Future development in extreme one-hour precipitation over Europe due to climate change

Transcription

1 Future development in extreme one-hour precipitation over Europe due to climate change A.N. Larsen 1 *, I.B. Gregersen 1, O.B. Christensen 3, J.J. Linde 2, P.S. Mikkelsen 2 1 Student, Technical University of Denmark, Anker Engelundsvej 1, DK-2800 Kgs. Lyngby, Denmark 2 Department of Environmental Engineering, Technical University of Denmark, Bygningstorvet, Bldg 115, DK Kgs. Lyngby, Denmark 3 Danish Meteorological Institute, Lyngbyvej 100, DK-2100 København Ø, Denmark *Corresponding author (agnethe.n.l@hotmail.com) ABSTRACT In this study the potential increase of extreme precipitation in a future warmer European climate has been examined. Output from the regional climate model HIRHAM4 covering Europe has been analysed for two periods, a control period and a scenario , the latter following the IPCC scenario A2. The model has a resolution of about 12 km, which makes the results unique in relation to extreme precipitation events critical to urban drainage systems. Extreme events with one- and 24-hour duration were extracted using the Partial Duration Series approach, a Generalized Pareto Distribution was fitted to the data and T-year events for return periods from 2 to 100 years were calculated for the control and scenario period in model cells across Europe. The analysis shows that there will be an increase of the intensity of extreme events generally in Europe; Scandinavia will experience the highest increase and southern Europe the lowest. A 20 year 1-hour precipitation event will e.g. become a 4 year event in Sweden and a 10 year event in Spain. Intensities for small durations and high return periods will increase the most, which implies that European urban drainage systems will be challenged in the future. KEYWORDS Climate factor; Europe, Extreme precipitation, Generalized Pareto Distribution, IPCC SRES A2 INTRODUCTION European countries have experienced very extreme rain events during the past decade, which have often been accompanied by an exceedance of the drainage capacity in urban areas causing flooding damages, traffic-related problems and even death. A relevant question is whether these events are yet another indication of a human caused climate change and if extreme precipitation events can therefore be expected to become even more frequent and extreme in the future. Many authors have evaluated the potential future change of extreme precipitation over Europe on the basis of the different climate models assembled under the European PRUDENCE project (Christensen & Christensen, 2007). Frei et al. (2006) e.g. compared six regional climate models (RCMs) in relation to their predictions of the changes in 1- and 5-day precipitation intensity with return periods from 5 to 50 years. Beniston et al. (2007) looked at summer maximum 1 day and winter maximum 5-day mean precipitation, and May (2007) analysed daily extreme precipitation. Common for all the mentioned studies is that they work with regional models with a spatial resolution of around 50 km, use a parametrical general extreme value (GEV) model to compute the extreme events and show significant increases of extremes in a future warmer climate for time scales of 24 hours or more. Jørgensen et al. Larsen et al. 1

2 (2006) analysed RCM output with a 12 km spatial resolution, however using a nonparametrical statistical method and constraining the analysed area to Denmark. The present article considers the future development in one- and 24-hour precipitation over Europe based on RCM output with a 12 km spatial resolution. 1-hour intensity is considered relevant to pluvial flooding occurring in urban areas and 24-hours intensity is included to bridge to fluvial flooding and other studies focusing on higher time scales relevant in nonurban hydrology. METHODS Climate Model and Scenarios Data was abstracted from the regional climate model (RCM) HIRHAM4 from DMI which is based on the global atmosphere model HadAM3. The boundary conditions for the control period are based on observed sea surface temperatures and atmospheric conditions (Christensen et al., 1996). The model area covering the whole of Europe has 131,760 cells in a 12 km spatial resolution. Two 30-year simulation periods are compared; a control period ( ) and a scenario period ( ) (Christensen & Christensen, 2007). The scenario follows IPCC SRES scenario A2, which is a high-middle emission scenario describing a heterogeneous world with little economical integration, a continuously increasing population and a slow technological development (IPCC, 2000). The adopted computational grid is a rotated regular latitude/longitude grid with the rotated South Pole at (27 o E, 37 o S) and a resolution of 0.11 degrees. Statistical methods Data was defined and extracted according to the Partial Duration Series (PDS) approach using Type 2 censoring (e.g.mikkelsen et al., 1995), and the extreme events were fitted to a Generalized Pareto Distribution (GPD) (Madsen et al., 2002; Madsen, 2005). The intensity of a T-year event was therefore estimated by equation 1: ˆ κ ˆ α 1 zˆ = ˆ T β + 1 (1) ˆ κ T λ where β is the location parameter, α is the scale parameter, κ is the shape parameter and λ, the expected number of extreme events per year, is pre-fixed. Further details about the employed statistical methods and the results may be found in Gregersen et al. (in preparation). From the equation above it can be derived that the location parameter is most important for low return periods and that the shape parameter is important for high return periods. The statistical analysis was accomplished by the use of the statistical software EVA (DHI, 2005). Definition of the climate factor A climate factor (CF) is a simple but very useful tool to describe the potential development in extreme precipitation due to climate change. It was defined in this study as the ratio between an extreme rain intensity in the scenario period and in the control period, following the equation given below: zˆ T, scenario ( x, y) CFT ( x, y) = (2) zˆ ( x, y) T, control If the climate factor is above one, the scenario will have higher intensity than the control period, and vice versa. 2 Development in extreme precipitation over Europe

