At the end of December 1999, two extremely severe storms only one day apart affected western Europe and caused considerable damage. A variable derived from satellite observations, the so-called temperature of the lower stratosphere (TLS), is used in this study for detecting and tracking the upper-level components of these storms. TLS is computed from a regression over five Television and Infrared Observation Satellite (TIROS-N) Operational Vertical Sounder (TOVS, aboard NOAA satellites) channels, with coefficients calculated from a climatological dataset [thermodynamical initial-guess retrieval (TIGR)], and provides information on the temperature near the tropopause. The objective of this paper is to assess the ability of TLS, in situations such as these two exceptional storms, to track and depict upper-tropospheric precursors of surface lows. After a brief synoptic description of the meteorological situation, TLS fields as well as the Action de Recherche Petite Échelle Grand Échelle (ARPEGE) model fields (mean sea level pressure, temperature, wind velocity, and geopotential height of the dynamical tropopause) are discussed concurrently for the period 23–27 December. Although the upper-level thermal fields are consistent overall, differences appear, especially during the incipient stage of the second storm. The forecast, which was poor in the operational context, is modified when a configuration close to the TLS one is adopted. Qualitative comparisons of TLS with Microwave Sounding Unit (MSU) channel-3 limb-corrected brightness temperatures and with the water vapor imagery are also shown. One advantage of TLS over these two other fields is the earlier detection of the upper-level precursor of the second storm. Because TLS computation is easy and fast, the suitability of TLS as a possible forecasting aid over midoceanic regions is promoted.
At the end of December 1999, two severe storms one day apart caused considerable damage over western Europe. Both storms were poorly forecast by models, with the skill of the forecast varying as a function of the forecast range, suggesting a problem in the initial conditions. The storms in fact formed over the Atlantic Ocean basin where few upper-level observations are available apart from satellites and aircraft.
Among the satellite sounders, the Television and Infrared Observation Satellite (TIROS-N) Operational Vertical Sounder (TOVS), an ensemble of instruments operating both in the infrared (High Resolution Infrared Radiation Sounder, HIRS-2) and in the microwave (Microwave Sounding Unit, MSU; Smith et al. 1979), onboard National Oceanic and Atmospheric Administration (NOAA) satellites, provides near-global coverage 2 times per day. Nevertheless, at the time of the storms, there was only one operational TOVS instrument, the second operational NOAA satellite carrying the ATOVS (Advanced TOVS) instrument. The TOVS data are mainly used for numerical prediction models in operational meteorological centers (Andersson et al. 1994). Another meteorological application is the diagnosis of upper-level features of cyclones. For example, Velden (1992) used MSU observations to describe the tropopause in the case of extratropical cyclones. However, Fourrié et al. (2000) have shown the advantage of simultaneously using HIRS-2 and MSU observations for such a description. They proposed a more precise determination of the temperature near the tropopause, hereinafter called the temperature of the lower stratosphere (TLS), based on a combination of two HIRS-2 and three MSU channels. They showed, for three cyclogenesis events during the Fronts and Atlantic Storm-Track Experiment (FASTEX; Joly et al. 1999), that TLS can be used to infer tropopause-level features such as ridges, troughs, and “tropopause breaks” on the cyclonic shear side of the jet streak along the jet stream. This study was followed by that of Chaboureau et al. (2001), which showed that systems characterized by high-level cloud patterns accompanied by rain can be discriminated with respect to the TLS field. In addition, these authors showed that the well-developed cases can be grouped into three families depending on the orientation of the warm TLS features: zonal, anticyclonic, and cyclonic.
Several studies of the December 1999 storms, relying on model simulations and conventional observations, have been performed (Pearce et al. 2001; Ulbrich et al. 2001; Wernli et al. 2002; Leutbecher et al. 2002). However there has been no attempt to assess the usefulness of satellite observations for the detection and the monitoring of the relevant dynamic upper-tropospheric features associated with these two storms. This paper aims to demonstrate the value of the satellite-derived TLS variable in the context of (i) providing novel lower-stratospheric thermal and dynamical information unavailable from other satellite sensors over a data-sparse oceanic region and (ii) assessing numerical weather prediction model initialization. The paper is arranged as follows: in the second section, the data used for the investigation, which are provided by the NOAA satellites and by the French operational model Action de Recherche Petite Échelle Grande Échelle (ARPEGE) analyses (Courtier et al. 1991), are described. The main features of the synoptic situation are presented in section 3. In section 4, the ability of TLS to describe the upper-level meteorological situation and to detect and track the upper-tropospheric precursors of these storms is discussed. Section 5 discusses the limitations and the advantages of the TLS variable for the monitoring and the forecasting of the two storms. Results are summarized in the last section.
