Spatial and Temporal Scales and Mechanisms of Extreme Precipitation Events over Central Europe.
2. State of the art, preliminary work
3. Objectives and work schedule
3.2 Work schedule
This project is focused on the quantitative estimation and mechanisms of variability of extreme precipitation. Our main task is to estimate long-term variability of extreme precipitation over Central Europe from high quality observations and to identify the major mechanisms of the observed changes. We will use the unique high density rain gauge daily observations of the DWD network covering Germany for the period from 1880 onwards. Using these data we will design homogenized gridded daily data set with spatial resolutions of 0.05-0.2 degree for 1880-2007. On that basis we will develop an ensemble of extreme precipitation statistics using known and newly developed statistical techniques based on the analysis of probability distributions of daily precipitation. This will allow us to obtain estimates of long-term variability of seasonal and possibly monthly extreme precipitation over Central Europe. These estimates will then be associated with cyclone activity, cyclone life cycle parameters, atmospheric moisture transport and regional moisture recycling. We will quantify the role of large-scale circulation conditions and local processes in long-term variability of extreme precipitation in order to establish predictability limits of extreme precipitation for different seasons over Germany. The project results will contribute to the improvement of seasonal and longer scale prediction of extreme weather to enable a more effective risk assessment and water management in Germany.go up...
During the last decades there has been a growing interest to improve the estimation of extreme precipitation including characteristics of variability and trends on global and even more on regional scales. Estimates of extreme precipitation can be obtained from high resolution (typically daily) rain gauge observations or from numerical simulations with regional or global general circulation models used in data assimilation mode (reanalyses). For Central Europe Klein Tank and Können (2003) using daily rainfall observations identified linear trends in extreme precipitation up to to 5 % per decade between 1945 and 1995. Similar conclusions were derived from continental scale and regional studies by Frei and Schär (2001), Groisman et al. 2005, Zolina et al. (2005, 2008), Brunetti et al. (2004, 2006), Moberg et al. (2006), Alexander et al. (2006) and others which also contributed to the IPCC Fourth Assessment Report (AR4) (Trenberth et al. 2007), and which confirmed an increasing number of heavy precipitation events over the last 50 years over Europe.
At the same time, these and other studies state large uncertainties when estimating the variability of extreme precipitation. First, the estimates of extreme precipitation trends over Europe are quite inhomogeneous in space with locations having even different signs of tendency being superimposed. Figure 1 shows a composite of published spatial patterns of linear trends in extreme precipitation characteristics over Europe for the last 50 years. Klein Tank and Können (2003) reported strong spatial noise in the trends of precipitation during very wet days (i.e. exceeding long-term 95th percentile). Frei and Schär (2001) noticed a strong spatial inhomogeneity in the trend estimates for the Swiss Alps on the centennial time scale. Likewise the gridded extreme precipitation indices from Alexander et al. (2006) do not reveal a robust pattern of tendencies in extreme precipitation over Europe. Similar noisy trend patterns were reported for the North American continent (Kunkel et al. 1999, 2003, Groisman et al. 2005) with spatial trend inhomogeneity especially pronounced in the mountain regions (Beniston 2005).
Figure 1: Compilation of the published spatial patterns of the estimates of linear trends in extreme precipitation characteristics over Europe for the last 50 years.
The spatial inhomogeneity and uncertainty of estimates of extreme precipitation variability can be partly explained by the superposition of quantitatively and qualitatively different tendencies in precipitation for different seasons (i.e. the seasonality in variability of European precipitation). Shabalova and Weber (1999), Datsenko et al. (2001), and Zveryaev (2004) demonstrated that the interannual variability and trends of mean European precipitation are seasonally dependent due to seasonal differences in the circulation patterns, which transport or generate moisture. Extreme precipitation is in addition influenced by local mechanisms which can be quite different during the cold and the warm seasons e.g. by different constrains of the local atmospheric moisture content (Allen and Ingram 2002). Increasing extreme precipitation during the cold season and opposite tendencies during summer over Central Europe were shown by Frei and Schär (2001), Brunetti et al. (2004, 2006), Zolina et al. (2005), Moberg et al. (2006) for centennial records and by Fowler and Kilsby (2003), Hundecha and Bardossy (2005), Zolina et al. (2008) for the last 5 decades. Although estimates generally agree qualitatively, they exhibit significant quantitative differences.
Large uncertainties of estimates of long-term variability in European extreme precipitation make it difficult to evaluate global and regional climate models and reanalyses. Compared to observations climate models (Zwiers and Kharin 1998, Hennessy et al. 1997, Yonetani and Gordon 2001, Semenov and Bengtsson 2002, Watterson and Dix 2003, Kiktev et al. 2003) and reanalyses (Zolina et al. 2004) show spatially more consistent variability and trend patterns, but the differences between the model outputs and/or reanalyses can be quite large. Most climate model experiments hint at increasing occurrences of extreme precipitation under anthropogenic warming conditions (Zwiers and Kharin 1998, Semenov and Bengtsson 2002, Palmer and Raelsaenen 2002, Watterson and Dix 2003). The credibility of these results requires an answer to the question, how capable these models are in simulating the observed changes in extreme precipitation. A prerequisite is, however, a sufficient certainty in the observed variability.
