Dry-air intrusions (DIs) are dry, deeply descending airstreams from the upper troposphere toward the planetary boundary layer (PBL). The significance of DIs spans a variety of aspects, including the interaction with convection, extratropical cyclones and fronts, the PBL, and extreme surface weather. Here, a Lagrangian definition for DI trajectories is used and applied to ECMWF interim reanalysis (ERA-Interim) data. Based on the criterion of a minimum descent of 400 hPa during 48 h, a first global Lagrangian climatology of DI trajectories is compiled for the years 1979–2014, allowing quantitative understanding of the occurrence and variability of DIs, as well as the dynamical and thermodynamical interactions that determine their impact. DIs occur mainly in winter. While traveling equatorward from 40°–50° latitude, DIs typically reach the lower troposphere (with maximum frequencies of ~10% in winter) in the storm-track regions, as well as over the Mediterranean Sea, Arabian Sea, and eastern North Pacific, off the western coast of South America, South Africa, and Australia, and across the Antarctic coast. The DI descent is nearly adiabatic, with a mean potential temperature decrease of 3 K in two days. Relative humidity drops strongly during the first descent day and increases in the second day, because of mixing into the moist PBL. Significant destabilization of the lower levels occurs beneath DIs, accompanied by increased 10-m wind gusts, intense surface heat and moisture fluxes, and elevated PBL heights. Interestingly, only 1.2% of all DIs are found to originate from the stratosphere.
Dry-air intrusions (or dry intrusions; DIs) are distinct large-scale airstreams that descend from the vicinity of the tropopause to middle and low tropospheric levels, while moving equatorward along the downward sloping isentropes. DIs are a fundamental component of the synoptic-scale atmospheric circulation, and as such their impact is revealed in a variety of aspects, over a range of scales, regions, and dynamical interactions, as elaborated in the following. Yet, their dynamical characteristics and occurrence frequencies are not quantified climatologically, which is the focus of the current study. At times, DIs are stratospheric in origin, and as such, they attracted attention for over six decades. DIs of stratospheric origin have also been referred to as stratospheric intrusions. The recognition of the circulation associated with tropopause folds, and their connection with intense cyclogenesis, was first formulated by Reed (1955) and Reed and Danielsen (1958), who diagnosed descending stratospheric air in the vicinity of the jet stream. Danielsen (1968) identified the importance of stratospheric intrusions in the context of chemical transport and mixing with tropospheric air. Indeed, deep stratospheric DIs influence the variability of surface ozone concentrations (e.g., Ebel et al. 1991; Stohl et al. 2000; Škerlak et al. 2014; Lin et al. 2015). Because of their low moisture content, DIs can be identified clearly in satellite imagery as “dry slots,” often with a hammer head structure (Browning 1993; Browning and Roberts 1994a; Browning and Golding 1995; Michel and Bouttier 2006). Accordingly, their impact on the radiative response of the atmosphere to solar forcing is strong, particularly if they reach the tropics (Cau et al. 2005).
DIs generally move equatorward, advecting dry and, in case of a stratospheric origin, high–potential vorticity (PV) air to latitudes where such low humidity (and high PV) values are highly anomalous. This aspect is highlighted in tropical and subtropical latitudes, and has been studied from a climatological perspective (Kiladis 1998), diagnosed as isentropic high-PV tongues that extend to the tropics (Waugh and Polvani 2000; Fröhlich and Knippertz 2008), and by analysis of selected events (Waugh 2005). This type of stratospheric intrusion is associated with Rossby wave breaking in the jet exit region in the subtropics, mainly in winter (Waugh and Polvani 2000; Waugh and Funatsu 2003). Tropical convection was triggered by the intrusions on the southeastern side of their tip in the Northern Hemisphere (Waugh and Funatsu 2003; Waugh 2005). Similarly, a trajectory analysis for one monsoon season demonstrated how extratropical DIs in the Sahelian midtroposphere intensify convection over West Africa (Roca et al. 2005). However, DIs can originate from various regions and descent routes and may, in some cases, also inhibit tropical convection and/or tropical cyclone activity by modifying the static stability and vertical wind shear (Cau et al. 2007; Dunion 2011; Zhang et al. 2016).
In midlatitudes, DIs occur as an element of baroclinic wave life cycles (Thorncroft et al. 1993) and serve as an important component of airflow around cyclones (Carlson 1980). A conceptual heuristic model that describes the role and structure of DIs in cyclone dynamics was summarized in Browning (1997). DIs have been observed to precede cyclone deepening (e.g., Uccellini et al. 1985; Young et al. 1987; Browning 1993). Case studies suggest that stratospheric DIs contribute to the intensification of cyclonic circulations by high-PV air in upper levels overrunning moist air and a low-level baroclinic zone (Young et al. 1987). Tropopause folds associated with descending airstreams have been shown to contribute significantly to the generation of potential instability by wind shear acting across moisture gradients (Griffiths et al. 2000) and to modulation of convective activity (Russell et al. 2008; Russell et al. 2009; Russell et al. 2012). Unique to stratospheric DIs is their impact through the advection of high-PV air down to lower levels (Browning and Golding 1995; Wernli 1997). At times, DIs may contribute directly to severe surface wind gusts (Browning and Reynolds 1994; Raveh-Rubin and Wernli 2015), as well as to intense precipitation in cyclones through their interaction with the moist air ahead of the cold front (Carr and Millard 1985; Browning and Golding 1995; Raveh-Rubin and Wernli 2016). The mesoscale interaction of the leading edge of the DI with the cold front and moist warm sector ahead of it have been studied by combining numerical model output and observational data for selected events (Young et al. 1987; Browning and Reynolds 1994; Browning and Roberts 1994b; Browning and Golding 1995). In this context, the DIs have a dual role in destabilizing the frontal region. DI may suppress continuous rain formed in a warm conveyor belt (WCB) by underrunning it in its northern part (Browning and Golding 1995) and by evaporating rain falling from the cloud head (Browning and Roberts 1994b). On the other hand, DIs may trigger convection in areas where it overruns moist air, thereby releasing potential instability and causing strong winds, precipitation, and thunderstorms (Young et al. 1987; Browning and Reynolds 1994; Browning and Golding 1995; Lagouvardos and Kotroni 2000; James and Clark 2003; Yang et al. 2009; Gao et al. 2010; Raveh-Rubin and Wernli 2016). However, the climatological relationship between DIs and extreme surface winds is unknown. Beyond their impact on surface winds, precipitation and convective activity, the DI modulation of the planetary boundary layer (PBL) gives rise to wildfires (Mills 2008; Pollina et al. 2013; Schoeffler 2013; Langford et al. 2015) and can potentially influence air–sea interaction.