2. Satellite data and model fields
a. Temperature of the lower stratosphere
The Improved Initialization Inversion (3I) method (Chédin et al. 1985; Scott et al. 1999) aims at interpreting NOAA/TOVS observations in terms of atmospheric thermodynamic variables (vertical temperature and moisture profiles, cloud field description, surface characteristics, etc.). As a first step, 3I looks for an optimal initial solution by comparing the observations to all the situations archived within the thermodynamical initial-guess retrieval (TIGR) dataset (Chédin et al. 1985; Achard 1991; Chevallier et al. 1998). This vast archive consists of 1761 situations sampling the whole world and spanning the whole year. The situations are described by their thermodynamic (temperature, moisture, etc.) and radiative properties (associated TOVS brightness temperatures) computed from the Automatized Atmospheric Absorption Atlas (4A; Scott and Chédin 1981). The 1761 situations are clustered into five airmass classes: tropical, temperate, cold temperate and summer polar, Northern Hemisphere very cold polar, and winter polar, hereinafter called tropical, mid-lat1, mid-lat2, and polar1 and polar2, respectively. The TLS variable, introduced by Chédin et al. (1985), is used to constrain the search for the best initial solution in the 3I method. It is obtained in each HIRS-2 spot through a combination of five TOVS channels, which are (i) the most sensitive to the temperature around the tropopause (Fig. 1) and (ii) not contaminated by clouds (MSU channel 2 is sensitive to extended precipitating clouds but it is corrected for the cloud liquid water):
where the regression coefficients ai were derived from TIGR for all observing conditions (viewing angle, airmass type, and surface pressure, to take into account elevated surfaces) and TBH(M)j is the brightness temperature of the HIRS-2 (MSU) channel j. MSU data used in Eq. (1) result from the interpolation of MSU limb-corrected brightness temperature fields within each HIRS-2 spot.
The determination of the regression coefficients ai was performed as follows: for each TIGR temperature profile, a TLS value has been determined through a weighted average of temperatures within a pressure range given in Table 1 for each airmass class. Once a TLS value has been ascribed to each TIGR profile, and since the brightness temperature values associated with each profile are also in TIGR, ai coefficients are determined using Eq. (1) for all possible observation conditions (minimization of a least squares criterion). This methodology provides TLS values with no bias and a standard deviation less than 1.9 K (Chédin et al. 1985).
For computing TLS in the HIRS-2 spots (30 × 30 km2), an offline calculation is performed just after the interpolation of MSU data into HIRS-2 spots and the airmass classification. TLS fields have already been used, for example by Claud et al. (1995) and then by Fourrié et al. (2000), for describing temperature patterns near the tropopause during cyclogenesis events. Because TLS represents a thermal field, the maximum value areas are associated with a low-level tropopause (such as troughs) and the minimum value regions are linked to ridges. In particular, for FASTEX intensive observing periods (IOP), we observed that the gradients displayed by the TLS field were associated with strong gradients exhibited by the 300-hPa geopotential height fields along the cyclonic shear side of the upper-level jet streak (Fourrié et al. 2000). These gradients are called tropopause breaks and correspond to the step between the high tropopause level south of the upper-level jet streak and the low tropopause level north of this jet streak. A tropopause fold can be observed in these regions and it is a deformation of the tropopause that leads to an intrusion of a thin wedge of stratospheric air through upper-level fronts into the mid- and lower troposphere. During the development of the tropopause fold, the tropopause discontinuity shifts from the cyclonic to the anticyclonic side of the jet stream and crosses under the jet stream axis (Reed 1955). A low tropopause level is generally characterized by a stratospheric air intrusion (Reed 1955) associated with high potential vorticity (PV) values. These tropopause breaks therefore represent upper-level PV anomalies, which are the upper-level components of the baroclinic interaction scheme. This scheme is involved in the deepening of surface lows larger than 9–10 hPa per 24 h (Ayrault and Joly 2000) and consists of the interaction of two PV anomalies, one located at the tropopause level and one near the surface. Quantitative comparisons between TLS and temperature fields deduced from the ARPEGE model for different potential vorticity unit (1 PVU = 1 × 10–6 K kg–1 m2 s–1) surfaces have shown that TLS is a good indicator of the temperature between the 4- and 8-PVU surfaces (Fourrié et al. 2000). These comparisons were performed for several IOPs of the FASTEX experiment.