Although comparisons between model results and observational data (Kiktev et al. 2003, Tebaldi et al. 2006) show some consistency with observed changes (Groisman et al 2005, Hegerl et al. 2007), they also reported significant differences in variability patterns. Similarly, large discrepancies with the observations were found for different reanalysis data sets (Zolina et al. 2004) and also in regional climate model simulations over Germany (Bachner et al. 2008). Kunkel et al. (2002) analyzed experiments with regional model forced by lateral boundary conditions from reanalysis over the contiguous United States and noticed a low correlation in the timing of individual heavy events between the model and observations, reflecting differences between model and observations in the speed and path of many of the synoptic-scale events triggering the precipitation. When model results do not compare well with observations, models cannot be used with confidence to analyze the processes; thus we can tell a little about the mechanisms driving extreme precipitation variability over Europe. Recently Scaife et al. (2008) have shown that European patterns of extreme precipitation associated with the North Atlantic Oscillation (NAO) are clearly detectable in model simulations but become very uncertain when derived from the observational data. Zolina et al. (2008b) used more stable metrics for quantifying extreme precipitation, however, the manifestation of NAO-related pattern was improved only marginally. The circulation characteristics associated with the NAO, such as cyclone activity, demonstrate an evident link with mean precipitation, but quite uncertain relationship with precipitation extremes.
We believe, that these discrepancies are to a large extent caused by uncertainties on the observational side and in the analysis techniques. The first source of the uncertainties is the quality of the rain gauge networks used in these comparisons. Continental scale “climate quality” records were surely properly validated and standardized. However, the resolution of these data is too coarse to allow for a sufficiently detailed and consistent analysis of the variability in extreme precipitation and its mechanisms due to the small scale of characteristics of rainfall in general. We can mention here European Climate Assessment (ECA) Data Set (Klein Tank et al. 2002) in which a typical station-to station distance varies from 50 to 200 km. Even large convective systems may pass such a network being unnoticed. These data sets are valuable for the development of gridded precipitation products with monthly resolution, such as CRU data set (New et al. 1999, 2000). Existing higher resolution data obtained by combination of different collections (e.g. NCDC data, Groismann et al. 2005) often portray biases due to different instrumentation and observational practices and due to numerous gaps in the data which can greatly affect estimates of precipitation indices (see, Zolina et al. 2005). In our project we will solve this problem by using a unique homogenized data set of extremely high resolution and quality over Germany. Furthermore, we will develop a regional long-term high resolution gridded data set which will be accompanied with precise estimates of all sources of uncertainties.
The second source of uncertainties originates from the usage of the metrics for quantification of extreme precipitation. Both, empirical extreme precipitation indices (e.g. Klein Tank and Könen 2003, Alexander et al. 2006) and parameters derived from fitted probability density functions (Groismann et al. 1999, Zolina et al. 2005, Panorska et al. 2007) have their strengths and weaknesses and lead to different types of uncertainties in the estimates. Empirically derived indices (Klein Tank and Können 2003, Alexander et al. 2006, Moberg et al. 2006) lead to high computational uncertainties due to the rare nature of extreme events. These indices can be only applied to the daily time series of the whole year to generate significant results, and not to seasonal or monthly series due to statistical reasons (see Zolina et al. 2008b). However, when the indices are based on the analysis of fitted analytical distributions such as Gamma or other PDFs (Wilks 1995, Groisman et al. 1999, Katz 1999, Zolina et al. 2004, Panorska et al. 2007), they also may not be suited to quantify extreme precipitation variability due to the problems of goodness of fit. Furthermore, being traditionally two-parametric, analytical PDFs do not always capture the variability which is associated with changes in the fractional contribution of heavy precipitation to the seasonal/monthly totals. Our project will try to resolve this problem by using newly developed extreme precipitation statistics (Zolina et al. 2008b) which have the advantages of applying the metrics either to empirical or analytical probability distributions.
Finally, the third source of uncertainties, which is specifically related to the understanding of physical mechanisms of precipitation extremes, is the use of too rough indicators of atmospheric circulation variability. We believe that the use of the NAO index (Scaife et al. 2008) or the other leading modes of atmospheric circulation in the Atlantic-European sector quantified through empirical circulation indices or empirical orthogonal functions (EOFs) (Haylock and Goodess 2004) is insufficient because the dominant circulation patterns do not contain all the factors affecting heavy precipitation, such as atmospheric moisture transport by midlatitudinal cyclones and the regional moisture convergence providing pre-conditioning for extreme precipitation (Trenberth and Dai 2003). In our project we will bridge this gap by the analysis of midlatitudinal storm tracks and characteristics of cyclone life cycle which will provide a more explicit characterization of regional atmospheric conditions leading to extreme precipitation in different seasons. Furthermore, the impact of other factors, such as the influence of changing amounts of aerosols (and the associated cloudiness changes by dimming and brightening) on mean and extreme precipitation (Lambert et al. 2008, Previdi and Liepert 2008), are not yet properly accounted for in the analysis of precipitation extremes.go up...