A Lagrangian approach provides crucial information on the dynamics of airstreams (Wernli and Davies 1997; Stohl 2001). Stohl (2001) constructed an ensemble of forward trajectories in the Northern Hemisphere for one year, and identified both DIs (as strongly descending trajectories) and events of stratosphere–troposphere transport. The present study builds on methodology similar to that of Stohl (2001) to construct a global, all-year, and long-term DI climatology with additional emphasis on dynamical characteristics. Stohl (2001) found that DIs start along the 30°–60°N belt with maxima in East Asia and western North America, almost exclusively in winter. In contrast, the spatiotemporal spread of stratosphere–troposphere transport is larger. Evidence suggests that trajectories that originate in the stratosphere typically do not experience a very deep descent (Wernli 1997; Stohl 2001), but when they do, they are confined to the storm-track region in winter (Wernli and Bourqui 2002). In a recent study, Škerlak et al. (2014) compiled a global climatology of deep stratosphere–troposphere transport using a refined Lagrangian algorithm. They confirmed that the mass flux is concentrated in the storm tracks, qualitatively related to an increased tropopause fold activity. In the DI case described in detail in Wernli (1997), dry descending trajectories were selected with a criterion of a minimal descent of 350 hPa in 48 h, further refined to identify different aspects of the DI, mainly considering the PV evolution and stratospheric origin of the trajectories. It was shown that the strongest descent occurs for DI trajectories originating from the upper troposphere and that descent associated with stratospheric air parcels crossing the tropopause is weaker. Nevertheless there is not yet an established Lagrangian definition of DIs that has been applied globally, and hence no systematic long-term Lagrangian climatology of DIs exists. Here, we construct a global Lagrangian climatological dataset of DIs for the years 1979–2014 and quantify their global occurrence and variability, as well as the dynamical and thermodynamical processes that determine their impact. Specifically, we address the following questions:
What is a suitable Lagrangian criterion for identifying DIs that descend substantially into the middle and low troposphere?
Where and when do DIs typically occur?
What are the typical dynamical characteristics of DIs? Do they vary in the different regions?
How many DIs are stratospheric in origin?
Do DIs mix into the atmospheric boundary layer? If so, what is their impact?
2. Data, Lagrangian methods, and DI selection criterion
Global atmospheric data from the ECMWF interim reanalyses (ERA-Interim; Dee et al. 2011) are used for the years 1979–2014. The data are horizontally interpolated to a 1° × 1° grid, with 6-hourly time intervals and 60 vertical hybrid levels.
The Lagrangian analysis tool (LAGRANTO), version 2.0 (Sprenger and Wernli 2015), is used to diagnose DIs. The tool calculates trajectories using the three-dimensional ERA-Interim wind fields with an iterative predictor–corrector procedure. The systematic calculation of trajectories was carried out for every 6-h time step between 0000 UTC 4 January 1979 and 1800 UTC 31 December 2014. At each starting time, 48-h long forward trajectories were initiated from a three-dimensional grid, with 80-km horizontal spacing, and 20-hPa vertical spacing, above the 600-hPa level. The three-dimensional location of the air-parcel trajectories is stored every 6 h along the 48-h interval. In a second step, a selection criterion of a pressure increase (i.e., a descent of at least 400 hPa during 48 h) is applied to identify automatically DI trajectories according to their strong descent (the choice of the 400-hPa threshold is discussed below in section 3a). The selected DI trajectories are further extended, 48 h backward and 72 h further forward in time, beyond the 48-h interval of strong descent, yielding a total DI trajectory length of 7 days. In addition to their location (i.e., latitude, longitude, and pressure), the following physical variables are traced along the DI trajectories: three-dimensional fields of potential temperature, specific humidity, relative humidity, potential vorticity, and velocity are interpolated to the trajectory location at every time step. Two-dimensional fields, such as surface sensible and latent heat fluxes and maximum 10-m wind gust, are traced according to the horizontal location of the trajectory. To reflect the maximum value of 10-m gust at each 6-h step, the value is taken to be the maximum in a 250-km radius circle around the horizontal location of the trajectory, which is a typical distance over which the trajectory travels during 6 h. Additional diagnostics are calculated, including the boundary layer height anomaly, the vertical gradient of equivalent potential temperature (in the 800–950-hPa layer), and the PV anomaly, as elaborated in the section 3.
Naturally, a trajectory can exhibit an increase in pressure that fulfills the DI criterion in a time shorter than 48 h. To avoid considering the same air parcel multiple times, an additional filter is applied to the DI trajectories, which deletes trajectories that descend at least 400 hPa during a shorter time window starting after time 0 h. By construction, the algorithm results in sets of DI trajectories that start their descent at time 0 h. This algorithm to avoid double counting of the same air mass has been applied previously to the Lagrangian dataset of WCBs, and was described in detail in Madonna et al. (2014). Finally, to diagnose DIs of stratospheric origin as a specific subset of DIs, the DI trajectories had to fulfill an additional objective selection criterion, namely, the absolute value of their PV during the whole 48-h period prior to their 400-hPa descent had to be larger than 2 PVU (1 PVU = 10−6 K kg−1 m2 s−1).
a. Lagrangian definition of dry intrusions and its sensitivity
To establish a suitable Lagrangian criterion of DIs, an appropriate vertical descent threshold and a time period have to be set. Because of the high case-to-case variability of the humidity field, as well as its regional and seasonal variation, no moisture criteria are imposed. Rather, the characteristics of the moisture along DIs and its spatiotemporal variability will be an outcome of this study using a selection based on vertical descent only. The chosen criterion is thus a 400-hPa increase during 48 h, which identified coherent sets of DIs. In the following, the sensitivity to variations in the selection criterion is discussed.