3 RESULTS The number of extreme events per year was set to 2.5 in this study, based on earlier knowledge and experience (Mikkelsen et al., 1999; Madsen 1998; Gregersen & Larsen, 2007). By utilizing a goodness-of-fit test it was also shown that the Generalized Pareto Distribution gives a statistically reasonable description of the extreme events across Europe. Figure 1. Return periods when utilizing both a ranking and a PDS method for a single cell in Denmark for control and scenario, respectively. The statistical uncertainty of the PDS estimates due to sampling variability is shown as the 95% confidence limits (dotted lines). Figure 1 shows two methods of evaluating the extreme events; the non-parametric ranking method commonly used in urban drainage and the parametric PDS-GPD method. Here it can be seen that the parametric method and the ranking method give similar results for return periods below one third of the length of the time series as also concluded by Arnbjerg-Nielsen (1997). For higher return periods the parametric approach is most usable to describe data, as extrapolation is possible and robust estimates insensitive to outliers can be calculated. It is however worth noticing that the uncertainty will become larger for higher return periods for both methods. Figure 2. The intensity of a 5-year event for control (left) and future scenario (right). The increased precipitation can be seen as a general shift in colours. Larsen et al. 3

4 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK, 2008 Figure 3. Climate factors for 2, 5, 10, 20, 50 and 100 year events for one hour intensity. Maps for 24-hour intensity are available in Gregersen & Larsen (2007). 4 Development in extreme precipitation over Europe

5 T-year events in Europe The 5-year event was calculated on the basis of equation 1 and mapped for control and scenario period, see Figure 2. The shift in colours indicates that there is a general change towards higher intensities in the scenario period, and a huge change in the alpine/mediterranean and Baltic areas is seen. One thing which also is worth noting is that the storm tracks shift from a North-west direction to a more western direction. Figure 3 shows the climate factor for the duration of one hour for return periods between 2 and 100 years. The climate factor is above one in most cases. When the return period increases, the maps become more scattered. This suggests that the climate factor becomes more specific for a certain area but is probably more likely due to variability caused by inherent chaotic features of the RCM. The highest climate factor is seen around the Baltic Sea with climate factors as high as 2. However the model has some problems simulating the temperature in the Baltic Sea (Christensen & Christensen, 2007). The intensity of the 5-year event is also seen to increase above the Alps. Climate factors for specific countries Data from a network of rain gauges has been recorded in Denmark since 1979 and analysis of these data show regional differences (Madsen et al., 1998). Jørgensen et al. (2006) however showed that for Denmark the spatial variation of T-year events do not exhibit the same tendencies if the resolution of the RCM is changed or if the RCM is run with varying boundary conditions. This observation is expected to be valid for Europe too and therefore, to illustrate the potential consequences of climate change for urban drainage practice in different countries average intensities for a given T-year event were calculated for France, United Kingdom, Sweden, Romania, Spain, Austria and Denmark. Figure 4 shows the average T- year event for these countries for control and scenario period, respectively. Figure 4. T-year event for specific European countries for control (left) and scenario (right) period. Table 1 quantifies what can be seen on Figure 4 for the 5, 20 and 100-year events in the control period by listing the climate factor and the corresponding return period of an event with the same intensity in the scenario, as previously suggested by Grum et al. (2006). Sweden stands out from the rest with a high climate factor for all return periods. Here the 100 year event in control period will happen every 13 year in the scenario according to Table 1. Denmark follows Sweden but with a slightly lower climate factor. The reason for this is (see Figure 3) that the areas around the Baltic Sea will have increasing extreme precipitation in the Larsen et al. 5

