This study investigates the abrupt and severe water vapour decline in the stratosphere beginning in the year 2000 (the “millennium water vapour drop”) and other similarly strong stratospheric water vapour reductions by means of various simulations with the state-of-the-art Chemistry-Climate Model (CCM) EMAC (ECHAM/MESSy Atmospheric Chemistry Model). The model simulations differ with respect to the prescribed sea surface temperatures (SSTs) and whether nudging is applied or not. The CCM EMAC is able to most closely reproduce the signature and pattern of the water vapour drop in agreement with those derived from satellite observations if the model is nudged. Model results confirm that this extraordinary water vapour decline is particularly obvious in the tropical lower stratosphere and is related to a large decrease in cold point temperature. The drop signal propagates under dilution to the higher stratosphere and to the poles via the Brewer–Dobson circulation (BDC). We found that the driving forces for this significant decline in water vapour mixing ratios are tropical sea surface temperature (SST) changes due to a coincidence with a preceding strong El Niño–Southern Oscillation event (1997/1998) followed by a strong La Niña event (1999/2000) and supported by the change of the westerly to the easterly phase of the equatorial stratospheric quasi-biennial oscillation (QBO) in 2000. Correct (observed) SSTs are important for triggering the strong decline in water vapour. There are indications that, at least partly, SSTs contribute to the long period of low water vapour values from 2001 to 2006. For this period, the specific dynamical state of the atmosphere (overall atmospheric large-scale wind and temperature distribution) is important as well, as it causes the observed persistent low cold point temperatures. These are induced by a period of increased upwelling, which, however, has no corresponding pronounced signature in SSTs anomalies in the tropics. Our free-running simulations do not capture the drop as observed, because a) the cold point temperature has a low bias and thus the water vapour variability is reduced and b) because they do not simulate the appropriate dynamical state. Large negative water vapour declines are also found in other years and seem to be a feature which can be found after strong combined El Niño/La Niña events if the QBO west phase during La Niña changes to the east phase.
Since the early 1980s balloon-borne stratospheric water vapour measurements (e.g. Hurst et al., 2011) and climate models have predicted a continuous increase in stratospheric water vapour concentrations (Stenke and Grewe, 2005; SPARC CCMVal, 2010; Gettelman et al., 2010). Satellite measurements have not yet observed such an increase (UARS/MLS, UARS HALOE, and SAGE II instruments; see for instance Solomon et al., 2010; Hartmann et al., 2013). However, if we look from the late 1980s/early 1990s to now, we actually find a decreasing trend from merged satellite observations in the lower stratosphere (see Hegglin et al., 2014). The explanation for this has become a large scientific challenge and a lot of discussion persists on whether Boulder balloon observations are representative or if there is an issue in the satellite data merging.
An increase in stratospheric water vapour with time is expected as a net result of global warming predicted for the 21th century by coupled CCMs (Gettelman et al., 2010). However, multi-year data sets show significant fluctuations on different timescales, which makes it difficult to assess robust trends (Hegglin et al., 2014).
In the year 2000, an extraordinary sudden drop of stratospheric water vapour
content was observed (e.g. Randel et al., 2006; Fueglistaler et al., 2005;
Rosenlof and Reid, 2008; Maycock et al., 2014), which again brought into
focus that temperature fluctuations
have a large potential to significantly impact the amount of water vapour in
the stratosphere. The strong and widely noticed water vapour drop in the year
2000 is particularly remarkable due to the fact that it is followed by
a 5-
Since water vapour is the most prominent greenhouse gas, and therefore is an
important contributor to variations and trends in climate, it is necessary to
better understand its large variability. Stratospheric water vapour
variations are connected with temperature changes in the tropical region,
especially with the cold point temperature (Randel et al., 2004; Fueglistaler
et al., 2013). Changes of stratospheric water vapour levels ranging from
interannual to decadal timescales are less well understood, in particular the
contribution of processes involved. Well known and understood is the “tape