The sensitivity to the selection criterion is tested for different pressure thresholds in the range between 350 and 500 hPa during time windows varying between 24 and 96 h. Figure 1 shows the selected trajectories fulfilling different criteria for one starting time. The descent starts generally from the eastern flanks of upper-tropospheric ridges (note that the Eulerian, instantaneous 2-PVU contour on 315 K is shown at the starting time of the descending trajectories). Five coherent sets of DIs are identified in the Northern Hemisphere, and two in the Southern Hemisphere (Fig. 1e). Of course, more trajectories are identified for a smaller pressure difference and for a longer time window. Consequently, the composition of a given coherent set of trajectories changes for different criteria (e.g., the set in the northeastern Atlantic that reaches North Africa is composed of different trajectories for different threshold criteria), as well as the globally identified number of such sets. However, the effect of varying the time window results in selecting some mutually exclusive trajectories. For example, a larger time window (for a given pressure difference threshold) allows for additional slowly descending trajectories but at the same time excludes trajectories that descend and later ascend, which is a common feature of the midlatitude flow (see section 3c). Therefore, different dynamical characteristics typify DIs that descend over a different time period. Figure 1f summarizes the global number of trajectories that fulfill the selection criterion for the different thresholds for a 1-week period, showing dramatic differences that span four orders of magnitude. Thus, computational considerations also play a role when a long-term computation is carried out. Evidence from case studies (e.g., Raveh-Rubin and Wernli 2016) suggests that both ends of the range of the time window are potentially interesting to investigate. The choice of 400 hPa descent during 48 h is appropriate both from a computational standpoint regarding the typical number of trajectories selected at every 6-h time step and from the desire to obtain coherent sets of trajectories, identified to descend over a synoptic time scale.
b. Global Lagrangian climatology: Seasonal and geographical variations
The automatic identification procedure is carried out every 6 h for the years 1979–2014, which results in more than 26 × 106 DI trajectories overall. Table 1 summarizes the number of identified DI trajectories, partitioned into the two hemispheres and seasons. In this section, the climatological spatiotemporal occurrence of DIs is quantified, in particular their monthly frequencies and geographical distributions.
The DI monthly distribution is shown by the black bars in Fig. 2. Globally, DIs are more prevalent in the winter months compared to the other seasons. This characteristic is more pronounced in the Northern Hemisphere, where fewer DIs occur in June–August (JJA), compared to Southern Hemispheric DIs in December–February (DJF; see also Table 1).
To present the geographical distribution of DI occurrence, we focus on the DI positions in the summer and winter seasons: DJF and JJA. Since DI air parcels move typically equatorward and eastward over thousands of kilometers, their positions at the different times relative to the 48-h time window for descent are differentiated. Time 0 h is defined as the time when DI air parcels begin their 48-h long descent. Every 6 h, the position of DI trajectories is set to 1 at every 1° × 1° regular grid point if at least one DI trajectory is present in the grid cell, anywhere in the vertical column. Otherwise, it is set to 0. This occurrence flag is summed over all 6-hourly steps in the relevant months for the years 1979–2014, and normalized by the number of 6-hourly steps, resulting in the DI occurrence frequency (Figs. 3 and 4). In DJF, the Northern Hemisphere experiences the maximum in DI frequencies. The DI air parcels are distributed at the start of their descent around 45°N latitude over the oceans, as well as in northeastern Asia and continental North America, and around 40°N in the Mediterranean Sea and central Asia. However, increased frequencies over northeast China, Japan, the western North Pacific, and continental North America are clearly evident (Fig. 3b). The pre-DI air, as seen by the occurrence frequency at −48 h (Fig. 3a), shows a large spread, within a belt between 30° and 75°N, with a pronounced frequency peak (10%) over Asia, and a second maximum (5%) over western Canada and the eastern North Pacific. As expected by the equatorward descent, at midjourney (i.e., 24 h) DIs are frequent in a belt between 25° and 50°N, with highest frequencies south of Japan and in the western and central North Pacific (>10%) and in the western North Atlantic (5%). The DI frequency at the end of the descent, at 48 h, peaks in the storm-track regions (i.e., in the western and central North Pacific, and in the North Atlantic) where extratropical cyclones are frequent (>20%; Wernli and Schwierz 2006). The geographical distribution in the Northern Hemisphere is generally consistent with results based on a single year and a different Lagrangian descent criterion (Stohl 2001). Yet, a major area with high frequency of DI starting positions in the western United States and eastern Pacific (Fig. 4 in Stohl 2001) is not pronounced here, compared to the other maxima. The geographical distribution in the winter storm-track regions is very similar to that of WCBs inflow locations [cf. Fig. 3d herein and Fig. 4d in Madonna et al. (2014)]. Additionally, high frequencies of DIs occur in the eastern North Pacific, off Baja California (Fig. 3d). Post-DI air (72 and 120 h) shows a spread to a wide zonal belt, indicating a spread of DI air both to the subtropics and to high latitudes. In the Southern Hemisphere, specific areas have a persistent (although low, ~1%) DI frequency, namely coastal areas in southern and western Australia, southern South America (on both sides of the Andes), and southern Africa. The differences among DIs in the different regions (red rectangles in Fig. 3c) will be addressed in section 3c(3).
In JJA, the NH cyclones migrate poleward and are less frequent (e.g., Wernli and Schwierz 2006), and WCB frequency is reduced by a factor of 3–4 (Madonna et al. 2014). Almost all DIs occur in the Southern (winter) Hemisphere, with the exception of a low (1%) and spatially confined region of DIs off the western North American coast (Fig. 4), consistent with Stohl (2001). In the Southern Hemisphere, a robust zonal belt of DI frequency larger than 3% is confined to latitudes higher than 30°S. A maximum of occurrence frequency of DIs at time 0 h is found in the South Atlantic, which likely end their descent at 48 h off Namibia and South Africa. Other regions with elevated frequency (most discernible at time 48 h; Fig. 4d) are the central South Pacific, eastern South Pacific (where there is a significant low-level DI concentration off the northern Chilean coast; Figs. 4d–f), and the south Indian Ocean. Interestingly, in the south Indian Ocean, post-DI air may cross the equator. Additionally, DIs descending off the Antarctic coast are relatively frequent (8%–10%) between 0° and 160°E. These regions will be examined in greater detail in section 3c(3).
c. Dynamical characteristics
During the course of the DI descent, the dynamical and thermodynamical characteristics of the air parcels change substantially. In this section we examine the DI dynamical characteristics and their temporal evolution. The general DI set includes all identified DI trajectories during 1979–2014, and is referred to as “reference” hereafter. We then focus on the subset of DIs with stratospheric origin and examine the regional variability of DI dynamical characteristics.
