WP1: CLIMATE

Some general considerations of regional climate modelling in Eastern Europe

The CLAVIER project follows a traditional approach in considering impacts of climate change. It is a sequential approach, as illustrated in Figure 1. WP1 is in charge of the three first steps: from emission scenarios to global climate scenarios and then to regional climate scenarios. In CLAVIER, this sequential chain was realized by two emission scenarios (B1 and A1B), two global models (MPI and IPSL) and three regional models (LMDZ, REMO5.0 and REMO5.7). Results confirm our general believing that all the three steps introduce spreading and dispersion with roughly an equal importance for each step.

In this work, we made a special experimental design to investigate the influence of the sequential approach on the simulation of regional climate. It consistes of interactively coupling a global climate model (LMDZ-global) and a regional climate model (LMDZ-regional). This is also referred to as two-way nesting, in contrast with the classical one-way nesting from global scale to regional scale. The two-way nesting runs LMDZ-regional and LMDZ-global in parallel with an exchange of information (atmospheric temperature and circulation) every two hours. In this configuration, LMDZ-global is named "master" and LMDZ-regional "slave". In running the two-way nesting system, we recorded the atmospheric temperature and circulation every 6 hours which were then used to force in a sequential manner LMDZ-regional. In this configuration, LMDZ-regional is called "regional", since the simulation is a classical one-way nesting regional simulation.

We can now compare the three simulations "master", "slave" and "regional". To simplify the comparison, we select only one point corresponding the most closely to Budapest. Figure 2 plots the precipitation rate for a period of one year. The upper panel compares "slave" against "master" and can be used to reveal the added values of high-resolution models. The lower panel compares "slave" against "regional", and will be used to validate the one-way nesting. From the upper panel, we can see that the two models are generally in agreement to produce the rainfall sequence. But two differences can be noted. Firstly, "slave" produces in general stronger precipitation events. This implies that using high-resolution model is essential if our interest of study is more directed to extreme events. It is well known that large-scale models under-estimate strong precipitations. Secondly, some discrepancies between "master" and "slave" can be mainly observed during the summer season, from days 120 to 240. This is also in agreement with our general understanding that the local weather in eastern Europe is mainly controlled by large-scale circulations in winter, but can be strongly generated by local conditions in summer. From the lower panel of Figure 2, we can see that the two curves are generally in a very good agreement. This implies that our sequential approach in doing climate downscaling is valid and the associated biases are small. However, some discrepancies can be found also for the summer season.

Figure 1: Schematic of the general approach in CLAVIER project for climate change impact studies.Figure 1: Schematic of the general approach in CLAVIER project for climate change impact studies.

Figure 2: Temporal evolution in a year of daily precipitation rate (mm/s) in Budapest. The upper panel shows results of "master" and "slave". The lower panel shows results of "slave" and "regional".Figure 2: Temporal evolution in a year of daily precipitation rate (mm/s) in Budapest. The upper panel shows results of "master" and "slave". The lower panel shows results of "slave" and "regional".

Climate change scenarios performed with LMDZ-regional, with comparison to REMO

In the CLAVIER project, LMDZ-regional was forced by the outputs of three global climate change scenarios from ECHAM-A1B, ECHAM-B1 and IPSL-A1B. All the three simulations cover the period from 2000 to 2050 and follow the IPCC-defined emission scenarios. Two additional simulations were performed for the period from 1951 to 2000 following the 20th-century simulations with the global climate models ECHAM and IPSL.

Figure 3 plots the temporal evolution of annual-mean surface air temperature, averaged for the CLAVIER domain (Hungary + Romania + Bulgaria). The black curve indicates the 20th century ECHAM simulation for the period 1951-2000. The counterpart from IPSL is represented in orange curve. We can observe a general warming trend for the last two decades of the 20th century for the two curves, but the IPSL result is about 2°C cooler than the ECHAM result. The green and yellow curves (from 2001 to 2050) are the A1B and B1 scenarios from ECHAM, respectively. The A1B scenario is generally warmer than the B1 scenario, but the difference is small for our considered time scale, around 2050. The red curve is the A1B scenario from IPSL for the period 2001-2050. Despite the general cool feature of IPSL in the 20th century, the future warming is more important, the surface air temperature reaches a very similar level as in ECHAM. This indicates that the temperature increase is about 2°C larger in IPSL than in ECHAM, which is directly related to a different behaviour of simulated climate sensitivity in the two IPCC-AR4 models developed and used in Hamburg (ECHAM) and Paris (IPSL) respectively.

