Volume 119, Issue 9 p. 5240-5256
Research Article
Free Access

How important is intensified evaporation for Mediterranean precipitation extremes?

Andreas Winschall

Andreas Winschall

Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Search for more papers by this author
Harald Sodemann

Harald Sodemann

Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Search for more papers by this author
Stephan Pfahl

Stephan Pfahl

Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Search for more papers by this author
Heini Wernli

Heini Wernli

Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Search for more papers by this author
First published: 21 April 2014
Citations: 49

Correspondence to: H. Sodemann,

[email protected]

Abstract

The moisture sources for heavy precipitation events in the northwestern Mediterranean are investigated with a Lagrangian moisture source diagnostic, which is based on the analysis of moisture changes during 6 h time intervals along backward trajectories, originating from the location of the heavy precipitation events. We consider the 50 strongest events in each season in the region 0–15°E, 42–46°N during the years 1989–2009. The results of the moisture source diagnostic indicate that during precipitation extremes the Mediterranean Sea surface is only one of several source regions in all seasons, and typically matched by moisture from the North Atlantic in autumn and winter and land evapotranspiration in summer. The event-to-event variability is large. Individual events can be dominated by land evapotranspiration, moisture transport from the North Atlantic, or an export of moisture from the tropics. The time of maximum moisture uptake varies between a few hours to more than a week before the precipitation event. The results from the moisture source analysis are used for investigating whether intense surface evaporation occurs prior to heavy precipitation. Surface evaporation anomalies at the moisture sources are positive over the North Atlantic and the European and African land surface, whereas no signal is present over the Mediterranean. Therefore, for the Mediterranean moisture contribution, convergence from the background moisture reservoir is essential, whereas for the remote sources anomalously intense surface evaporation is required to foster the moisture supply for Mediterranean heavy precipitation events.

Key Points

  • Identified moisture sources for 200 Mediterranean heavy precipitation events
  • Individual cases with highly variable moisture source patterns and timing
  • North Atlantic moisture associated with anomalously intense surface evaporation

1 Introduction

Heavy precipitation events (HPEs) are natural hazards that frequently affect the northwestern Mediterranean. Due to the topographic setup with several mountain barriers like the Alps, the Massif Central, and the Pyrenees, this region is favorable for intense precipitation events and flash floods, especially in the autumn season [e.g., Buzzi et al., 1998; Doswell et al., 1998; Frei and Schär, 1998; Ricard et al., 2012, and references therein]. According to the classical perception, a key factor for the seasonality is the high sea surface temperature of the Mediterranean in autumn, which results from intense insolation during summer. In cases of cold-air advection, high sea surface temperatures can induce intense ocean evaporation, which serves as a potential moisture supply for precipitation events. In addition, the advection of colder air masses over a warm ocean surface leads to unstable stratification of the atmosphere, which supports the formation of convective rain systems [e.g., Romero et al., 1998a; Homar et al., 1999; Davolio et al., 2006, 2007].

The Massive Central and the Spanish Mediterranean coast are areas that are frequently affected by HPEs [e.g., Romero et al., 1998b; Homar et al., 2002; Nuissier et al., 2008; Ducrocq et al., 2008; Nuissier et al., 2011]. Another “hot spot” of HPEs in the western Mediterranean is the southern part of the Alps [Frei and Schär, 1998]. Several studies investigated the mechanisms leading to southern Alpine HPEs on different spatial scales, focusing either on smaller-scale processes [e.g., Fehlmann and Quadri, 2000; Richard et al., 2003; Langhans et al., 2011] or the large-scale setup [e.g., Massacand et al., 1998; Martius et al., 2006; Schlemmer et al., 2010]. Some studies explicitly dealt with the moisture supply for southern Alpine precipitation events. Most of them emphasized the importance of moisture evaporated from the Mediterranean for the formation of precipitation events [Romero et al., 1997; Massacand et al., 1998], in particular in the context of Alpine lee cyclogenesis [Tibaldi et al., 1990], while others pointed out that also moisture from the North Atlantic can have an impact for southern Alpine precipitation events [Reale et al., 2001; Turato et al., 2004; Winschall et al., 2012; Pinto et al., 2013]. Using the Lagrangian method of James et al. [2004], Drumond et al. [2011] compared variations of moisture transport to several Mediterranean regions during wet and a dry seasons, concluding that North Atlantic moisture sources may be important for precipitation variability. Sodemann and Zubler [2010] studied the moisture supply for Alpine precipitation climatologically with a quantitative Lagrangian moisture source diagnostic. They found a contribution of about 40% from the North Atlantic, 20% from the Mediterranean, 15% from the North and Baltic Seas, and 20% from the European land surface to the annual mean precipitation in a domain covering the Southern Alps.

In this study the moisture sources for the most intense large-scale HPEs in the northwestern Mediterranean during the years 1989–2009 are investigated. We consider the 50 days in every season with the strongest daily accumulated area-averaged precipitation in the target region displayed in Figure 1. This region corresponds roughly to the main target region of the first Hydrological cycle in the Mediterranean Experiment special observation period (HyMeX SOP1) that took place in autumn 2012. We aim to locate and quantify the moisture sources for these 200 HPEs utilizing the Lagrangian moisture source diagnostic developed by Sodemann et al. [2008], which is based on analyzing the moisture change along backward trajectories from the region of the HPE (see 2 for more details). Since our methods and data are restricted to larger scales, we specifically focus on large-scale HPEs in our analysis, and thus, potentially very dangerous flash floods which are governed by small-scale processes are not covered by our results.

Details are in the caption following the image
The black frame marks the target region of this study, where HPEs are identified (0–15°E, 42–46°N). The red and blue shading denotes the regions in the North Atlantic and western Mediterranean used to calculate the histograms in Figures 5 and 6, respectively.