1) General characteristics
By construction, a coherent descent of at least 400 hPa is the fundamental characteristic of every DI trajectory. The variability of the pressure level of DIs during their descent, as well as during their pre-DI and post-DI stages, is shown in Fig. 5. On average, DIs descend from 400 to 800 hPa. The variability of the evolution of pressure (and other quantities) incorporates regional and seasonal differences [section 3c(3)], as well as variability among trajectories in any coherent set of trajectories [see, e.g., section 3d(1)]. During the DI descent, marked between the vertical lines in Fig. 5, the vertical positions of 90% of DIs vary by approximately 300 hPa at every time step. However, this variability is almost doubled at time −48 h, or even larger at time 120 h, indicating a large diversity in the pre- and post-DI stages, compared to the more coherent DI descent, partly dictated by the definition of DIs. The separation between the median and mean pressures at time −48 h indicates a long tail of the pressure distribution toward the mid-to-low levels, while the main part (>75%) of DI air originates above the 600-hPa level. At 120 h, the majority of DI air remains below the 700-hPa level, while a small fraction rises again to 400 hPa.
Additional parameters are traced along the DI trajectories, namely, potential temperature θ, specific humidity q, relative humidity (RH), PV, and PV anomaly (PVA) with respect to the local long-term monthly climatology (Fig. 5). The pressure p increase along DIs dictates a substantial adiabatic warming. In terms of potential temperature, however, a small diabatic cooling of the order of a few kelvin is observed. The diabatic cooling is likely a result of radiative cooling and/or evaporation of water into the dry air mass. Radiative cooling is expected to be a dominant factor during the first 24 h, when evaporation is minimal, as typical values of radiative cooling rates are of the same order (e.g., Savijärvi 2006). The originally dry air indeed gains substantial amounts of moisture, as seen by the increase of q starting at 18–24 h. This increase is possibly due to mixing with the relatively moist PBL air and/or a contribution from evaporation of sedimenting hydrometeors from overhead clouds. The evolution of RH suggests that there are two typical stages during the DI descent. First, during the first 18–24 h, the DI air parcels warm adiabatically, causing a decrease of RH to minimal levels. In the second phase, as the DI air mixes into the PBL, the RH increases because of the gain in water vapor. A large variability of q and RH exists at this stage, because of the variability of PBL characteristics and the intensity of mixing.
It provides insight to follow the evolution of PV and to diagnose a possible stratospheric origin of the DI air, PV production/destruction by nonadiabatic processes, and the DIs’ dynamical effect on the lower-tropospheric environment. For this purpose, both PV and PVA are traced along the DI position, where values in the Southern Hemisphere are multiplied by −1. It is evident that the PV of DIs is largely tropospheric (smaller than 2 PVU in the Northern and larger than −2 PVU in the Southern Hemisphere). PV remains constant or slightly decreases, possibly by the aforementioned mixing into the low-PV air in the lower troposphere. Yet, PV anomalies advected by the DI to the lower troposphere impact the circulation and stability there. Calculation of the PV difference from the local long-term monthly climatology (PVA) along DIs shows, however, that high-PV anomalies (e.g., with magnitudes larger than 1 PVU) occur rarely. It is interesting to note that prior to the DI descent the PVA of the DI is mainly negative (i.e., 75% of the trajectories at −12 h have negative PVA), indicating their location within a ridge in the upper troposphere. Nonetheless, 1.2% of DIs have PV magnitudes larger than 2 PVU during the 48 h prior to the DI descent, and can therefore be considered to be of stratospheric origin. This subset is examined in more detail in the next subsection.
2) Stratospheric DIs
Stratospheric DIs (SDIs) are identified according to their PV during the two days before descending (i.e., between −48 and 0 h). The criterion for an SDI is that its PV magnitude remains larger than 2 PVU during this time period. As noted in section 3c(1), 1.2% of DIs fulfill this criterion. The numbers of SDIs are summarized for the different seasons and each hemisphere in Table 1 (in parentheses). The numbers indicate that, when comparing the two hemispheres, winter SDIs are slightly more frequent in the Northern than in the Southern Hemisphere (77 145 vs 62 999 trajectories), and in summer SDIs are much more frequent in the Southern Hemisphere (12 250 trajectories vs 2103 in the Northern Hemisphere). This property occurs in the general DI population but is amplified for SDIs.
The relative monthly distribution of SDIs is presented in Fig. 2 (gray bars). Although the winter peak is apparent, SDIs also occur in the transition months. Compared to all DIs, the relative monthly occurrence is higher for SDIs during the months outside of DJF in the Northern and JJA in the Southern Hemisphere.
The geographical frequency distributions of SDIs in DJF and JJA are shown in Figs. 6 and 7, where an arbitrary contour (4% occurrence frequency) from the general DI distribution is shown as a reference. The distribution of SDIs in DJF differs significantly from the overall DI distribution. As expected from their different origin, the distribution prior to 0 h (Fig. 6a) differs substantially from the general distribution pattern, as the SDI subset is strictly above the dynamic tropopause. A major stratospheric origin region for SDIs is over China, shifted by 5°–10° northeastward compared to all DIs (black contour in Fig. 6a). Moreover, distinct from the general DI occurrence, a significant source region of SDIs is the Hudson Bay, from where SDIs start to descend over North America, similar to the reference DI pattern (Fig. 6b). The SDIs then descend over the United States, and reach the lower troposphere east or west of the United States, dominating the overall SDI frequency distribution (Fig. 6d). In contrast, it is the western North Pacific that hosts the majority of the general population of DIs. In the Southern Hemisphere, in JJA, the geographical distribution of the SDI subset is generally aligned with the general DI distribution, both being zonally homogeneous with peaks off the coast of the continents, and over Antarctica (Fig. 7). In MAM, the SDI frequency peaks over the western United States (0.2%), with frequent occurrence over East Asia, southern Australia, and Antarctica (near 60°E; not shown). These areas encounter fewer SDIs during SON, compared to MAM. During SON, SDIs are frequent in the North Atlantic, off the northern Chilean coast and southern Australia (not shown).