Immediate science questions which will be answered by the project are (i) how quantitatively the extreme precipitation has changed over Germany during the last 127 years? and (ii) which mechanisms are responsible for these changes? To answer these questions we will first address the problems of data homogenization and gridding as well as of the estimation of parameters of extreme precipitation. Further we will consider whether the observed changes occurred due to a linear shift of the precipitation distributions, or due to the changing distribution shape, or due to changes in fractional contribution of very wet days to the total. Being armed with this knowledge, we will determine whether the variability in extreme precipitation results from changing the proportions between convective and frontal rainfall or whether it is driven primarily by the changing cyclone activity, particularly by the shift of European storm tracks. In this analysis we will account also for the latest findings by Wentz et al. (2007) and Previdi and Liepert (2008), who attributed long-term changes in maritime precipitation to global “dimming” and “brightening” caused by the increase and decrease of anthropogenic aerosol and its effects on the precipitation efficiency of clouds (second indirect aerosol effect). It is an important question, whether this is also translated to the European continent and can result in a decrease and increase of extremes, respectively. Being super-imposed on the shorter-scale variability, these effects might mask partially the attribution to dynamical effects. The main goal of the project is to estimate as precise as possible long-term variability of extreme precipitation over Central Europe of the past century from high quality observations and to identify the major mechanisms of the observed changes in order to assess regional predictability of extreme precipitation over Central Europe. In order to achieve this goal the following objectives are to be met:
- Development of an ensemble of quality-checked gridded high spatial resolution daily data sets for Germany for the period 1900-2007, based on the DWD archives of rain gauge observations;
- Analysis of long-term (interannual to centennial) variability in the major precipitation statistics including extreme precipitation over Germany for the period 1900-2007;
- Analysis of the characteristics of cyclone activity, cyclone life cycle and atmospheric moisture transport on the basis of regional and global reanalyses including the R100 NOAA reanalysis, as well as upper air sounding over Germany;
- Identification of the major mechanisms driving precipitation extremes over Central Europe including their relative importance, and establishing predictability limits of extreme precipitation for different seasons over Germany.
We will comprehensively analyze extreme precipitation over Germany using the high resolution daily precipitation data set of DWD in order to understand and discriminate the role of large scale and local processes in forming climate variability of precipitation extremes over Central Europe in different seasons. To this goal we will design high resolution gridded precipitation data sets on the basis of DWD daily rain gauge records which will cover the periods 1900-2007, 1930-2007, 1950-2007 and 1960-2007. For each of these periods we will generate homogenized time series in which the impact of time dependent biases (e.g. caused by increasing station density) will be minimized.
The strength of our approach is that we use a very dense precipitation network. This will allow us to produce regional high resolution long-term daily data sets, which will be on a regional scale a step forward compared to the monthly CRU data (New et al. 1999). Furthermore our gridded data set will achieve a necessary improvement compared to the operational DWD product REGNIE (Rudolph 2006). REGNIE is based on approximately 1-kilometer analysis (distance-depending approach with accounting for the altitude dependence) of deviations of the “current daily precipitation” derived from ~600 operational gauges from the reference climatology for the period 1961-1990. In contrast to REGNIE which did not include homogenization of sampling density, we will use many more data points, will perform data homogenization, will account for significantly different number of stations for different historical periods, and will apply novel procedures (see below) for an objective analysis which will account for the local spatial structure of precipitation. When developing gridded data sets we will use a range of statistical methodologies, some of which were developed in our previous work (Zolina et al. 2004, 2005, 2008) e.g. the estimation of uncertainties due to gaps and methods of homogenization for daily precipitation records. Although methodologies to generate gridded data are well developed for spatial meteorological and hydrological data (Cressie 1996, Klein Tank et al. 2002, Mitchell and Jones 2005, Efthymiadis 2006), we will enrich them by introducing semivariograms and the adaptation of local procedures (Akima 1970). These new elements will allow both for improvement in the accuracy and for providing estimate of uncertainties.
In the next step we will design an ensemble of long-term precipitation statistics for each data set. These statistics will include characteristics of mean precipitation (precipitation totals, number of wet days and precipitation intensity) as well as parameters of extreme precipitation, such as high percentiles of probability distributions, distribution steering parameters and contributions to precipitation totals by given numbers of wet days. We will in particular analyze the duration of extreme precipitation events by considering two-dimensional statistical distributions of intensity and duration of precipitation events. In order to derive long-term time series of precipitation characteristics we will estimate precipitation statistics based on known initial value probability distributions, such Gamma PDF, Weibull PDF and others (e.g. Wilks 1995, Katz 1999, Groismann et al. 1999, Zolina et al. 2004) as well as extreme value distributions (e.g. Panorska et al. 2007). To quantify the role of the most wet days in forming precipitation extremes we will apply our new methodology (Zolina et al. 2008b) to better estimate the contribution by particular wet days. Furthermore, we will derive new statistical methods, e.g. to quantify the distributions of precipitation of different durations.