A central aim of this study is to investigate whether anomalously intense surface evaporation is a precursor for northwestern Mediterranean HPEs, or whether the convergence of moisture from a climatologically “normal” surface evaporation is the driving mechanism for such precipitation events. In other words, is anomalously intense evaporation essential to moisten the air and for subsequently producing a HPE, or is convergence from the background moisture reservoir the critical mechanism? Romero et al. [1997] found intense Mediterranean evaporation prior to a HPE in northeastern Spain in 1988. However, their results from numerical sensitivity experiments did not show a large influence of this evaporation on the precipitation event. The study of Winschall et al. [2012] showed a considerable contribution of moisture from a region of intense evaporation over the North Atlantic to a southern Alpine heavy precipitation event in November 2002. The same dynamical precursor, an upper level potential vorticity streamer, triggered both the intense surface evaporation over the North Atlantic along its western flank and the HPE along its eastern flank a few days later. From a climatological point of view, their study indicated that intensified evaporation over the North Atlantic might be an important precursor for Alpine HPEs. However, Winschall et al. [2012] did not deduce a climatological link of evaporation and precipitation by means of a moisture source diagnostic. Therefore, in this study, the moisture sources for northwestern Mediterranean HPEs and the link between surface evaporation and HPEs are studied systematically with a Lagrangian moisture source diagnostic.

After introducing the data and methods (2), we discuss the synoptic conditions for HPEs in the northwestern Mediterranean (3) and the moisture sources of these events (4) with composite maps. The case-to-case variability of the moisture uptakes is addressed in 5. In 6, the link of intense evaporation and HPEs is investigated and conclusions are presented in 7.

2 Data and Methods

This study is mainly based upon data from the ECMWF (European Centre for Medium-Range Weather Forecasts) ERA-Interim reanalysis [Dee et al., 2011]. Six-hourly accumulated surface precipitation and surface latent heat flux are calculated from short-term forecasts. The values at 0000 and 0600 UTC were taken from forecasts started at 1200 UTC of the previous day, whereas the values at 1200 and 1800 UTC are based on forecasts started at 0000 UTC of the same day.

It is not straightforward to devise a definition of HPEs which allows to identify them objectively from such a reanalysis data set. In particular, the decisive atmospheric factors for flash floods, i.e., HPEs at time scales of shorter than 24h, are hardly resolved. For such events, it may be of critical importance to use numerical weather prediction models at convection-permitting resolution, as well as to have a better resolution of the main topographical features [Romero et al., 1997; Davolio et al., 2006, 2007; Nuissier et al., 2008]. Finally, knowledge on the catchment properties are key to predict flash floods. All of these requirements are clearly beyond the possibilities of the ERA-Interim data set, and thus necessarily not in the focus of this study.

We instead employ a definition of HPEs which matches the capabilities of our data set and focuses on larger-scale events. Such regionally extended HPEs are a matter of considerable economic and societal importance in the Mediterranean as well. For instance, the Piemont flood of October 2000 (fourth most intense event in September, October, and November (SON)) had large societal impacts [Turato et al., 2004], as had the strongest event in the data set in September 2006 [Lambert and Argence, 2008]. The main reasons for our focus on HPEs that affect larger-scale areas are that (i) typically, the dynamics of these events are better represented by the ERA-Interim data than for small-scale events, and (ii) large-scale events are represented by more air parcel trajectories leading to more robust results from the moisture source diagnostic (see below). For this study, it is more important that the timing of heavy precipitation events matches with observations, rather than absolute precipitation amount. The reasonable timing of events is supported by the findings of Pfahl and Wernli [2012].

Thus, for each season (which are defined as March, April, and May (MAM, spring); June, July, and August (JJA, summer); September, October, and November (SON, autumn); and December, January, and February (DJF, winter)) during the years 1989–2009, the 50 most intense precipitation days are considered. These most intense precipitation days are determined by the highest value of daily accumulated precipitation, averaged over a prespecified target region in the northwestern Mediterranean (Table 1). The region extends from 0 to 15°E and 42 to 46°N (Figure 1). In the following, we refer to the 200 considered days as the “HPEs.” This number of cases is a compromise between having a representative sample of cases and keeping the required computations feasible. Several of the HPEs identified according to this definition occur on consecutive days. However, this does not affect the main conclusions of our analysis (see below).

