Global changes in seasonal means and extremes of precipitation from daily climate model data

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117,, doi: /2011jd016260, 2012 Global changes in seasonal means and extremes of precipitation from daily climate model data Simone Russo 1,2 and Andreas Sterl 3 Received 17 May 2011; revised 2 November 2011; accepted 5 November 2011; published 12 January [1] We investigate simulated changes of seasonal precipitation maxima and means in a future, warmer climate. We use data from the ESSENCE project, in which a 17-member ensemble of climate change simulations in response to the SRES A1b scenario has been carried out using the ECHAM5/MPI-OM climate model. The large size of the data set gives the opportunity to detect the changes of climate extremes and means with high statistical confidence. Daily precipitation data are used to calculate the seasonal precipitation maximum and the seasonal mean. Modeled precipitation data appear consistent with observation-based data from the Global Precipitation Climatology Project. The data are split into six time periods of 25 years to get independent time series. The seasonal peaks are modeled by using the generalized extreme value distribution, while empirical distributions are used to study changes of the seasonal precipitation mean. Finally, we use an empirical method to detect changes of occurrence of very wet and dry periods. Results from these model simulations indicate that over most of the world precipitation maxima will increase in the future. Seasonal means behave differently. In many regions they are decreasing or not increasing. The occurrence of very wet periods is strongly increasing during boreal winter in the extratropics and decreasing in the tropics. In summary, wet regions become wetter and dry regions become drier. Citation: Russo, S., and A. Sterl (2012), Global changes in seasonal means and extremes of precipitation from daily climate model data, J. Geophys. Res., 117,, doi: /2011jd Introduction [2] Changes in extreme weather and climate events have significant impacts and are among the most serious challenges to society in coping with a changing climate [U.S. Climate Change Science Program, 2008]. Indeed, confidence has increased that some extremes will become more frequent, more widespread and/or more intense during the 21st century [Intergovernmental Panel on Climate Change, 2007]. Of particular interest are changes in rainfall distribution, as water-related disasters, i.e., floods and droughts, can cause devastating impacts on human lives and economic damages. [3] The intensity of precipitation extremes is projected to increase under global warming in many parts of the world, even in regions where mean precipitation decreases [e.g., Kharin and Zwiers, 2000, 2005; Semenov and Bengtsson, 2002; Voss et al., 2002; Wilby and Wigley, 2002; Wehner, 2004]. Simulations with global coupled ocean atmosphere general circulation models (CGCMs) forced with projected greenhouse gas and aerosol emissions are the primary tools 1 Laboratoire de Physique des Oceans, UMR6523, IRD/IFREMER/ UBO, Université de Bretagne Occidentale-UFR, Brest, France. 2 Now at Institute for Environmental Protection and Research, Rome, Italy. 3 Royal Netherlands Meteorological Institute, De Bilt, Netherlands. Copyright 2012 by the American Geophysical Union /12/2011JD for studying possible future changes in climate mean, variability and extremes [Kharin et al., 2007]. Some results on the ability of atmospheric general circulation models to simulate temperature and precipitation extremes are documented by, e.g., Kharin and Zwiers [2005] and Kiktev et al. [2003, 2007]. [4] Contrary to many recent studies that focus on changes of yearly precipitation means and/or maxima [e.g., Kharin and Zwiers, 2005, Kharin et al., 2007, Min et al., 2011], we here zoom in on the seasonal scale. Using the daily precipitation output of a 17-member model ensemble of the ECHAM5/MPI-OM climate model run under the SRES A1b emission scenario (for details see Sterl et al. [2008]) we investigate for two selected seasons (DJF and JJA) changes in the probability distributions of mean and maximum daily precipitation. We demonstrate that the future behavior of 100 year return levels of seasonal precipitation maxima differs significantly from that of the seasonal precipitation means. We introduce a simple empirical definition for very wet (dry) periods by considering joint events of high (low) seasonal means and peaks. With this definition we show that the occurrence of wet periods is increasing in wet areas and decreasing in dry areas. 2. Data Set and Statistical Methods 2.1. The ESSENCE Data Set [5] In the present study we use 24 h precipitation amounts from the 17 runs of the Ensemble Simulations of Extreme 1of11