recorder” effect (Mote et al., 1996), describing the annual cycle of the
tropical stratospheric water vapour amount in accordance with the seasonally
varying cold point temperature (e.g. Fueglistaler et al., 2005). Moreover,
variations of the tropopause temperatures are clearly related to tropical
upwelling, the equatorial quasi-biennial oscillation of stratospheric zonal
winds (QBO), and the El Niño–Southern Oscillation (ENSO) as, for
example, discussed by Randel et al. (2004). Recent analyses of the observed
stratospheric water vapour record show that many of the variations on
timescales of 1 to several years can be linked to changes in tropical
tropopause temperatures, but some discrepancies still exist (e.g. Schoeberl
et al., 2012; Fueglistaler et al., 2013). Randel and Jensen (2013) state that
the water vapour fluctuations observed by satellite instruments over the last
20-
Here we present results of a set of four simulations with different model set-ups with the state-of-the-art CCM EMAC (ECHAM/MESSy Atmospheric Chemistry model), indicating that it is possible to retrace the observed water vapour fluctuations in the stratosphere (including the millennium drop). In the following section the CCM EMAC is briefly described, the investigated model simulations and the used observational data sets are presented. The millennium water vapour drop as represented in one of the model simulations is compared to observations in Sect. 3. To clarify which part of the millennium drop we refer to, we define two different phases of “the drop”: phase 1 is the short period of the steep decline between the drop onset, i.e. the water vapour maximum and its subsequent minimum. Phase 2 is the period of low values between the minimum and the start of the recovery. In Sect. 4 all model simulations are compared with respect to their ability to represent the millennium drop. Section 5 provides an analysis of other large moisture anomalies in the lower stratosphere and their relation to preceding El Niño/La Niña events. An overall discussion of our findings is given in Sect. 6.
The ECHAM/MESSy Atmospheric Chemistry (EMAC) model is a numerical chemistry
and climate simulation system that includes submodels describing tropospheric
and middle atmosphere processes and their interaction with oceans, land, and
human influences (Jöckel et al., 2010). It uses the second version of the
Modular Earth Submodel System (MESSy2) to link multi-institutional computer
codes. The core atmospheric model is the fifth-generation European Centre
Hamburg general circulation model (ECHAM5, Roeckner et al., 2006). For the present study we analysed
EMAC (ECHAM5 version 5.3.02, MESSy version 2.50) in the T42L90MA-resolution,
i.e. with a spherical truncation of T42 (corresponding to a quadratic
Gaussian grid of approx. 2.8 by 2.8
Overview over the chemistry–climate model simulations used for this analysis.
The multi-year simulations have been performed with the CCM EMAC in the
framework of the ESCiMo project (Earth System Chemistry integrated Modelling,
Jöckel et al., 2016). Within ESCiMo, so-called reference
simulations (RCs) have been carried out, as
defined by the IGAC/SPARC Chemistry-Climate Model Initiative (CCMI) and
described in detail by Eyring and Lamarque (2012). The forcing of the
transient reference simulations in either free-running (RC1; from 1960 to
2011) or nudged mode (RC1SD, RC1SDNT; from 1980 to 2012) are similar
(hindcast simulations). They are taken from observations or empirical data,
including anthropogenic and natural forcing based on changes in trace gases,
solar variability, and volcanic eruptions (see Table 1 for an overview). The
sea surface temperatures (SSTs) and the sea ice concentrations (SICs) are
from observations or reanalysis data (RC1: HadISST, RC1SD and RC1SDNT:
ERA-Interim. In the case of RC1SD,
the model prognostic variables (vorticity, divergence, the logarithm of the
surface pressure, the temperature, and additionally the mean temperature –
wave number zero in spectral space) are nudged by Newtonian relaxation
towards ERA-Interim reanalysis data. RC1SDNT is nudged similarly with the
exception that the mean temperature is
The internal generation of a QBO is a feature of the L90MA set-up of EMAC
(Giorgetta et al., 2002). Therefore, in all simulations a QBO is internally
generated. Nevertheless, the zonal winds near the equator are nudged towards
a zonal mean field with a Gaussian profile in the latitudinal direction and
with a relaxation timescale of 58 days in our simulations to get the correct
phasing of the observed QBO (Jöckel et al., 2016). The nudging is applied
in the altitude range between 10–90 hPa, with full nudging weights (i.e.