Table 1. Summary of the 50 Strongest HPEs for Each Season in the Target Domain, Sorted by Daily Mean Precipitationa
MAM P (mm) JJA P (mm) SON P (mm) DJF P (mm)
20040416 17.7 20000610 16.3 20060914* 24.1 19901209 18.3
19930301 16.9 19970601 14.1 19930923 23.3 19990110 16.9
20020508 16.3 20020715 10.6 19930924 19.9 19961209 15.5
19990503 14.4 20020810 10.3 20001015 19.3 20081214 14.3
19950424 14.1 19970605 10.2 20021010 18.7 20030106 13.7
19960315 12.8 20020825 10.0 19941105 17.7 20081215 13.6
19950512 12.5 20020605 9.9 19991021 17.7 19971218 13.1
20040429 12.4 20020714 9.7 20001106 17.0 20091224 13.1
19930425 12.0 20000611 9.3 19991018 16.9 19890225 12.7
20020403 11.9 19910607* 8.8 20021009 16.8 20060128 12.7
20010302 11.8 20090601 8.2 19921005 16.5 19970103 12.7
20090426 11.7 19920607 8.2 20081128 16.2 19951230 12.4
20090329 11.6 19930827 8.1 19940923 16.2 20081210 11.9
20090427 11.5 19920809 8.0 19990920 16.0 20031201 11.9
19940518 11.4 19910731 7.9 20091022 15.7 20090202 11.8
19910325 11.3 20020608 7.6 20081104 15.7 19951231 11.4
20050517 11.3 19970811 7.6 20060925 15.4 20031203 11.3
19980528 11.2 19970628 7.5 19950919 15.0 19940204 11.2
19980430 11.1 20020606 7.4 20000930 14.8 19961208 11.2
20090304 11.0 19940611 7.3 19961117 14.8 20041226 11.1
19990326 11.0 19920604 7.2 20090916 14.8 20091222 11.1
19890412 10.8 20020604 7.0 19921004 14.7 19921205 11.1
19890403 10.5 19940612 6.9 19941104 14.3 20080103 10.9
19890425 10.3 20020826 6.9 19961015 14.2 19921208 10.8
20070504 10.3 19920610 6.9 20081028 14.1 19951216 10.5
19910324 10.2 20020828 6.8 19921003 14.0 19940106 10.4
20020411 10.1 19920601 6.8 20000929 13.8 19921207 10.4
19990504 9.9 19950623 6.7 19941020 13.5 19981203 10.3
19920502 9.9 19970629 6.3 20060915 13.3 19960123 10.3
19990517 9.7 19900830 6.2 19971106 13.3 20001227 10.3
19910509 9.6 20020811 6.0 20010923 13.3 20030121 10.2
19900406 9.6 20050821 5.9 19991113 13.2 19931225 10.2
20020412 9.5 19940626 5.8 20091108 12.5 20041208 10.1
19940416 9.5 19890828 5.7 19901125 12.5 20051202 9.9
19970421 9.5 19920705 5.7 19991020 12.3 20040221 9.9
19920331 9.4 19910730 5.6 20081031 12.0 19910114 9.8
19910405 9.3 20070821 5.6 19931002 11.9 19971219 9.6
19890404 9.2 19890703 5.6 19911012 11.9 20020206 9.5
20040503 9.2 20090621 5.6 20091021 11.9 20080104 9.5
19930402 9.1 20020809 5.5 20031031 11.8 20061208 9.4
20090305 9.1 19920623 5.5 20031018 11.8 19961219* 9.4
20050410 9.0 19960621 5.5 20021011 11.7 19991215 9.2
19980415 8.9 20020824 5.4 20001123 11.5 19960107 9.1
20090411 8.8 20070606 5.3 19941106 11.3 20081216 9.0
20090331 8.8 20000710 5.3 19901019 11.3 19970102* 9.0
19920326 8.7 20020609 5.2 19990915 11.3 19940205 8.9
20070501 8.7 19940625 5.2 19931105 11.2 19960112 8.9
20040504 8.7 19930622 5.1 20001014 11.2 19960202 8.9
20000409 8.6 20020827 5.1 20081102 11.1 20060129 8.9
19990304 8.6 19920611 5.0 20011111 11.1 19960110 8.9
  • a Dates marked with an asterisk (*) indicated cases described in more detail in 5. Dates printed in bold have more than one third contribution from the Mediterranean Sea. Dates printed in italics have less than 10% contribution from the Mediterranean Sea.

For all HPEs, kinematic backward trajectories are calculated from the target region (see Figure 1) starting at 0000, 0600, 1200, and 1800 UTC on each day, using the trajectory tool developed by Wernli and Davies [1997] and three-dimensional wind fields at all 60 model levels of the ERA-Interim data set. In order to trace back the transport path of the air mass above the target domain, trajectories are started from a regular mesh spanning the target region with 0.5° horizontal resolution and from 14 vertical layers (from 950 to 300hPa, every 50hPa). Spacing the trajectory starting locations in fixed pressure intervals facilitates the conversion of specific humidity changes into precipitation amount later on. The period of backward calculation is 10 days and only precipitating trajectories, i.e., trajectories with a relative humidity larger than 80% at the starting time t = 0 h and a decrease in specific humidity during the 6 h from t =− 6 to 0h, are considered.

Moisture sources are diagnosed using the quantitative Lagrangian moisture source diagnostic by Sodemann et al. [2008]. The diagnostic is based on the analysis of changes in specific humidity between 6 h intervals along the trajectories. It is assumed that moisture changes in an air parcel at this time scale correspond to the net difference of evaporation and precipitation. A positive moisture change during a 6 h interval therefore corresponds to evaporation exceeding precipitation, and vice versa for a negative moisture change [James et al., 2004]. Thus, also such processes as precipitation recycling are identified if they are effective at a time scale larger than 6 h. Figure 2 illustrates the moisture changes along an exemplary trajectory that contributes to the precipitation of one of the considered events by a moisture loss of 5gkg−1 in the final 6 h. When evaporation exceeds precipitation, the air parcel experiences a positive moisture change, as found for the periods from −36 to −30 h, −30 to −24 h, and −18 to −12 h. The impact of each of these so-called moisture uptakes on the precipitation at the starting point of the backward trajectory is weighted, considering the intermediate moisture changes between the uptake and the final moisture loss. For a more detailed description of the Lagrangian moisture source diagnostic, the reader is referred to Sodemann et al. [2008].

Details are in the caption following the image
Schematic illustrating the Lagrangian moisture source diagnostic along an exemplary trajectory. The solid black line corresponds to the vertical position of the air parcel (the trajectory path), the dashed line indicates the height of the atmospheric boundary layer. The time axis at the top indicates the time since the start of the backward trajectory. Six-hourly moisture changes along the trajectory are indicated in gkg−1. Blue and green shading at the bottom indicate ocean and land, respectively.