2 Weather Events Under Nonlinear Climate Change (ESSENCE) [Sterl et al., 2008] to detect changes in seasonal mean and extremes. In the ESSENCE project a 17-member ensemble of runs with a state-of-the-art climate model was generated by perturbing the initial state of the atmosphere. Gaussian noise with an amplitude of 0.1 K was added to the initial temperature field. The initial ocean state was not perturbed. The model used was the ECHAM5/MPI-OM coupled climate model developed at the Max Planck Institute for Meteorology in Hamburg. It is based on version 5.2 of the ECHAM model with a T63 spatial resolution, equivalent to a resolution in latitude and longitude. The two component models, ECHAM5 for the atmosphere and MPI-OM for the ocean, are well documented (ECHAM5 [Roeckner et al., 2003] and MPI-OM [Marsland et al., 2003]), and a special section of the Journal of Climate (19(16), ) devoted to the coupled model and its validation. Model runs start in 1950 and end in For the historical part ( ) the concentrations of greenhouse gases (GHG) and tropospheric sulfate aerosols are specified from observations, while for the future part ( ) they follow the SRES A1b scenario [Nakicenovic et al., 2000] The GPCP Observations [6] To evaluate our precipitation model output we use daily Global Precipitation Climatology Project (GPCP) data ( The GPCP One Degree Daily (1DD) Precipitation data set provides daily, global 1 1 gridded fields of precipitation totals from October 1996 onward and is regularly updated. The 1DD set combines several different data sources covering different areas of the globe. Every attempt has been made to make the complete record homogeneous, given the different available input sources ( oa/wmo/wdcamet-ncdc.html) Test of Significance [7] During 25 years the climate change signal is small. Hence, with good approximation, we can treat 25 years as stationary. Furthermore the 17 realizations are independent as their pairwise correlations are small, and year-to-year values are independent as the autocorrelation decay time is shorter than 1 year. We have used the first period ( ) as the reference to measure the future seasonal precipitation changes. Each sample for both mean and extreme precipitation is a set of 17 (ensembles) times 25 (seasonal peaks or means) equals 425 events defined as follows: M i ¼ [17 e¼1 E i ¼ [17 e¼1 y[ max;i S mean ðe; yþ; ð1þ y¼y min;i y[ max;i S ext ðe; yþ; ð2þ y¼y min;i for i =1,,6, with (y min,1,y max,1 ) = (1951, 1975),,(y min,6, y max,6 ) = (2076, 2100). Here denotes the union of sets, e the model ensemble member, and S mean and S ext denote the seasonal mean and maximum of daily precipitation, respectively. The sets M i and E i are defined for each studied season. [8] We will only show results for periods 3 ( ), 5 ( ) and 6 ( ). Changes before the year 2000 are small, and those in period 4 ( ) look like a linear interpolation between periods 3 and 5. From our analysis we exclude the very dry regions, defined as the areas where at least in one of the three selected periods the seasonal mean and peaks are below 0.1 and 1 mm/d, respectively. These areas are masked by dotted black patterns in Figures 4 9. The exclusion of the arid regions from our analysis is because future precipitation changes there are difficult to interpret. Models have difficulty in simulating very dry conditions [Sterl et al., 2008]. As many authors [Giannini, 2010; Biasutti and Giannini, 2006; Cook and Vizy, 2006; Douville et al., 2006; Christensen et al., 2007] have discussed, projections of 21st century change are very uncertain in these regions, with equal numbers of models predicting a significantly wetter or drier future, or no significant change with respect to present conditions at all. Moreover in these areas, as detected by the goodness-of-fit analysis described in section 2.4, the GEV model can only be accepted at a lower confidence (level of significance between 5% and 20%). [9] To test whether precipitation in period i is significantly different from that in the reference period (period 1) we determine whether the two data sets M 1 and M i (and E 1 and E i ) are different, precisely, whether or not they are drawn from the same distribution at a chosen level of significance (5%). To do so we use the Kolmogorov-Smirnov (K-S) test [e.g., Von Storch and Zwiers, 2003; Iacus and Masarotto, 2007], which has the advantage of making no assumptions about the distribution of the data and is nonparametric. For those grid points where we detect changes, i.e., the null hypothesis that M 1 and M i (E 1 and E i ) are drawn from the same population is rejected at the 5% significance level, we quantify changes in seasonal precipitation means and extremes. [10] For