b. Microwave Sounding Unit channel-3 brightness temperature
As the maximum of the MSU channel-3 (MSU3) weighting function is located at about 300 hPa (Fig. 1), MSU3 limb-corrected brightness temperature imagery is commonly used to infer characteristics of tropopause-level thermal anomalies (e.g., Velden 1992). MSU3 analyses allow the tracking of midlevel synoptic-scale baroclinic waves (Hirschberg et al. 1997). Moreover, Hirschberg and Fritsch (1991a, b) have shown that stratospheric temperature anomalies are dynamically linked to PV anomalies and are also related to the hydrostatic structure and development of baroclinic waves and cyclones. A drawback of MSU3 is that its weighting function is fairly broad and straddles the midlatitude tropopause. In addition, the horizontal resolution varies from 109 km at nadir to 323 km at the edge of the swath.
c. Water vapor imagery
Uccellini et al. (1985) demonstrated how satellite water vapor (WV) imagery is useful in identifying tropopause breaks characterized by relatively dry stratospheric air intrusions. Rodgers et al. (1985) referred to dark bands formed by strong subsidence as dry intrusions. These features correspond to the vertical motions associated with tropopause folds. The WV imagery is thus useful in identifying upper-level PV anomalies. Forecasters use this information to depict upper-level PV anomalies associated with black cores (dry areas) in the WV imagery for comparison with upper-level model fields in the case of a discrepancy between successive forecast fields. In this study, we have used the Meteosat WV imagery available over the eastern part of the Atlantic Ocean basin.
d. ARPEGE fields
ARPEGE analysis fields resulting from a four-dimensional variational (4DVAR) data assimilation scheme have been used for the meteorological description of the storms. These analyses were performed in a nonoperational context after the events occurred. The assimilated observations came from conventional data such as radiosondes, TOVS NOAA National Environmental Satellite, Data, and Information Service (NESDIS) preprocessed brightness temperatures, wind derived from geostationary satellite imagery, measurements from aircraft, and surface data. The ARPEGE model grid is a stretched grid that corresponds to a grid spacing of about 50 km near Newfoundland and 20 km over France. The fields considered in this study for the upper levels are the temperature on the 2-PVU surface (T-2PVU), the wind velocity, and the geopotential height of the same surface. In addition, the vorticity at 850 hPa and the mean sea level pressure have been used to follow the meteorological features at the surface or close to it. The low-level vorticity has been found to be a better field than the mean sea level pressure field for identifying synoptic systems, because it allows systems to be identified much earlier in their life cycle.
As mentioned earlier, the TLS method represents the temperature of an atmospheric layer above the dynamical tropopause level defined as the 2-PVU surface away from the deep Tropics by Hoskins and Berrisford (1988). However in order to keep with the spirit of published works on upper-troposphere/lower-stratosphere dynamics, the synoptic description makes use of the temperature and the wind velocity on this surface. Moreover, the geopotential height of the dynamical tropopause is useful in characterizing the tropopause height in the midlatitudes and it will be studied.
3. Synoptic description of the storms
Before analyzing TLS fields, the major meteorological features of the period are described using ARPEGE fields. A full synoptic description of the storms can be found in Wernli et al. (2002), Ulbrich et al. (2001), and Pearce et al. (2001). Figures 2 and 3 show mean sea level pressure fields and the 2-PVU surface geopotential height field, respectively, for the period ranging from 1800 UTC 23 December to 1800 UTC 27 December every 12 h. The wind velocity on the 2-PVU surface, when larger than 70 m s–1, is also presented in Fig. 3. Three periods were considered: the period before the formation of the lows, the 24–26 December period during which the first storm occurred, and the 26–27 December period during which the second storm occurred.
a. 1800 UTC 23 December–0600 UTC 24 December
At 1800 UTC 23 December, there was a strong upper-level jet stream across the North Atlantic Ocean from 90° to 15°W, with maximal wind speeds larger than 90 m s–1 (Fig. 3a). The shape of the jet stream is characteristic of a zonal weather regime, which provides suitable conditions for cyclogenesis events over western Europe. The 2-PVU surface geopotential height field displayed a strong gradient [400 dam (5° latitude)–1] indicating a tropopause break along the cyclonic shear side of this jet streak (Fig. 3a). The two storms affected France and western Europe because they were driven by the upper-level jet streak which was unusually far south and had unusual strength. Indeed, the upper-level jet was stronger than usual; from 1800 UTC 25 December to 0600 UTC 26 December, it crossed the Atlantic Ocean from 70°W to almost 10°W, which means that it extended more to the south and to the east (partly into Europe) than usual (Fig. 1 from Ayrault et al. 1995). At 0600 UTC 24 December, the eastern tip of the upper-level jet streak (48°N, 18°W; Fig. 3b) is above an open wave at the surface (Fig. 2b) that is southwest of Ireland (20°W), which may also have contributed to the jet stream extension but which will not be discussed further.