Figure 4 gives the geographic distribution of annual-mean changes in surface air temperature and precipitation for the A1B scenario. ECHAM and IPSL can be compared against each other. For surface air temperature, the warming in IPSL is much more important in IPSL with maxima in the Northeast of the domain. The warming in ECHAM is modest and with a more uniform spatial distribution. Concerning the precipitation, a general tendency of decrease is depicted in the South part of the domain and an increase in the North. The variation in the CLAVIER domain is small. Again we can observe that IPSL shows larger changes than ECHAM does.

To make some quantitative comparison, Table 1 shows the averaged surface air temperature for the three countries and for the two emission scenarios respectively. Table 2 gives the results on precipitation.

Results presented here confirm our initial intuitive expectation that important uncertainty still remains among different global ocean-atmosphere coupled models used for future projection of global climate evolution. Analyses of uncertainties related to regional models through the comparison between LMDZ and REMO5.0 models are presented in Tables 3 and 4, Figures 5 and 6. This comparison reveals a weak spread among the two regional models. How to explore the uncertainty issue in climate change prediction? How to use the uncertainty consideration into impact studies? These issues constitute actually a main challenge.

Figure 3: Annual-mean air temperature at 2m (upper, °C) and precipitation rate (lower, mm/day) in function of time from 1951 to 2050. The spatial average was performed for the CLAVIER region (Hungary, Romania and Bulgaria).Figure 3: Annual-mean air temperature at 2m (upper, °C) and precipitation rate (lower, mm/day) in function of time from 1951 to 2050. The spatial average was performed for the CLAVIER region (Hungary, Romania and Bulgaria).

Figure 4: Changes of surface air temperature (left, °C) and precipitation rate (right, mm/day) as predicted by LMDZ-regional (2001/2050 - 1951/2000). The upper panels are from LMDZ-regional forced by the MPI global climate model and the lower pannels forcFigure 4: Changes of surface air temperature (left, °C) and precipitation rate (right, mm/day) as predicted by LMDZ-regional (2001/2050 - 1951/2000). The upper panels are from LMDZ-regional forced by the MPI global climate model and the lower pannels forc

Table1: Spatial average of surface air temperature for the two A1B scenarios (from respectively MPI and IPSL) and the B1 scenario from MPI global climate model:

1961/1990 2021/2050 B1 2021/2050 A1B
LMDZ forced by ECHAM Romania 8.53 9.67 (+1.14) 10.25 (+1.72)
Hungary 10.48 11.57(+1.09) 12.15 (+1.67)
Bulgaria 10.63 11.87(+1.25) 12.41 (+1.78)
LMDZ forced by IPSL Romania 6.39 9.35 (+2.96)
Hungary 8.82 11.59 (+2.77)
Bulgaria 8.5 11.31 (+2.81)

Table2: Spatial average of precipitation rate (mm/day) for the two A1B scenarios (from respectively MPI and IPSL) and the B1 scenario from MPI global climate model:

1961/1990 2021/2050 B1 2021/2050 A1B
LMDZ forced by ECHAM Romania 1.92 1.90 (-0.02) 1.98 (+0.06)
Hungary 1.59 1.57 (-0.02) 1.57 (-0.02)
Bulgaria 1.50 1.41 (-0.09) 1.52 (+0.02)
LMDZ forced by IPSL Romania 2.22 2.22 (0)
Hungary 1.55 1.62 (+0.07)
Bulgaria 2.05 1.91 (-0.14)

Table3: Spatial average of surface air temperature for the A1B scenario of ECHAM, used in LMDZ and REMO5.0:

1961/1990 2021/2050 A1B
LMDZ forced by ECHAM Romania 8.53 10.25 (+1.72)
Hungary 10.48 12.15 (+1.67)
Bulgaria 10.63 12.41 (+1.78)
REMO forced by ECHAM Romania 9.13 10.53 (+1.40)
Hungary 10.53 11.89 (+1.36)
Bulgaria 11.81 13.34 (+1.53)

Table4: Spatial average of precipitation rate (mm/day) for the A1B scenario of ECHAM, used in LMDZ and REMO5.0:

1961/1990 2021/2050 A1B
LMDZ forced by ECHAM Romania 1.92 1.98 (+0.06)
Hungary 1.59 1.57 (-0.02)
Bulgaria 1.50 1.52 (+0.02)
REMO forced by ECHAM Romania 2.30 2.33 (+0.03)
Hungary 1.84 1.81 (-0.03)
Bulgaria 1.88 1.83 (-0.05)

Figure 5: As in Figure 3, but for LMDZ and REMO5.0 forced by the same ECHAM A1B scenario.Figure 5: As in Figure 3, but for LMDZ and REMO5.0 forced by the same ECHAM A1B scenario.

