The Madden–Julian oscillation (MJO) and convectively coupled equatorial waves are the dominant modes of synoptic-to-subseasonal variability in the tropics. These systems have frequently been examined with proxies for convection such as outgoing longwave radiation (OLR). However, upper-tropospheric water vapor (UTWV) gives a more complete picture of tropical circulations because it is more sensitive to the drying and warming associated with subsidence. Previous studies examined tropical variability using relatively short (3–7 yr) UTWV datasets. Intersatellite calibration of data from the High Resolution Infrared Radiation Sounder (HIRS) has recently produced a homogeneous 32-yr climate data record of UTWV for 200–500 hPa. This study explores the utility of HIRS UTWV for identifying the MJO and equatorial waves.
Spectral analysis shows that the MJO and equatorial waves stand out above the low-frequency background in UTWV, similar to previous findings with OLR. The fraction of variance associated with the MJO and equatorial Rossby waves is actually greater in UTWV than in OLR. Kelvin waves, on the other hand, are overshadowed in UTWV by horizontal advection from extratropical Rossby waves.
For the MJO, UTWV identifies subsidence drying in the subtropics, poleward of the convection. These dry anomalies are associated with the MJO’s subtropical Rossby gyres. MJO events with dry anomalies over the central North Pacific Ocean also amplify the 200-hPa flow pattern over North America 7 days later. These events cannot be identified using equatorial OLR alone, which demonstrates that UTWV is a useful supplement for identifying the MJO, equatorial waves.
Divergent circulations in the tropics play a major role in weather patterns around the globe (Matthews et al. 2004; Roundy et al. 2010; Weare 2010). On intraseasonal (30–60-day) and shorter time scales, these tropical circulations are largely associated with the Madden–Julian oscillation (MJO; Madden and Julian 1994; Zhang 2005) and convectively coupled equatorial waves (Takayabu 1994; Kiladis et al. 2009). The MJO is a planetary-scale convective system that moves eastward around the globe at about 5 m s−1 with a period of 30–60 days.
Equatorial waves are synoptic-scale convective systems whose horizontal structures and dispersive characteristics correspond to shallow water equatorial wave solutions (Matsuno 1966). Equatorial Rossby (ER) waves and Kelvin waves are among the most prominent of these waves. ER waves move westward with wavelengths of 4000–15 000 km and periods of 10–40 days, which somewhat overlaps the scale of the MJO. Kelvin waves are eastward-moving synoptic-scale systems with wavelengths of 2000–4000 km and periods of 3–10 days (Kiladis et al. 2009). Despite their smaller scale, Kelvin waves can produce significant impacts because of their large amplitudes and their ability to circumnavigate the globe (Straub et al. 2006; Schreck and Molinari 2011; MacRitchie and Roundy 2012). Owing to their long lifespans and broad impacts, the MJO, ER waves, and Kelvin waves represent key sources of predictability for the global circulation.
The MJO and equatorial waves are frequently identified with proxies for convection such as outgoing longwave radiation (OLR) and rainfall (Wheeler and Kiladis 1999; Roundy and Frank 2004; Masunaga et al. 2006; Schreck et al. 2011). These proxies are useful over the tropical Indian Ocean and western Pacific Ocean where the MJO and equatorial waves modulate deep convection. The convection weakens in the Western Hemisphere, even if the associated circulations remain strong (Hendon and Salby 1994; Rui and Wang 1990). The MJO, for example, propagates more uniformly around the globe in upper-tropospheric velocity potential χ than in OLR (Knutson and Weickmann 1987), so 200-hPa χ is often used to track the MJO in the Western Hemisphere (Barrett and Leslie 2009).
Schmetz et al. (1994, 1995) found that upper-tropospheric divergence corresponded with enhanced moisture at the same levels, while upper-tropospheric convergence coincided with upper-level drying. Following their example, we will demonstrate the utility of upper-tropospheric water vapor (UTWV) for diagnosing the divergent circulations associated with the MJO and equatorial waves. UTWV has the advantage of being observed directly by satellites, whereas χ must be calculated from numerical models.