heavy-tailed distributions the K-S test may not be optimal as it gives relatively little weight to the tail. There is a large family of related tests, such as the Anderson-Darling test [Anderson and Darling, 1954; Von Storch and Zwiers, 2003] that give more weight to the tails than does the K-S test. It uses statistics that are more difficult to compute, but that are also more powerful and more sensitive to departures from the hypothesized distribution in the tails of the distribution. The K-S test is distribution-free in the sense that the critical values do not depend on the specific distribution being tested. The Anderson-Darling test makes use of the specific distribution in calculating critical values. This has the advantage of allowing a more sensitive test and the disadvantage that critical values must be calculated for each distribution. We prefer to use the K-S test for both mean and extremes Generalized Extreme Values and Empirical Distributions [11] To quantify changes in the mean precipitation we use the 50th percentile (i.e., the median) of the empirical distribution for the set of 425 seasonal means. Given an ordered sample of independent observations x ð1þ x ð2þ :: x ðnþ 2of11

3 from a population with distribution function F, the empirical distribution function is defined by [Coles, 2001] ~F ðx i Þ ¼ i n þ 1 for x ðþ i x < x ðiþ1þ : ð3þ [12] Changes of extremes are quantified by the difference of the 100 year return value of precipitation between the two periods. The return time T(x) for a value x is the average time between two occurrences of x. To estimate the 100 year return values we use the generalized extreme value (GEV) distribution. According to the extremal types theorem [Coles, 2001] the GEV provides a model for the distribution of block (annual or seasonal) maxima. It is given by h Gx ðþ¼exp 1 þ x x m s i 1=x : ð4þ The parameter m describes the location of the distribution, the scale parameter s its width, and the shape parameter x its asymmetry, determining the behavior of the tail. G(x) is defined for {x :1+x x m s > 0} where the parameters satisfy < m <+, s > 0 and < x <+. The cases x =0, x > 0 and x < 0 define the widely known Gumbel, Frechét, and Weibull distribution families, respectively. [13] The GEV is fitted to the sets E i of 425 block (seasonal) maxima, using the maximum-likelihood fitting implemented in the R package ismev ( packages/ismev/index.html). The T-return value is given by the 1 1/T(x) percentile of G, i.e., 1 Tx ðþ¼ 1 Gx ðþ : Inserting (4) and inverting gives for the 100 year return value ( T 100 ¼ m s h i 1 ln 0:99 x for x 0 m slnð ln 0:99Þ for x ¼ 0 [14] This equation shows that the increase of any of the three parameters leads to an increase of the 100 year return value. An increasing location parameter describes a shift of the whole distribution toward higher values, an increasing scale parameter a widening, indicating higher variability, and an increasing (decreasing) positive shape parameter a lengthening (shortening) of the high-end tail, indicating a tendency toward rare, but extreme events. This is again an increasing variability, but only toward the high extremes. The 100 year return level uncertainty was estimated as the 95% level of confidence by using the delta method as described by Coles [2001] Extreme Periods [15] To define very wet (dry) periods we use the 90th (10th) percentile of data sets M 1 and E 1 (equations (1) and (2)) from the reference period For each studied season (DJF and JJA) we define two bivariate thresholds, C wet and C dry.c wet is that point in the (mean, extreme) plane corresponding to the 90th percentiles of the sets M 1 and E 1, whereas C dry corresponds to the 10th percentiles. Figure 1 shows a graphical example of our calculation for the grid ð5þ ð6þ point corresponding to the location of De Bilt, Netherlands (5.6 E, 51.3 N). A season (DJF or JJA) is a wet (dry) event if both its seasonal mean and peak are above (below) the (mean, extreme) coordinates of C wet (C dry ). We count the number of such wet (dry) events for each 25 year period and estimate their future behavior. Summarizing, a wet (dry) event is a season (DJF or JJA) in some year in which both mean and maximum precipitation are extraordinarily high (low). A wet (dry) period is a 25 year period in which the number of wet (dry) events is higher (lower) than the number of wet (dry) events in the reference period. 