Figure 6: As in Figure 4, but for LMDZ and REMO5.0 forced by the same ECHAM A1B scenario.Figure 6: As in Figure 4, but for LMDZ and REMO5.0 forced by the same ECHAM A1B scenario.

Assessment of uncertainties

The outcome of the climate change impact studies presented within CLAVIER relies to a large extent on the performance of the climate models, which were used to provide their input. Each information provided by a climate model has a certain amount of uncertainty, which has to be considered when it is used to project future climate conditions. Three types of uncertainties have to be underlined:

  • The choice of the emission scenarios
  • Systematic biases which are typical for the individual climate models.
  • Uncertainties due to internal variability of climate models.

Choice of the emission scenario

The human and its political and societal evolution is one main contributer of uncertainty. Human behaviour enters climate change simulation via the prescribed future global emissions of greenhouse gases. An emission scenario providing the input for climate models for future climate projections is a source of uncertainty since we can only estimate the trend of human greenhouse gas emissions in the future. Within CLAVIER, one emission scenarios of the IPCC has been chosen, namely the A1B being a medium scenario (applied for REMO and LMDZ). This scenario was adopted within CLAVIER as obligatory. A second IPCC emission scenario has been used for a few studies and is named B1, which can be interpreted as a rather an optimistic scenario (used for LMDZ).

Systematic biases

Another source of uncertainty are the global and regional models which are used for the simulation. Although most climate models are based on the same underlying physical principles and account for the most important processes relevant for weather and climate, their physical parametrizations of certain processes as well as the applied numerical methods can differ. A systematic model bias is the result of an unrealistic description of physical processes and can be detected by model validation and comparison with observation data. To account for the uncertainty caused by the choice of the emission scenario and the used models, it is advisable to use a multi-scenario, multi-model ensemble of simulations, as it was done in CLAVIER. Figure 7 shows the resulting range of the annual cycles for the temperature change signal of Hungary up to the year 2050 for the three regional models used in CLAVIER, driven by two different global models (ECHAM5 and IPSL), and for the IPCC emission scenarios A1B and B1.

Within CLAVIER, various empirical-statistical error correction methods have been applied and investigated to mitigate regional climate model errors. In general, these methods are highly useful in climate change impact studies, but cannot replace model improvements which rest on the precise identification of the origin of a bias. Further, they rely on observational datasets, which themselves can differ from each other and therefore are another source of uncertainty. As an example, a comparison of annual cycles of four precipitation observational datasets are shown for Hungary, Romania, and Bulgaria in Figure 8. It can be seen that the ECA observations of the ENSEMBLES project which were used for the error correction, are significantly drier than the other datasets.

The most severe systematic error relevant for the CLAVIER domain is known as the "Summer Drying Problem" (SDP), and is characterized by the too dry and too warm simulation of climate over Central and Eastern Europe during (late) summer [Hagemann et al., 2004, Jacob et al., 2008]. It is typical for many regional and also some global climate models. The SDP shows up in REMO, however it is visible only to a much less extent in LMDZ. The 2m temperature difference of REMO, LMDZ, and error corrected REMO simulation results with the ECA observations in summer are shown in Figure 9. The origin of the SDP is still unknown in general, although a sensitivity study with REMO MPI-M suggest that it might be caused by an unrealistic simulation of atmospheric moisture and heat transport in the Carpathian Basin. A 5-years simulation with a more realistic numerical atmospheric diffusion scheme resulted in an improvement of the simulated 2m temperature in the CLAVIER countries (Figure 10): The overestimation in summer and autumn is mitigated with the new scheme.