UTWV anomalies are generally associated with quasi-isentropic advection in the extratropics (Soden and Fu 1995) and near the tropopause (Sassi et al. 2001). In the tropical troposphere, however, convection becomes the dominant mechanism. Convection releases diabatic heat and advects moisture from the lower troposphere into the upper troposphere (Soden and Fu 1995). The heating and moistening have competing effects on UTWV brightness temperatures, but the net result is usually a decrease (Wu et al. 1993). Meanwhile, cumulus detrainment ejects dry air into the surrounding regions (Sassi et al. 2001; Salby et al. 2003). As convective activity strengthens and spreads, the compensating dry regions become even drier (Sohn and Schmetz 2004). The combination of this drying with adiabatic warming increases UTWV brightness temperatures. The result is that UTWV is somewhat more sensitive to subsidence than to ascent. This sensitivity is opposite to that of OLR. Deep convection produces cold cloud tops that reduce the OLR. Apart from convection, OLR responds to the temperature of Earth’s surface or to the temperature of low clouds. These temperatures are relatively constant within the tropics, making OLR insensitive to the magnitude of subsidence. UTWV is therefore more useful than OLR for assessing both the ascending and descending branches of the circulation.
On interannual time scales, UTWV anomalies are strongly correlated with El Niño–Southern Oscillation, the Pacific decadal oscillation, and the Pacific–North American index (Bates et al. 1996, 2001; Shi and Bates 2011). UTWV is particularly useful for diagnosing the descending branches of the Walker circulation along the equator. Additionally, the UTWV anomalies in the subtropics tend to be out of phase with those near the equator, consistent with compensating vertical motion that responds to the tropical convection (Bates et al. 1996). During La Niña, moist UTWV anomalies are observed in the subtropical eastern North Pacific (Bates et al. 2001). These anomalies are associated with extratropical systems that penetrate farther equatorward within an enhanced westerly duct (Webster and Holton 1982). During El Niño, the westerly duct weakens and dry anomalies develop in this region.
On intraseasonal (30–60-day) time scales, UTWV is strongly modulated by the MJO (Sassi et al. 2001; Waugh 2005; Wong and Dessler 2007; Schwartz et al. 2008). Moist anomalies have been observed in the convective phase and dry anomalies in the suppressed phase. Tian et al. (2006, 2010) examined moisture profiles in the MJO using the Atmospheric Infrared Sounder (AIRS). The strongest equatorial signals associated with the MJO occurred in the lower troposphere up to around 400 hPa. However, they also found compensating dry signals at 648 hPa in the subtropics to the north and south of the convection. Schwartz et al. (2008) found similar signals at 261 hPa using the Aura Microwave Limb Sounder (MLS).
The AIRS and MLS datasets in the studies described above only extend back to 2002 and 2004, respectively. We will use a recently developed climate data record of High Resolution Infrared Radiation Sounder (HIRS) channel 12 brightness temperatures. This channel has a central wavelength near 6.7 μm, which is primarily sensitive to the amount of water vapor integrated from 500 to 200 hPa (Soden and Bretherton 1996). It also varies with the air temperature over this layer, but that sensitivity is only about one-third as large as the water vapor sensitivity (Wu et al. 1993). The HIRS UTWV data extend back to 1979, more than two decades longer than the AIRS and MLS data used by previous studies (Tian et al. 2006, 2010; Schwartz et al. 2008). Furthermore, HIRS UTWV has been calibrated for sensor changes between satellites, producing the most homogeneous dataset possible (Shi and Bates 2011).
This study explores the value of HIRS UTWV over OLR for identifying the MJO and equatorial waves. In particular, the seasonal and regional patterns of the MJO and equatorial wave activity will be explored. By applying the same diagnostics to both OLR and UTWV, we will illustrate the differences between these datasets. We will also test whether UTWV might provide information about the extratropical responses to the MJO that is lacking in OLR. The results will show that UTWV is a useful complement to OLR for examining the MJO, equatorial waves, and their impacts.