In Sodemann et al. [2008] it was assumed that an uptake can only then be directly connected to evaporation from the underlying surface, if the air parcel is located in the atmospheric boundary layer at the time of the uptake (which is the case during the periods from −36 to −30 h and −18 to −12 h in the example). A scaled boundary layer height was used, determined by the ERA-Interim boundary layer height multiplied by a factor of 1.5. Uptakes diagnosed above the scaled boundary layer height (in the free troposphere) are less straightforward to interpret. They are likely connected to (shallow) convection, but might also be due to horizontal turbulent mixing, evaporating precipitation, positive moisture innovations during data assimilation, or errors in the trajectories. However, a comparison of results considering only boundary layer versus all uptakes (boundary layer and free tropospheric) shows very similar uptake patterns in all season (not shown). For almost all events considered here, the fraction taken up in the boundary layer is very high, with seasonal averages ranging between 84.9 and 91.0%, free tropospheric uptakes between 5.3 and 8.7%, and only 3.6–6.4% moisture without identified source regions. Hence, in the following, we will always consider the total of both uptake categories. Note that this does not entail to disregard the separation capability of the Lagrangian diagnostic into boundary layer and free troposphere moisture uptakes in general; for other studies, it may be important to separate the more reliable uptakes identified within the atmospheric boundary layer (BL) from the ones in the free troposphere, as the fraction of free tropospheric uptakes can be substantially larger [Sodemann and Zubler, 2010].

3 Synoptic Composites

In this section we analyze the average synoptic conditions in terms of precipitation and atmospheric circulation leading to western Mediterranean HPEs by investigating composite plots for the 50 HPE days in each season. Figure 3 shows the seasonal composites of precipitation accumulated over the 50 HPE days. In all seasons except summer, the precipitation values show a distinct maximum at the southern slope of the Alps, exceeding 22mmd−1. In summer the values are more equally distributed between the French Atlantic coast and the Adriatic Sea with weak maxima over the Alps and the Pyrenees of 12–14mmd−1. In summer the lowest precipitation values can be found, and in autumn the highest. In autumn and winter the region with intense precipitation extends far to the west and south of the target region, indicating that in these seasons HPEs are typically of larger scale than in summer.

Details are in the caption following the image
Seasonal composites of the daily accumulated precipitation (in mmd−1) averaged for the 50 HPEs in (a) MAM, (b) JJA, (c) SON, and (d) DJF, respectively. The black frame marks the target region.

Figure 4 shows composites of potential vorticity (PV) on the 330K isentrope, 10 m wind vectors where their magnitude exceeds 5ms−1, and surface latent heat flux (negative values correspond to evaporation). For the surface flux, values have been averaged during the 3 days before each HPE day. In all seasons, a more or less pronounced upper level trough structure is visible in the PV contours over France (red), which corroborates the findings of Massacand et al. [1998] and Martius et al. [2006]. The trough is of similar intensity in summer, spring, and winter and more pronounced in autumn. It induces a flow of warm and moist air from the south to the target region and toward the Alps and the Massif Central, which supports the formation of orographic precipitation. Strong northerly or northwesterly low-level winds are found beneath the western side of the upper level trough over the eastern North Atlantic in all seasons. These regions of strong low-level winds are roughly consistent with regions of strong evaporation. This agrees with the results of Winschall et al. [2012], who revealed the connection of strong low-level winds and intense evaporation along the western flank of an upper level trough, which triggered a southern Alpine HPE. This mechanism appears to be particularly important in autumn (Figure 4c), when a clear evaporation maximum is visible to the west of Portugal at the back of the trough. In summer the strongest evaporation occurs over the land surface. In autumn and winter, strong surface winds and surface latent heat flux values of below −100 Wm−2 occur over the North Atlantic and in addition over the western Mediterranean.

Details are in the caption following the image
Seasonal composites for the 50 HPE in (a) MAM, (b) JJA, (c) SON, and (d) DJF, respectively, of the surface latent heat flux averaged over the 3 days before the HPEs (shading, in Wm−2), PV on a 330K isentrope (red contours for 2 and 4potential vorticity units) and 10m wind vectors for winds stronger than 5ms−1. Black contours denote 10m for velocities of 6 and 8ms−1 at 00 UTC of the HPE days. The black frame marks the target region.

To address the question if surface evaporation is anomalously intense during the days preceding a HPE, histograms of the surface latent heat flux are calculated for the blue North Atlantic grid points and the red Mediterranean grid points in Figure 1. The resulting histograms are presented in Figures 5 and 6 for the North Atlantic and the Mediterranean, respectively. For the North Atlantic region the histograms show the surface latent heat flux averaged over 72 h, which is regarded as a reasonable upper limit of the time required to advect moisture from the North Atlantic box to the target region [Winschall et al., 2012]. Histograms are shown for the 50 HPE days in each season (blue area) and, as a reference, also for all other days in the respective season during the years 1989 to 2009 (black line). In all seasons except for summer, the histogram for HPEs is shifted toward more negative (intense) values of the surface latent heat flux compared to the histogram for non-HPE days. This indicates that indeed higher than normal evaporation occurs over the North Atlantic during the 3 days prior to Mediterranean HPEs, which corroborates the findings of Winschall et al. [2012]. The exception is the summer season, for which the distributions are quite narrow and the difference between the fluxes on HPE and non-HPE days is relatively small.

Details are in the caption following the image
Histograms of the domain-mean surface latent heat flux in Wm−2 at grid points in the eastern North Atlantic (blue region in Figure 1), averaged over the 3 days before the 50 HPEs (blue shading) and before all other days between 1989 and 2009 (black line).
Details are in the caption following the image
Histograms of the domain-mean surface latent heat flux in Wm−2 at grid points in the western Mediterranean (red region in Figure 1), averaged over the day before and the day of the 50 HPEs (red shading) and before all other days between 1989 and 2009 (black line).

Figure 6 shows similar histograms of surface evaporation in the Western Mediterranean (see Figure 1). Here the considered time period includes only the day before and the day of the HPE. This shorter time period (compared to the analysis in the North Atlantic) is justified since Mediterranean evaporation is a more local source for HPEs in the considered target region. As before, HPEs (red area) are compared to non-HPE days (black line). Again, the evaporation before HPE days is intensified compared to the reference. Compared to the results for the North Atlantic, the Mediterranean histograms are more skewed and the mean values of evaporation are lower. Intense evaporation occurs mainly in SON and DJF.