The distinct origin of SDIs implies that their dynamical characteristics may be expected to differ from the general (largely tropospheric) set of DIs. Referring to the mean evolution of the physical parameters of the general DI population (solid lines in Fig. 5), here the same variables are traced along SDI trajectories. At every time step, the difference from the reference DI mean is calculated for the different variables and denoted with a Δ (dashed line in Fig. 8). Systematic differences between the mean SDI evolution and all DIs are evident. In accordance with the SDIs’ high-altitude origin, the SDI pressure is systematically lower compared to the DI reference mean. The dip in the Δp evolution of SDIs suggests that the SDI descent is slower, compared to the reference mean in the first 24 h but faster in the second stage until 48 h. The evolution of θ shows a small positive difference compared to the reference mean prior to 0 h but almost no difference thereafter, suggesting again a higher initial elevation as well as more pronounced radiative cooling. Consistent with the high initial elevation and the pressure evolution, specific and relative humidity are systematically lower for SDIs. Finally, by construction of the SDI dataset, PV and PVA values are more than 2 and 1.5 PVU higher, respectively, compared to the reference mean prior to 0 h. At later times, although PV decreases, ΔPVA indicates that on average, significant PV anomalies with an amplitude of about 0.5 PVU reach the PBL, suggesting a potential impact on the stability and circulation in their vicinity.
3) Regional variability
The global reference profiles (Fig. 5) embrace large variability, partly due to systematic regional differences. Here, we examine the mean evolution of key parameters in nine regional subsets of DIs, which correspond to peaks in winter DI frequencies. The selected subsets are summarized in Table 2, and the regions are marked with red rectangles in Figs. 3c and 4c. A DI is attributed to a regional subset if it is located in the predefined rectangle at least once during its 48-h descent. As done for the SDI subset in the previous subsection, here we investigate the differences of the various quantities of the subset mean from the global reference mean (i.e., the solid lines in Fig. 5). The evolution of the differences, denoted with a Δ, are shown for the different parameters for seven out of the nine subsets (Fig. 8). The characteristic evolution of the SP and SWAF subsets resemble other subsets, as elaborated in the following, and are therefore not shown in Fig. 8. The NA, NP, and SP subsets share a similar evolution of pressure difference from the reference, by starting the descent about 20 hPa higher. Moreover, beyond 48 h, a lower mean pressure (by 30–50 hPa) indicates some ascent of post-DI air in those regions. In contrast, post-DI ascent is less common for DIs in the WUS, SWAM, SWAF, and SIND. ANT DIs start distinctly at higher pressure and descend more slowly compared to the reference in the first phase but faster in the second phase of the descent. Finally, ANT DIs remain on average at higher pressure in the post-DI stage.
The DI subsets follow a similar evolution of Δθ but differ according to their initial θ values. At 48 h the subsets differ in their Δθ slope, indicating increasing θ for the NA, NP, and SP DIs but opposing tendencies for WUS, SWAM, and SWAF. Other subsets exhibit an evolution similar to the reference mean, while ANT DIs are systematically colder. The moisture content along DIs differs dramatically among the different regions. Both q and RH evolutions exhibit the large diversity of low-level humidity fields, as most variability arises after 18–24 h, when some DI trajectories start mixing with moist air in the lower troposphere. At 48 h, DIs in NA and NP are moister than average, while SWAF, SWAM, WUS, and especially MED DIs are anomalously dry. The MED DIs remain particularly dry for the following three days (pink line in Fig. 8) while the others tend to equilibrate toward the reference, which can be explained by the unique continental environment of the MED DIs after the descent, which is partly over the Sahara Desert (not shown). Last, PV and PVA regional variability is insignificant for the mean values. Yet, ANT DIs (black line in Fig. 8) have a mean PV magnitude of 1–2 PVU during their descent; however, this mean value reflects climatological values and is not apparent in their mean PVA signal. Most likely, this diabatic PV modification during the descent is related to mixing processes near the very stable Antarctic boundary layer. Overall, the mean evolution suggests that three groups of subsets have different characteristics, namely Antarctic DIs, storm-track DIs (NA, NP, and SP), and DIs in regions outside of the storm tracks that may also interact with continental orography and variable PBL characteristics (WUS, SWAF, SWAM, MED, and SIND).
d. Impact on the planetary boundary layer
It was shown in the previous section that DI trajectories reach the top of the boundary layer after about 18–24 h on average. The relatively dry air of the DI is then mixed within the well-mixed PBL, thereby changing its characteristics. Here the DI impact on the static stability, surface fluxes, and wind gusts is examined first with an illustrative case study, and further placed in a climatological context.