3. Results 3.1. Model Output Evaluation [16] Comparisons of seasonal precipitation means and extremes as derived from ECHAM5 model output and GPCP data are shown in Figures 2 and 3. In order to calculate the ESSENCE-GPCP differences the GPCP data were interpolated onto the ECHAM5 grid (T63). The time period used for model output evaluation is between 1997 and 2009, when it is possible to get daily GPCP observations (see section 2.2). For each of the 13 years of the evaluation period we calculate the mean and maximum for each studied season (DJF and JJA) from both ECHAM5 and GPCP and compare their medians (Figures 2 and 3). Qualitatively, observation and model data compare very well. The dry and wet patterns, for both seasonal mean and extremes, are well reproduced (Figures 2a 2d and 3a 3d). Quantitatively, the absolute differences (Figures 2e, 2f, 3e, and 3f) over 80% 90% of the global area do not exceed 2 mm/d and 10 mm/d for seasonal means and extremes, respectively. The relative differences (Figures 2g, 2h, 3g, and 3h) between model and Essence GPCP observations, defined as GPCP 100, are estimated to be around 30% in the extratopics with higher values in dry areas/seasons. Over very dry regions (e.g., Sahel) where we compare numbers very close to zero, Figures 2e 2h and 3e 3h show very small ( 0) absolute, but very large (between 80% and 160%) relative differences. Such differences are well within the range of published model/data discrepancies [Kharin et al., 2007, Sun et al., 2007, Min et al., 2011, Allan and Soden, 2008]. Sun et al. [2007] found that the ECHAM5 model underestimates the observed heavy precipitation (>20 mm/d) and overestimates the intensity and amount of light to moderate precipitation (<20 mm/d). This is confirmed here. For instance, we find the model to overestimate mean summer precipitation (Figure 2) over much of the continental landmasses north of about 40 N, where seasonal means are moderate, while the heavy winter extremes in the same areas are underestimated (Figure 3), as are the tropical maxima. [17] In general, the model shows a fair degree of correspondence with the observations. Therefore we have some confidence that the model output can be used to infer precipitation changes in a future climate GEV Goodness of Fit [18] To test the goodness of fit a K-S test was carried out between the empirical distribution function of a sample E i (i = 1,,6, see equation (2)) and the fitted GEV distribution function defined by the parameters m, s and x. The K-S test is a popular goodness-of-fit statistic that measures the distance 3of11

4 Figure 1. The calculation method used to define a dry or a wet period. The example is for the winter (DJF) season at the location of De Bilt, Netherlands (5.6 E, 51.3 N). On the x and y axis we report the seasonal mean and extreme values, respectively, expressed in mm/d. The two black open circles are the points corresponding to the C dry and C wet bivariate thresholds. The open gray circles and gray pluses are the precipitation values within reference period 1 ( ) and period 6 ( ), respectively. The dashed black lines define areas with seasonal mean and maximum precipitation values below the C dry or above the C wet thresholds. The black pluses represent the occurrence of dry (values below the C dry point) and wet periods (values above the C wet point) counted in period 6. The number of black pluses above C wet is 48 (11.3%), while the number of gray open circles above C wet is equal to 12 (2.8%), meaning that in De Bilt in the future the probability of having a wet DJF period is 4 times larger than during the reference period. A similar discussion applies for the dry periods. between the empirical distribution function and the specified distribution [Von Storch and Zwiers, 2003]. The p values of the goodness of the GEV fit for the last studied period ( ) are shown in Figure 4. In the other periods the results look quite similar. [19] Except for the very dry regions, where the p values range between 0.1 and 0.15, the GEV model fits the ESSENCE precipitation data with high confidence, showing a level of significance of at least 5%. The grid points where the GEV model fits with a lower confidence ( p values between 0.1 and 0.15) are localized in the surrounding of the very arid regions masked out from our analysis (represented in Figure 4 as the dotted black area). In these regions also the 100 year return level uncertainty, estimated by using the 95% level of confidence (see section 2.4) and expressed as percent error of the 100 year return value, is higher (up to 10%) than in the extratropical and subtropical regions, where it does not exceed 6% (not shown). In general we can say that due to the large number of samples per time slice (425) the resulting estimates of the distribution parameters have small error bars Mean and Extremes Intensity Changes [20] Figures 5 (left) and 6 (left) show relative changes of the median of seasonal mean, and Figures 5 (right) and 6 (right) show 100 year return values for periods 3 ( ), 5 ( ), and 6 ( ), expressed as percentage of the reference