1.0) from 20–50 hPa, levelling off to 0.3 (0.2) at the upper (lower) edge
of the nudging region. Full nudging is utilized between
7
For comparison with our EMAC simulation with specified dynamics RC1SD (nudged mode), we use (i) the water vapour data from combined HALOE (Halogen Occultation Experiment) and MLS satellite measurements as described in Randel and Jensen (2013), (ii) a merged data set from seven limb-viewing satellite instruments, which were compiled into a long-term record (Hegglin et al., 2014), and (iii) a combination of satellite observations performed by HALOE and MIPAS (Michelson Interferometer for Passive Atmospheric Sounding) instruments (Russell et al., 1993; Fischer et al., 2008). For more details see the Appendices A1–A3.
With the data set (iii) we performed a novel analysis on several zonal mean characteristics of phase 1 of the millennium drop as a function of altitude and latitude (Appendix A4). Based on derived start and end dates of this phase 1 (i.e. from its maximum to its minimum anomaly), we calculate the length (duration) of phase 1, the start date (months since January 2000), and the size (amplitude). The same analysis was applied to the simulated stratospheric water vapour of RC1SD. This analysis is based solely on running annual means and does not imply an intercomparison of periods before (2000) with periods after (2001) the decline, in contrast to previous studies (Randel et al., 2006; Maycock et al., 2014).
This study is motivated by a chart which is shown in Fig. 1 (the original
figure was published by Randel and Jensen, 2013): the near-global lower
stratospheric water vapour anomalies were derived from multi-year satellite
measurements (1992–2012) from HALOE and Aura/MLS at the 83
Interannual changes of the near-global mean
(60
The HALOE data (see Fig. 1) show a step-like change from an enhanced water
vapour period before the drop (1993–2000) into a phase with reduced water
vapour. The recovery from this phase (our phase 2) with prevailing negative
anomalies starts in 2007.
The RC1SD simulation (nudged, including mean temperature nudging) closely
reproduces the water vapour fluctuations as observed. The timing of relative
minimum and maximum water vapour values is reproduced very well. Evidently
the model underestimates the strength of the interannual fluctuations (only
about 0.3
Figure 2 (also corresponding to Fig. 2 of Randel and Jensen, 2013) moreover shows that the cold point temperature anomalies of RC1SD follow those of the radiosonde data. This can be expected due to the nudging of EMAC towards ERA-Interim data.
Cold point temperature anomalies in
the tropics (20
There are expectations that the water vapour drop in observations
exhibits different characteristics at different latitudes and
altitudes with respect to the start date, the drop size (amplitude),
and the length (duration)
of the anomaly. For example, Urban et al. (2014) showed that in the
tropics the significant reduction of water vapour started in the
altitude range from 16.5 to 18.5
Here, we perform a novel, comprehensive analysis to quantify the characteristics of the water vapour drop (phase 1): (a) the amplitude (drop size), (b) the duration of the drop (drop length), and (c) the onset of the drop (drop date). The results are shown in Fig. 3 as a function of latitude and altitude for both, the combined HALOE/MIPAS data set (Schieferdecker, 2015) and for the RC1SD simulation. The details of the methodology are described in Appendix A4.
Characteristics of the millennium water vapour decline (phase 1) with respect to height (hPa). The analysis approach of the water vapour decline is described in the Appendix A4. Right: satellite observations. Left: RC1SD simulation. Top: drop size (amplitude)(unit: ppmv), middle: drop date (months since January 2000), bottom: drop length (duration) (unit: months). White boxes with crosses indicate that the analysis failed to find a water vapour decrease that fulfilled the criteria listed in the Appendix A4.
The amplitude (drop size) of the drop consistently maximizes in the tropical lower
stratosphere in observations and the RC1SD
simulation (Fig. 3, top). However, the amplitude in the tropics is larger in
the observations. Towards higher latitudes and altitudes up to about
20
A similar pattern can be seen for the start date of the millennium
drop (Fig. 3, middle). Up to 40
The duration of phase 1 is the less consistent quantity (Fig. 3, bottom). Values typically range from 6 months (inherent from the approach) to about 20 months. In the simulation the length is in the order of 9 months in the lowermost tropical stratosphere. The observations exhibit here longer drops related to the larger drop amplitudes.
In conclusion, Fig. 3 nicely reflects that the cold point tropopause anomaly reduces water vapour and that this signal propagates into the upper stratosphere and further to the poles. The propagation is a bit asymmetric due to the different branches of the Brewer–Dobson circulation.