For both regions and all seasons, except the North Atlantic during summer, more intense evaporation prior to HPEs is clearly evident. A Kolmogorov-Smirnov test on the similarity of the respective distributions for HPE and non-HPE days yields p values of less than 1% for all cases except for the North Atlantic in summer, when the p value exceeds 5%. Note, however, that this simple statistical approach cannot reveal a direct physical link between the moisture evaporating in the considered areas and the moisture precipitating in the target region. A refined analysis is presented in 6 based on the results of the Lagrangian moisture source diagnostic.

4 Composite Moisture Sources for HPEs

Figure 7 shows the results of the moisture source diagnostic for all HPEs in the four seasons. As discussed in 2, the total uptakes (boundary layer and free tropospheric uptakes) are shown. The units for the moisture uptakes are mmd−1, i.e., the values correspond to the daily accumulated precipitation averaged over the target region and then spread out to their source locations. In descriptive terms, the information depicted in Figure 7 is the contribution of total evaporation at the moisture sources to precipitation in the target region (contributing evaporation).

Details are in the caption following the image
Composites of the diagnosed moisture sources combined from boundary layer and free tropospheric uptakes (shading, in mmd−1) for the 50 HPEs in (a) MAM, (b) JJA, (c) SON, and (d) DJF.

At first glance, the results look similar for all seasons with a maximum moisture uptake over the western Mediterranean, but a closer look reveals distinctive differences. In spring, the moisture sources are mostly confined to the Mediterranean Sea and adjacent coastal regions (Figure 7a). The maximum is located between Sardinia and Tunisia. In summer (Figure 7b), the maximum is shifted to the west, and probably, due to the intense land evapotranspiration, moisture sources extend further into Spain, France, and Italy than during spring. In autumn, strong moisture sources are diagnosed in the Mediterranean west of 20°E (Figure 7c). Furthermore, the Lagrangian diagnostic indicates that the eastern North Atlantic is contributing with evaporation to Mediterranean HPEs in autumn, which is in line with the strong North Atlantic evaporation in general (Figure 4c). In winter, a widespread uptake area exists over the North Atlantic, while over land fewer moisture contributions are diagnosed. The maximum over the Mediterranean is located in the Balearic Sea. The identification of the North Atlantic moisture source in autumn is in line with other studies. For instance, Pinto et al. [2013] found that for the most intense HPEs in northwestern Italy, which occur predominantly in autumn, moisture advection from the North Atlantic is essential and much stronger than for less intense events in the same region.

Seasonal moisture uptake composites are compiled by averaging the moisture uptakes from all precipitation events in the respective season. We checked whether uptake composites are biased by the strongest HPEs (because the integral over all uptakes of one event scales with the total precipitation in the target area). Composite maps for which the accumulated uptakes of individual events have been normalized to one do not show a different pattern compared to the weighted results (not shown). Also, note that the fact that HPEs on consecutive days are included in the analysis does not bias the results toward these events (Figure S1 in the supporting information).

Generally, the diagnosed moisture sources are in qualitative agreement with the evaporation patterns shown in Figure 4. In summer the moisture contribution from land surfaces and in autumn and winter the North Atlantic sources match with the enhanced evaporation in the respective regions. In all seasons, the Mediterranean appears as the most important source. However, the moisture sources for individual events can strongly differ from these seasonal composites, and the Mediterranean may simply be the region which most consistently contributes to HPEs. This aspect of case-to-case variability is addressed in 5.

In addition to the location of the moisture sources, also the time when the uptakes occur relative to the onset of the HPE can be studied with the Lagrangian diagnostic. This allows to address the question if the uptakes take place shortly before the event (maybe associated with the same dynamical evolution that triggers the HPE), or if the moisture has already been present in the atmosphere several days before. This temporal dimension is illustrated for all seasons by the histograms of the times when the diagnosed moisture uptakes occur relative to the onset of the HPE (Figure 8). Hour zero corresponds to 0000 UTC of the HPE day. The maximum of each histogram is indicated by a solid black vertical line. In all seasons, it occurs at 0900 UTC on the day preceding the HPE. The maximum is most pronounced in MAM, while in SON earlier uptakes are more important (Figures 8a and 8c). The dashed black vertical lines mark the median, i.e., the time at which 50% of the moisture uptakes have happened, varying from t =− 57 h in spring to t =− 81 h in summer. The histograms show a daily cycle in spring, summer, and autumn with an uptake maximum in the afternoon, most probably due to the peak of insolation and evaporation at this time of the day. This effect of insolation on the daily cycle of evaporation is most pronounced over land, which explains the missing daily cycle in winter when the land surface is a negligible moisture source (cf. Figure 7d). Overall, Figure 8 reveals that most of the moisture uptakes contributing to the investigated HPEs occur 2 to 3 days before the event. The longest tail occurs in summer, when a substantial fraction of the uptakes occurs more than 10 days prior to the onset of the HPEs. Comparably long moisture residence times for warm conveyor belt humidity during summer have been found by Pfahl et al. [2014]. Note that this temporal scale does not straightforwardly translate into a spatial scale of the moisture transport, in contrast to the findings of Pfahl et al. [2014]. Indeed, Figure 7 indicates fairly local moisture sources in summer, i.e., in the season when the median uptake time is largest.

Details are in the caption following the image
Histograms of the moisture uptake times of all uptakes for the HPEs in (a) MAM, (b) JJA, (c) SON, and (d) DJF. The times of the most frequent uptake time is marked with a solid line, the median is indicated by a dashed line. Hour zero corresponds to 0000 UTC of the HPE day.

In comparison to the climatological results for southern Alpine precipitation from Sodemann and Zubler [2010, Figure 8], the moisture sources identified here are quite similar in location, but are up to 2 times more intense locally, and have larger contributions from the central North Atlantic during autumn and winter. This is in agreement with Drumond et al. [2011], who concluded that both moisture source extent and intensity in the North Atlantic should be important for seasonal precipitation variability in the Mediterranean, even though their study did not allow for a quantitative moisture source identification and only covered a short time period.