period ( ). [21] Away from the very dry regions the percentage of global area showing significant seasonal precipitation changes increases from 40% in period 3 ( ) to 90% in period 6 ( ). The extremes increase nearly everywhere. Decreases occur mainly in JJA and in regions that are much smaller than those for which mean precipitation is projected to decrease. Similar results were obtained by Kharin et al. [2007] for the annual maximum precipitation. Changes in the period have the same patterns as those in the later periods, but lower magnitude, a result that was also found for temperature [Sterl et al., 2008; Russo and Sterl, 2011]. Over the wet extratropical areas the largest changes occur in winter (DJF). The dry tropicalsubtropical areas show the same intensity of changes during both seasons, but in summer (JJA) the dry pattern is more extended than in winter (DJF) (Figures 5 (left) and 6 (left)). [22] During both winter and summer seasons, seasonal mean precipitation change is generally negative between 45 S and 45 N, while the magnitude of extreme precipitation is mainly increasing in the same zonal band. An exception to this general rule is the Intertropical Convergence Zone (ITCZ), where both mean and extreme precipitation increase. These results are consistent with those of Kharin et al. [2007] and Sun et al. [2007]. Using a multimodel ensemble, Kharin et al. [2007] calculated the median of yearly mean precipitation and 20 year return values of yearly maximum precipitation. They found that 20 year return values are increasing nearly everywhere, while mean precipitation increases in the tropics and in the midlatitudes and high latitudes, but decreases in the subtropics. Sun et al. [2007] showed that light to moderate precipitation increases over most of the globe, except for subtropical regions, where it decreases significantly. As nearly all precipitation in these areas is light to moderate, this fits with our results. 4of11

5 Figure 2. Median of seasonal mean precipitations (mm/d) in (left) DJF and (right) JJA for (a and b) Essence and (c and d) GPCP data. Regions where the median of seasonal precipitation means is less than 0.1 mm/d are masked out in white. ESSENCE-GPCP (e and f) absolute and (g and h) relative differences ( Essence GPCP GPCP 100). [23] In our study the band with decreasing mean and increasing extremes is zonally moving by about 10 during the annual cycle. This is the same movement as the present climatological seasonal cycle. To summarize, the dry (subtropical) areas will become even drier and may also expand poleward. [24] As an example (Figure 6) we focus on a particularly vulnerable (highly populated) area that is affected by this drying, namely southern Europe. While the mean precipitation on the north shore of the Mediterranean seems to remain fairly constant during DJF, the southern shore experiences strong drying. During summer, the drying area moves northward as far as southern Scandinavia. At the same time, extreme rainfall tends to increase, with a larger increase in winter (DJF) than in summer (JJA). Other areas around 40 N/S experience similar changes. 5of11

6 Figure 3. As Figure 2 but for the median of the seasonal maxima (in mm/d). In Figures 3a 3d the regions where the median of seasonal precipitation maxima is less than 1 mm/d are masked out in white. [25] Outside of the 45 S 45 N geographical band, both mean and extreme precipitation increase throughout each season. This corroborates results for annual values as found by Kharin et al. [2007], Sun et al. [2007] and Min et al. [2011]. The magnitude of the increase grows with time. The increase of the maxima (100 year return values) is up to ten times higher than that of the seasonal means Correlation Between Means and Extremes [26] In section 3.3 we showed that in some regions mean and extreme precipitation both increase (decrease), while in others they change in opposite direction. We now assess the significance of this result by using the Kendall rank correlation coefficient, commonly referred to as Kendall s tau (t) coefficient [Ruppert, 2011]. It is a statistic used to measure the association between two data sets even if the number of data points is small. In our case we have six mean and peak values in each grid point, one for each of the six 25 year periods. [27] In this analysis we only consider correlations and corresponding p values at every grid point. This is called the local test approach because a local null hypothesis is tested 6of11