In the last section, we showed that the millennium water vapour drop is reasonably well reproduced by the RC1SD simulation with nudged mean temperature. In the following we investigate whether the other simulations RC1SDNT, RC1, and RC2 (see Table 1) are also capable of simulating the variability of lower stratospheric water vapour and, in particular, the drop in the year 2000.
Near-global mean (60
The RC1SDNT (without mean temperature nudging)-simulated water vapour anomaly
time series amplitude is too small by a factor of about
The magnitude of interannual variability in water vapour in the tropical lower stratosphere is overall far lower in the free-running simulation RC1 (Fig. 4). We will discuss and quantify this at the end of this section. However, a decrease in water vapour around the year 2000 is found also in RC1. The strength of the drop (phase 1) is underestimated by a factor of 2. The minimum period (phase 2) is visible but the minimum is far too high. Compared to RC1SDNT, the free-running RC1 simulation does not simulate the observed atmospheric dynamical state. Yet, this also seems to be important for reproducing the observed water vapour variability, in particular the millennium drop. This is consistent with the results of Garfinkel et al. (2013b), who showed, with model simulations forced by observed SSTs only, that SSTs alone cannot explain the timing and the subsequent recovery of the millennium drop.
Cold point temperature anomalies (K; deseasonalized, 12-month running mean) derived from RC1SD, RC1SDNT, RC1, and RC2 simulations.
The main difference between the RC2 and RC1 simulation is that RC2 uses simulated instead of observed SSTs. Because of this difference, RC2 shows neither the water vapour decline nor the long period with low water vapour values after 2000. Accordingly, no low cold point temperature anomalies are visible in Fig. 5.
The effect of the correct cold point temperature on the saturation water vapour value is also demonstrated for the RC1 simulation (Fig. 6). We took the temperature variability of RC1 as shown in Fig. 5, but used the actual cold point mean temperature as simulated in RC1SD. Thus, we shifted the cold point temperature anomalies. Then we calculated the corresponding saturation moisture over ice for RC1SD (just for comparison), for RC1 (original simulation) and for RC1 shifted (shifted cold point temperature anomaly). The results show that a corrected absolute cold point temperature of RC1 (i.e. RC1 shifted) is expected to improve the representation of phase 2 of the drop.
Saturation water vapour anomaly over ice (deseasonalized, 6-month
running mean) calculated from the respective cold point temperatures
(10
We conclude that (1) without observed SSTs the millennium drop (phase 1) cannot be simulated. There are indications that observed SSTs also contribute to phase 2. (2) The specific atmospheric dynamical state as simulated by RC1SD and RC1SDNT seems to be important for the representation of the millennium drop (phase 2). Note that the period of low cold point temperature after 2001 has no distinct signature in observed tropical SSTs. (3) The correct cold point temperature is necessary to simulate the correct minimum of low water vapour values (phase 2) and the amplitude of the drop (phase 1).
In this section we analyse other sudden stratospheric water vapour declines (phase 1) and try to understand what they have in common with phase 1 of the millennium drop. In particular, we examine the role of preceding El Niño/La Niña events, related upwelling anomalies and the QBO.
Moisture anomalies in ppmv (detrended, deseasonalized,
12-months running mean) derived from RC1SD and RC1 simulations at
80
It is well understood that El Niño/La Niña events have the potential to affect stratospheric variability through SST anomalies (Scaife et al., 2003; Randel et al., 2009; Calvo et al., 2010; Garfinkel et al., 2013a). The La Niña event which started in the autumn of 1998 (after the unusually strong El Niño in 1997/1998) was quite unusual in its duration and intensity. Strong La Niña conditions were present for 2 straight years (both the winters of 1998/1999 and 1999/2000 were strong La Niña). It did not fully decay until the summer of 2001, though it was largely gone by spring of 2000 (when the drop started). In the tropical lower stratosphere, the QBO is the dominant dynamic feature (Rosenlof and Reid, 2008; Dessler et al., 2014) which contributes to the extraordinary temperature fluctuation in the tropical tropopause region. It appears as a reversal of the tropical zonal wind direction with a mean period of about 28 months (ranging from 22 to 34 months) and is a primarily wave-driven stratospheric phenomenon.
We have analysed the time evolution of water vapour anomalies for the RC1SD
and RC1 simulations at 80
Although there are two other large water vapour declines in the RC1SD simulation starting in 1994 and 1996, we neglect this time period, because the eruption of Mt. Pinatubo (1991) had a significant impact on temperature and water vapour in our simulations (Löffler et al., 2016). Likewise, we cannot exclude that the eruption of Mt. Chichon in 1982, although less strong than the eruption of Mt. Pinatubo, had an influence on the results.
