In this study, the authors seek large-scale signals that may distinguish MJO from non-MJO convective events before they start over the Indian Ocean. Three such signals were found. Low-level easterly anomalies extend from the surface to the midtroposphere and move from the western to eastern Indian Ocean. Surface pressure anomalies exhibit a zonal structure of wavenumber 1 with an equatorial low-pressure surge penetrating eastward from Africa through the Indian Ocean and reaching the Maritime Continent. Negative temperature anomalies in the middle to upper troposphere start over the Indian Ocean and move eastward. All of them emerge 20 days before convective initiation of the MJO and move eastward at speeds close to that of the MJO without any direct connection to MJO convection. They are not obviously related to the extratropics in any discernible way or any preceding MJO events. They are absent in non-MJO convective events. These signals provide useful information for forecasting MJO initiation over the Indian Ocean. They can be signatures of a dry dynamics mode of the MJO, if it exists.
Convective initiation is perhaps the most understudied and least understood aspect of the Madden–Julian oscillation (MJO; Madden and Julian 1971, 1972). MJO prediction by some models suffers from particularly low skill at MJO initiation (Seo et al. 2009; Kim et al. 2010). As summarized in Zhang (2005), possible mechanisms for MJO initiation include continuation and amplification of a preexisting event (Knutson et al. 1986; Knutson and Weickmann 1987), extratropical influences (Lau and Peng 1987; Frederiksen and Frederiksen 1997; Hsu et al. 1990; Lin et al. 2007; Ray et al. 2009; Ray and Zhang 2010), local energy discharge–recharge processes (Hendon 1988; Blade and Hartmann 1993; Kemball-Cook and Weare 2001), and convective stochastic forcing (Neelin and Yu 1994; Yu and Neelin 1994).
Initiation of the MJO has drawn more research attention lately. Matthews (2008) classified the MJO events into two types: the successive MJO, which immediately follows a preceding MJO event, and the primary MJO, which starts without any immediate preceding MJO event. He found that 40% of total MJO events are primary. Most variables relevant to the MJO, such as humidity and boundary layer convergence, show no evident signal during initiation of the primary MJO; they only become significant once the MJO is generated. Webber et al. (2012) proposed that initiation of the primary MJO can be related to westward propagation of a downwelling oceanic Rossby wave. Straub (2013) found that slowly eastward-propagating 850-hPa easterly anomalies over the Indian Ocean precede initiation of the primary MJO by at least 10 days. Seo and Kumar (2008) and Zhao et al. (2013) suggested that low-level moisture convergence plays a key role in MJO initiation. Zhang and Ling (2012) documented the structure and evolution of potential vorticity (PV) through convective initiation to mature stages of the MJO over the tropical Indian Ocean. The MJO PV structure cannot be explained by eastward PV advection from Africa; it is locally generated by MJO and stochastic convection over the Indian Ocean. Seo and Song (2012) showed that the midtropospheric PV anomaly acts as a dynamic forcing and plays an important role in the initiation of boreal summer intraseasonal oscillation.
Most, if not all, previous studies on MJO initiation focused on the MJO alone. In fact, there are two issues related to convective initiation of the MJO. One is when, where, and how active convection may occur in a large area over the tropical Indian Ocean. The other one is whether such a large-scale convective activity would lead to an MJO event. All large-scale convective events over the tropical Indian Ocean do not lead to the MJO. We cannot fully understand MJO convective initiation until we know why some large-scale convective events become the MJO while others do not.
This study seeks signals that may distinguish large-scale convective events over the tropical Indian Ocean that lead to the MJO from those that do not. We anticipate that MJO and non-MJO convective events may share common features. We are particularly interested in possible precursors in large-scale fields prior to the convective initiation of the MJO that are missing in non-MJO convective events. Such signals must be identified when MJO and non-MJO convective events are directly compared.
We first defined convective initiation over the tropical Indian Ocean and identified MJO and non-MJO events (section 3) using satellite and reanalysis data (section 2). Then we compared composites of MJO and non-MJO events, with an emphasis on time prior to their convective initiation (section 4). We choose to examine precipitation, low-level zonal wind, surface pressure, tropospheric temperature, moisture, diabatic heating, potential vorticity, sea surface temperature, and gross moist stability. A summary and a discussion are given in section 5.
2. Data and method
The following data were used in this study over the period of 1998–2009.
Daily rainfall data (0.25° × 0.25°) from the Tropical Rainfall Measuring Mission (TRMM; Kummerow et al. 2000) Multisatellite Precipitation Analysis (Huffman et al. 2007) were used to identify large-scale convective events over the tropical Indian Ocean and to classify them into different categories (MJO versus non-MJO events).
Daily data from the global European Centre for Medium-Range Weather Forecasts (ECMWF) Interim Re-Analysis (ERA-Interim; Dee et al. 2011). The horizontal resolution is 1.5° × 1.5°. The vertical resolutions are 25 hPa for 1000–750 hPa, 50 hPa for 750–250 hPa, and 25 hPa for 250–100 hPa. Diabatic heating was estimated using the data as a residual of the thermodynamic equation following the approach of Yanai et al. (1973).