6 scenario. It is therefore expected that the Baltic countries has similar tendencies as Sweden. France is also seen to have a high climate factor, albeit lower than for the area around the Baltic Sea. The United Kingdom, Romania, Austria and Spain however have lower climate factors for the different return periods. Table 1. T-year event, 1-hour intensity, for the control period for selected countries, together with the corresponding climate factor and return period for the scenario period. Z 5 control [mm/hour] CF T scenario [year] Z 20 control [mm/hour] Impact of duration and return period on the climate factor Figure 5 shows, for Denmark as an example, that the climate factor is highest when the duration is low and the return period is high. It is seen that for a 1-year event the duration is not important and the climate factor will be about 1.1, independent of the duration. For a low duration the climate factor will however be higher, meaning that Denmark will get more highintensive rain for short durations. The tendency shown for Denmark is assumed to be valid for whole Europe too. CF T scenario [year] Z 100 control [mm/hour] CF T scenario [year] [-] [-] [-] France UK Sweden Romania Spain Austria Denmark Figure 5. Duration curve for the average climate factor in Denmark. Climate factors were only available for 1-hour and 24-hour precipitation, and the connecting lines between pairs of data do therefore not represent any deterministic relationship. Annual variation of extreme events Figure 6 shows the change in the yearly variation of the number of extreme events. The number of extreme events is seen to increase in the winter-spring period. For Denmark more than a doubling of the number of extreme event is observed in January. Although the number will increase more in the winter-spring period than during summer, it should be noted that the number of extreme events is still highest in the summer period. This conclusion is in 6 Development in extreme precipitation over Europe

7 agreement with Frei et al. (2006), which (based on 1-day and 5-day precipitation data from RCMs) shows that both the frequency and intensity of the extreme events will increase in the winter, whereas only the intensity will increase in the summer. Figure 6. The factor for the number of monthly extreme precipitation events for different countries. Calculated as the number of events in scenario in a given month divided by the number of events in control in the same month. Uncertainty The uncertainty of the climate factor for Denmark is shown in Figure 7. The uncertainty increases for growing return periods, and this tendency is assumed to be valid for the whole of Europe. The climate factor is above one for return periods below 20 years with a probability of 68%. This means that in the A2 scenario there will be an increase in extreme precipitation for 20 years return period with a certainty of 68%. Figure 7. Confidence intervals for the average climate factor for Denmark. Having the uncertainty on the PDS T-year event estimates in control and scenario, the confidence limits for the climate factor are found by use of the law for propagation of uncertainty. DISCUSSION The climate factor was calculated for areal rainfall (12x12km) at the hourly time scale, whereas rainfall measurements for urban drainage applications are mostly point measurements with 5- or 1-minutes resolution. The intensities of extreme rainfalls at the minute-time scale are of course higher than those of longer durations, and temporal downscaling of the RCM precipitation data prior to calculating climate factors has therefore Larsen et al. 7

