Europe:
The overall patterns of change are the same (although the RCM exhibits an enhanced drying) with a relatively large decrease over Central and Southern Europe and a general increase in the northernmost regions. But there are also marked differences on the regional scale. Note in particular the orographically induced additional increase along mountain slopes of Western Norway and the change in the geographical position and magnitude caused by the representation of the Alps This is accounted for mainly by the differences in model topography. Thus, the relatively large area with an increase in precipitation over Italy in the ECHAM4 simulation is reduced to a small region around the Po valley, likewise, along the east coast of the Iberian Peninsula. These kinds of resolution-dependent differences are seen in all seasons. The enhanced drying is in agreement with findings in other studies using RCMs with a European focus (Machenhauer et al., 1998; Jones et al., 1997; Rummukainen et al., 2001).

In comparing the two experiments, it is also seen that the ECHAM4-driven run shows a tendency for a larger shift towards more intensive precipitation. This is in agreement with the HadAM3H-driven runs showing a larger reduction in mean precipitation.

From the analysis so far, there is no guarantee that the spatial patterns of increase are temporally correlated; that is, the statistic is done for individual grid points. [The authors] have therefore carried out the same analysis after lumping the precipitation data over eight different European river catchments. For those of the rivers, which have their origin and a sufficiently large subcatchment within the Alps (the Rhine, Elbe, and Danube), we have further discriminated against high altitude, in that we only consider the areas which are located more than 400 m above sea level. This may seem as a somewhat arbitrary choice but is the result of a compromise between having enough grid points to sample from and still represent the upper catchment areas reasonably well. This way, we furthermore focus on areas like those causing the recent floodings in this region.

To summarise the behaviour we show the relative change in average precipitation and in exceedence of the 95th and 99th percentiles for eight important and geographically different river systems: The Torne River in Northern Sweden, the Odra, the upper Rhine, the upper Danube, the upper Elbe, and the Rhône rivers representing Central Europe, plus the Ebro and Po rivers representing Southern European rivers. [ECHAM boundaries and HadAM3H boundaries have been differenciated]. A running 3-month window has been applied in the analysis in order to have sufficient data points. The seasonality of the relative change of average precipitation is similar in the two experiments. The two experiments have different seasonality in the relative change of high exceedences however. The average precipitation is generally decreasing in the summer period. But higher percentiles have increasingly more positive values resulting in some large increases of extremes, especially for the important central-European rivers. The ECHAM-driven experiments show rather large increases in late summer, but the HadAM3H-driven experiments do not share this feature; nevertheless, also this set of experiments has more positive values for the high intensities than for the mean in the summer. The ECHAM/OPYC B2 scenario, which is not shown here, generally exhibits the same signals as the corresponding A2 scenario, but with the expected smaller magnitude.

Model setup: In [this] approach, [the authors] have selected [...] the scenarios A2 and B2 in order to test the sensitivity of climate change to the degree of global warming. In addition, [they] have chosen to study the sensitivity of [the] results to the driving boundary conditions by conducting experiments using boundaries from two different global models. Two time slices were chosen to represent the current climate (1961–1990) and the future (2071–2100), respectively. [The GCM used were the ECHAM 4/OPYC and the HadAM3H].

The studies were carried out using a horizontal resolution of 0.44° corresponding to 50 km. The initial atmospheric and six-hourly updates of lateral boundary conditions were taken from the ECHAM4/OPYC and HadAM3H simulations, respectively. Sea surface temperatures and sea–ice conditions were updated daily by spatially interpolated fields from the same model simulations. Before either integration, the model was spun up for 2 years following the procedure proposed by Christensen (1999) to ensure a well-balanced initial state.

Analysis of extreme precipitation:
We consider daily precipitation events on a gridpoint level exceeding certain thresholds of the simulated cumulative precipitation distribution function f. Here, we address the occurrence of events by analysing the exceedence of the 99th percentile of f corresponding to return periods of approximately 1 year; we will also examine the exceedence of the 95th percentile briefly. The results of investigating exceedences rather than direct percentiles are minor; there is however a tendency of more robust signals and a higher statistical significance when an average quantity like the exceedence is considered.

We concentrate on intensive summer time conditions, which in Central and Southern Europe peak during July–August–September (JAS). Although we concentrate on this period here, we have expanded our analysis to all seasons for a more general assessment where the total precipitation over several European river catchments is investigated. First, each experiment and each grid point are treated separately, and all 2700 daily precipitation values of the 30-year period are sorted, leading to information about all exceedences. Next, the same procedure is repeated, considering precipitation within specified entire river catchments.

(4) - Remarques générales

Although global models have provided indications of an intensified global hydrological cycle in a warmer climate, indications of the changes in the extreme precipitation events at the regional scale remain rather unconvincing to date [see, however, Huntingford et al. (2003), Palmer and Räisänen (2002), and Christensen and Christensen (2003)]. This is particularly so in regions where a reduction in the total precipitation is simulated (May et al., 2002; Voss et al., 2002). However, the resolution of these global models precludes their simulation of realistic synoptic or regionally induced circulation that may lead to the extreme episodes, which are observed in nature. A 2001 assessment by the IPCC (Houghton et al., 2001) concludes that because of increasing greenhouse gas concentrations ". . .over many Northern Hemisphere mid- to high-latitude land areas, more intense precipitation events are very likely. For other areas, there are either insufficient data or conflicting analyses", reflecting the limited numbers of studies which have addressed this issue on a regional scale.

Recent research is targeting this unfortunate situation; for example, the EU project PRUDENCE (Christensen et al., 2002) will provide several centuries worth of regional model simulations with a daily temporal resolution for all to study. In this study, which is a follow-up on similar analyses in Christensen and Christensen (2003) with one set of boundary conditions, the relation between extreme daily-precipitation events and climate change has been explored in the present work with the regional climate model (RCM) HIRHAM4 (Christensen et al., 1996) using boundary conditions from two different GCMs. We have furthermore studied a number of major European river catchments characterised by a broad regime of precipitation climates under the present conditions.

Using the combined information from the two driving models and from the B2 and A2 scenario simulations, we find that CO2-induced warming can lead to a shift towards heavier intensive summertime precipitation over large parts of Europe. In general, the change in precipitation exceeding the 99th percentile is different from the changes in the mean. Even when a reduction in total mean precipitation is simulated, the amounts of precipitation in the intensive events are much less reduced and even increase in many places. The higher the percentile considered, the larger are the areas that show a positive change.

There is a solid physical mechanism which can support this finding (Frei et al., 1998). In a warmer climate, more moisture will evaporate particularly over sea than at present due to higher sea surface temperature, and hence result in a higher saturation mixing ratio (described by the Clausius–Clapeyron relation). This will facilitate latent heat release during the buildup of weather systems, thereby both possibly intensifying the systems and make more water available to rain out. In order to see if this attribution of cause is correct, further case analyses are necessary, which will happen within the PRUDENCE project.