1) An illustrative case
An illustrative example is given for a coherent set of 191 DI trajectories, which started their descent on 0000 UTC 21 October 2014 near southern Greenland (Fig. 9a), crossed over the United Kingdom, France, and central Europe, and reached the lower troposphere in the western Mediterranean Sea, before spreading farther across North Africa (while the eastern branch ascended toward the Black Sea). During the DI lifetime, 10-m maximum wind gust (parameterized in ERA-Interim) exceeded 20 m s−1 below the DI (not shown). During this time, severe weather along the DI pathway was reported to the European Severe Weather Database (ESWD, which provides detailed and quality-controlled information on severe convective storm events over Europe; see Dotzek et al. 2009). Widespread severe wind gust impacts were reported along the DI pathway over the United Kingdom and continental Europe, including wind gusts exceeding 25 m s−1 and causing damage resulting from falling trees. Farther downstream, heavy precipitation and severe convective wind gusts were reported in eastern Europe, with local flooding of main roads and approximately 100 houses resulting from several hours of heavy precipitation and flash flooding of a local stream in Koprivnica, Croatia, on 22 October. Thus, to assess the relation between the DI to the surface weather, the static stability in the atmospheric column is investigated. The conditions in the atmospheric column along the DI are shown with a Lagrangian vertical cross section (i.e., a vertical cross section along the set of DI trajectories; Figs. 9b,c) such that at every time step the atmospheric profiles are taken at the horizontal position of the 191 trajectories and then averaged. Therefore, the x axis represents time (in hours), and the locations of the trajectories within the vertical column are marked with black dots. The vertical structure of specific and relative humidity is presented in Fig. 9b. In this case, DI trajectories start their descent from a layer between 300 and 600 hPa, within a dry air mass. The first 18–24 h are typified by a dry air mass overriding a moist layer. After this time, some DI trajectories enter the moist PBL and mix down to low levels until 48 h, and closer to the surface at 60 h. As the DI air is advected to the southeast, the specific humidity content in the lower troposphere increases, but the vertical gradient of RH weakens. In the first descent phase, the DIs follow the downward-sloping isentropic surfaces, whose slope is determined by the meridional temperature gradient (Fig. 9c). Moreover, the downward advection of very dry air creates a strongly negative θe gradient with height and a potentially unstable layer below 800 hPa, which can facilitate the downward mixing of DI air into the PBL in its second phase of descent if the instability is released. Noteworthy is the lack of convective motion along the DI, as parameterized in the reanalysis. The DI originates mainly in the upper troposphere and PV mostly remains below 2 PVU throughout the DI lifetime. Yet, the DI creates a tongue of relatively high PV (0.8–1 PVU), a possible signature of mixed stratospheric fraction, that reaches the top of the mixed PBL. This anomaly, although weak, consistently also reduces the static stability below, favoring the downward mixing of high-momentum DI air.
The evolution of p, θ, q, RH, PV, and PVA along the DI from October 2014 (Fig. 10) is consistent with the average evolution depicted for the DI climatology (Fig. 5), and serves also as an example for the variability within one coherent set of DI trajectories. The trajectory velocity of DIs (VEL; calculated in a Lagrangian framework) shows jet stream wind speeds at 0 h, followed by a monotonic deceleration (Fig. 10). To quantify the evolving PBL characteristics, the following quantities are traced along the DI trajectories: 10-m gust maximum within a 250-km radius circle around the DI position, and sensible and latent surface heat fluxes. To quantify the DI impact on the static stability below it, and on the PBL height, two additional quantities are traced along DIs: the vertical difference in equivalent potential temperature between 800 and 950 hPa at the horizontal position of the trajectory, θe,800 − θe,950, and the anomaly of the top of boundary layer pressure (in hectopascals), calculated with respect to the long-term monthly mean and to the diurnal cycle. The PBL height itself is parameterized in the ERA-Interim data by the mass flux/eddy diffusivity concept, using different K closures for different PBL regimes (ECMWF 2007; Dee et al. 2011). The two quantities provide measures for the low-level potential instability and for the deepening of the well-mixed boundary layer, respectively (Fig. 10). During the DI descent, the high 10-m wind gust signature is evident, with the majority of DI trajectories located where gusts exceed 20 m s−1 but is less pronounced in the pre- and post-DI stages. The evolution of θe,800 − θe,950 shows indeed the development of the negative values during the DI descent, and its absence before and after (Fig. 10). The deepening of the well-mixed PBL is evident clearly by the evolution of the PBL pressure anomaly, which shows an anomalously deep PBL (by more than 100 hPa compared to climatological conditions). The mixing of DI air into the PBL and with the reduced static stability and the high gusts all provide ideal conditions for intense heat fluxes into the atmosphere, especially above the ocean. Indeed, significant surface heat fluxes are found, with an extreme case of surface latent heat flux during the second DI phase, reaching 800 W m−2 (Fig. 10) over the Mediterranean Sea. It is probable that the strong heat fluxes into the atmosphere further enhance and sustain turbulent mixing in the PBL, as well as significantly impact the Mediterranean water mixing. This case exemplifies the high variability of the surface flux response to the approach of a DI, shown by the large spread among the DI trajectories. This is expected from the diverse PBL characteristics along the DI pathway, from the North Atlantic, to Europe with complex topography, to the warm Mediterranean Sea, and finally to northern Africa.
2) Climatological impact on the PBL
To evaluate the generality of the processes delineated in the case above, the same PBL state parameters are traced along the position of all DI trajectories in the study period (Fig. 11). A coherent behavior is found for the Lagrangian mean velocity and 10-m wind gust in the presence of DIs, showing a monotonic decrease in velocity, accompanied by increased mean 10-m wind gusts around 18 h. The mean increase in wind gusts is concurrent with the development of a potentially unstable atmospheric column (i.e., negative θe,800 − θe,950 values) after 0 h in Fig. 11 and a deepening of the well-mixed PBL, as shown by the dominance of negative PBL pressure anomaly with increasing magnitude until 18 h. During the second phase when DI air mixes within the PBL, the signals weaken on average. As an important reinforcement mechanism for PBL mixing, surface heat fluxes are shown to play a considerable role. The spread among different DIs’ evolution of surface sensible and latent heat fluxes shows a common dramatic effect during DI descent. On average, surface sensible heat flux peaks after 12–18 h after the start of descent, such that the environment of 25% of DI trajectories is accompanied by sensible heat flux less than −100 W m−2 and of 5% by sensible heat flux less than −200 W m−2. After this time, the fluxes decrease to background levels. The response of surface latent heat flux is more pronounced and the peak is reached later, compared to sensible heat flux. In more than 50% of DIs, values of surface latent heat flux less than −200 W m−2 are reached. This intense flux, together with the large variability, with 5% of DIs exceeding −400 W m−2, indicate that in many cases DIs are accompanied with very intense ocean evaporation. For comparison, Winschall et al. (2012) defined a value of surface latent heat flux less than −250 W m−2 as an oceanic evaporation “hot spot,” key for a consequent heavy precipitation event downstream.