Based on the gridded data sets, we will analyze the climate-scale variability in precipitation statistics over Germany. Linear trends will be quantified in mean precipitation as well as in extreme and heavy precipitation. For this task we will estimate the trend significance based on both parametric and non-parametric tests (Walker 1975, Hayashi 1982, Keppel and Wickens 2004, Wilks 2006). Our trend analysis will specifically target the seasonality of trend estimates shown by Zolina et al. (2008a) assuming that during the last 120 years periods both with and without seasonality existed. In order to separate these periods we will use field significance estimates (Livezey and Chen 1983, Perneger 1998, Wilks 2006) as well as so-called piecewise trend statistics implemented by Weisse et al. (2005) for the analysis of the North Sea storminess. Shorter period interannual variability in precipitation characteristics will be quantified on the basis of SVD and EOF decompositions of mean and extreme precipitation space-time series. This analysis will allow quantifying the leading modes of precipitation variability, which might represent the responses of regional precipitation fields to large-scale and local atmospheric processes. The analysis of the EOF patterns and the corresponding principal components will quantify the dominant variability on time scales from several years to decades.
Now the quantitative estimates of variability in extreme precipitation for the last 120 years can be associated with cyclone activity and regional atmospheric processes. For this purpose we will use the existing cyclone statistics including parameters of the cyclone life cycle from reanalyses by Zolina and Gulev (2002, 2003). These data sets will be updated and extended to 2007. Regional European reanalysis based on the REMO model (Jacob et al. 2005) will provide the basis for regional storm tracks over the Atlantic-European region for the period from 1960 onwards. Higher resolution of this data set makes it better suited than the NCEP or ECMWF reanlayses and will allow for a more accurate association of regional precipitation statistics with characteristics of cyclones over Europe. For the long-term period from 1880 to 2007 we will use the NOAA R-100 reanalysis (Compo et al. 2006) which will be of a lower accuracy and resolution, but it will be used only for the long-term precipitation statistics. Thus, we will have the data sets of precipitation statistics and cyclone activity consistent with each other covering the periods 1880-2007 (coarse resolution and lower accuracy), 1930-2007 (intermediate resolution) and 1950(60)-2007 (high resolution and accuracy).
Being armed with these data sets we will analyse the mechanisms responsible for forming extreme precipitation variability over Central Europe. We will associate annual and seasonal parameters of cyclone life cycles with extreme precipitation time series on different time scales. Besides the basic cyclone life cycle characteristics (intensity, deepening rate, propagation velocity, life time) we will use also parameters of the atmospheric moisture transport by cyclones and characteristics of the local atmospheric moisture recycling (Brubaker et al. 1993, Elathir and Bras 1996, Trenberth 1999, Trenberth et al. 2003). On one hand, indeed, extreme precipitation can not come from the local atmospheric column or from the local evaporation (except very light precipitation) but has to originate from the transport by storm-scale circulation (Trenberth et al. 2003). On the other hand, synoptic-scale circulation impacts on the local precipitation extremes are strongly conditioned by the regional mechanisms of precipitation, such as orography, smaller-scale regional processes of water vapor advection and regional transformation of the large-scale storm track. On the longer time scale of interest is also the impact of changes in aerosols (and their effects on cloudiness by dimming or brightening), mentioned above. This effect is probably more pronounced over oceans (Lambert et al. 2008) and may result in changes of the water balance of the Atlantic cyclone transforming their ability to transport moisture to Europe. On the other hand, this effect can be also translated to the land areas where changes in aerosols can be also strong. All these require a detailed analysis of the regional thermodynamic processes on different space-time scales. For this purpose we will additionally (to the cyclone characteristics) use long-term series of the tropospheric temperature and moisture available from the radiosonde profiles at selected locations over Germany. Thus, we will be able to quantitatively assess the mechanisms forming extreme precipitation in different seasons and hopefully to provide estimates of predictability limits for extreme precipitation over Germany.
Expected project results will consist of scientific outputs (new methods and research results) and derived data products. Scientific results will include the following major aspects:
- Methodologies for the development of gridded precipitation data sets, including methods of data homogenization and estimation of errors and time-dependent uncertainties in gridded precipitation time series based on rain gauge data;
- New methods for the estimation of most accurate extreme precipitation statistics from daily rain gauge data;
- Estimates of long-term variability of extreme precipitation over Germany for different seasons for the period from 1900 to 2007, including estimates of linear trends together with their statistical significance for different periods, piecewise trend statistics and major variability patterns quantified through EOF and SVD analysis;
- Long-term variations in cyclone activity and cyclone life cycle over Europe, including atmospheric moisture transport and moisture recycling for the period 1900-2007;
- Estimates of the effect of changing dimming and brightening due to aerosols onto extreme precipitation changes over Europe;
- Identification and quantification of the mechanisms generating extreme precipitation over Central Europe for different seasons.
These results will be presented as papers published in peer reviewed journals. The following derived data products will have a technological value and will be made available to climate, environment and economy researchers as well as to environmental managers:
- Long-term homogenized gridded space-time series of daily precipitation for Germany, covering the periods 1900-2007, 1930-2007 and 1960-2007 with different spatial resolution and accuracy;
- Time series of the extreme precipitation statistics for Germany for different seasons from 1900 to 2007, including percentiles of heavy and extreme precipitation, fractional contribution due to very wet days and other relevant statistics;
- Long-term archive of the regional characteristics of cyclone activity for the period 1900-2007, including storm track density, parameters of cyclone life cycle as well as characteristics of atmospheric moisture transport and local moisture recycling;
- Ensemble of gridded high spatial resolution daily data sets for Germany for the period 1900-2007, based on the DWD archive of rain gauge observations.