The dominant effect of El Niño/La Niña events on the tropical surface temperatures (including land and sea surface temperatures) are clearly visible in Fig. 8a in all simulations. The data derived from the RC1 simulation indicate strong temperature signals related to the El Niño and La Niña episodes (1: 1969/1970, 2: 1973/1974, 3: 1982/1983, 4: 1986/1987, 5: 1997/1998, 6: 2009/2010). The RC1SD simulation only covers El Niño and La Niña events from no 3 to no 6, but the surface temperatures are similar to RC1.
The SSTs for the RC2 simulation were taken from a coupled ocean–atmosphere simulation of the HadGEM model, and the tropical surface temperatures are generally lower than in observations (Fig. 8b). However, the simulated surface temperature represents similar fluctuations (in magnitude) as observed, but originates in different periods of time and often with a longer time duration. As mentioned above (Sect. 2), in all four EMAC simulations the QBO is nudged to zonal mean winds with respect to the amplitude and phase. Therefore the signature of the QBO in the temperature anomaly (Fig. 9b, RC1 as representative for all simulations), propagating downwards to the TTL, is present in all EMAC simulations (for the RC1SD simulation see Fig. S2 in the Supplement). Although the QBO nudging set-up is equal in all simulations presented, the resulting winds are not the same in RC1SD, RC1, and RC2. The QBO nudging does not force a one-by-one representation of the nudged data by the model, the model still develops its own dynamical state. Note that the QBO at roughly 90 hPa is key for the temperature signal affecting water vapour, i.e. at an altitude, where the QBO nudging strength is already reduced and therefore relies on signal propagation. Only for RC1SD (and RC1SDNT), where divergence and vorticity and the logarithm of the surface pressure are also nudged, the wind profiles are close to those of ERA-interim (Fig. S1, Supplement). The RC1 and RC2 simulations, however, show a smaller amplitude and the QBO is less visible at 90 hPa.
Around a strong El Niño event (black vertical lines, Fig. 9b) we find
a positive moisture (Fig. 9a) and temperature anomaly throughout the
troposphere up to about 100
Correlation of anomalies (detrended, deseasonalized) for RC1SD,
RC1 and RC2 at 90 and 70
In Fig. 9, in a narrow layer between 100 and 50
Positive and negative temperature anomalies in the narrow layer are related
by a large part to changes in upwelling (Fig. 10), which directly modifies
the tropopause temperature through lifting of air masses. Additionally,
a positive upwelling anomaly (cooling) is accompanied by a negative ozone
anomaly (cooling; not shown). For this reason upwelling anomaly and ozone
anomaly are anti-correlated with a Pearson's correlation coefficient of about
Temporal evolution of tropical upwelling anomalies in the
tropics (20
In the TTL, positive temperature anomalies always result in positive water
vapour anomalies propagating upwards into the stratosphere
(Fig. 9).
This is independent of a heating and moistening of the tropical troposphere
during El Niños and also occurs under La Niña conditions. Because El
Niño (La Niña) conditions lead to an increase (decrease) in upwelling
(Fig. 9) a cooling (warming) of the TTL region can often be found (El
Niños 1, 2, 4, 5, 6, La Niñas: 2, 4, 5, 6). A moistening can occur in
cases, where the mature phase of an El Niño is over and positive TTL
anomalies appear. This is consistent with the results of Garfinkel et
al. (2013a) who also find a moistening of the stratosphere after La Niña
events. TTL temperature anomalies are an indicator of the regional dynamical
properties (Mote et al., 1996; Randel et al., 2004). The travelling time for
water vapour anomalies in the lower stratosphere calculated from the maximum
correlation between temperature at 100
We find a similar result only for RC1SD, but RC1 and RC2 exhibit the maximum
correlation for lag
Every El Niño event is generally accompanied by a strong positive
upwelling anomaly (Fig. 10) followed by a period with reduced upwelling and
thus positive temperature anomalies in the TTL. Many of these positive
temperature anomalies mark the onset of strong drops in temperature and water
vapour. Note the double maximum in the temperature anomaly after the 1972/73
(no 2) El Niño (Fig. 9b), which is related to the reduced upwelling
(Fig. 10). This confirms that upwelling plays the other important role in
generating temperature anomalies around 100–60
Although the SSTs of the RC1SD and RC1 simulation are similar, the period
with a positive upwelling anomaly after the year 2001, leading to the
observed low tropopause temperatures and low water vapour values in the lower
stratosphere (Randel et al., 2006) is not adequately simulated in the RC1
simulation. We performed an episode analysis for the previously selected four
(RC1SD) and five (RC1) strong El Niño events, followed by a La
Niña event (Fig. 8) and strong declines in water vapour respectively, to
emphasize the conditions that favour these large variations.