2. Data and methods
Daily interpolated OLR data for 1979–2010 were obtained from the National Oceanic and Atmospheric Administration’s Office of Oceanic and Atmospheric Research, Earth System Research Laboratory, Physical Sciences Division (NOAA/OAR/ESRL/PSD). These data were observed with the Advanced Very High Resolution Radiometer (AVHRR) sensor on the NOAA polar-orbiting satellites. They were binned into daily 2.5° latitude–longitude grids, and missing values were filled following Liebmann and Smith (1996). For any given day, only one satellite was used, generally the one with an equatorial crossing time closest to 1430 LST. Because of OLR’s strong diurnal cycle, day and night passes were binned and interpolated separately. The resulting grids were then averaged together to produce the daily values.
The HIRS channel 12 UTWV data for 1979–2010 were obtained from NOAA’s Comprehensive Large Array-Data Stewardship System (CLASS). As part of NOAA’s Climate Data Record (CDR) program, the UTWV data were calibrated for intersatellite differences following Shi and Bates (2011). The calibrated data were then binned into daily 2.5° grids to match the OLR. The intersatellite calibration allowed us to combine data from multiple satellites so that less than 1% of the time series at any given grid point was missing. These missing values were interpolated using a similar scheme (Liebmann and Smith 1996). For UTWV, however, the daytime and nighttime passes were not treated separately because of its smaller diurnal cycle (Lindfors et al. 2011). This smaller diurnal cycle, combined with the intersatellite calibration, also allowed us to achieve maximum coverage by binning data from all available satellites, regardless of their equatorial crossing times.
The UTWV brightness temperatures are also sensitive to clouds in the upper troposphere and stratosphere, so previous studies generally discarded or corrected data in regions with deep clouds (Wu et al. 1993; Soden and Bretherton 1996; Bates et al. 1996). Deep clouds are abundant in the tropics, so discarding these data would have left too many gaps for the current analysis. Since we are treating water vapor as a passive tracer, rather than being concerned with its radiative effects, these clouds should not significantly influence our results. Only clouds with overshooting tops were removed from our analysis, as these could cause spurious signals in our limb-correction scheme. Several algorithms exist for converting the brightness temperatures to upper-tropospheric humidity for comparison with models and in situ observations (Soden and Bretherton 1993, 1996; Jackson and Bates 2001). For simplicity, however, we use the observed brightness temperatures as a proxy for water vapor as in Wu et al. (1993).
Winds, geopotential height, and velocity potential at 200 hPa were obtained from the Climate Forecast System Reanalysis (CFSR; Saha et al. 2010), developed by the National Centers for Environmental Prediction (NCEP). These data are available for 1979–2009 from NOAA’s National Operational Model Archive and Distribution System (NOMADS), which is maintained at NOAA’s National Climatic Data Center (NCDC).
b. Variance analysis
The fraction of variance associated with the MJO and each equatorial wave type was calculated as follows. First, the 32-yr mean and the first three harmonics of the seasonal cycle were subtracted from the original data. Second, these daily anomalies were divided by each grid point’s standard deviation. The resulting standardized anomalies were then filtered for the wavenumbers and frequencies associated with each wave type. The variance of these filtered values represents the fraction of the total variance that is associated with each wave type.
c. Compositing procedure
Composite means were used to examine the spatial structure and evolution of the MJO and Kelvin waves. First, the time series of filtered UTWV was obtained at a base point. To reduce seasonal variability, only data from November to April (boreal winter) were considered. Next, we identified the dates of all maxima or minima in the remaining time series that exceeded two standard deviations (±2σ). Lagged composites for a given field were then constructed by simply averaging the data relative to these dates, such that the maximum amplitude occurs on day 0. Previous studies produced similar composites using linear regression (e.g., Wheeler et al. 2000; Roundy and Frank 2004), but that method forces the composites to be symmetric between positive and negative phases. By simply averaging the data relative to extrema of the same sign, we allowed the composites to be asymmetric in time as in Roundy (2008).