The project results and data products will benefit many scientific and economic areas. They will contribute to the improvement of seasonal and longer scale prediction of extreme weather and climate events, to effective risk management, and to better water management in Germany. The consequences of unexpected precipitation-associated flooding on German rivers during the last years identified a general unpreparedness for such events in many sectors of German society and economy. Our results will put risk assessment on a better quantitative basis and add to the credibility of forecasts for the different time scales. A revision of the currently assumed risk levels for flooding, which have not yet taken into account, as well as effects of climate variability and change on the statistics of extreme events, can be made. E.g. for the Rhine river catchment the declaration of land for building is revised due to recent events, unfortunately without a solid quantitative basis, which could be obtained with the results from the project. Furthermore, the project results will contribute to the planning of soil protection from erosion due to heavy rains. According to the World Bank, up to 43% of costs of soil protection from erosion by flooding and from degradation due to droughts are spent on emergency actions, undertaken to mitigate adverse effects of poorly predicted weather and climate events. We do hope that the project results will significantly abate these costs.go up...
In order to meet the stated objectives and to carry out the research programme outlined above, the project will be organized in several work packages (WPs).
WP1. Homogenization and critical analysis of gaps in the daily records form the DWD archive.
This activity will target data quality control, quantitative analysis of missing data and homogenization of daily rain gauge observations form the DWD daily station archive. The rain gauge network of DWD is one of the densest and most properly maintained regional precipitation networks (Zolina et al. 2008). It currently consists of 5454 stations (Fig. 2) which have been digitized, controlled and included into the digital data base (MIRAKEL-Datenbank).
Figure2. Operational rain gauge network of DWD. Red points show the rain gauges covering the period 1950–2004 and blue points show the rain gauges which do not cover 1950–2004 (adopted from Zolina et al. 2008).
The station-to-station distance varies from 1-7 km in the Southern and Central parts to 3-20 km in the Northern part. All stations are equipped with the Hellmann rain gauge which is shielded with a cover to protect from evaporation and equipped with built-in heating devices controlled by thermostats. Their collecting surface is 1 m over the ground. The accuracy of a daily precipitation measurement is 0.1 mm. Since the late 1990s the manual reading procedure continuously changed to automatic by installation of PLUVIO devices. This transformation did not lead to a time change reading attribution. For this WP we are planning to carry out two major sub-packages:
WP1.1. Data record control, correction and homogenization.
In the first step we will analyse the homogeneity of the records and apply if necessary corrections for windy conditions. Corrections which have been applied to the records include adjustment of the reading time (if possible) and account also for instrument exposition indicated by quality flags which for the majority of stations read “0”, implying “repeatedly checked, no errors remained”. However, the wind correction was not applied yet. In order to properly apply the wind correction (Sevruk 2000, Nešpor and Sevruk 1999, Michelson 2004, Sieck et al. 2007) we will select a subset of the so-called synoptic stations (about 300 locations over Germany) co-located in space and time with the rain gauge data. Then we will apply the correction according to the guidelines of Nešpor and Sevruk (1999) and Michelson (2004) based on statistical analysis and numerical modelling. This will return at least 300 stations with fully corrected records for the period after 1948. Estimates of the corrections will be analysed with the multivariate methodology of Cressie (1996) in order to obtain estimates of the potential magnitude of the correction for the other stations which were not equipped with wind speed measurements. Under this activity we will also critically analyse all cases with solid precipitation (primarily snow) which will be identified from weather codes and temperature records and will apply the correction of Goodison (1981) and Goodison and Metcalfe (1989, 1994). Finally, we will analyse the timing of readings. For most stations precipitation measure occurs daily at 07:30 local time, corresponding to CET. However, for some stations over former DDR measurements were taken at different times (e. g. at 12:00 with attribution to the previous date). We will analyse the metadata for these stations in order to attribute these measurements to the correct dates, wherever possible.
WP1.2. Analysis of data coverage and homogenization with respect to the sampling density.
In the next step we will analyse the temporal coverage and the number of gaps in the station records and will provide the homogenization of records with respect to the missing data. Zolina et al. (2008) analysed approximately 5000 stations and showed that about 2000 of DWD stations cover the period from 1950 onwards and about 3000 gauges operated after 1960. A considerable amount of rain gauges also exists for the period 1930-1950. During 1900-1930 the DWD collection contains only 104 stations. Given these inhomogeneities in temporal coverage, we will produce the gridded climatologies for four streams, shown in Table 1. Thus, we will develop homogeneous gridded daily precipitation data sets for different periods with different spatial resolution. For longer periods (centennial and longer) we will design a coarser resolution climatology, based on a limited number of stations taken for gridding and continuously providing the observations over the whole 100-yr or longer period. For the last few decades (1960-2007) we will develop a data set of very high spatial resolution. Thus, we will avoid the time-dependent impact of sampling, when analysing long-term secular climate variability.