Due to the large spatial extent of the sources, it is not immediately obvious from Figure 7 what the percent contribution of, e.g., the Mediterranean and the North Atlantic to the precipitation during the HPEs is. Figure 9 provides this information as a cumulative distribution of all HPEs ranked according to the respective Mediterranean Sea contribution. Almost all HPEs have less than 50% moisture contribution from the Mediterranean Sea. Only 16–38% of all HPEs, depending on season, have more than one third of Mediterranean Sea moisture sources (Table 1, boldface). In turn, between 4 and 16% of all HPEs have less than 10% Mediterranean moisture sources (Table 1, italic). Contributions are highest in spring and autumn and lowest during winter, when long-range moisture transport dominates. Summer contributions are also lower due to the larger share of moisture contributed from land evapotranspiration. This analysis underlines that the Mediterranean is only one of several moisture sources during HPEs, and in fact a minor one. This does not necessarily mean that the Mediterranean moisture source is irrelevant, since precipitation processes are highly nonlinear, as has been argued in Winschall et al. [2013].

Details are in the caption following the image
Cumulative distribution of all HPEs according to the moisture contributions from the Mediterranean Sea surface for each season. The four selected HPE cases and their Mediterranean contributions are 7 June 1991 (4.5%), 19 December 1996 (7.6%), 2 January 1997 (18.4%), and 14 September 2006 (35.8%).

5 Moisture Sources for Individual HPEs

The composite analysis in the previous section characterized the average moisture sources of western Mediterranean HPEs. However, these composites do not provide information about the variability of the moisture sources for individual HPEs. Nevertheless, it is important to consider the variability between events to identify the decisive factors from a meteorological viewpoint. For example, tropical moisture sources can be relevant for some events, as had been speculated by Turato et al. [2004] and later confirmed by, e.g., Knippertz et al. [2013]. In other situations, the combination between Mediterranean evaporation and more remote moisture contributions appears as an important aspect of an HPE. To illustrate these aspects of the substantial case-to-case variability, and to underline the potential of detailed further investigation of each HPE, moisture sources for four Mediterranean HPEs (one for summer and autumn and two for winter) are presented and discussed with respect to the underlying meteorological situation in this section (cases marked with an asterisk in Table 1).

The event on 7 June 1991 has less than 5% moisture contribution from the Mediterranean (Figure 9), in contrast to about 71% from the North Atlantic, and in part extends into tropical latitudes. The moisture source maximum in the subtropical western North Atlantic (Figure 10a) is related to evaporation connected to a very deep, quasi-stationary subtropical cyclone that was present near 50°W, 35°N already about 10 days before the HPE (not shown). As the cyclone weakened, a large anticyclone formed with a center over the Canary Islands, leading to southwesterly moisture advection. The moisture was taken up rather continuously along the air parcel trajectories, which is evident from the histogram of the uptake times (Figure 11a). Fifty percent of the moisture uptake occurs more than 5 days before the onset of the HPE (median =−129 h). Shortly before the HPE, a small, intense cyclone formed west of France and precipitated the subtropical moisture in the target region (see Figure S2 in the supporting information), including contributions from evapotranspiration and precipitation recycling over northern Spain as indicated by the moisture uptake maximum there (Figure 10a). This “handover” of moisture between sequential cyclones has recently been identified as a potentially important mechanism for far-reaching meridional moisture transport in the North Atlantic [Sodemann and Stohl, 2013].

Details are in the caption following the image
Diagnosed moisture sources, combined from boundary layer and free tropospheric uptakes, for four individual HPEs (in mmd−1): (a) 7 June 1991, (b) 19 December 1996, (c) 2 January 1997, and (d) 14 September 2006.
Details are in the caption following the image
Histograms of the moisture uptake times of all uptakes for the four selected HPEs (a) 7 June 1991, (b) 19 December 1996, (c) 2 January 1997, and (d) 14 September 2006. The times of the most frequent uptake time is marked with a solid line, the median is indicated by a dashed line. Hour zero corresponds to 0000 UTC of the HPE day.

The event on 19 December 1996 shows a continuous band of moisture sources extending from the tropical Atlantic along the coast of western Africa into the western Mediterranean (Figure 10b). About 81% of the moisture are contributed by North Atlantic sources, and less than 8% originate from the Mediterranean. The spatial distribution of the moisture sources suggests that it may be related to a tropical moisture export [Knippertz et al., 2013]. It should, however, be noted that the tropical moisture is only responsible for a part of the total moisture. Synoptically, this HPE is related to a subtropical moisture advection induced by an unusually long-lived midlatitude cyclone (Figure S2 in the supporting information). The system formed in the western North Atlantic about 10 days before the HPE, then resided in the central North Atlantic at about 45°W, 35°N for 3 days, before moving slowly poleward while deepening strongly up to a pressure minimum of 965hPa. Both the maximum and the median of the corresponding histogram of uptake times occur about 6 days before the HPE (Figure 11b), which highlights the importance of long-range transport of tropical moisture to the Mediterranean for this event.

The event on 2 January 1997 shows a complex moisture source pattern with three maxima, one each over the eastern North Atlantic, Morocco, and the western Mediterranean (Figure 11c). Corresponding to these distinct moisture sources, the histogram of uptake times has a nearly bimodal distribution (Figure 11c) with one maximum 4 days prior to the onset of the HPE connected to the North Atlantic and the Moroccan sources, and the second maximum the day before the precipitation event connected primarily to the Mediterranean moisture contributions. The synoptic situation prior to the HPE was dominated by a large-scale trough over the eastern North Atlantic (Figure S2 in the supporting information), similar to the case studied by Winschall et al. [2012]. The HPE itself was related to a narrow PV streamer associated with a surface cyclone over Spain on the day of the event. While still dominated by North Atlantic moisture contributions (~53%), land regions (~28%) and the Mediterranean (~18%) contributed significant shares (see Figure 9). A notable characteristic of the uptake time histogram is the fact that almost all contributing moisture evaporated less than 7 days before the HPE.