b. The “Lothar” storm
Figure 4 exhibits the evolution of the surface pressure minimum during the life cycle of the so-called “Lothar” storm. Three periods are delimited: an incipient phase (0600–1800 UTC 24 December), a transition stage (1800 UTC 24 December–1800 UTC 25 December), and a deepening phase (after 1800 UTC 25 December). The surface cyclogenesis initiated at 0600 UTC 24 December offshore of Florida (37°N, 64°W; Fig. 2b). It took place on the southern side of an intense baroclinic zone that formed across the North Atlantic (Wernli et al. 2002; Ulbrich et al. 2001). Connected to this strong baroclinicity, the upper-level jet streak was characterized by wind speeds as large as 90 m s–1 (Fig. 3b). The low (Fig. 3b) formed in a region of upper-level divergence associated with the thermally indirect right-rear quadrant of the upper-level jet streak [according to the “four-quadrant conceptual model” of vertical velocity around a straight jet streak; Bjerknes (1951); Riehl et al. (1952)]. A 10-hPa pressure deepening was observed between 0600 and 1800 UTC on 24 December (Fig. 4).
At 0600 UTC 25 December, the surface low (with a minimum value of 998 hPa) intensified while moving northeastward to about 43°N, 40°W (Fig. 2c). At upper levels, the jet was extending from the American coast to western Europe (Fig. 3c). At 1800 UTC 25 December, the surface low became an open wave located at 48°N, 20°W (Fig. 2d): it deepened slowly (−5 hPa during the last 24 h; Fig. 5), followed by an explosive deepening phase [using the criterion of Sanders and Gyakum (1980)] beginning at 1800 UTC 25 December [−15 hPa (12 h)–1]. The upper-level jet streak remained quasi-stationary over several days (from 0600 UTC 24 December to 1800 UTC 26 December), implying that the surface low was moving through the jet streak, with the incipient stage occurring in the entrance region. The rapid deepening occured as the low translated beneath the exit region (Fig. 3e), a well-known feature that induces rapid cyclone growth (Uccellini 1990; Baehr et al. 1999). The low hit northern France with its maximum deepening early on 26 December (Fig. 2e; Ulbrich et al. 2001), with the minimum surface pressure value (969.5 hPa) being reached at 0600 UTC (Fig. 4).
c. The “Martin” storm
This storm, reaching France on 27 December in the evening, corresponds to a classical cyclogenesis event involving the interaction between an identified upper-air trough and a weak thermal wave along the baroclinic region [type B of Petersen and Smebye (1971)]. It initiated at 0600 UTC 25 December offshore of North America at 38°N, 60°W (Fig. 2c) on the southern edge of the 70 m s–1 region (shaded) and at the entrance of the jet streak (Fig. 3c). At 1800 UTC, there was, in the 2-PVU surface geopotential height field (Fig. 3d), a minimum located on the cyclonic shear side of the upper-level jet. The surface low (Fig. 2d) moved northeastward along the trough's flank, which itself was slowly progressing southeastward from Newfoundland (Ulbrich et al. 2001). The jet stream now spread out between 70°W and 10°E (Fig. 3d). The surface low began its deepening (Fig. 2e) at 42°N, 45°W at 0600 UTC 26 December (−2 hPa between 0600 and 1800 UTC 26 December; Fig. 5) and moved 10° eastward (Figs. 2e and 2f) between 0600 and 1800 UTC 26 December. In contrast to the Lothar storm, which did not change the intense upper-level jet streak, the so-called Martin low modified the large-scale airflow (Ulbrich et al. 2001), leading to a decrease of the jet stream velocity. Also, the jet was situated more to the east than previously (Fig. 3f). Its western tip moved from 40°N, 50°W to 45°N, 37°W between 1800 UTC 26 December and 0600 UTC 27 December. During the same period, the surface low deepened by 13 hPa (Fig. 5) and reached a minimum pressure value of 964 hPa (Fig. 2h) at 1800 UTC 27 December when hitting France (deepening of −27 hPa in 12 h from 0600 to 1800 UTC 27 December; Fig. 5).
