Two different objective GWC classification methods were compared to describe the situation in Schleswig-Holstein (SH) during December 2014: (1) the modified Lamb weather types (LWTs; Jenkinson and Collison, 1977) used at the Federal Maritime and Hydrographic Agency (BSH; Löwe, 2005) with a model centre over the central North Sea, and (2) the objective classification (OWTC; Dittmann et al., 1995; Bissolli and Dittmann, 2001) of the German Meteorological Service (DWD) with a model centre over central Germany. Further differences are the input parameters: while LWT is based solely on sea level pressure data, here the NCEP/NCAR Reanalysis 1 (Kalnay et al., 1996), at 16 grid points over northern Europe, OWTC input data include air pressure, temperature, wind, and water vapour content on different height levels derived from the current operational GME (Global Model Extended) of DWD (http://www.dwd.de/EN/ourservices/wetterlagenklassifikation/wetterlagenklassifikation). Further, OWTC output parameters include cyclonality on two height levels (950 and 500 hPa) and a humidity index (“wet” and “dry”) that describes the precipitable water content of the atmosphere compared to the long-term daily mean. LWT output, however, includes a gale index in four categories (from “no gale” to “very severe gale”) derived from the strength of the geostrophic flow and the vorticity.

Schröter et al. (2015) have investigated 76 nationwide flood events that affected at least 10 % of the German river catchments over a period from 1960 to 2009 and ranked them concerning their severity. Further included were the floods from 1954 and 2013 (Blöschl et al., 2013). The investigations are based on the data set from Uhlemann et al. (2010) using time series of daily mean discharge records at 162 gauge stations of the German Water and Shipment Administration (WSV) and the German Federal Institute of Hydrology (BfG). Additionally, Schröter et al. (2015) used daily precipitation sums from the REGNIE data set (see, e.g., Rauthe et al., 2013) provided by DWD with a spatial resolution of 1 × 1 km to describe the meteorological situation of these events. The basic idea in using this approach is the assumption that a combination of extreme initial wetness (i.e. oversaturation of the soil) and a strong but not extraordinary event precipitation leads to flooding. These factors were evaluated by means of two indices: (1) the maximum R3d as a trigger of the flood, calculated at each grid point separately within a window of ±10 days around the onset of the flood event, and (2) the initial API, calculated from the sum of daily precipitation at each grid point Ri(xy) and weighted with respect to the time span (m=30 days) of rainfall occurrence prior to the R3d to ensure a clear separation of both indices; see Eq. (1): APIx,y=∑i=130kiRi(xy,(m-i)). Here, i marks the day prior to the R3d and k=0.9 a depletion constant that approximates the decrease in soil moisture due to evapotranspiration and percolation to deeper soil layers. Using this approach, the rainfall at day 1 prior the R3d is weighted highest.

Both indices were calculated using REGNIE's daily precipitation sums for the December 2014 flood. The constant k was not changed; however, future investigations could include regional soil types at a high resolution (if accessible). Further, it should be noted that the coastal regions were excluded by Schröter et al. (2015) since floods might be affected by the water level conditions in the North and Baltic seas, i.e. the possibilities for drainage (K. Schröter, personal communication, 2016). Since sea gauge data did not show any extremes and drainage was possible at all times during the December 2014 flood (see Sect. 3.4), a comparative analysis is justified and might help to point to regional risk potentials, even for spatially limited flood events.

Furthermore, the Matlab toolbox WAFO (WAFO-group, 2000) was used for the statistical evaluation: according to Schröter et al. (2015), the yearly maximum 3-day precipitation sums and the respective 30-day antecedent precipitation were calculated at each REGNIE grid point. Then, 5 to 100-year return periods (5–100yRPs) were derived at each grid point using the Gumbel distribution over the base period 1960–2009.

Mean trends above the 95 % significance level for the five highest R3d (R3dfivemax) and API (APIfivemax) values per year were calculated over 30-year running intervals from 1960–1989 to 1985–2014 for the Kiel Canal catchment (EZG NOK), SH, and all of Germany (D). Instead of the yearly maximum alone, the five highest events per year were chosen so as to obtain more reliable and robust statistics. A modified version of the Mann–Kendall test (see Hamed and Rao, 1998) was used to determine significant trends, avoiding misleading results due to autocorrelation (in case autocorrelation is greater than 0). All trends were calculated at each grid point separately. Then, area means were derived.

Results from gauge data from the LKN–SH and LLUR–SH (2015) report will also be presented to investigate the applicability of the chosen precipitation indices. A map showing maximum water levels at all catchment gauges in Schleswig-Holstein during the December 2014 flood is kindly provided by Thomas Hirschhäuser (LLUR–SH). Furthermore, in-depth analyses of the hydrological situation including discharges and statistical evaluations can be found in the LKN–SH and LLUR–SH (2015) report. The most relevant results are presented in Sect. 3.6.