Systematic regional differences are expected to arise from persistent regional climatological characteristics and forcing of the PBL as well as land–sea contrasts. The same regions as in section 3c(3) are examined here, as well as the subset of stratospheric DIs, and the difference of the subset mean from the reference mean (solid lines in Fig. 11) is shown in Fig. 12 for the same variables discussed above. Consistent differences can be attributed to an overall enhanced impact of DIs on the PBL, when increased 10-m wind gusts occur together with more negative values of θe,800 − θe,950, PBL pressure anomaly, and surface heat fluxes, or vice versa. Northern ocean storm-track DIs (NA and NP) exhibit enhanced mean peak surface heat fluxes, while MED and WUS DIs show consistently reduced impact with lower gusts, PBL deepening, and surface fluxes. The diverse response is expected for the MED and WUS sets, which interact with various PBL regimes over the Mediterranean and continental areas. In the Southern Hemisphere, SP DIs are associated with greater destabilization and deepening of the PBL but with no mean impact on the heat fluxes beyond the reference mean (not shown). The SWAM DIs exhibit reduced PBL impact in all measures, and SWAF DIs deviate weakly from the reference mean in a mixed response (not shown). It is relevant to note that, by convention, the PBL height is set to cloud base under the occurrence of stratocumulus or cumulus, which corresponds to the climatologically low PBL height in the SWAM and SWAF regions (von Engeln and Teixeira 2013). However, it is yet unclear why the PBL interaction with the incoming DI differs there compared to other regions. Uniquely, ANT DIs are associated with both stronger velocities and 10-m gusts but reduced PBL impact in all other measures. Finally, stratospheric DIs start from relatively lower mean velocities (−7 m s−1 at 0 h) but maintain higher horizontal velocities (+5 m s−1 at 24 h) consistent with the higher initial altitude and slower initial subsidence. Noteworthy is stronger than average 10-m wind gusts below SDIs (+1 m s−1 at 36 h) co-occurring with the highly anomalous PBL pressure anomaly of SDIs, reaching mean values of 15 hPa lower than the reference, indicating a pronounced and delayed PBL response in comparison to the reference DIs.
4. Summary and discussion
Dry intrusions are studied globally from a Lagrangian perspective, based on ERA-Interim data for the years 1979–2014. Such an approach allows a direct quantification of DI occurrence frequency together with the dynamical characteristics of the DI air parcels and their environment, along their descending path. Here the key findings are summarized, in light of the questions (i)–(v) posed in the introduction, followed by a discussion of the caveats of the utilized methods and an outlook for further research.
The Lagrangian criterion to identify DIs requires a pressure increase (i.e., descent) of at least 400 hPa along 48-h air trajectories. This definition selects coherent sets of deeply descending DI trajectories and was used in this study, for the first time, to compile a global Lagrangian DI climatology based on the ERA-Interim data. This definition is comparable with the study in Stohl (2001), where a descent of more than 5000 m from any level at or above 5500 m, during 48 h, was required. An additional criterion that identifies DIs with a stratospheric origin requires the magnitude of PV of DI air parcels to be at least 2 PVU at all times during the 48 h prior to the start of the DI descent.
DI occurrence frequencies are largest in winter, and almost negligible in summer (especially in the Northern Hemisphere), as expected from the sharp seasonality found in Stohl (2001). During NH winter, DIs start within a zonal belt around 40°–45°N, with peaks that exceed 10% frequency in East Asia, the western North Pacific, and over North America. DIs reach the lower troposphere in the NH storm-track regions, as well as the eastern ocean basins, the Mediterranean Sea, North Africa, and the Arabian Sea. The post-DI air is diffluent and thus extends both poleward and equatorward. In the SH winter, DIs start from a zonal belt around 45°S, with increased frequencies in the South Atlantic and south Indian Oceans. When reaching low levels, DIs occur preferentially in the eastern ocean basins (i.e., west of Namibia, west of northern Chile, and west of Australia). Additionally, DIs descend across the Antarctic coast, mainly in the Eastern Hemisphere.
DIs, as defined here, descend at least 400 hPa from the upper to the lower troposphere during 48 h. They start from 300–500 hPa and descend to 700–900 hPa. The descent is nearly isentropic, yet a mean decrease of approximately 3-K for potential temperature is found, possibly a result of variable contributions of longwave radiative cooling and evaporation. Two typical phases of DI descent are suggested. During the initial descent, the moisture content is negligible, and the adiabatic warming of the air parcels lowers the relative humidity further. In the second phase, after 18–24 h, an increase in the moisture content of the air parcel starts as a result of mixing with the underlying PBL. The PV along DIs is mostly tropospheric, and decreases during its descent. The PV anomaly along DIs is negative prior to DI descent, indicating its location in the upper troposphere where the tropopause is anomalously high (such as in a ridge). The variability among DIs increases after their descent, in terms of location, vertical level, and moisture content. The large variability at this stage is due to the diffluence of the post-DI air at the low levels, partly rising back poleward or remaining at low levels, consistent with case studies in which DIs stretch along cold fronts in extratropical cyclones (Browning 1997 and references therein).
Systematic variations of DI characteristics occur in different regions. Notable are mean deviations arising from climatologically different moisture contents in the PBL, leading to, for example, drier than average DIs over North Africa and west of continents.
Stratospheric DIs constitute only a small proportion of 1.2% of the total DI population, having distinctive climatological occurrence, dynamical characteristics, and impact. For example, a hot spot region for SDIs is the western United States, a region known to experience enhanced deep stratosphere–troposphere transport (Škerlak et al. 2014). Although SDIs also peak in winter, higher fractions of SDIs occur in the transition seasons compared to DIs. Having a higher magnitude of PV initially, SDIs advect high-PV values to lower latitudes and altitudes where they are anomalous, modulating the circulation and static stability there. In fact, SDIs are shown to be associated with 50% stronger PBL height anomalies, on average, compared to the overall DI impact [see (v)]. The SDIs may correspond partially to stratosphere–troposphere transport trajectories crossing the 800-hPa level in Škerlak et al. (2014), which account for 5% of the total stratosphere–troposphere transport mass flux across the tropopause.