4. TLS field analysis
TLS fields (displayed in Fig. 6) are now examined for these two severe storms. For comparison, in Fig. 7, the T-2PVU fields are shown with the vorticity at 850 hPa superimposed in order to depict the surface low location.
a. Establishment of the upper-level jet stream across the Atlantic Ocean
At 1800 UTC 23 December, TLS fields (Fig. 6a) displayed an elongated warm zone from the Great Lakes to about 53°N, 40°W. A similar structure was found in the field of T-2PVU (Fig. 7a). The TLS gradient located in the southern area of this structure between 45°N, 65°W and 49°N, 37°W (Fig 6a, 226 K falling to 220 K) corresponded to the tropopause break displayed by the geopotential height field (Fig. 3a) and located along and slightly beneath the cyclonic shear side of the upper-level jet stream [as already observed for FASTEX cases by Fourrié et al. (2000)]. This property, as shown in the following, can be used to display the establishment of the jet stream across the Atlantic Ocean before the occurrence of Lothar.
At 0600 UTC 24 December (Fig. 6b), the TLS warm zone was more elongated than previously and reached 20°W. Between 1800 UTC 23 December and 0600 UTC 24 December, the eastern tip of the upper-level jet streak (see isotachs above 80 m s–1; Figs. 3a and 3b) moved 15° eastward from 35° to 20°W. Two maxima were located at 49°N, 40°W (226 K) and south of the Great Lakes at 39°N, 86°W (230 K). This warm TLS band is thinner south of Iceland at 55°N, 20°W and south of Newfoundland (48°N, 55°W). These features are in general agreement with the shape of the T-2PVU fields. Both fields display the tropopause break associated with the upper-level jet (Fig. 3b).
In conclusion, TLS, which describes temperature patterns around the tropopause, depicts the tropopause break along the upper-level jet streak and therefore shows the establishment of the strong upper-level jet along the Atlantic Ocean.
b. The Lothar storm
At 0600 UTC 25 December, a curved structure with high values of TLS (229 K; Fig. 6c) was present along the east coast of the United States. At its northern tip (49°N, 49°W), there was a zone with a maximum value of 222 K. Note that the maximum value of the 850-hPa vorticity core located at 45°N, 38°W (Fig. 7c) has increased by 4 times while the surface pressure has deepened by only 1 hPa (Fig. 3c) during the 12-h period.
At 1800 UTC 25 December, a TLS maximum (221 K; Fig. 6d) was located at 52°N between 30° and 40°W and corresponded to the northern edge of a jet streak (Fig. 3d). A second TLS maximum was present at 50°N, 44°W and in agreement with a maximum on the T-2PVU field (Fig. 7d). The latter does not play a direct role in the cyclogenesis since it is situated too far from the surface low [according to the scale relationship between the vertical and the horizontal length of Hoskins et al. (1985)].
At 0600 UTC 26 December, areas where TOVS data were missing were present over France. Unfortunately the intensification phase of the Lothar storm was short, and because of this lack of data, no detailed TLS analysis was possible during this stage.
In conclusion, TLS variations allows us to follow the tropopause break associated with the jet streak but do not show any upper-level PV anomaly. These results are well corroborated by the studies of Wernli et al. (2002) and Ulbrich et al. (2001), who show that between 0000 UTC 24 December and 1800 UTC 25 December there was no prominent upper-level disturbance along the jet axis nor at the location of the evolving cyclone. Only during the intensification stage, which started at 0600 UTC 26 December, do Wernli et al. (2002) show a narrow filament of stratospheric PV from European Centre for Medium-Range Weather Forecasts (ECMWF) analysis fields, which cannot be observed in TLS fields because of the lack of data.
c. The Martin storm
At 0600 UTC 25 December (Fig. 6c), a warm TLS region was located along the North American coast and centered at about 40°N, 70°W; it corresponded to the upper-level trough on the T-2PVU field (Fig. 7c) and on the geopotential height gradient on the 2-PVU surface (Fig. 3c). The spatial temperature gradient was steeper for the T-2PVU field than for the TLS field [8 K (10° latitude)–1 vs 5 K (10° latitude)–1]. This is in agreement with the difference between TLS and T-2PVU: a 12–14-K difference between both fields was observed for the maxima. The 222-K TLS zone corresponded indeed to a 236-K T-2PVU core. Let us recall that TLS represents the temperature of an atmospheric layer above the 2-PVU surface, which explains the discrepancy in terms of amplitude between TLS and T-2PVU (Fourrié et al. 2000).
At 1800 UTC 25 December, a 229-K TLS maximum (Fig. 6d) can be observed southeast of the American coast at 38°N, 70°W. It depicted the front of the upper-level trough (Fig. 3d) at the northern entrance of the jet stream.
