremes (pint, px1d, px3d, px5d) are relatively weak and do not show a consistent trend among the different model. As an example, Figure 9 shows the projected changes for the mean annual cycle of maximum 1-day precipitation for Hungary. An exception is Bulgaria, where most models show by trend decreasing values of precipitation extremes for summer and increasing values for winter and where generally the projected changes are stronger. The most striking feature of the projected precipitation extremes is a strong increase in extreme winter precipitation values for the LMDZ-IPSL-A1B simulation.

Figure 10: DJF: change in maximum 5-day precipitation [mm/5days]Figure 10: DJF: change in maximum 5-day precipitation [mm/5days]
As an example, Figure 10 gives the mean winter changes of the maximum 5-day precipitation as projected by the different models. This strong signal of the LMDZ-IPSL-A1B simulation is however drastically reduced by the applied error correction (see section 3).

Figure 11: change in the mean annual cycle of the maximum number of consecutive dry days for BulgariaFigure 11: change in the mean annual cycle of the maximum number of consecutive dry days for Bulgaria
The projected changes in the number of consecutive dry days (pxcdd) for Bulgaria are shown in Figure 11. As compared to the precipitation extremes, the changes in the number of consecutive dry days show a relatively clear annual cycle with decreasing maximum lengths of dry periods for winter and increasing maximum lengths of dry periods for spring, summer, and partly also for autumn. This feature is strongest for Bulgaria but also visible for Hungary and Romania.

back to top

3. Influence of the error correction on projected changes of extremes

Bias corrected values of mean daily precipitation, daily mean, minimum and maximum temperature have been available for the REMO-MPI-A1B and for the LMDZ-IPSL-A1B simulations. In the following, extreme indices derived from the original REMO-MPI-A1B and LMDZ-IPSL-A1B simulation results are opposed to those derived from the bias corrected results to see by how much the bias correction affects the projected changes of extreme events.

Figure 12: effect of the bias correction on the projected changes of tnfd, HungaryFigure 12: effect of the bias correction on the projected changes of tnfd, Hungary
The number of frost days is more affected by the applied bias correction for the REMO-MPI-EC5 simulation than for the LMDZ-IPSL-A1B simulation, which remains basically unchanged for Romania and Bulgaria. For Hungary, the LMDZ-IPSL-A1B projected decrease in tnfd is by 1-2 days/month stronger for the bias corrected values than for the original models output from October to March. The REMO-EC5-A1B simulation of tnfd is stronger affected by the bias correction, resulting in increasing changes of tnfd for the bias corrected projections for all months and all countries. This is shown in Figure 12 for Hungary.

Figure 13: JJA: effect of the error correction on the projected changes of tnn20Figure 13: JJA: effect of the error correction on the projected changes of tnn20
The projected increase in the number of tropical nights (tnn20) is reduced by the error correction for both models. This is shown in Figure 13, which gives the projected JJA change in tnn20 for the original models (upper panels) and for the bias corrected data (lower panels).

Figure 14: effect of the bias correction on the projected changes of px1d for RomaniaFigure 14: effect of the bias correction on the projected changes of px1d for Romania
The main effect of the bias correction on the projected changes of the mean precipitation intensity on a rainy day (pint), the maximum 1-day, and 3-day precipitation (px1d, px3d) is the reduction of the strong projected increase of precipitation extremes for the winter months by the LMDZ-IPSL-A1B simulation. This is for pint only the case for Bulgaria, for px1d and px3d for all countries. The REMO-EC5-A1B projected changes of pint, px1d, and px3d remain basically unchanged by the bias correction. As an example, Figure 14 shows the effect of the error correction on the projected changes of the maximum 1-day precipitation for Romania.

Figure 15: effect of the bias correction on the projected changes of px5d for BulgariaFigure 15: effect of the bias correction on the projected changes of px5d for Bulgaria
For the maximum 5-day precipitation, again the strong projected changes by LMDZ-IPSL-A1B are reduced drastically in the winter and early spring months. In contrast to px1d and px3d, the projected changes of px5d by REMO-EC5-A1B are also affected by the bias correction. This results mainly in a reduction of the projected increases of px5d and an intensification of the projected decreases of px5d (see Figure 15 for Bulgaria).

back to top

References

Haylock, M.R., Hofstra, N., Klein Tank, A.M.G., Klok, E.J., Jones, P.D. and New, M., 2008 A European daily high-resolution gridded data set of surface temperature and precipitation for 1950-2006.Journal of Geophysical Research, 113, D20119, doi:10.1029/2008JD010201