Another important issue is the homogenization of records with respect to the amount of missing data. Zolina et al. (2005) showed that up to 10% of gaps in daily records do not affect estimates of long-term variability and that for the range 10-30% of gaps linear trend estimates remain quite robust, although the characteristics of interannual variability can be biased. In Zolina et al. (2005) we suggested a methodology for the homogenization of daily precipitation records with respect to the missing values. This methodology is based on the simulation of gaps in the data according to the existing sampling structure. In this procedure, first we will estimate the frequency distributions of gaps of different continuity. Then the gaps in the time series will be simulated according to a prescribed gap structure. Figure 3 shows the number of gaps in the limited subset of data for 1950 to 2004 (Zolina et al. 2008).
Figure 3. Spatial (left) and temporal (right) distribution of the percentage of gaps in the limited subset of data for 1950 to 2004 (adopted from Zolina et al. 2008).
For most stations the number of gaps does not exceed 5-10% for the last five decades. For earlier decades gaps may increase up to 20-30% especially for the stations located in former DDR. In our project we will perform the homogenization of the time series using different thresholds for different streams (Table 1). After application of all homogenization procedures we will obtain four different data sets of daily rain gauge observations. These four data sets will have different quality and different spatial resolution. However, each data set will be homogeneous in this respect. In other words, data covering the period 1900-2007 will be equally erroneous in 1970s and 1900s. Compared to this stream, data included in Stream 4 (1960-2007) will be of better quality. Thus, we will obtain homogenized time series which will be further used for the development of gridded daily precipitation products for Germany.
WP2. Development of gridded high resolution daily climatologies on the basis of DWD station data.
This activity will be focused on the development of several high resolution long-term gridded climatologies of daily precipitation over Germany for different historical periods. In order to provide effective averaging of individual rain gauge observations for selected grid boxes we will apply statistical procedures, developed for gridding of meteorological variables. The theory of gridding, including kridging is described in detail by Cressie (1996). For precipitation gridding procedures were developed e.g. by Osborn and Hulme (1997), Huffman et al. (2001), Gyalistras (2003), Auer et al. (2005), Efthymiadis et al. (2006) and others reviewed by Efthymiadis et al. (2006). Most of these works target high resolution climatologies for monthly data sets characterized by much weaker spatial mesoscale variability compared to the daily data. During the last years DWD (Rudolph et al. 2007) provided daily high resolution operational product REGNIE which is based on operationally reported at GTS daily data whose number is changing over time. Since no temporal and spatial homogenization was applied this product is used primarily for case studies. A limited subset of 1700 stations from DWD collection has been used by Grieser et al. (2007) to produce gridded climate means of extreme precipitation return values on the basis of Weibull PDF. Few further attempts for daily gridded products were primarily dealing with satellite data (Huffman et al. 2001), blended products (Kottek and Rubel 2007) or with very limited time periods of gauge observations (Rubel and Hantel 2001). We will develop for the first time long-term high spatial resolution daily data set. This activity will include two sub-packages:
WP2.1. Analysis of mesoscale variability of daily precipitation for different time periods and estimation of the parameters for the gridding procedure.
For this purpose we will use the semivariogram approach. In this method the difference between simultaneous measurements at different points is considered as a function of point-to-point distance. When it is extrapolated to zero distance (where spatial variability does not contribute to the total variance) it should represent the error variance o2, which has to be divided by two to get the squared measurement error m2 = o2/2. Polynomial or linear fits are used to extrapolate the error to zero distance.
Figure 4. An example of the application of the semivariogram method for the region centered at 8.5ºE, 51ºN (shown on a small map). Polynomial approximation is shown in green and its extrapolation onto zero-distance is in pink. Error bars show uncertainties of estimation of the squared variance for different 2-km classes of distance.
We will extend this approach assuming that the obtained estimates are responsible not only for the error variance but also for the variance associated with the processes not accounted for by the network (i.e. small scale local processes, associated with convective precipitation or impact of regional orography). An example of the application of this procedure for the region centered at 8.5ºE, 51ºN is presented in Figure 4. The squared double variance decreases by about 5 times in the range of intra-station distances from 50 to zero kilometers, resulting in the estimate of variance of ~1.7 mm/day. We will estimate the squared variances for every station for different time periods (and, hence, different station density). This will allow us to establish a time varying variance parameter which will be further used for gridding of daily precipitation along with the de-correlation function characteristics, which will be quantified in the same way as semivariograms.
WP2.2. Development of gridded daily precipitation data sets for different periods.
We will apply procedures of spatial averaging in order to derive gridded daily precipitation space-time data sets based on the modified method of local procedures (Akima 1970). In this method the spatial interpolation parameters vary in space and time according to the estimates of variances and de-correlation functions derived in WP2.1. In other words, the influence radii will vary and thus increase or decrease the number of stations influencing the averaging for particular grid cells and times depending on the spatial variance and de-correlation function. As a result of this step we will produce daily precipitation fields of different spatial resolutions for different periods (see Table 1). Before these data sets are used for the further analysis of precipitation extremes and their variability we will extensively compare the derived products with the REGNIE data of Rudolph (2007), as well as with climatologies for limited periods by Grieser (2007) and Kottek and Rubel (2007). Importantly, the project will be carried out in a close co-operation with DWD and GPCC which we will provide cross-pollination of ideas on application of gridding methodologies.