Additionally, 4 smaller declines in water vapour of simulation RC1SD, where
no ENSO event preceded, were selected and analysed. The results are presented
in the Supplement, together with the analysis of the RC1 simulation. Here we
focus on the RC1SD simulation.
The onset of the individual temperature declines at 80
Episode analysis for the normalized upwelling anomaly
(black) for (10
All onsets of the temperature drops of RC1SD are associated
with a minimum in the large-scale upwelling anomaly and a west phase
of the QBO (Fig. 11).
Furthermore, the time evolution of the upwelling anomaly is strongly
correlated with SST anomalies during El Niño and La Niña periods (El
Niño region 3.4, Figs. 11 and S13 for RC1) except for the 1982/1983
(no 3) El Niño event, which had its maximum already before the maximum of
surface temperature was reached. However, for the whole simulation period in
RC1SD upwelling anomalies and surface temperature anomalies in the tropics
are only correlated with
We use results of four different simulations performed with the CCM EMAC to analyse the millennium drop in stratospheric water vapour. The simulations differ with respect to the prescribed SSTs and whether nudging is applied or not (see Table 1). We find that a nudged set-up (RC1SD, including nudging of the global mean temperature) performs best compared to observations. A nudged set-up excluding the mean temperature from nudging (RC1SDNT) also reproduces the millennium drop, however, with a smaller amplitude and too-high water vapour values during drop phase 2. This is solely related to the cold point temperature bias, because this is the only difference between RC1SD and RC1SDNT. The free-running RC1 simulation with observed SSTs grossly underestimates the drop, but can capture some elements of it, and the free-running simulation with simulated SSTs (RC2) shows no drop at all. The analysed gradual degradation of the drop signal from RC1SD(NT) over RC1 to RC2 is further augmented by the difference in the QBO signal between the different simulations.
Our first conclusion is that the correct SSTs are important to trigger the drop (i.e. phase 1) and also, at least partly, for the period of low values in phase 2. However, the simulation of some of the characteristics of the millennium drop (phase 2) in RC1 does not give full confidence that the SSTs contribute significantly to the drop phase 2. The drop phase 2 might only be simulated by chance. Here, more realizations with the model set-up of this free-running simulation RC1 are necessary to confirm our suggestions. Second, the specific atmospheric dynamical state as simulated by RC1SD and RC1SDNT contributes to the characteristics of the millennium drop. This is especially true for phase 2, a period of increased upwelling after 2001, which has no corresponding pronounced signature in SSTs anomalies in the tropics. Finally, the correct absolute cold point temperature is necessary to simulate the correct minimum of low water vapour values (phase 2) and thus the amplitude of the drop (phase 1). The millennium drop of stratospheric water vapour of RC1SD in phase 1 is correlated with a strong negative tropical SST fluctuation from La Niña 1999/2000 (after an unusual strong positive tropical SST anomaly from El Niño 1997/1998) with reduced upwelling at the onset of the decline and a positive phase of the QBO changing to the negative phase and stronger upwelling.
We also analysed the time series of water vapour anomalies in order to understand if there are similarities in the processes leading to large amplitudes in water vapour anomaly. In the RC1SD simulation strong drops in temperature and water vapour at the tropopause (phase 1) and above can also be found after other El Niño events (e.g. 1982/1983 and 2009/2010) followed by a La Niña, when conditions comparable to the millennium drop occur: reduced upwelling due to a La Niña event coincides with a west phase of the QBO (warming) followed by an increase in upwelling in connection with the east phase of the QBO (cooling). The reduced upwelling induces a positive ozone anomaly (warming) and vice versa.