7 Figure 4. The p values of the Kolmogorov-Smirnov test performed to evaluate the goodness of GEV fit for period 6 ( ). The areas marked by pluses and crosses are very dry regions where precipitation changes are not investigated (see section 2.3). at each grid point [VonStorchandZwiers, 2003]. As explained by, e.g., Von Storch and Zwiers [2003], Ventura et al. [2004], Wilks [2006], and Sterl et al. [2007], the local test decision cannot be used at a regional or global scale when a univariate hypothesis test is used at multiple locations. This means that we cannot average locally calculated p values over a region, and when we talk about changes in a region, we effectively look at a composite of single points. To calculate the level of significance ofa region or globally we should apply multivariate tests as explained by Von Storch and Zwiers [2003]. [28] Kendall s tau correlation values are shown in Figure 7 (top left, winter (DJF) and bottom left, summer (JJA)). The corresponding p values, expressed as percentages, are shown Figure 5. Relative change in DJF of (left) the median of seasonal means and of (right) the 100 year return period of seasonal maxima for (top) period 3 ( ), (middle) period 5 ( ), and (bottom) period 6 ( ). The values are expressed as percentage of relative change with respect to period 1 ( ). The color bar is in percent, and changes that, according to the results of the Kolmogorov- Smirnov test applied to both seasonal precipitation mean and maximum values, are not statistically significant at the 5% level are masked out in white. The areas marked by pluses and crosses are very dry regions where precipitation changes are not investigated (see section 2.3). 7of11

8 Figure 6. As Figure 5 but for JJA. in Figure 7 (right). Over the extratropical areas we see a strong correlation signal. It is largest in Eurasia and North America during winter (DJF), when Kendall s tau values exceed 0.8 with a significance level of 5% (Figure 7, top). In summer the correlation values are lower (Figure 7, left), and the area where they are significant is smaller (Figure 7, right). High positive correlations with corresponding high p values are also found near the equator. Here the signal is Figure 7. (left) Kendall s tau correlation between (top) winter (DJF) and (bottom) summer (JJA) median and 100 year return level of seasonal maxima and (right) corresponding p values. 8 of 11

9 Figure 8. Relative change (in percent) of occurrence of wet periods with respect to reference period 1 ( ) for (top) period 3 ( ), (middle) period 5 ( ), and (bottom) period 6 ( ). The grid points where, according to the results of the Kolmogorov-Smirnov test applied to both seasonal precipitation mean and maximum values, precipitation changes are not statistically significant at the 5% level are masked out in white. most pronounced in JJA, indicating a strengthening and/or slight meridional shift of the ITCZ at its most northerly position. In the dry subtropical areas the mean-extreme correlation values differ from those in the rest of the world. They are generally negative, but with a low confidence during both seasons, reflecting the fact that the percentages of change for both mean and extreme precipitation are lower in these areas than elsewhere. [29] Figure 7 confirms the findings from section 3.3. In the extratropics and along the equator both seasonal means and extremes increase, and in the subtropical areas the means are decreasing while the extremes increase. The negative correlation shows high significant values only in very localized areas (for example southern Australia and northwestern Pacific in DJF). In areas where the p value does not exceed 0.1 (Figure 7, left) the correlation coefficient has only positive values. Together this means that there are only very small regions on the globe with significant opposite changes between seasonal means and extremes Occurrence of Wet and Dry Periods [30] We have seen that in many extratropical regions, mainly during boreal winter, the intensities of both seasonal means and peaks are strongly increasing. We now describe present and future changes in the occurrence of wet and dry periods as defined in section 2.5. For this study we consider only those grid points where seasonal means and peaks are at the same time changing at the 5% significance level as determined by the Kolmogorov-Smirnov test (see section 2.3). Results for wet and dry periods are shown in Figures 8 and 9, respectively. [31] Over the northern extratropical areas the occurrence of wet periods is increasing more during DJF (Figure 8, left) than during JJA (Figure 8, right). This corroborates the findings from the mean and extreme intensity changes in section 3.3. [32] In the same zonal band the occurrence of dry periods is decreasing (the changes are negative) by between 1% and 10% during both seasons. Thus the wet regions are becoming less dry, and in the future the probability to have a season with seasonal precipitation mean and maximum below the values given by C dry of the reference period ( ) is decreasing. So we can say that the 25 year bivariate distributions of seasonal mean and peaks are shifting in the direction of wettening. Furthermore, the area showing significant changes is larger in DJF than in JJA. [33] Over the tropics and subtropics areas where the occurrence of dry periods increases are mixed with areas where it decreases. This again is in accordance with corresponding results from section 3.3. While the decrease of the 9of11