In general, both classification methods show predominant westerly GWCs from 5 December onwards with north-westerly (NW) situations during the heavy precipitation event from 21 to 23 December (Table 1): OWTC shows humid conditions; LWT shows “gale”. However, differences are apparent during the first precipitation event: the cores of the low-pressure systems are centred far north, categorized by LWT as severe gale (Alexandra) and gale (Billie) with south-westerly (SW) cyclonic flow (Fig. 2a, b). OWTC on the other hand categorized a NW anticyclonic flow and dry conditions. An explanation provides the relevant model centre; OWTC is focused over central Germany, while LWT is ideally centred in the North Sea. Since most of southern and central Germany was unaffected by this precipitation event, most of the model domain was indeed dry.

It becomes obvious that the use of the classification method is subject to several factors. One of these specifies that the method should be suitable to the region of interest and capture its unique features. In this case, LWT provides slightly better results due to its focus on the North Sea and the fact that OWTC misses wet days during the first precipitation event. Nevertheless, both GWC analyses clearly show that not only one specific weather type but the succession of similar (westerly) types was important to the overall high soil moisture conditions, i.e. in generating prolonged rainfall, especially in northern Schleswig-Holstein. Additionally, the extreme precipitation event in southern Schleswig-Holstein was caused by a succession of NW types from 19 to 23 December (5 days; LWT) and 17 to 23 December (7 days; OWTC). Considering the mean life time of the NW type of 1.82 days (base period 1971–2000; Löwe et al., 2013, their Table 2-10), the event was extraordinary for this region.

Above-average monthly precipitation amounts between 80 and 160 mm were recorded at the German coasts during December 2014; local monthly means were exceeded by more than double that value and old records were broken. In SH, values of 175 mm up to 225 mm were reached (Fig. 3a), which corresponds to about 225–300 % of the long-term means (Fig. 3b). All over the rest of Germany, December 2014 was unremarkable, with maximum mean values around or clearly below those of the reference period (1961–1990).

Looking at daily precipitation sums from the REGNIE data set, two main rainfall periods can be distinguished: one from 10 to 12 December, more pronounced in northern SH, and one from 18 to 24 December (Fig. 4). Maximum daily precipitation was detected from 22 to 23 December in southern SH and HH, with local values exceeding 50 mm corresponding to the standard monthly mean values.

As seen in Fig. 4a–c, the first rainfall period begins in far north-eastern (NE) SH on 10 December slowly progressing to the south. It further shows that not only SH was affected during this event: pronounced rainfall was detected north of the Eifel region on 12 December. Figure 4d–f display the main precipitation event from 21 to 23 December. Now mainly northern Germany is affected, especially southern SH on 22 December. Values are comparable to daily precipitation sums from selected DWD stations presented in Fig. 5 showing that REGNIE (solid black lines) fits the station data (bar plots) well. Only the highest maximum values are slightly underestimated: highest values are found at Wittenborn, north of Hamburg, with over 50 mm on 22 December (see Fig. 4e), about 30 mm at Schleswig in central SH, and over 20 mm at Leck in northern SH on 22 December (see Fig. 4f). The first rainfall period is captured as well, with maximum values on 11 and 12 December (see Fig. 4b, c).

Additional investigations using modelled soil moisture data from DWD's Agrometeorological Research Centre (ZAMF) for sandy loam soil and cultivation with sugar beets show highest values in the SH region, with up to 139 % nFK (effective field moisture capacity) in the north for 21 December 2014 (start date of the corresponding event precipitation). Values decrease southward but never below 100 % nFK except in the south of SH (Fig. 6a). The unit % nFK describes the saturation in percent of effective field capacity of the upper 60 cm of soil. If soil moisture exceeds 100 % nFK, the actual water content is higher than is usable for plants (DWD, 2016), i.e. as in most of northern and central SH at the onset of the main precipitation event. The south to north gradient is in accordance with precipitation data, showing a slow progression of rainfall events from north to south (see Sect. 3.2).

ZAMF also provides soil moisture data for loamy sand soil and cultivation with winter grain. Using this data, values in northern and central SH are between 105 and 110 % nFK for the same date (Fig. 6b). It should be noted that neither the actual soil differentiation nor the degree of sealing is part of the model chain, and locally, this might be of importance (see, e.g., Apel et al., 2016). Nevertheless, both soil types show the same oversaturated regions in SH with some minor differences in the Fehmarn area (eastern SH).

Figure 7 shows the 3-day precipitation sum R3d for the December 2014 flood in Schleswig-Holstein (Fig. 7a) and its corresponding ratio to the 5-year return period (5yRP, Fig. 7b). The scaling for R3d is set according to Schröter et al. (2015). Clearly, the main contiguous part of the event precipitation is restricted to northern Germany, with some spots in central and southern Germany. R3d shows rather moderate maximum values of 109 mm north of Hamburg compared to the flood in 2013 with maximum values up to 300 mm (see Schröter et al., 2015, their Fig. 5, left). These differences can be explained mainly with the origin of both events: the flood 2013 was triggered by a quasi-stationary trough over central Europe in May and June leading low-pressure systems with hot and humid air masses at their flanks from SE Europe northwards. Additional orographic effects caused by the mountain ridges in central Europe, large-scale uplifting downstream from the low-pressure systems, and embedded convective processes finally led to prolonged and extended rainfall (e.g. Belz et al., 2013, 2014; Stein and Malitz, 2013). The December 2014 flood was triggered by low-pressure systems with North Atlantic air masses exclusively and appeared in winter when relatively cold air cannot hold as much water.