The delayed descent of SDIs compared to the rest of DIs is qualitatively consistent with the descent route of 33 cases of stratospheric air reaching the surface in southern Japan (Itoh and Narazaki 2016). In these cases, the initial descent within a tropopause fold is moderate, but it becomes stronger during the following coherent descent along sloping isentropes. An analysis of the mechanisms responsible for DI descent is beyond the scope of the current study and is an interesting area of future research.
During the second phase (second day) of the DI descent, the moisture content of the air parcels increases, indicative of mixing with the moist PBL. A case study exemplified the environment in which DIs mix into the PBL. Advection of low-θe air above a layer of relatively high θe, because of the low-level high moisture content, created a potentially unstable layer from the surface up to 800 hPa. Furthermore, the mixing of the descending relatively cold and dry DI air into the PBL and the high wind speeds induced intense upward sensible and latent heat fluxes. The latter may potentially allow the release of the potential instability and enhance vertical mixing and thickening of the PBL. In addition, from the mechanical turbulence perspective, the high wind speeds of the DI trajectories when hitting the top of the PBL (>25 m s−1) created a vertical wind shear favorable for potential downward momentum transfer by turbulent mixing. While all these mechanisms are consistent with an increase in the surface wind gusts, additional mechanisms unrelated to the DIs may contribute to the damaging winds, especially in the vicinity of extratropical cyclones and given the 250-km radius employed for tracing the wind gusts along DIs. From a climatological perspective, more than 75% of DIs impact the PBL in a similar manner and cause its deepening, involving enhanced surface sensible and latent heat fluxes of −70 and −200 W m−2, respectively, on average, and increased 10-m wind gusts.
The DI impact on the PBL deepening has far-reaching implications beyond increased wind gusts. With the arrival of the DI, the temperature and moisture content as well as their vertical gradients within the PBL are strongly modified, and the PBL height typically increases. Thus, over land favorable conditions for wildfires are created by DIs (Mills 2008; Pollina et al. 2013; Schoeffler 2013; Langford et al. 2015). For DIs over the ocean, intense episodes of ocean fluxes, as the one presented here for the Mediterranean Sea, can possibly promote deep-water formation (e.g., Josey 2003), influencing individual events and the climatological state alike. Moreover, the intense evaporation loads newly evaporated moisture into the PBL, which, if transported and lifted, may contribute to precipitation and even flooding downstream (Winschall et al. 2012). Climatological “venting,” namely the transport of air mass (and aerosols) from the PBL to the free troposphere by clouds, is dominated in the extratropics by cyclones and fronts through upward transport in WCBs (Cotton et al. 1995). This study adds the contribution of DIs to such exchange by their descending motion into the PBL, and the interaction with clouds there. In this context, an indirect impact on the PBL stems from potential DI association with cyclogenesis and frontogenesis. Furthermore, the potential emergence of stratocumulus, shallow cumulus, or deep convection by the release of the aforementioned potential instability can play a central role in switching between PBL regimes (Medeiros et al. 2005; von Engeln and Teixeira 2013).
Three caveats arise from the data and method in the present study. DI identification is highly sensitive to the choice of the Lagrangian criterion, such that weakly descending trajectories are not considered as DIs in the present study. Therefore, the current results may be considered to represent strongly descending DIs, during a time scale that represents a typical synoptic-scale system passage. Examination of individual cases with a modified Lagrangian criterion results in similar qualitative dynamical characteristics (e.g., potential temperature and, specific and relative humidity evolutions; not shown). The DI climatological frequencies agree with respect to their geographical location with a previous study that employed a slightly different DI definition (Stohl 2001). Thus, only moderate differences can be expected for modified DI definitions, although the actual quantitative frequencies may differ significantly, as shown by the sensitivity tests. Notably, the Lagrangian DI definition is adapted for the resolution of the ERA-Interim dataset for use in a global climatology. For regional studies, or for diagnosis in different datasets or individual case studies, the criterion may be refined further, as was done for case studies of Mediterranean region DIs represented in mesoscale model simulations (Raveh-Rubin and Wernli 2016).
The second caveat arises from the limitation of the model turbulent diffusion scheme in representing PBL characteristics, including wind gusts and height, under changing atmospheric stability regimes (ECMWF 2007). Therefore the degree of impact on the PBL may differ for different model parameterization choices.
A third caveat of the present climatological results is the validity of the mean behavior for representing individual events. Climatological occurrence frequencies, along with evolution of physical parameters of percentiles of a large DI population, do not cover the full variability and do not necessarily characterize individual DI events in a meaningful way. Moreover, such analysis does not highlight case-to-case variability or the uniqueness of extremes. Nonetheless, the case study presented here proves that the mean temporal behavior is dynamically meaningful and can be observed in single cases.
The present climatology can be compared to other feature-based climatologies based on ERA-Interim data for the same study period (Sprenger et al. 2017). The two main maxima in the climatological frequency of NH winter DIs are generally seen in other features as well, such as cyclones (Wernli and Schwierz 2006), blocking (Croci-Maspoli et al. 2007), WCBs (Madonna et al. 2014), and jet streams (Koch et al. 2006), although with variable locations and frequencies. It can be insightful to explore the relationship among these features and the DI climatology, for better understanding of the mutual interactions. Furthermore, important questions arise for future investigation: How often is the potential instability released below DIs, and how does this affect surface weather extremes? What mechanisms initiate and maintain the DI descent? Do DIs, and particularly SDIs, occur together with tropopause folds? What are the dynamical precursors of DIs, and what is their dynamical impact downstream? How often do DIs interact with extratropical cyclones? Understanding the dynamical mechanisms that control DIs will help to improve the predictability of DI impact by improving their representation in numerical weather prediction models that provide the forecast.
The work is funded by the Swiss National Science Foundation Marie Heim-Vögtlin Programme (PMPDP2_158347/1). MeteoSwiss and ECMWF are acknowledged for providing access to ERA-Interim data. The European Severe Weather Database (ESWD) of the European Severe Storms Laboratory (http://www.essl.org) is acknowledged for allowing online access to reports from 20 to 24 October 2014. I thank Michael Sprenger (ETH) for the ongoing technical help. I am grateful to Heini Wernli (ETH) for inspiring and fruitful discussions, and for feedback on the manuscript. Finally, I thank Keith Browning and two anonymous reviewers for their constructive comments.