At 0600 UTC 26 December (Fig. 6e), two TLS maxima were present east of the United States: the first one, M1 (characterized by a maximum value of 227 K), is located offshore of Nova Scotia (43°N, 58°W) and the second one, M2 (225 K), is present at 38°N, 65°W. This fact has been mentioned by Ulbrich et al. (2001), who showed that, at that time, the southerly part of the trough weakened, while farther north, a small-scale upper-air depression moved eastward from Newfoundland. The T-2PVU fields (Fig. 7e) exhibit only the southern maximum M2 (248 K) at 32°N, 70°W. However, only the M1 upper-level maximum can interact with the surface low situated at 42°N, 45°W, which began to deepen (Fig. 2e). The southern one was indeed too far away [according to the scale relationship between the vertical and the horizontal length of Hoskins et al. (1985)]. Moreover, there is another discrepancy between the TLS and T-2PVU fields: at about 52°N, 26°W, TLS exhibited a cold region, suggesting a ridge, while T-2PVU isolated a warm core, corresponding to a trough. These differences between the two fields are discussed in more detail in the next section, where the verification analysis and the improvement of the forecast of the Martin storm are reviewed.
At 1800 UTC 26 December (Fig. 6f), the TLS maximum, M1, was located south of Greenland (49°N, 42°W) with a value of 228 K. This maximum corresponded to the upper-level component of the Martin storm. The second maximum, M2 was located at 39°N, 56°W and reached 224 K on the TLS field (Fig. 6f). The corresponding T-2PVU field (Fig. 7f) displays a large maximum value area with two embedded maxima (244 K) fitting in with those of the TLS field: the warm T-2PVU core at 40°N, 55°W corresponding to the TLS M2 maximum and the other one at 47°N, 46°W corresponding to M1. Moreover, the jet stream velocity had decreased, its western tip moved eastward (Fig. 3f) while the surface low moved through the jet streak.
At 0600 UTC 27 December, the M1 maximum exhibited by the TLS field had moved to 48°N, 22°W and its amplitude had decreased (222 K; Fig. 6g). On the T-2PVU field, M1 was located at 50°N, 25°W and its maximum value was 244 K (Fig. 7g). The second maximum, M2, had also decreased in amplitude (222 K on TLS field at 43°N, 47°W; 236 K at 47°N, 40°W on T-2PVU field). Note the strong 850-hPa vorticity increase in Fig. 7g that indicates a baroclinic interaction between the upper-level anomaly and the low-level vorticity core.
At 1800 UTC 27 December (Fig. 6h), M1 hits Brittany and was located at 47°N, 5°W just above the surface low (Fig. 2h). However, TLS values were below 204 K in the middle of the TLS maximum temperature core, while at that time the corresponding T-2PVU values were more than 240 K. These unexpected very low TLS values are due to the airmass misclassification of the TLS maximum core and will be discussed in the next section.
A synthesis is given in Fig. 8 that shows the positions of TLS and T-2PVU maxima as well as the surface low center every 12 h for the period between 1800 UTC 25 December and 1800 UTC 27 December. Good agreement between the TLS and T-2PVU maximum trajectories is obtained after 1800 UTC 26 December. Prior to this, the maxima observed in T-2PVU were located too far away from the surface low to interact with it. This is not the case for TLS, which isolated the upper-level precursor as early as 1800 UTC 25 December at 38°N, 70°W.
In the previous section, disagreements between the TLS and T-2PVU fields have been pointed out to describe tropopause-level dynamical features. For this reason, we compare here TLS fields with MSU3 brightness temperatures and WV imagery, which represent classical tools for this type of characterization. This way, both the limitations and the advantages of TLS will be discussed.
a. Comparison with MSU channel-3 brightness temperature and WV imagery
Figure 9 displays the MSU3 brightness temperature field and the corresponding WV imagery provided by Meteosat at 1800 UTC 25 December (Figs. 9a and 9b), 0600 UTC 26 December (Figs. 9c and 9d), and 1800 UTC 27 December (Figs. 9e and 9f), corresponding to the three different stages considered in the previous section.
At 1800 UTC 25 December, the WV imagery (Fig. 9b) exhibited a dry area corresponding to the tropopause break along the cyclonic shear side of the upper-level jet stream between 15° and 20°W but there was no dry intrusion above the Lothar low (48°N, 20°W). The MSU3 field displays a meridional temperature gradient (Fig. 9a) along 50°N over the cyclonic shear side of the upper-level jet (Fig. 3d) but this gradient is large scale and is difficult to associate with the rather thin tropopause break structure. As no direct baroclinic interaction occured during the transition stage, none of these upper-level fields are very relevant for the tracking of upper-level features.