The ENSO index from the Climate Prediction Center, which is the 3-month running mean of SST anomalies in the Niño-3.4 region based on Extended Reconstructed Sea Surface Temperature, version 3b. A threshold of ±0.5 K was used to define warm and cold ENSO events.
NOAA High-Resolution Blended Analysis of daily SST (Reynolds et al. 2007) with a horizontal resolution of 0.5° × 0.5°. The Indian Ocean dipole (IOD) index (Saji et al. 1999) was calculated using the data smoothed by a 100-day running mean.
For a given variable, its long-term trend was first removed, and then its annual climatology (all-time mean for each calendar day) was removed from its daily time series to generate anomalies. A 7-day running mean was then applied to the anomalies to remove their high-frequency variability. The resulting anomalous time series are dominated by intraseasonal variability.
3. Case selections
a. Convectively active episode
The RMM index (Wheeler and Hendon 2004) does not always accurately describe the location of active convection associated with the MJO, especially for individual events at their convective initiation stages over the Indian Ocean. The RMM index is based on global tropical coherent patterns of outgoing longwave radiation (OLR) and upper- and low-level zonal wind. Contributions from OLR are generally much less than the wind (Straub 2013). When MJO convection is weak or absent and convection–circulation coupling is feeble, such as prior to and during convective initiation of the MJO over the Indian Ocean, RMM phases are determined mainly by global-scale distributions of zonal wind instead of MJO convection over the tropical Indian Ocean. In these cases, the location and time of convective initiation of the MJO cannot be accurately determined by the RMM index. Other MJO indices based on global tropical patterns of convection (e.g., Matthews 2008) suffer from similar problems. To better describe MJO convective initiation over the Indian Ocean, we need local convective signals. The RMM index, however, provides reliable information of global wind perturbations of the MJO and can be used to distinguish MJO initiation over the Indian Ocean with and without circumnavigating propagation of previous MJO events, namely, the successive and primary MJO (Matthews 2008).
In this study, the anomalous precipitation time series of TRMM was used to define general convective variability over the tropical Indian Ocean. Let P be anomalous TRMM precipitation averaged over the Indian Ocean of the deep tropics (hereafter tropical Indian Ocean; 10°S–10°N, 60°–90°E) and δP its standard deviation. A large-scale convective event over the tropical Indian Ocean was detected if P ≥ δP lasted continuously for at least 3 days. The first day of P ≥ δP was defined as the day of convective initiation (day 0).
b. MJO and non-MJO
MJO signals in precipitation were extracted from the TRMM data by using time–space Fourier transform following Wheeler and Kiladis (1999). Spectral coefficients in the eastward intraseasonal (20–80 days) and large-scale (zonal wavenumber 1–10) band were used to reconstruct a time series of MJO precipitation. It should be pointed out that this method suffers from a known but often-forgotten caveat: all eastward spectral power does not belong to eastward-propagating perturbations. Only that incoherent with its westward counterparts does (Hayashi 1982). For precipitation, the eastward power within the MJO band is about 4 times greater than its corresponding westward power (Zhang 2005). The possible maximum error of taking all eastward power of precipitation to represent the MJO would be overestimating the MJO strength by 25%. This calls for a refined spectral method of extracting MJO signals. We used the commonly practiced but somewhat flawed Wheeler and Kiladis (1999) method in this study for the ease of comparison to results from many others using the same method.
Once a convective episode was identified (section 3a), the following steps were taken to determine whether it was part of convective initiation of the MJO (Fig. 1). First, averaged MJO precipitation over 10°S–10°N was calculated. Second, a threshold of 1.45 mm day−1 was chosen as the noise level of . This value is the averaged standard deviation of over the Indian Ocean and Maritime Continent (60°–120°E). Third, if 1) was greater than 1.45 mm day−1 at day 0 at any longitude between 60°–90°E and 2) an area of > 1.45 mm day−1 (within the thick black contour in Fig. 1a) extended eastward through 100°E (the eastern boundary of the Indian Ocean) in the following days, then an MJO event was declared. This is equivalent to the 1.45 mm day−1 contour extending eastward passing 100°E on the Hovmöller diagram of . Examples of thus-identified MJO events are given in Fig. 1a as cases C2 and C3. If one of the two criteria was not satisfied, no MJO initiation was identified over the Indian Ocean and the case was regarded as a non-MJO event (e.g., case C1 in Fig. 1a).
Because of the influence of the Asian summer monsoon, boreal summer MJO events, with their northward propagation component, are more complicated than winter ones (Wang and Rui 1990; Li et al. 2001). To avoid this complication and make our analysis clean, boreal summer cases were not included in this study. A total of 23 MJO events and 13 non-MJO events were identified during October–April of 1998–2009.