Areas with R3d exceeding the 5yRP are centred north of Hamburg, the eastern NOK region, the catchments Stör and Krückau, and at the coasts of Mecklenburg-Vorpommern (Fig. 7b). Higher return periods can be found north of Hamburg in the area of Wittenborn (see Fig. 4). Here, values even exceed 100yRPs locally, but due to the fact that the base period only spans 50 years, return periods over 100 years are becoming increasingly uncertain (rule of thumb: 2 times the observational time span gives the maximum return period to be statistically sound). Therefore, the cut has been made at 100 years. Nevertheless, this shows how extraordinary this event was for this region.

Figure 8 shows the corresponding values for the API, again, scaled according to Schröter et al. (2015). Maximum API values of 41.5 mm (Fig. 8a) are well below those of the flood 2013 (see Schröter et al., 2015, their Fig. 7, left) and can be found in NW SH, which is in fair agreement with the soil moisture data (see Fig. 6a). In contrast to R3d, the 5yRPs for API are exceeded only in NW SH (Fig. 8b), with maximum values corresponding to 20yRP.

It is obvious that antecedent precipitation in combination with the maximum precipitation event led to SH-wide flooding in 2014 (Figs. 7, 8): areas that were struck by heavy rainfall did not need additional initial wetness to be flooded; areas with high antecedent precipitation were saturated already and needed only small amounts of additional event precipitation. In this regard, the importance of both indices to describe this flood accurately is illustrated. Furthermore, ongoing investigations in the NOK catchment area suggest that R3d and API are promising indicators or predictors to describe problematic situations in the canal's operational routine. However, since many other influencing factors like wind surge, locking of ships, dewatering, ferry traffic, etc. are involved, pinpointing the respective factors to one single event is difficult, and therefore, accurate estimations of the consistency for regional investigations cannot be given at this point yet.

The LKN–SH and LLUR–SH (2015) report points out that more than a third (66 out of 184) of inland gauges in SH exceeded the, thus far, highest high-water level (HHW) during the December 2014 flood (Fig. 9). Almost all of these gauges are located in areas affected by high R3d values in southern and north-eastern SH or high API values in north-western SH (see Figs. 7, 8).

Furthermore, 14 gauges exceeded high waters with low probability, i.e. 200 yr return periods (HW200; according to HWRM-RL, 2007). Two more gauges exceeded their 100 yr return periods (HW100) and eight gauges their 10 yr return periods (HW10; LKN–SH and LLUR–SH, 2015, their Fig. 46). Regarding discharges, three gauges exceeded their respective 200 yr return periods, four more their 100 yr return periods, and five more their 50 yr return periods (LKN–SH and LLUR–SH, 2015, their Fig. 80). All of these gauges are located in areas affected by R3d or API.

More than 80 % exceeded the mean high-water level (MHW), while gauges not reaching the MHW were mainly sea gauges located in the North Sea. Return periods of half a year were hardly exceeded here (Jens Möller, BSH, personal communication, 2015). Therefore, the December 2014 flood could have been much worse if an additional storm surge had hindered the drainage of the SH catchments into the North Sea.

Figure 10 shows significant mean 30-year running trends for the North and Baltic seas Canal catchment area (EZG NOK), SH, and D for the five highest R3d and API events per year. Obviously, trends are not only highly dependent on the respective base period, showing considerable interannual variation, but on the area under investigation as well: while trends for R3d are positive in SH (with one exception) and the EZG NOK during the whole period, they become negative during recent years looking at Germany as a whole. Keeping in mind that cyclone pathways and connected extreme precipitation events are shifting north-eastwards (e.g. Stendel et al., 2016), SH and the EZG NOK will probably experience more and heavier extreme situations in the future. Furthermore, a clear separation of NE and SW Germany regarding significant R3d trends is evident in recent years, exemplarily shown for the period 1983–2012 (Fig. 11a): trends are positive in NE Germany and negative in SW Germany with only some local spots (e.g. mountainous areas) showing opposing trends. This separation also is in accordance with the shifting cyclone pathways.

API trends (Fig. 11b) are negative for all areas in the beginning, change to high positive values during the 1980, and have settled at lower values since. Again, D shows the smallest values since the NE–SW separation is also evident but not as articulated as for R3d. Nevertheless, API can also be expected to increase more strongly in the north-east under climate change conditions, leading to wetter soil and an increased risk of flooding in these areas. Combined with the higher probability of extreme precipitation events, especially for northern Germany (SH, EZG NOK), the risk increases even further.