The statistical significance of the composite anomalies was evaluated using a bootstrap test similar to Matthews and Kiladis (1999). In this test, null cases were identified using dates with the same months and days as the original composite, but from the other 31 years. For example, if a maximum filtered value occurred on 11 February 1980, then its corresponding null cases would be every 11 February from 1979 and from 1981 to 2010. Null composites were calculated by randomly selecting the same number of dates from the null cases as had been in the original composite. This procedure was repeated to generate 1000 null composites. Using a two-tailed test, the anomaly at a given grid point in the original composite was considered 95% significant if it was either greater than or less than the same grid point in 975 of the 1000 null composites. This statistical test is advantageous because it accounts for the sample size and the possibility that variance may change with the time of year.
a. Comparing OLR and UTWV
Figure 1 shows the mean (top), total variance (middle), and correlation with 200-hPa χ (bottom) for OLR (left) and UTWV (right). The first three harmonics of the annual cycle were removed before calculating the variance and correlation plots. The mean and variance are for 1979–2010, while the correlations (bottom) are for 1979–2009 because of the availability of the CFSR data.
Consistent with previous studies (Soden and Fu 1995; Bates et al. 1996; Sassi et al. 2001), OLR and UTWV share similar spatial patterns in their means (Fig. 1, top). Both fields are depressed in the monsoon and ITCZ regions with the lowest values occurring near the Maritime Continent. These lower values are consistent with deep convection, which produces cold cloud tops (lower OLR) and advects moisture into the upper troposphere (lower UTWV brightness temperatures). Meanwhile, the subtropical regions feature elevated values for both fields, which are consistent with the broad subsidence that occurs in these regions. The high OLR values can be attributed to the warm surface temperatures and the absence of clouds, while the higher UTWV brightness temperatures are consistent with subsidence drying and warming in these regions.
The mean (Fig. 1a) and variance (Fig. 1c) for OLR roughly mirror each other, with larger variance collocated with lower mean values of OLR. Deep convection is highly variable by nature, and the large variance in the ITCZ and monsoon regions can be attributed to the sharp contrast between cold cloud tops during convective periods and the warm surface during clear periods.
UTWV shows a very different pattern of variance (Fig. 1d). The variance is relatively low in the convective regions. Increased variance instead occurs poleward of the convective regions. Two factors may contribute to this poleward shift: First, UTWV is more sensitive to large-scale advection in the subtropics than to local convection in the tropics (Sherwood 1996; Pierrehumbert 1998; Dessler and Sherwood 2000; Chuang et al. 2010). Second, the UTWV data are more sensitive to subsidence than OLR (Wu et al. 1993; Sohn and Schmetz 2004), so the shift could be related to variations in compensating subsidence on the poleward sides of tropical convection. Another important distinction between the variances of OLR and UTWV is their longitudinal distribution. To a large degree, the OLR variance is confined to the Indian Ocean and western Pacific. Meanwhile the UTWV variance is more evenly distributed at all longitudes. These results are consistent with those of Bates et al. (1996), who compared longitude–time Hovmöller diagrams of OLR and UTWV.
The bottom panels of Fig. 1 show the correlations between each variable and the coincident 200-hPa χ. Velocity potential is well correlated with OLR. Deep convection corresponds with upper-level divergence, and subsidence is associated with upper-level convergence. Consistent with Schmetz et al. (1994, 1995), Fig. 1f shows that UTWV is also correlated with χ. In fact, the average correlation with χ between 15°S and 15°N is 0.31 for OLR and 0.34 for UTWV. The larger correlations for UTWV are particularly evident in the Western Hemisphere. These results suggest that UTWV, like χ, can be useful for identifying vertical motion in regions with less convective activity.
Figure 2 shows a two-dimensional histogram or density plot comparing the values observed between the UTWV and OLR. The shading indicates the fraction of all the data from 30°S to 30°N at which any given pair of UTWV and OLR values coincide. Both datasets are negatively skewed, as evidenced by the larger fractions in the upper-right portion of the figure. The skew is smaller for UTWV. The relationship between UTWV brightness temperature and OLR is roughly linear for the lower values (UTWV < 240 K, OLR < 250 W m−2) associated with deep convection, consistent with previous findings (e.g., Soden and Fu 1995; Sassi et al. 2001). The relationship changes for higher (nonconvective) values. The distribution becomes more horizontal with a broader range of UTWV values possible for any given OLR value. As suggested by previous studies (Wu et al. 1993; Bates et al. 1996), UTWV contains more information than OLR in regions of subsidence.
b. Spectral analysis
Figure 3 shows the zonal wavenumber–frequency spectra for OLR (top) and UTWV (bottom). Following the methodology of Wheeler and Kiladis (1999), these spectra have been divided by an estimate of the red background to highlight significant peaks in power. The OLR spectra (Fig. 3, top) are virtually identical to Fig. 6 from Wheeler and Kiladis (1999) and are included here for comparison. Signals associated with tropical cyclones were removed following Schreck et al. (2011, 2012) in order to focus on the MJO and equatorial waves.