The event on 14 September 2006 is the HPE with the most intense area-averaged precipitation value in the target domain (Table 1). Very strong moisture contributions are apparent in a large area of the western Mediterranean and the surrounding land surface with additional moisture supply from northern Europe and the eastern North Atlantic (Figure 10d). A high-pressure system, characterized by strong insolation and weak winds, resided over the Mediterranean and Western Europe during the 5 days preceding the HPE. On the day of the heavy precipitation, a PV streamer formed rapidly over Spain, associated with a weak surface cyclone, gathering much of the available moisture and precipitating it out effectively (Figure S2 in the supporting information). Corsica was affected by intense flooding that day [Lambert and Argence, 2008]. This event had a relatively large Mediterranean moisture contribution (~36%, see Figure 9) and a small North Atlantic contribution (~12%) but was dominated by land evapotranspiration (50%). The uptake time histogram (Figure 11d) is generally similar to the seasonal climatological pattern (cf. Figure 8c) but highlights a larger contribution of moisture that has been present in the atmosphere more than 3 days before the event. The pronounced diurnal cycle apparent in the moisture uptake time histogram underlines the substantial contribution from insolation-driven land evapotranspiration prior to the event.

The high variability in the moisture sources and uptake times between the different events is striking. In this context, it is, however, important to stress again that the cases discussed above are a subjective choice showing the substantial variability from case to case and are not intended to be representative for seasons or typical patterns. Such an analysis is beyond the scope of the present study.

6 Link of Evaporation and Precipitation

In this section we address in more detail the question if intense evaporation is a precursor of Mediterranean HPEs. We use the results of the moisture source diagnostic presented in the previous sections to link the diagnosed moisture uptakes with anomalies in the surface latent heat flux, following a similar approach as Pfahl et al. [2014]. The standard deviation of the 6-hourly averaged surface latent heat flux in the four seasons is used as a reference to determine if a 6-hourly flux anomaly is significant. This quantity is plotted in Figure 12. It shows some characteristic seasonal and spatial variations. In all seasons except summer, the maximum standard deviation in the considered domain is located in the North Atlantic storm track, west of 40°W, with values up to 130Wm−2 in winter. In summer this region only shows a standard deviation of 60Wm−2 while the largest values, exceeding 100Wm−2, occur over Southern European land areas. Over the Mediterranean Sea, values are quite uniform in spring and summer, but distinct maxima larger than 100Wm−2occur in autumn and winter, west of the Balearic Islands and in the Aegean Sea.

Details are in the caption following the image
Standard deviation of the 6-hourly averaged surface latent heat flux in Wm−2, during (a) MAM, (b) JJA, (c) SON, and (d) DJF.

To assess whether the diagnosed moisture uptake events are associated with surface latent heat flux anomalies, the 6-hourly grid point values of the surface latent heat flux are divided into two groups: the first group consists of all 6-hourly periods when a specific threshold for the total moisture uptakes as diagnosed in the previous section is exceeded at the grid point (so-called “uptake periods”), and the second group contains all other 6-hourly periods during the considered 20 years (“no-uptake periods”). The threshold to distinguish uptake from nonuptake periods has been set to 0.01mm(6h)−1. The sensitivity of the results to the choice of this threshold is small and will be discussed later.

Figure 13 shows the difference of the mean surface latent heat flux for the two categories defined above in units of the seasonal standard deviation (see Figure 12). Grid points are only considered if at least three uptakes are diagnosed that exceed the threshold. Negative values indicate lower values of the surface latent heat flux, i.e., enhanced evaporation during periods when uptakes are diagnosed compared to no-uptake periods. Furthermore, at each grid point, a t test with the null hypothesis of the surface latent heat flux being equal for the two categories has been performed. In Figure 13 values are only plotted at grid points where the null hypothesis is rejected with a confidence level larger than 95%.

Details are in the caption following the image
(a–d) Seasonal composites of the difference in 6-hourly averaged surface latent heat flux between uptake time steps (6-hourly periods that contribute more than 0.01mm(6h)−1 versus all time steps without uptakes in the respective season. The values are given in units of seasonal standard deviation of the surface latent heat flux. Only grid points are shown for which a t test rejects the null hypothesis of the means of uptake and no-uptake time steps being equal with a confidence level of 95%.

The figure panels show interesting seasonal and regional differences. In spring (Figure 13a), the area with frequent uptakes is confined to the Mediterranean and the surrounding land surface. The Mediterranean shows no signal of intensified evaporation during uptake periods. The values are near zero or slightly positive in the eastern Mediterranean. In contrast, over the surrounding European and African land surface evaporation is increased by about 1 standard deviation during uptake periods. The same is true for summer (Figure 13b). The difference of uptake versus no-uptake periods over the Mediterranean is close to zero, while for the more remote source regions over Northern Africa, Europe, and parts of the eastern North Atlantic, intensified evaporation is diagnosed. Over the North Atlantic west of France, the anomaly has an amplitude of more than 1 standard deviation. In autumn and winter uptakes over the North Atlantic also contribute importantly to the HPEs (Figures 7c and 7d). In these regions the difference in the latent heat flux between uptake and no-uptake periods is clearly negative (Figures 13c and 13d), indicating anomalously intense North Atlantic evaporation associated with Mediterranean HPEs. However, in the Mediterranean itself, the values are again close to zero as in the other seasons, except for more remote uptake regions in the eastern Mediterranean. Pronounced negative differences also occur in autumn over southeastern Europe and the western part of the Black Sea.