WP3. Derivation of the ensemble of extreme precipitation statistics.
With our gridded daily precipitation data sets we will estimate long-term series of precipitation statistics for different seasons. These statistics will include parameters characterizing mean precipitation and its intensity (monthly seasonal and annual precipitation totals, number of wet days and precipitation intensity). For the characterization of extreme precipitation we will analyse the probability distributions such as Gamma PDF, Fisher-Tippett PDF, Weibull PDF and Generalized Pareto Distribution (GPD). We will use the Kolmogorov-Smirnov (k-s) test and other statistics (e.g. Pierson test) to quantify the goodness of fit and select the most appropriate distribution to characterize the statistical structure of precipitation. From the fitted PDFs the distribution parameters (e.g. scale and shape parameters of the Gamma distribution), the upper percentiles (90%, 95%, 99%) of the distribution as well the return values corresponding to 10, 20, 50 and 100-year periods will be derived. Finally we will derive extreme precipitation indices (Klein Tank and Können 2003) quantifying the precipitation fraction due to the most wet days. These indices, computed from the raw data, will be further compared to the metrics, based on a new distribution for the ratio between the daily precipitation and the precipitation total developed by Zolina et al. (2008b). This distribution allows us to explicitly estimate fractional contribution of the wettest days to precipitation totals. Furthermore, it can be applied to seasonal and monthly time series when the computation of traditional indices based on the raw data becomes uncertain. Quantitative comparison of empirically derived and PDF-based indices will be one of the important deliverables for the WP. Two-dimensional distributions characterizing intensity of precipitation events of different durations will provide integrated estimates of the precipitation which occurred continuously during time periods from 1 to 10-15 days. Considering impacts of extreme precipitation, even moderate precipitation continuously happening during several adjacent days may result in more disastrous flooding compared to a single day extreme. As a result we obtain an ensemble of statistical characteristics of daily precipitation for the period 1900-2007 derived from different streams of daily gridded data of different resolution and accuracy. All derived statistics will be accompanied by the estimates of the uncertainties inherent in the computation. Besides the space-time series of extreme precipitation statistics, we will provide the maps of uncertainties which will regionally quantify the confidence level of estimates. Since we will be testing several different distributions, we will for the first time classify the skills of these distributions in quantifying precipitation extremes for different regions and potentially for different climate periods. Finally, we will characterize changes in the structure of the precipitation probability distributions over the years. These statistics will help to further quantify the trends and shorter period variability in the intensity of precipitation of different intensity classes and, thus, analyse the evolution of daily precipitation PDFs.
WP4. Analysis of climate variability of extreme precipitation.
Linear trends in extreme and heavy precipitation will be estimated by least squares techniques, and the trend significance will be tested using e.g. the student t-test along with the Hayashi (1982) reliability ratio which considers the confidence intervals of the statistical significance of trends. These statistics are similar to the signal-to-noise statistics (Buishand et al. 1988) used by Klein Tank and Können (2003) for analyzing the significance of their estimates. The latter confidence intervals can be rather wide even if the Student’s t-test is formally satisfied for a given percentage point implying that the reliability of the trend estimate will then be questionable. We will also use the non-parametric Mann-Kendall test (Mann 1945, Kendall 1975), which was widely adopted for climate and hydrological time series (see, e.g. Schertz et al. 1991, Maugeri and Nanni 1998, Turkes 1999, Wang and Swail 2001, Brunetti et al. 2004, De Jongh et al. 2006 and others). The trend estimates will be considered significant at a specified significance level (e.g. 95%, 99%) if they satisfy simultaneously the Student t-test, the Mann-Kendall test, and the Hayashi reliability ratio, as adopted in Zolina et al. (2005, 2008). In addition to these traditional significance tests we will also estimate the spectra of interannual time series for different seasons at very low frequencies. These estimates will be compared with those derived after the linear trends are removed from the data. Statistically significant differences between the estimates of spectral power can be considered as an indication of the presence of trends in the time series. By analysing trends for different periods, we will assess the trend sensitivity to the impact of specific parts of the records, i.e. the beginning and the end parts of the record. This problem will be addressed in two ways. First, we will compute trends over shorter time series representing all possible combinations of starting and ending years within the first and the last periods of the time series. In order to estimate the field significance of the trend patterns we will analyse the field significance according to Livezey and Chen (1983) from the binominal distribution. We will also extend the field significance test to a more powerful test of global significance by accounting for the actual values of the single tests and considering their minimum (Katz 2002, Wilks 2006). Additionally we will use the so-called piecewise linear trend estimates (Weisse et al. 2005) for the analysis of the persistence of trends over long periods. This procedure estimates trends for time periods starting and ending at a given year (T), which then becomes an additional parameter of the trend model. Finally we will detect the leading modes of interannual to decadal scale variability using empirical orthogonal function (EOF) and singular value decomposition (SVD) analysis. These procedures will identify the dominant patterns of variability. We hypothesize that in winter and summer extreme precipitation has opposite tendencies in the northern regions (influenced by the storms propagating over Northern Europe) and in the southern part (where the role of regional processes and orography can be quite pronounced). The EOF analysis will identify spatial patterns, which reflect this variability and quantify its temporal scales. The EOFs will be computed using procedures of matrix algebra (e.g. von Storch and Zwiers 1999) and accompanied by estimates of accuracy which indicate how good the leading modes are separated form each other (North 1982).