In the RC1 simulation we also find large amplitudes in water vapour at the tropopause (phase 1) after ENSO events. However, the QBO anomalies are often not in phase with the temperature or water vapour decline. This affects the timing of declines displayed in Fig. 5, which is slightly different compared to RC1SD. In RC1 the temperature variability seems to be dominated more by upwelling, which is in absolute terms, also larger in RC1 than in RC1SD (Figs. S5–S7). This is at least not in contradiction to Dessler et al. (2014), who found that the BDC provides the largest part to the water vapour variability in the lower stratosphere. Nevertheless, from our nudged simulation RC1SD, which is more in accordance with ERA-Interim, we find the co-occurrence of reduced upwelling and the QBO west phase anomaly changing to east in connection with the large declines (Figs. S3 and S5). In principle it might be that other causes also trigger such declines as well, although we did not find it in our data.
During periods of strong surface forcing of a successive El Niño/La Niña event, the trend in the upwelling anomaly is often (but not always) strongly correlated to the SSTs in the El Niño 3.4 region (Figs. 11 and S13). This connection was already stated by Calvo et al. (2010), and Deckert and Dameris (2008). Thus, large water vapour declines (phase 1) are quite robustly associated with strong El Niño followed by La Niña, but phase 2 is associated more with the dynamical state of the atmosphere and not with the previous ENSO event. The analysis of the detailed characteristics of the dynamical state in phase 2 in RC1SD and RC1 is beyond the scope of the paper. Tropical upwelling, which strongly controls temperature in the tropopause layer, is influenced by the ENSO (see e.g. Calvo et al., 2010). We find that in the free-running simulation RC1, the QBO does not propagate downward far enough into the tropopause region. Furthermore, the relation of tropical SSTs/ENSO to upwelling is stronger in RC1 compared to the nudged simulation.
This raises the question of whether there are processes or forcing which are missing or under-represented in the RC1 and the RC2 simulations. Because SSTs are prescribed from similar observations, RC1SD and RC1 differ mainly with respect to the nudging (of temperature, vorticity, divergence, the logarithm of surface pressure), and the temperatures of land surfaces, which are not prescribed, but can evolve interactively. RC2 uses simulated SSTs, which are colder than those used for RC1. Therefore RC2 can be expected to show different results at least for the time evolution.
So far it is not clear how many of the processes of the obtained cause and effect relationship are insufficiently described or parameterized. More investigations are needed to clarify whether an inaccurate representation of these processes and feedback mechanisms in EMAC is responsible or if it is a matter of model resolution that leads to the disagreement regarding the strength of year-to-year fluctuations of water vapour and temperature. Moreover, a general problem of free-running models is that the cold point is slightly too high (Gettelman et al., 2009) and therefore a little too cold compared to observations, which already leads to a reduced variability in absolute humidity.
Looking at the now 22-year-long global water vapour record constructed on satellite instrument measurements, there is another severe water vapour drop of similar size apparent after 2011 (Urban et al., 2014). Once longer records of global measurements become available in the future, it might turn out that such significant stratospheric water vapour fluctuations occur regularly. Natural changes that affect the stratospheric water vapour content are modified by climate change itself and may impact future climate. This demonstrates that robust climate predictions need realistic fluctuations of SSTs and an adequate representation of the QBO to reproduce the observed stratospheric water vapour fluctuations. Obviously severe changes can have a “memory” effect, impacting climate change on a decadal timescale (Solomon et al., 2010).
The variability of tropopause temperatures is dominated on an interannual period by modulations of the El Niño–Southern Oscillation, the tropical upwelling, and the stratospheric QBO. Variations in ozone amplify the impact of those drivers. In our analysis this relationship seems to be sufficient to show the connection between large water vapour drops, QBO phases, and preceding El Niños. While these processes are understood (Randel et al., 2006, 2009; Fueglistaler and Haynes, 2005; Jones et al., 2011; Urban et al., 2014; Fueglistaler et al., 2013; Randel and Jensen, 2013), the moisture variability can also be influenced by horizontal transport, supersaturated regions, cirrus, and overshooting ice during convective events. All these processes were neglected in our analysis.
From Urban et al. (2014) we know that a period exists where the variability of lower stratospheric water vapour is uncorrelated to the mean zonal temperature (2008–2011). The reason is so far unknown. Here, we omitted to analyse this period, because it is beyond the scope of this paper.