In summary, the Mediterranean Sea surface does not show a signal of a positive or negative evaporation anomaly during uptake periods associated with HPEs in any of the seasons (except for the Aegean Sea in autumn and winter). The anomaly values are either weak (in the range of −0.5 to 0.5 standard deviations) or not significant according to the t test. In contrast, remote sources like land evaporation (in all seasons except winter) and moisture uptake regions over the North Atlantic (mainly in autumn and winter) experience significantly positive anomalies of surface evaporation. Especially in autumn, values exceed 2 standard deviations over the eastern North Atlantic, the western Atlas, the Balkan, and the Aegean Sea.

The results of Figure 13 depend on the choice of the threshold used to distinguish 6-hourly grid point values of the surface latent heat flux that belong to the uptake and the no-uptake group, respectively. Most of the uptake regions shown in Figure 7 exceed the chosen threshold of 0.01mm(6h)−1, at least a few times. However, because of the reduction of the contribution of early (and typically more remote) uptakes due to subsequent raining out during transport (see 4), an uptake of 0.01mm(6h)−1 over the North Atlantic has to be considered as more extreme than the same value over the Mediterranean. We therefore tested the sensitivity of the results shown in Figure 13 and repeated the evaluation with lower and higher threshold values of 0.005mm(6h)−1 and 0.025mm(6h)−1, respectively. The results remain qualitatively similar even though most structure is retained with the current threshold (Figure S3 in the supporting information). With the lower threshold, more 6-hourly periods contribute to the group of uptake periods, in particular over Europe and the North Atlantic. Still, negative surface latent heat flux anomalies result for the remote sources and no anomalies are found over the Mediterranean. With the higher threshold, the different signal is limited to the Mediterranean Sea surface because more remote sources never contribute with such strong uptakes to Mediterranean HPEs. As above, the Mediterranean does not show an anomaly in evaporation, which indicates that the results of this detailed analysis of surface evaporation anomalies in moisture uptake regions leading to Mediterranean HPEs are robust. The Mediterranean itself does, on average, not show intensified evaporation prior to HPEs, while in remote source regions a (strongly) positive anomaly of evaporation can be found. Note that these are statistical results for a set of 50 HPEs in every season, which might not be valid for every single HPE, due to the strong case-to-case variability in the moisture sources and moisture transport conditions mentioned before. Finally, we also recall that this study focuses on fairly large-scale precipitation events. It could be interesting to perform a similar analysis for HPEs identified in smaller regions. This, however, would require the availability of high-quality high-resolution atmospheric analysis data with a detailed representation of topographical features, and convection-permitting grid spacing, covering the eastern North Atlantic and the western Mediterranean. Unfortunately, such a data set is currently not yet available for a multidecadal time period.

7 Conclusions

In this study the evaporative moisture sources for the 50 strongest heavy precipitation events (HPEs) in each season in the northwestern Mediterranean have been studied with a previously developed Lagrangian moisture source diagnostic. Composite maps of the synoptic conditions associated with the considered events show a characteristic upper level trough upstream of the HPEs, which induces a flow of moist air from the Mediterranean to southern France and Switzerland, and northern Italy, where the maximum precipitation values were located.

The Lagrangian diagnostic indicated that in all seasons, the Mediterranean Sea surface is only one of many moisture sources for the considered HPEs. In almost all cases, additional sources are needed to generate a precipitation extreme. In summer, evapotranspiration from the European land mass surrounding the western Mediterranean is an important moisture source for HPEs, and in autumn and winter, evaporation from the eastern North Atlantic supplies substantial moisture for the events. Most of the moisture uptakes occur during the 2 days before an HPE. However, the distribution of uptake times relative to the onset of HPE has a long tail and a considerable fraction of uptakes occurs more than 5 or even 10 days prior to the HPE. Whereas moisture uptakes that occur only a few hours or days prior to the raining out can be regarded as causally related to the dynamics of the HPE itself (see the detailed case study of Winschall et al. [2012]), the much earlier uptakes are probably physically unrelated to the meteorological evolution triggering the HPE.

Considering individual events revealed the high event-to-event variability in moisture sources of HPEs. Some events gain moisture from tropical moisture exports, leading to a maximum moisture uptake more than a week before the event in the (sub)tropics. For some events, precipitation mainly consists of North Atlantic moisture or of moisture from localized land evapotranspiration. Thus, Mediterranean HPEs with often similar dynamical configurations shortly before the onset of the event, involving the formation of an upper level trough over western Europe, can be related to strongly varying moisture supply due to the preceding meteorological evolution.

The second part of the study focused on the question whether the identified moisture source regions of HPEs show an anomaly in surface latent heat flux at the time of the moisture uptake, or, in other words, is anomalously intense surface evaporation at the local and/or remote source regions an important precursor of Mediterranean HPEs? To address this question, the surface latent heat flux during uptake periods associated with HPEs has been compared with periods when no uptakes for HPEs were diagnosed. The results reveal an interesting spatial pattern: no anomaly of evaporation can be found in the more local source regions over the Mediterranean in any season. Thus, for the moisture supply from the Mediterranean convergence of moisture from normal (i.e., average), evaporation is sufficient to generate HPEs. However, for HPEs with remote sources, either over the surrounding land surfaces or in the eastern North Atlantic, moisture uptakes for Mediterranean HPEs are associated with anomalously intense evaporation. We conclude that when these remote regions contribute to a Mediterranean HPE, very intense surface evaporation is essential to provide the required moisture supply. This result also indicates that it is important to better investigate the dynamics of very intense evaporation events upstream of HPEs, further away and farther back in time than may have been assumed previously.

Acknowledgments

MeteoSwiss and the ECMWF are acknowledged for granting access to the ERA-Interim data set. We would like to thank Oreste Realse, Roberto Rudari, and Joaquim Pinto for their constructive reviews, which helped to improve the manuscript.