WP5. Development of the archive of cyclone trajectories and characteristics of the cyclone life cycle and analysis of climate variability in cyclone activity.
In order to analyze the impact of atmospheric circulation onto precipitation characteristics we will characterize the cyclone activity in the Atlantic-European sector. We will use 6-hourly and 3-hourly output from global reanalyses such as NCEP, ERA-40 and NOAA R-100 (Compo 2006) as well as from regional high resolution simulations with REMO and COSMO-CLM. For some of these data sets cyclone tracking is already completed. The cyclone tracking will be performed using the algorithm developed by Zolina and Gulev (2002, 2003). The tracking will be performed on the polar orthographic projection. It includes a dynamical interpolation of sea level pressure to finer time steps to allow for the separation of spatial scales associated with cyclone migrations during one time step and the distances between the cyclone centers. The tracking procedure uses the method of “closest neighbors” (employed also in the schemes of Murray and Simmonds 1991, König et al. 1993, Blender et al. 1997, Sinclair 1997, Hodges 1994 and others) only as a first guess for cyclone identification. It follows a 3-pass analysis of cyclone propagation velocities, a sorting of crossing trajectories, and a separate analysis of the stationary cyclones. Cyclone activity will be characterized by cyclone numbers and frequencies and characteristics of the cyclone life cycle, such as cyclone intensity, deepening rates, propagation velocities and life time (see Zolina and Gulev 2002). These parameters will allow us to classify cyclones and to estimate atmospheric moisture transport by different types of cyclones. Several data sets of cyclone activity will be developed covering the periods 1900-2007, 1950-2007 and 1960-2007. These data sets will be derived from atmospheric data of different accuracy and resolution (NOAA R-100 reanalysis, NCEP and ERA-40 reanlayses, REMO, COSMO-CLM) and, thus, will be consistent with the different precipitation statistics (WP2, WP3). Long-term time series of the characteristics of cyclone activity will be further analysed in order to quantify linear trends and interannual variability patterns in the cyclone characteristics in the same manner as for precipitation statistics. This will form the basis for the association of cyclone activity with extreme precipitation statistics at different space-time scales.
WP6. Association of variability in precipitation extremes with circulation changes and local processes for different seasons.
In order to associate the characteristics of atmospheric synoptic activity with extreme precipitation over Germany we will perform a joint statistical analysis which couples the leading modes of variability in extreme precipitation with the dominant patterns of cyclone activity and associated atmospheric moisture transport. Canonical correlation analysis (von Storch and Zwiers 1999) and more advanced methods will be used to quantify the relationships, e.g. the censored quantile regression approach recently developed by Friederichs and Hense (2007). Besides cyclone activity we will also use radiosonde profiles at 12 locations over Germany and up to 30 over Central Europe. These data along with the tropospheric variables from reanlayses will make it possible to quantify the regional importance of the large-scale and local mechanisms of extreme precipitation by considering characteristics of the local atmospheric moisture recycling on the basis of regional estimates of convergence of the atmospheric water vapor (Brubaker et al. 1993, Elathir and Bras 1996, Trenberth 1999, Trenberth et al. 2003). This analysis will allow us to quantify the extent to which extreme precipitation is constrained by the local energy budget of the troposphere and by the regional atmospheric moisture content (Allen and Ingram 2002). This analysis will be performed through the consideration of cyclone and/or regional circulation composites for different cases of extreme precipitation and different seasons. Furthermore we will consider the impact of global dimming and brightening (Lambert et al. 2008) onto the changing character of extreme precipitation. This consideration will account for both the direct effect of continental aerosols onto the local precipitation and the indirect (remote) effects of marine aerosols onto the moisture budget of Atlantic cyclones. Altogether, these will help us to quantify the mechanisms driving the so-called disproportional (with respect to the changes in precipitation totals) changes in extreme precipitation over Central Europe. Results of this analysis will help to associate extreme precipitation with large scale circulation variability (cyclone tracks and moisture advection by synoptic transients) and regional processes (local convection, impact of the orography) and to answer the question which characteristics of precipitation, including precipitation extremes are predictable and which are not for different seasons over Germany.
some activities which are independent on each other (e.g. WP1/2 and WP5) can be done in parallel. Similarly, the other activities which are interdependent (like development of gridded daily data sets and computation of precipitation statistics) can be also carried out simultaneously, since these require many test computations which should imply the adjustments in the methodologies used. We think that the presented time line of the project is quite balanced and should guarantee successful execution of the project.go up ...