In our analysis, we further neglected any possible changes in the transport of water vapour into the TTL and the presence of supersaturated regions or cirrus clouds in the TTL. Since temperature and water vapour are non-linearly dependent, a monthly mean temperature does not give any information about the actual frequency distribution of saturation values of water vapour. In our simulations, the actual water vapour values are generally lower than the saturation values. It points to a lack of certain processes which are important for the budget of water vapour in the lower stratosphere (for instance convective overshooting). This is a topic of further research.
The simulation results analysed here are archived at the German Climate Computing Center (DKRZ) and are available on request.
HALOE was deployed on UARS (Upper Atmosphere Research Satellite) and
performed measurements from September 1991 to November 2005. The
measurements were based on the solar occultation technique. Absorption
spectra were obtained in specific spectral bands in the wavelength
range between 2.5 and 11
To fill some observational gaps that are inherent of the solar occultation
technique employed by the HALOE instrument we also consider MIPAS limb
observations of thermal emission. Those provided typically more than 1000
individual measurements per day, lasting from June 2002 to April 2012. MIPAS
was carried by Envisat (Environmental Satellite) which used a sun-synchronous
orbit with full latitudinal coverage on a daily basis. The measurements
covered the spectral range between 4.1 and 14.6
The combination is based on monthly zonal mean time series from the individual data sets. In the overlap period a time-independent shift is determined that minimizes the offset between the time series in a root mean square sense. This shift is derived for every altitude level and latitude bin considered and subsequently applied to the MIPAS time series. Applications of the combined HALOE-MIPAS time series can be found in Eichinger et al. (2015) or Schieferdecker et al. (2015).
The basic data for the analysis presented in Fig. 3 are monthly zonal mean
data covering the time period from July 1998 to December 2005. The
HALOE-MIPAS data set is interpolated in time to fill a few gaps. The data are
averaged over a latitude range of 20
The analysis is performed separately for every pressure level and latitude bin using the steps listed below. Figure A1 shows an example.
An example of the millennium drop (phase 1) characteristics analysis
considering the HALOE/MIPAS time series at 100
In a first step we calculate a running average over 1 year. In Fig. A1 the averaged time series is given by the black line. Based on that time series, in the next step we calculate the gradient in water vapour along every data point.
Subsequently we look for periods with sequences of at least six data
points that have a negative gradient allowing one data point
in between to have a positive or zero gradient. Typically we find
several such periods, as seen in the example in Fig. A1. We only
consider those periods that have started within a certain time
interval. For 100
To decide which of the periods represents the millennium drop, we rely on two parameters: one, the absolute change in water vapour and, two, its overall gradient. These parameters are calculated for every period. Subsequently the periods are ranked according to these parameters with the largest absolute value gaining the highest rank. The ranks for a period are summed up and the period with the lowest sum is considered as the period that most likely represents the millennium drop. In the example shown in Fig. A1 the first period is chosen to represent the millennium drop as it exhibits both the largest decrease and the strongest negative gradient among the possible periods.
Pearson's correlation coefficient is determined by the following:
Sabine Brinkop prepared the manuscript and analysed the model results. Martin Dameris was involved in the discussion of the data analysis results and supported the writing of the manuscript. Hella Garny calculated the residual circulation based on the model simulations. Patrick Jöckel performed the simulations with EMAC. Gabriele Stiller provided the MIPAS data and participated in the discussion. Stefan Lossow performed the novel analysis of the water vapour drop characteristics with the HALOE/MIPAS and the model simulation data.
This work has been funded by the Deutsche Forschungsgemeinschaft (DFG) within the research unit Stratospheric Change and its Role for Climate Prediction (SHARP-FOR 1095). The presented investigations were carried out within the water vapour project SHARP-WV. The model simulations were performed within the DKRZ project ESCiMo (Earth System Chemistry integrated Modelling). The merged data set of stratospheric water vapour was kindly provided by M. Hegglin. We would like to thank the Deutsches Klimarechenzentrum (DKRZ) for providing computing time and support and Christoph Kiemle for valuable comments on the manuscript. We thank S. Fueglistaler, M. Schoeberl and an anonymous referee for valuable comments on the manuscript. The article processing charges for this open-access publication were covered by a Research Centre of the Helmholtz Association. Edited by: P. Haynes