10 Figure 9. As Figure 8 but for dry periods. number of wet seasons is generally less than 5%, the number of dry periods is increasing in some (sub)tropical areas by up to 30%. 4. Discussion and Conclusion [34] In this study we used a large ensemble from one model, driven by one emission scenario. Unfortunately, available computing resources did not allow us to perform a second ensemble using another emission scenario. However, as can be seen from Figures 5 and 6, the response of the model to increasing GHG levels is roughly linear suggesting that a more rapid increase of GHGs would produce the period 6 results earlier in time. The use of only one model is a more serious limitation. Precipitation, and especially heavy precipitation, is strongly dependent on details of the parametrization [e.g., Emori et al., 2005]. A multimodel ensemble or a perturbed physics ensemble might be preferable and yield different results. However, our results are in line with those of Kharin et al. [2007], who use a multimodel ensemble, and with those of Sun et al. [2007], who use 14 climate models with three emissions scenarios (SRES B1, A1B and A2). Complementing these studies our use of a large singlemodel ensemble has the advantage of strongly reducing the statistical uncertainty. [35] Summarizing our results, in all seasons the mean precipitation is increasing poleward of 40 and decreasing equatorward of these latitudes, except for the ITZC, where it increases. Dry areas become drier and wet areas become wetter. The position of the boundary between areas becoming drier/wetter moves zonally by about 10 during the annual cycle. While the mean winter (DJF) precipitation over southern Europe experiences only small changes over the simulation period, summer (JJA) precipitation declines as far northward as southern Scandinavia. [36] Precipitation peaks behave differently. They are increasing over most of the globe and in all seasons. They decrease, or increase only slowly, only in a small fraction of the tropical-subtropical area, where seasonal mean precipitation decreases. Increases in the intensity of precipitation extremes exceed those of mean precipitation nearly everywhere. [37] The magnitude of the precipitation changes increases with time. Maximum changes are found in the respective winter hemisphere. Throughout the simulation period the spatial patterns of change are the same with dry areas becoming drier and wet areas becoming wetter. [38] Acknowledgments. This work was funded by the EU (contract MRTN-CT ) as part of the SEAMOCS research training network. The ESSENCE project, led by Wilco Hazeleger (KNMI) and Henk Dijkstra (UU/IMAU), was carried out with support of DEISA, HLRS, SARA, and NCF (through NCF projects NRG , CAVE , and SG ). We thank the DEISA Consortium (cofunded by the EU, FP6 projects and ) for support within the DEISA Extreme Computing Initiative ( The authors thank Camiel Severijns (KNMI) and HLRS and SARA staff for technical support. The plots have been produced using the free Ferret software developed at NOAA/PMEL ( 10 of 11

11 References Allan, R. P., and B. J. Soden (2008), Atmospheric warming and the amplification of precipitation extremes, Science, 321, , doi: / science Anderson, T. W., and D. A. Darling (1954), A test of goodness of fit, J. Am. Stat. Assoc., 49, Biasutti, M., and A. Giannini (2006), Robust Sahel drying in response to late 20th century forcings, Geophys. Res. Lett., 33, L11706, doi: /2006gl Christensen, J. H., et al. (2007), Regional climate projections, in Climate Change 2007: The Physical Science Basis, Contribution of Working Group I to the Fourth Assessment Report on the Intergovernmental Panel on Climate Change, edited by S. Solomon et al., pp , Cambridge Univ. Press, Cambridge, U. K. Coles, S. (2001), An Introduction to Statistical Modeling of Extreme Values, 208 pp., Springer, Berlin. Cook, K. H., and E. K. Vizy (2006), Coupled model simulations of the West African monsoon system: Twentieth- and twenty-first-century simulations, J. Clim., 19, , doi: /jcli Douville, H., D. Salas-Melia, and S. Tyteca (2006), On the tropical origin of uncertainties in the global land precipitation response to global warming, Clim. Dyn., 26, Emori, S., A. Hasegawa, T. Suzuki, and K. Dairaku (2005), Validation, parameterization