As with OLR (Fig. 3, top), many of the UTWV signals (Fig. 3, bottom) fall between the dispersion curves for shallow water equatorial waves with equivalent depths between 8 and 90 m (black lines). The UTWV signals are particularly strong for Kelvin waves and ER waves, both of which are symmetric about the equator. Meanwhile, the antisymmetric signals such as n = 0 mixed Rossby–gravity (MRG) and eastward inertio-gravity (EIG) waves are weaker in the UTWV data. The parabolic curves near the top of each panel in Fig. 3 are the shallow water dispersion curves for n = 1 inertio-gravity waves, sometimes called 2-day waves (Haertel and Kiladis 2004). The signals associated with these waves are weak in Fig. 3, in part because the once-daily data used here cannot fully resolve these higher-frequency waves.
Tropical depression (TD)-type disturbances are not associated with any shallow water solution, but their signals appear as westward propagation with periods of 2–10 days (Fig. 3). These systems are commonly known as easterly waves, but we use the term “TD-type disturbances” to distinguish them from signals that align with linear equatorial wave solutions (Takayabu and Nitta 1993; Wheeler and Kiladis 1999). TD-type disturbances do not cross the equator, so their signals appear in both the antisymmetric (Fig. 3, left) and symmetric (Fig. 3, right) spectra. Their signals are somewhat stronger above the red background in UTWV (Fig. 3, bottom) than in OLR (Fig. 3, top). The MJO is another signal that is not associated with any shallow water dispersion curve. Like TD-type disturbances, the MJO appears prominently in both the symmetric and antisymmetric spectra for OLR and UTWV alike.
c. The MJO
1) Geographic distribution
Figure 4 shows the fraction of the total OLR and UTWV variance (Fig. 1c,d) that falls within the MJO filter band during November–April (boreal winter, Fig. 4a,b) and May–October (boreal summer, Fig. 4c,d). These maps are calculated as described in section 2b, and they illustrate the geographical distribution of MJO activity in each dataset. Consistent with previous studies (Zhang and Dong 2004; Roundy and Frank 2004), the MJO’s OLR signal (Fig. 4a) is concentrated in the equatorial Indian Ocean and the northern coast of Australia during boreal winter. These signals also appear in UTWV (Fig. 4b) along with some notable differences. In general, the fraction of variance associated with the MJO is larger in UTWV compared to OLR. The MJO also accounts for more than 12% of the variance over a band that nearly encircles the globe around 20°N. This signal is absent in OLR.
During boreal winter, tropical intraseasonal convection generally propagates eastward, but this propagation turns northeastward during boreal summer (Yasunari 1979; Krishnamurti and Subrahmanyam 1982; Knutson and Weickmann 1987; Wang and Rui 1990). The latter is sometimes referred to as the boreal summer intraseasonal oscillation (BSISO; Kikuchi et al. 2012). These signals are weaker in Figs. 4c and 4d than their boreal winter counterparts, in part because we are only filtering for the eastward-propagating component. The OLR signals are still concentrated on the Indian Ocean where the BSISO modulates the monsoon activity. A secondary signal appears over the eastern North Pacific and plays an important role in modulating tropical cyclone activity in that region and also the Gulf of Mexico (Molinari et al. 1997; Maloney and Hartmann 2000a,b; Aiyyer and Molinari 2008; Kossin et al. 2010). The UTWV signals are more diffuse, and the eastern North Pacific signal is lacking. Interestingly, the MJO band accounts for at least 9% of the UTWV variance over broad portions of the subtropical Southern Hemisphere during both seasons.