At 0600 UTC 26 December, the MSU3 field (Fig. 9c), as well as the TLS field (Fig. 6e), are unavailable over France. The WV imagery (Fig. 9d) exhibits a dark zone over northern France at 48°N, 0°W that corresponds to a dry intrusion during the explosive stage of the first storm. For the second storm, TLS (Fig. 6e) exhibits two maxima in the trough offshore of North America, highlighting the northern one. In contrast to the TLS field, which shows a warm closed anomaly, MSU3 (Fig. 9c) displayed only a thermal gradient between 38°N, 62°W and 52°N, 45°W. MSU3 is thus less relevant than TLS in describing the upper-level features associated with the Martin low.
At 1800 UTC 27 December (Fig. 6h), TLS displays a warm zone close to Brittany at 46°N, 5°W (within this warm area, TLS displays some low values below 204 K, a point that will be discussed in more detail in the next section). The MSU3 field (Fig. 9e) isolated a warm anomaly at 47°N, 5°W, the center of which reaches 226 K. The WV imagery (Fig. 9f) displayed a dark zone at 47°N, 0° corresponding to the dry intrusion located over the Martin low.
In conclusion, these three sources of data give complementary information on the meteorological situation. TLS fields display a warm closed structure earlier than MSU3 brightness temperature fields for the detection of the upper-level precursor of the Martin storm. In the French operational context, Meteosat WV imagery is only available east of 40°W. TLS is thus more relevant for the description of the features during the initiation stage over the western Atlantic Ocean basin, while the WV imagery is more useful for the intensification phase description.
b. Influence of the airmass classification
At 1800 UTC 27 December (Fig. 6h), there was a warm zone (corresponding to M1) with values of TLS that are very low to the west of France at 47°N, 3°W. These low values correspond to a tropical airmass type (not shown), when a midlatitude airmass type is expected. The tropical airmass TLS pressure range was quite different (45–86 hPa; see Table 1) from those of the other airmass types and this leads to a discontinuity in the TLS field, in particular, between the tropical and the mid-lat1 types. This classification can be explained by the fold of the stratospheric layer between 4- and 6-PVU surfaces as can be seen in Fig. 10, which exhibits the cross section between 50°N, 10°W and 40°N, 0° of the PV surfaces with respect to pressure. There is a low tropopause surface (below 400 hPa) located at about 46°N but the atmospheric layer above, between the 4- and 6-PVU surfaces, was folded and this layer was situated between 300 and 400 hPa and higher at 150 hPa. Above, the 8-PVU surface is roughly located between 150 and 100 hPa along the cross section. The fold of the 4–6-PVU layer leads to a tropical airmass-type classification because the level of the uppermost layer (the first layer “seen” by the satellite from space) is typical of a tropical airmass type and not of a midlatitude airmass type (at 30°N the 4–6-PVU surfaces lie above 150 hPa). Fourrié et al. (2000) have already found that TLS fields are more representative of the thermal structure in the area of 4–8 PVU for a cold-air cyclogenesis case from FASTEX. Figure 10 suggests that TLS fields could correspond more to the layer included between the 4- and 6-PVU surfaces for midlatitude cyclogenesis than to a layer including the 2-PVU surface.
An artifically prescribed mid-lat1 type instead of a tropical type has been introduced (Fig. 11) with the other airmass types remaining unchanged. In this configuration, at 1800 UTC 27 December (Fig. 11b), the cold values in the middle of the warm TLS anomaly disappear as do the horizontal discontinuities between the tropical and mid-lat1 airmass types; as for MSU3, there is a warm zone.
Additionally, the results at 0600 UTC 26 December (when TLS isolated two maxima along the American coast and emphasized the northern one) remain unchanged: the northern maximum (Fig. 11a) remained warmer than the southern one and the disagreement between the TLS and T-2PVU fields persists. TLS exhibits only one maximum instead of two for T-2PVU. As expected, the horizontal discontinuities in the TLS field between 35° and 45°N have disappeared.
c. Verification of the analysis and improvement of the forecast
During the first phase of the second storm, the TLS field emphasizes the northern maximum, which is indeed the upper-level precursor of this low (Ulbrich et al. 2001). On the contrary, the ARPEGE model highlights the southern feature of the trough until 1200 UTC 26 December (not shown). Moreover, TLS exhibits a ridge while T-2PVU exhibits a trough in front of the system. The features revealed by T-2PVU have led to a rather poor operational forecast based on the model initialized at at 1200 UTC 26 December. We have verified that the forecast of the storm was sensitive to the location and the intensity of the upper-level structure located offshore of North America (G. Hello and P. Arbogast 2001, personal communication). Therefore, the PV inversion method (Chaigne and Arbogast 2000), which allows for the modification of the PV field of the initial state, was used. The PV field was modified as follows in order to be qualitatively in better agreement with the TLS field: a potential vorticity core was added at the place where TLS exhibits the M1 maximum and the ridge in front of the trough was reinforced. A new 30-h forecast was run with this modified initial state at 1200 UTC 26 December. The fields of the new forecast are in better agreement with the position and intensity of the cyclone with the 4DVAR reanalysis at 1800 UTC 27 December than with the operational forecast fields. For instance, the pressure minimum of the Martin storm was closer to the one given by the reanalysis (P. Arbogast 2001, personal communication).