8 been suggested. However, the existing temporal downscaling methods (e.g. Olsson, 1998; Onof et al., 2000; Güntner et al. 2001; Arnbjerg-Nielsen, 2008) have not yet proved to preserve the extremity of rainfall at the minute-scale but only at the hourly-scale. An extreme event recorded as a point measurement is furthermore higher than an areal measurement (e.g. Einfalt et al., 1998; Jørgensen et al., 2006), and space-time downscaling, possibly using stochastic high-resolution rainfall time-space simulators (e.g. Willems, 2001), should therefore ideally be performed prior to calculating climate factors. This approach would furthermore potentially allow some level of calibration (or at least comparison) of the extreme characteristics of downscaled rainfall output from RCMs with those calculated based on ground observations of extreme rainfall data, in addition to the comparisons with yearly or seasonal rainfall, temperature and other atmospheric conditions that are conducted today. Such approaches have however not yet been tried out. We therefore argue that climate factors calculated straight-forward as in this study, based on 12 km, 1-hour RCM data representing the highest resolution so far used in time-slice climate change simulations, represent the most trustworthy estimate of the effects of anthropogenic climate change on future extreme precipitation intensities relevant for urban drainage. Figure 7 shows only the uncertainty in Denmark, but we expect that similar patterns exists for other regions in Europe. The confidence interval is seen to increase with an increasing return period, which is natural due to the inherent uncertainty associated with estimating extreme events corresponding to return periods at or exceeding the 30-years simulated data used to represent both the present and future climate. Figure 7 essentially means that the probability of the climate factor being larger than one is 68% for return periods between 2 and 20 years, and lower for higher return periods, and we therefore cannot argue with great statistical confidence that the extreme precipitation will increase. An important precondition is also the climate change scenario chosen as background for the RCM scenario simulations. The IPCC A2 scenario represents a high-middle scenario in terms of atmospheric CO 2 concentration, and there are several other IPCC scenarios that may, or may not be taken as a starting point for an analysis. Such scenario uncertainty is probably a larger contributor to the total uncertainty related with extreme rainfall projections based on RCM simulations than the lack of calibration with extreme rainfall data and the uncertainty represented by Figure 7. The point is however not whether the actual climate factors presented in this study are accurate but the general conclusions that (i) it is the most extreme rainfall properties in terms of (high) return period and (low) duration that will be affected the most, and (ii) with high uncertainty comes high potential consequence to society. Whether urban drainage designers should react on the climate factor suggested in this study is therefore basically a philosophical question related to the interpretation of the Precautionary Principle. We cannot (yet) say for sure that the future warmer climate in Europe will bring more extreme rainfall intensities, but there is a lot of scientific theory as well as practical evidence (i.e. observations) that point in this direction. This calls for a complete change of philosophy in terms of future planning, design and rehabilitating urban sewer systems in the direction of flexible/adaptive approaches that are robust to an uncertain future, as well as further research both in the underlying phenomena and in methods for risk-based planning and design. 8 Development in extreme precipitation over Europe

9 CONCLUSION In this study the potential increase of extreme precipitation in a future warmer climate has been examined, based on data from RCM simulations with 12 km resolution covering Europe. Two periods were analysed, a control period and a scenario , the latter following the IPCC scenario A2. The number of extreme events generally increases in the winter and spring season, however the total amount of extreme events will still be highest in the summer period. A climate factor expressing the relative difference between future and present extreme 1- and 24-hours precipitation intensities was furthermore calculated for return periods from years. The climate factor is generally above one across Europe, i.e. increasing extreme precipitation can be expected anywhere in Europe. The factor is however high especially around the Baltic Sea (1.5-2 for a 1-hour duration) and in the Middle of Europe, whereas it is more moderate in southern Europe (around 1.2 in Spain). Generally, the climate factor increases with increasing return period and decreasing duration, i.e. it is the storms that are most critical to urban drainage systems that are affected the most. Another way to express this is by looking at the future return period of storms that today are considered 100 years storms. A 100-year event in the control period for Sweden will have a return period of 13 years in scenario. Damages related to urban flooding will therefore occur more frequently if this prediction is valid, because most urban drainage systems are not made to handle 100-year events. ACKNOWLEDGEMENT We thank the Danish Meteorological Institute for providing the RCM data analysed in this work. Also thanks to Henrik Madsen, DHI, who helped setting up the EVA software and to Alex T. Jørgensen, PH-Consult, for providing basic Matlab programs for treatment of the raw RCM data. REFERENCES Arnbjerg-Nielsen, K. (1997) Statistical analysis of urban hydrology with special emphasis on rainfall modeling. Ph.D thesis. Department of Environmental Science and Engineering, Technical University of Denmark Arnbjerg-Nielsen, K. (2008): Quantification of climate change impacts on extreme precipitation used for design of sewer systems. 11th Int. Conf. on Urban Drainage, Aug-Sep 2008 Beniston, M., D. B. Stephenson, O. B. Christensen, C. A. T. Ferro, C. Frei, S. Goyette, K. Halsnaes, T. Holt, K. Jylhä, B. Koffi, J. Palutikof, R. Schöll, T. Semmler & K. Woth (2007) Future extreme events in European climate: an exploration of regional climate model projections, Climatic Change, 81, DOI /s z Christensen, J. & O. B. Christensen (2007) A summary of the PRUDENCE model projections of changes in European climate by the end of this century. Climatic Change, Special Issue: PRUDENCE. DMI, Copenhagen, Denmark. Christensen, J. H., O. B. Christensen, P. Lopez, E. V. Meijgaard & M. Botzet (1996). The HIRHAM4 Regional Atmospheric Climate Model, Scientific Report DMI, Copenhagen, Denmark (More at ) DHI (2005). DHI software. Online Einfalt, Th., G. Johann & A. Pfister (1998), On the spatial validity of heavy point rainfall measurements, Water Science & Technology, 37(11), 21-28, DOI: s (98) Frei, C., R. Schöll, S. Fukutome, J. Schmidli and P. L. Vidale (2006), Future change of precipitation extremes in Europe: Intercomparison of scenarios from regional climate models, Journal of Geophysical Research, 111 D06105, doi: /2005JD Larsen et al. 9