2) Tropical and subtropical signals
To identify the subtropical MJO-filtered UTWV signals, Fig. 5 shows a composite based on events with a +2σ (dry) MJO-filtered UTWV anomaly at 15°N, 175°W. This base point was selected because it is the peak of the boreal winter signal in the central Pacific, which is unique to UTWV. The compositing method was described in section 2c. Figure 5d shows the time of the maximum MJO-filtered UTWV anomaly (lag 0). The negative OLR anomalies (cool; blue shading) over the Maritime Continent and equatorial western Pacific indicate the convective envelope of the MJO. Moist UTWV anomalies (blue contours) coincide with the deep convection. Dry anomalies (red contours) to the north and south, including the large values near the base point, are consistent with compensating subsidence surrounding the MJO’s deep convection. These moist and dry anomalies move coherently eastward from the Indian Ocean at −14 days (Fig. 5b) to the central Pacific at +7 days (Fig. 5e). This pattern is comparable to the composites of 215-hPa water vapor from MLS by Schwartz et al. (2008).
The MJO’s convection dissipates around +14 days (Fig. 5f), when the convectively suppressed phase of the MJO is near western Indonesia. The subtropical dry UTWV anomalies have given way to moist anomalies. At +21 days (Fig. 5g), the MJO’s convection begins to redevelop over the Indian Ocean in association with the western end of the moist UTWV anomalies. These UTWV anomalies are less prominent at +28 days, but the unfiltered OLR shows the MJO’s convective envelope over the eastern Indian Ocean. This redevelopment of the MJO following a circumnavigation of its upper-level signals is consistent with previous studies (Rui and Wang 1990; Hendon and Salby 1994; Kikuchi and Takayabu 2003).
The Wheeler–Hendon Real-time Multivariate MJO (RMM) index (Wheeler and Hendon 2004) is one of the most frequently used metrics for identifying the MJO. It uses OLR, 850-hPa zonal wind, and 200-hPa zonal wind, each averaged over 15°S–15°N, to determine the phase and amplitude of the MJO. Figure 6a shows the RMM phase for the day 0 dates that were composited in Fig. 5d. Figure 6b is similar, but it selects the dates based on events with −2σ MJO-filtered OLR anomalies at 0°, 160°E, which corresponds with the core of OLR anomalies in Fig. 5d. Remarkably, nearly half (13 of 27) of the dates in Fig. 6a fall within RMM phases 6 and 7, even though the RMM is essentially independent of the subtropical UTWV used to select those dates. These phases are consistent with the enhanced convection observed over the western Pacific in Fig. 5d. Not surprisingly, the OLR-based dates in Fig. 6b are even more tightly clustered in phases 6 and 7. We will now explore how the extratropical response changes between the dates selected based on subtropical UTWV as in Fig. 6a and those based on equatorial OLR as in Fig. 6b.
3) Extratropical response
Figures 4 and 5 illustrate that UTWV contains subtropical MJO signals that are absent from OLR. These signals could be useful for identifying tropical–extratropical interactions within the MJO. To explore this possibility, Fig. 7 shows lagged composites of UTWV (shading), 200-hPa geopotential heights (contours), and 200-hPa winds (vectors). Figure 7 is composited relative to +2σ (dry) MJO-filtered UTWV anomalies at 15°N, 175°W, as in Fig. 5. As before, a moist anomaly propagates along the equator, while a dry anomaly is observed in the subtropics. Figure 7 illustrates that the dry anomaly is associated with an enhancement of the subtropical ridge. Previous studies observed this pattern with MJO-enhanced convection over the western Pacific (Liebmann and Hartmann 1984; Kiladis and Weickmann 1992; Matthews et al. 2004; Weare 2010). The westerly anomalies on the poleward side of the ridge correspond with a strengthening of the Pacific jet. As the ridge moves eastward, an amplified meridional wind pattern develops over North America (Fig. 8c). This pattern features a ridge over the western part of the continent and a trough to the east, consistent with the composites for RMM phase 8 in Zhou et al. (2012). Figure 6a showed that day 0 was generally in phases 6 or 7, so the RMM may be expected to be in phase 8 on day +7.