dependence, and future projection of daily precipitation simulated with a high-resolution atmospheric GCM, Geophys. Res. Lett., 32, L06708, doi: /2004gl Giannini, A. (2010), Mechanisms of climate change in the semiarid African Sahel: The local view, J. Clim., 23, Iacus, S. M., and G. Masarotto (2007), Laboratorio di Statistica con R, 2nd ed., 396 pp., McGraw-Hill, Milan, Italy. Intergovernmental Panel on Climate Change (2007), Climate Change 2007: The Physical Science Basis, Contribution of Working Group I to the Fourth Assessment Report on the Intergovernmental Panel on Climate Change, edited by S. Solomon et al., 996 pp., Cambridge Univ. Press, Cambridge, U. K. Kharin, V. V., and F. W. Zwiers (2000), Changes in the extremes in an ensemble of transient climate simulations with a coupled atmosphere ocean GCM, J. Clim., 13, Kharin, V. V., and F. W. Zwiers (2005), Estimating extremes in transient climate change simulations, J. Clim., 18, Kharin, V. V., F. W. Zwiers, X. Zhang, and G. C. Hegerl (2007), Changes in temperature and precipitation extremes in the IPCC ensemble of global coupled model simulations, J. Clim., 20, , doi: / JCLI Kiktev, D., D. Sexton, L. Alexander, and C. Folland (2003), Comparison of modeled and observed trends in indices of daily climate extremes, J. Clim., 16(22), Kiktev, D., J. Caesar, L. V. Alexander, H. Shiogama, and M. Collier (2007), Comparison of observed and multimodeled trends in annual extremes of temperature and precipitation, Geophys. Res. Lett., 34, L10702, doi: /2007gl Marsland, S. J., H. Haak, J. H. Jungclaus, M. Latif, and F. Roske (2003), The Max-Planck-Institute global ocean/sea ice model with orthogonal curvilinear coordinates, Ocean Modell., 5, Min, S., X. Zhang, F. W. Zwiers, and G. C. Hegerl (2011), Human contribution to more-intense precipitation extremes, Nature, 470, , doi: /nature Nakicenovic, N., et al. (2000), Special Report on Emissions Scenarios: A Special Report of Working Group III of the Intergovernmental Panel on Climate Change, 599 pp., Cambridge Univ. Press, Cambridge, U. K. [Available at Roeckner, E., et al. (2003), The atmospheric general circulation model ECHAM 5. Part I: Model description, Rep. 349, 127 pp., Max Planck Inst. for Meteorol., Hamburg, Germany. Ruppert, D. (2011), Statistics and Data Analysis for Financial Engineering, 638 pp., Springer, New York. Russo, S., and A. Sterl (2011), Global changes in indices describing moderate temperature extremes from the daily output of a climate model, J. Geophys. Res., 116, D03104, doi: /2010jd Semenov, V. A., and L. Bengtsson (2002), Secular trends in daily precipitation characteristics: Greenhouse gas simulation with a coupled AOGCM, Clim. Dyn., 19, , doi: /s Sterl, A., G. J. van Oldenborgh, W. Hazeleger, and G. Burgers (2007), On the robustness of ENSO teleconnections, Clim. Dyn., 29, , doi: /s z. Sterl, A., C. Severijns, W. Hazeleger, G. Burgers, B. van den Hurk, G. J. van Oldenborgh, P. van Velthoven, H. Dijkstra, P. J. van Leeuwen, and M. van den Broeke (2008), When can we expect extremely high surface temperatures?, Geophys. Res. Lett., 35, L14703, doi: / 2008GL Sun, Y., S. Solomon, A. Dai, and R. Portmann (2007), How often will it rain?, J. Clim., 20, , doi: /jcli U.S. Climate Change Science Program (2008), Weather and climate extremes in a changing climate: Regions of focus: North America, Hawaii, Caribbean, and U.S. Pacific islands, a report by the U.S. Climate Change Science Program and the Subcommittee on Global Change Research, Syn. Assess. Prod. 3.3, edited by T. R. Karl et al., 164 pp., NOAA, Washington, D. C. Ventura, V., C. J. Paciorek, and J. S. Risbey (2004), Controlling the proportion of falsely rejected hypotheses when conducting multiple tests with climatological data, J. Clim., 17, Von Storch, H., and F. Zwiers (2003), Statistical Analysis in Climate Research, 484 pp., Cambridge Univ. Press, Cambridge, U. K. Voss, R., W. May, and E. Roeckner (2002), Enhanced resolution modelling study on anthropogenic climate change: Changes in extremes of the hydrological cycle, Int. J. Climatol., 22, Wehner, M. F. (2004), Predicted twenty-first-century changes in seasonal extreme precipitation events in the Parallel Climate Model, J. Clim., 17, Wilby, R. L., and T. M. L. Wigley (2002), Future changes in the distribution of daily precipitation totals across North America, Geophys. Res. Lett., 29(7), 1135, doi: /2001gl Wilks, A. (2006), On field significance and the false discovery rate, J. Appl. Meteorol. Climatol., 45, S. Russo, Institute for Environmental Protection and Research, Via Curtatone 3, Rome I-00185, Italy. (simone.russo@isprambiente.it) A. Sterl, Royal Netherlands Meteorological Institute, PO Box 201, NL-3730 AE De Bilt, Netherlands. (sterl@knmi.nl) 11 of 11