10 Gregersen, I. B. & A. N. Larsen (2007). Development in extreme precipitation over Europe due to climate change. B.Sc. Thesis. Institute of Environment & Resources, Technical University of Denmark. Gregersen, I. B., A. N. Larsen, O. B. Christensen, J. J. Linde & P. S. Mikkelsen (in preparation) Grum, M., A.T. Jørgensen, R.M. Johansen, and J.J. Linde (2006), The effect of climate change in urban drainage: An evaluation based on regional climate model simulations, Water Science & Technology, 54(6-7), doi: /wst Güntner, A., Olsson, J., Calver, A. and Gannon, B., Cascades-based disaggregation of continuous rainfall time series: the influence of climate. Hydrology and Earth System Sciences, 5(2): IPCC (2000) Emissions Scenarios. Technical report, Intergovernmental Panel on Climate Change. Jørgensen, A.T., J.J. Linde, O.B. Christensen and P.S. Mikkelsen (2006). Future extreme precipitation in Denmark: An analysis based on future climate scenario simulation with a 12x12 km regional climate model. Proc. 7 th international workshop on precipitation in urban areas, St. Moritz, Switzerland, 7-10 December Madsen, H. (1998). Ektremregn i Danmark, Statistisk bearbejdning af nedbørsdata fra Spildevandskomiteens regnmålersystem (translated: Extreme precipitation in Denmark, A statistical analysis of data from the Danish Water Pollution Committee s rain gauge system ). Technical report.,technical University of Denmark, Copenhagen, Denmark. Madsen, H, P.S. Mikkelsen, D. Rosbjerg and P. Harremoës (2002) Regional estimation of rainfall intensityduration-frequency curves using generalized least squares regression of partial duration series statistics, Water Resources Research, 38(11). doi: /2001WR Madsen, H. (2005) EVA Extreme Value Analysis Technical Reference and Documentation. DHI Water and Environment. Denmark. Madsen, H, P.S. Mikkelsen, D. Rosbjerg and P. Harremoës (1998) Estimation of regional intensity-durationfrequency curves for extreme precipitation, Water Science and Technology, 37(11), doi: /98 May, W. (2007) Potential future changes in the characteristics of daily precipitation in Europe simulated by the HIRHAM regional climate model, Climate Dynamics, DOI /s y Mikkelsen, P.S., P. Harremoës and D. Rosbjerg (1995) Properties of extreme point rainfall II: Parametric data interpretation and regional uncertainty assessment. Atmospheric Research, 37. Mikkelsen, P.S., H. Madsen, K. Arnbjerg-Nielsen, H.K. Jørgensen, D. Rosbjerg and P. Harremoës (1999) Regional variation af ekstrem regn i Danmark (translated: Regional variation of extreme precipitation in Denmark). IDA, The Society of Danish Engineers Danish Water Pollution Committee, Paper no. 26. ISBN: Olsson, J., Evaluation of a scaling cascade model for temporal rainfall disaggregation. Hydrology and Earth System Sciences, 2(1): Onof, C., Chandler, R.E., Kakou, A., Northrop, P.J., Wheater, H.S., Isham V. (2000). Rainfall modelling using Poisson-cluster processes, Stochastic Environmental Research and Risk Assessment, 14, Willems, P., A spatial rainfall generator for small spatial scales. J. Hydrol. 252, Development in extreme precipitation over Europe