TLS is currently not used operationally (2001) and is independent of the forecasting operational suite due to its use of TIGR for the calculation of the regression coefficients. If implemented, it could be used by forecasters in case of disagreement between the model analyses and observations. Tools have indeed been developed at Météo-France in order to allow forecasters to modify the operational forecasts by changing the initial conditions, when they are not confident in the forecast. Currently, the decision to change the initial conditions is driven by the field of WV imagery. TLS could, therefore, be employed in the same way by providing additional information about the tropopause-level meteorological situation given that this study (and also other previous studies, e.g., Fourrié et al. 2000) suggests the potential of such a variable through comparisons with model analysis in a critical way.
6. Summary and conclusions
Two violent storms only one day apart caused serious damage over western Europe at the end of December 1999. The temperature of the lower stratosphere (TLS) approach, derived from TOVS observations, can detect the upper-level precursors of the storms as a warm anomaly in such exceptional circumstances. The ability of TLS has been assessed for these violent conditions, extending the work of Fourrié et al. (2000) conducted in the context of the FASTEX experiment, where most cases were “close” to normal (Joly et al. 1999). TLS is obtained by combining two infrared and three microwave TOVS channels, which are weighted by a set of regression coefficients calculated once and for all from the TIGR dataset. TLS, which describes the temperature of a layer around the tropopause level, has been compared with ARPEGE model fields at upper levels: the temperature, the wind velocity, and the geopotential height of the 2-PVU surface considered as the dynamical tropopause. The position of the strong upper-level jet stream across the Atlantic Ocean on 22–23 December, unusually farther south and east, is exceptional. The TLS approach exhibits the position of the jet as a thin band of high values of TLS clearly showing the tropopause break along this jet. During some stages of the first storm, the unfortunate lack of data over the region of interest limits the usefulness of the TLS approach for describing upper-level meteorological features.
However, in the case of the second storm, which hit the French coast during the evening of 27 December. TLS permits isolation of an upper-level precursor in a trough and the tracking of it. Interestingly, TLS and the temperature on the 2-PVU surface disagree during the initiation stage of this storm. Numerical simulations, forced by the upper-level anomaly to be situated at the location given by TLS (G. Hello and P. Arbogast 2001, personal communication), show a modified forecast both in location and intensity, emphasizing the value of such a variable.
TLS fields have also been compared with MSU channel-3 fields and WV imagery. The complementarity of TLS and WV imagery, with an advantage in TLS over the WV imagery during the incipient stage of the second low, has been shown. TLS can isolate the upper-level feature earlier because it shows the feature while the dark intrusion in the WV imagery, which is associated with vertical motion and therefore results from the upper-level feature, appears later.
TLS, which is obtained independently from the operational forecast suite by the use of the TIGR dataset, could therefore provide additional information for the detection of forecast failures in the same way as does WV imagery derived from geostationary satellites. Moreover, it could be used as an indication of a need for potential vorticity modification within the context of forecast improvement. Last, the MSU sounder has been replaced or will be replaced (depending on the satellite) by the Advanced Microwave Sounding Unit (AMSU) (e.g., Li et al. 2000) on board the NOAA satellites with a larger number of channels. Because of the increased horizontal and vertical resolution of the AMSU in comparison with MSU, an “advanced TLS” should prove even more useful with this sounder.
The authors thank J.-N. Thépaut who ran the 4DVAR reanalysis of the studied period, G. Hello, P. Arbogast, and A. Joly for the review of a previous version of the manuscript. The authors are particularly grateful to J. Donnadille for useful comments and discussions on the revised manuscript. In addition, they also thank the three anonymous reviewers for their helpful criticisms and the improvement of the language. A part of the explanations of the synoptic situation is taken from the following World Wide Web address: http://www.cnrm.meteo.fr/fastex/recyf_temp/index.html. FASTEX has been contractually supported by the Programme Atmosphère et Océan à Moyenne Echelle (PATOM) of the Institut National des Sciences de l'Univers and by the European Commission (Contract ENV4-CT96-0322).
Corresponding author address: Nadia Fourrié, CNRM/GMAP, Météo-France, 42 Av. G. Coriolis, 31057 Toulouse Cedex, France. Email: Nadia.Fourrie@cnrm.meteo.fr