Previous studies have used equatorial indices to identify the extratropical impacts of the MJO. It is natural to wonder if additional information is gained from the subtropical UTWV signal used here. As a test, Fig. 8 shows a complementary set of composites that are based on −2σ (convective) OLR anomalies at 0°, 160°E. This is the same base point that was used in Fig. 6b. UTWV is still shaded as in Fig. 7, and these anomalies are similar between Figs. 7 and 8. The primary difference is that the equatorial OLR index (Fig. 8) identifies a stronger convective signal. The extratropical response is distinctly different between the two composites. The enhanced subtropical ridge is weaker in Fig. 8, as is the response over North America. These results suggest that the subtropical UTWV signals may distinguish between MJO events that have extratropical responses and those that do not.
d. Equatorial waves
Figure 9 identifies the fraction of variance in the ER band for OLR and UTWV. The differences are subtler than they were for the MJO (Fig. 4). Both datasets show twin bands of activity along 15° latitude in both hemispheres. As in past studies (Roundy and Frank 2004; Huang and Huang 2011), the OLR ER wave signals (Figs. 9a,c) are stronger in the summer hemisphere, nearly disappearing in the Southern Hemisphere during boreal summer. Meanwhile, the UTWV signals are more symmetric about the equator and less seasonal.
Figure 10 shows the fraction of variance that falls in the Kelvin band in each dataset. Equatorial Kelvin waves share similar propagation characteristics with extratropical Rossby waves, and the two frequently interact with one another (Straub and Kiladis 2003). The largest fraction of OLR variance associated with the Kelvin band actually occurs along the poleward edges of the map (Figs. 10a,c). The weaker signals along the equator are associated with equatorial Kelvin waves. A lull in the filtered variance around 15°–20° latitude in both hemispheres provides some separation between the signals associated with equatorial Kelvin waves and those of extratropical Rossby waves. In UTWV (Figs. 10b,d), however, no such separation is possible. The Kelvin band appears to be dominated by the extratropical signals, particularly in the winter hemisphere.
Figure 11 shows composites based on −2σ (moist) Kelvin-filtered UTWV anomalies at 137.5°W. This longitude is chosen to coincide with the maximum Kelvin-filtered variance near the equator during November–April. Only lag 0 is shown, but the composites are based on Kelvin-filtered anomalies at three different latitudes. The composite for 20°N (Fig. 11a) is distinctly extratropical. The moist anomalies have a southwest–northeast tilt that coincides with 200-hPa southwesterlies. Both the wind and geopotential height anomalies suggest that these southwesterlies are part of a broader extratropical wave train. The composite for 10°N (Fig. 11b) is similar, although the wave train is somewhat weaker. The equatorial composite (Fig. 11c), however, resembles composite Kelvin waves from previous studies (Wheeler et al. 2000; Straub and Kiladis 2002; Kiladis et al. 2009). The dominant 200-hPa winds are easterlies and westerlies diverging out of the moist anomaly. Horizontal advection by extratropical Rossby waves is the primary source of UTWV variability in the subtropics (Pierrehumbert 1998), which may explain why these equatorial Kelvin waves are overshadowed in UTWV at higher latitudes.
4. Summary and discussion
This study demonstrates the utility of the HIRS UTWV climate data record for identifying the MJO and equatorial waves. UTWV is complementary to proxies for convection like OLR and rainfall, but UTWV contains more information in regions of large-scale subsidence. In the absence of convection, OLR values are driven by the temperature of low clouds or the surface. Similarly, rainfall data are constant (zero) in nonraining areas. UTWV, on the other hand, is sensitive to both the drying and warming that are associated with subsidence. This distinction allows UTWV to provide information about the strengths of both the ascending and descending branches of the divergent circulation.
The MJO and equatorial waves stand out above the red background variability in UTWV (Fig. 3), similar to other convective proxies. The fraction of variance associated with the MJO is actually larger in UTWV than in OLR (Fig. 4). For ER waves, the signals are more symmetric about the equator in UTWV than in OLR (Fig. 9). Kelvin waves, on the other hand, are eclipsed in UTWV by horizontal advection from extratropical Rossby waves (Fig. 11). Kelvin waves often interact with these extratropical systems (Straub and Kiladis 2003), but OLR is less sensitive to horizontal advection and provides a clearer separation between these disturbances (Fig. 10).