Figure 7: Annual cycles of the climate change signal for the temperature for the period 2021-2050 (reference period 1961-1990); regional models: REMO MPI-M, REMO OMSZ, LMDZ; global models: ECHAM5, IPSL; emission scenarios: A1B, B1Figure 7: Annual cycles of the climate change signal for the temperature for the period 2021-2050 (reference period 1961-1990); regional models: REMO MPI-M, REMO OMSZ, LMDZ; global models: ECHAM5, IPSL; emission scenarios: A1B, B1

Figure 8: Annual cycles of different observational datasets in Hungary, Romania, and Bulgaria (1961-1990): ECA [Haylock et al., 2008], CRU [Mitchell and Jones, 2005], GPCC [Fuchs et al., 2007], dataset of the Hungarian Meteorological Service (for Hungary only)

Figure 9: Summer Drying Problem. 2m temperature difference between simulation results and ECA observations in summer (1981-2000); yellow/red means an overestimation of the model. Left: REMO-MPI, center: LMDZ, right: REMO-MPI error correctedFigure 9: Summer Drying Problem. 2m temperature difference between simulation results and ECA observations in summer (1981-2000); yellow/red means an overestimation of the model. Left: REMO-MPI, center: LMDZ, right: REMO-MPI error corrected

Figure 10: Sensitivity experiment with REMO-MPI: Two numerical diffusion schemes ("sigma": Realized in standard REMO, "z": improved scheme) have been compared with each other and with the ECA observational dataset (1984-1988). Annual cycles of the 2m temperature are shown for Hungary, Romania, and Bulgaria

Model internal variability

In contrast to systematic biases, uncertainties due to internal variability are an essential feature of climate models and do not degade model results. In a regional climate model, the lateral boundaries are fixed by the externally given drving fields (in CLAVIER: ERA40 reanalysis data, ECHAM5 and IPSL global model simulations). The internal variability is a measure for the model's degree of freedom to develop its own dynamics in the interior of the model domain.

To quantify this type of uncertainty, a new method has been developed within CLAVIER in the recent months [Jacon et al., 2008]. A physically consistent perturbation of the lateral boundary data as well as the surface boundary data has been created. An ensemble of 9 REMO simulations was carried out for one given set of forcing data (ERA40) over a period of 7 years, for each of which the model domain is shifted about 1/10th of a grid box into a different direction relatively to the first ensemble member. The simulations are sensitive to the slightly perturbed input data, and the differences between the individual simulations can be attributed to the characteristic internal variability of the RCM. A large European domain was chosen which is the same that has been used for the REMO simulations provided for CLAVIER. Figure 11 shows the mean monthly standard deviations of temperature in summer and winter for the full simulation domain. The results show that the uncertainty related to internal variability of REMO is large in the CLAVIER domain in summer, where the deviation between monthly mean temperatures can exceed 0.5°C. This hints at a high sensitvity of the region to the internal model parameters and boundary conditions and thus for the occurrence of systematic biases like the SDP. In the context of uncertainty assessment, this method is complementary to the multimodel and multi-scenario approach as mentioned above.


Figure 11: Model internal variability of the 2m temperature (REMO-MPI, 1990-1997) in degrees Celsius: Standard deviation of the 9-member ensemble of simulations. Left: summer, right: winter

References

    Fuchs T., U. Schneider, and B. Rudolf : "Global Precipitation Analysis Products of the GPCC. Global Precipitation Climatology Centre (GPCC)." Deutscher Wetterdienst, Offenbach, Germany. (2007)

    Haylock, M.R., Hofstra, N., Klein Tank, A.M.G., Klok, E.J., Jones, P.D. and New, M.: “A European daily high-resolution gridded data set of surface temperature and precipitation for1950-2006.” Journal of Geophysical Research, 113 (2008), D20119, doi:10.1029/2008JD010201

    Hagemann, S., B. Machenhauer, R. Jones, O.B. Christensen, M. Deque, D. Jacob, and P.L. Vidale: “Evaluation of water and energy budgets in regional climate models applied over Europe.” Climate Dynamics, 23, pp. 547-567 (2004)

    Jacob D., L. Kotova, P. Lorenz, Ch. Moseley, and S. Pfeifer: “Regional climate modelling activities in relation to the CLAVIER project.” Quarterly Journal of the Hungarian meteorological society, pp. 141-153 (2008)

    Mitchell T. D. and P. D. Jones: “An improved method of constructing a data base of monthly climate observations and associated high-resolution grids.” Int. J. Climatology, 25, pp. 693-712 (2005)

    Zängl, G.: “An Improved Method for Computing Horizontal Diffusion in a Sigma-Coordinate Model and Its Application to Simulations over Mountainous Topography.” Monthly Weather Review, 130, pp. 1423–1432 (2002)