In regions with weaker convection, such as the Western Hemisphere, 200-hPa χ is often used to track tropical disturbances. Like UTWV, χ produces anomalies that are associated with both ascent and subsidence. This similarity probably explains why χ is better correlated with UTWV than with OLR, particularly in the Western Hemisphere (Fig. 1, bottom panels). Velocity potential cannot be observed directly, however, and must be obtained from numerical model analyses. While these models are increasingly reliable, they are still prone to errors, particularly in divergent circulations (Newman et al. 2000; Kinter et al. 2004). UTWV brightness temperatures have the advantage of being observed directly by satellites.
Previous studies have examined the MJO using other satellite-derived estimates of UTWV (Sassi et al. 2001; Waugh 2005; Tian et al. 2006, 2010; Wong and Dessler 2007; Schwartz et al. 2008), but these had shorter periods of record. The climate data record used herein has the distinct advantage of providing a 32-yr homogeneous dataset (Shi and Bates 2011). At the same time, HIRS UTWV is a 200–500-hPa layer average, so additional insight may be gained by comparing the results with those from shorter datasets that had greater vertical resolution. For example, Fig. 5 shows that equatorial convection is collocated with moist UTWV anomalies and that subtropical dry anomalies surround the convection. Schwartz et al. (2008) found a similar pattern at 215 hPa using Aura MLS data, and Tian et al. (2006) demonstrated that these signals extend downward to at least 648 hPa in AIRS water vapor. The patterns change above and below these levels, however. In the lower stratosphere, dry anomalies not only surround the convective region but are also observed directly above it (Schwartz et al. 2008). Meanwhile, in the lower troposphere Tian et al. (2006, 2010) observed large moisture anomalies in the equatorial eastern Pacific associated with the MJO. These anomalies were attributed to blocking by the Andes, which explains why they do not extend high enough to be observed in the HIRS UTWV.
The subtropical UTWV anomalies are particularly interesting because OLR lacks these signals. Schwartz et al. (2008) attributed the subtropical UTWV anomalies to Rossby gyres forced by the MJO’s convection (Hendon and Salby 1994). Figure 7 supports this interpretation by showing an enhanced subtropical ridge over the central North Pacific coincident with the dry UTWV anomalies. The ridge occurs northeast of the enhanced convection, consistent with the MJO’s typical horizontal structure (Liebmann and Hartmann 1984; Rui and Wang 1990; Kiladis and Weickmann 1992; Matthews et al. 2004; Weare 2010). Composites based on subtropical UTWV (Fig. 7) show these gyres prominently, while they are less apparent in composites based on equatorial OLR (Fig. 8). In this way, UTWV provides a method for identifying the MJO’s Rossby gyres independent of model or reanalysis data. The subtropical UTWV anomalies and Rossby gyres are also associated with a more amplified flow pattern over North America. Rui and Wang (1990) found that stronger MJO events are more likely to propagate northeastward to North America, so the subtropical UTWV anomalies may be indicative of these stronger events.
The subtropical signals identified by UTWV could play an important role in the initiation and maintenance of the MJO. Ray et al. (2009) demonstrated the ability of a mesoscale tropical channel model to simulate the MJO—as long as the lateral boundaries at 21°S and 21°N were forced with reanalysis data. When the extratropical and subtropical forcing was limited by holding these boundaries constant, the MJO failed to develop. Figures 5a and 5f show moist UTWV anomalies extending from the Arabian Peninsula to India that precede the MJO’s convective development. These anomalies can be traced backward across the Saharan Desert to the Atlantic Ocean (Figs. 5d–f). Much of this pathway lies poleward of the 21°N boundary used by Ray et al. (2009), so the UTWV anomalies may be associated with a critical forcing for the MJO. Future research should explore this possibility with the goal of using UTWV to extend our forecast capabilities for the MJO.
We are grateful to three anonymous reviewers for their insightful comments. This research benefited from discussions with Ken Knapp. We are also grateful to Darren Jackson for his assistance in obtaining, processing, and interpreting the UTWV data. Ethan Shepherd and Lisa Krolak provided additional support. Schreck received support for this research from NOAA’s Climate Data Record (CDR) Program through the Cooperative Institute for Climate and Satellites (CICS-NC).