There is another complication. Some events started very weak ( < 1.45 mm day−1) over the Indian Ocean, but they intensified rapidly as they moved over the Maritime Continent and western Pacific, where their values became greater than 1.45 mm day−1. These cases were classified as the non-MJO events over the Indian Ocean based on the procedure described earlier. However, conditions over the Indian Ocean may have influenced their subsequent development into MJO events. Including them in non-MJO composites may potentially produce misleading results. These events (four) were not included in this study.
c. Primary and successive MJO
Based on conditions prior to convective initiation of the MJO over the Indian Ocean, each MJO event can be classified as primary or successive (Matthews 2008). Signals of a preceding MJO event prior to a successive MJO event mainly exist in the circulation. They can be identified through the RMM index on its phase diagram as a counterclockwise rotation through phases 7, 8, and 1 and into 2 and 3 with an amplitude greater than 1 (e.g., Fig. 1b, red curve before day 0 of C3). In contrast, a primary MJO event can be identified if the amplitude of the RMM index started less than 1 in phase 7, 8, or 1, and it becomes greater than 1 as it rotates counterclockwise from phases 2 and 3 onward to 4, 5, 6, and 7 (e.g., Fig. 1b, blue curve before day 0 of C2). Of the 23 MJO events, 10 (~43%) were thus identified as successive and 13 (~57%) as primary. The percentage of successive MJO events is different from that (60%) found by Matthews (2008). The discrepancy may come from different sources. We used a combination of the RMM index with an amplitude threshold of 1 (i.e., one standard deviation) and TRMM precipitation to define MJO events over the Indian Ocean for October–April during 1998–2009; Matthews (2008) used an OLR-based index with a threshold of 0.4 over the entire tropics for all seasons during 1974–2005.
To examine possible modulation of convective initiation events of the MJO by ENSO and IOD, we show in Fig. 2 their timing relative to ENSO and IOD phases. These events sometimes cluster together, but without any apparent connection to particular phases of ENSO and IOD. These events appear to be randomly distributed in all phases of ENSO. Influence of ENSO in the present study should therefore be minimal.
There might be signs of IOD influences. Successive MJO (triangles) occurred only in the neutral phase of IOD (within the two dashed lines). There were more primary MJO (open circles) in the negative phase (below the lower dashed line) than positive phase (above the upper dashed line). There are more non-MJO events (dots) in the positive than negative phases. This analysis period is, however, too short to establish any statistics for IOD modulations of the different types of convective events.
Our selection and classification procedures include subjective choices, although with physically based justifications. Their results need to be scrutinized. In this subsection, we compare composites of the three types of convective events: the primary MJO, successive MJO, and non-MJO. Composite Hovmöller diagrams of precipitation anomalies averaged over 10°S–10°N for these three types of events are shown in Fig. 3. Eastward-propagating positive anomalies in precipitation 30–40 days before day 0 (convective initiation) of the successive MJO are clearly seen (Fig. 3b). They are absent before convective initiation of the primary MJO (Fig. 3a). Interestingly, eastward propagation of negative anomalies in precipitation appears 10–20 days before day 0 of both primary and successive MJOs, but signals of statistical significance are very limited for the primary MJO. There is no eastward propagation signal for the non-MJO, but there are negative precipitation anomalies over the Indian Ocean before day 0, suggesting a local intraseasonal stationary oscillation.
The general distinction between the two types of the MJO and non-MJO is also clearly displayed by their composites of the RMM index (Fig. 4). The amplitude of the RMM index is always smaller than 1 for the non-MJO (blue curve) but larger than 1 for the MJO over the Indian Ocean and part of the Maritime Continent (phases 2–4) after their convective initiation (day 0). The RMM index of the successive MJO (black) maintains its amplitude greater than 1 in phases 6–1 before convective initiation over the Indian Ocean (phase 1), whereas for the primary MJO (red), its amplitude starts close to 0 at day −20 and then quickly intensified to be greater than 1 after day −5 over the Indian Ocean (phase 2) and beyond. The inability of the RMM index to accurately identify convective initiation of the MJO is also illustrated in Fig. 4. Positive anomalies in precipitation of both primary and successive MJO reach their maxima at 85°E on day 4 (Figs. 3a,b). But day 4 is in RMM phase 3 for the primary MJO and phase 2 for the successive MJO.
For both non-MJO and primary MJO, there is no obvious precursor in precipitation before its positive anomalies reach maxima and start to show different behaviors in zonal propagation. The rest of this article is focused on differences in preconditions for convective initiation between the primary MJO (hereafter simply referred to as the MJO) and non-MJO events. It is equally interesting to see what distinguishes MJO events that lead to successive events from those that do not. We leave this to another study.
Composites were made for the 13 (primary) MJO events and 13 non-MJO events selected by the procedure described earlier in this section. All fields used for the composite are anomalies as described in section 2. A Student's t test was performed to identify results that are significant at the 95% confidence level. When discussing the composites, we will focus mostly on the significant results. However, all results with or without statistical significance will be presented together to provide a continuous spatiotemporal context for the significant results.
In this section special attention is given to signals prior to convective initiation (day 0) and to precursors that may distinguish the MJO from non-MJO events before and during their convective development. The eastward-propagating nature of the MJO and its absence in non-MJO events are evident not only in precipitation (Fig. 3), but also in many other fields shown in Fig. 5. A useful exercise of inspecting this and other time–longitude Hovmöller diagrams would be to block its upper portion (after day 0) and view only the lower portion (before day 0) to appreciate the difference, if any, in large-scale conditions prior to convective initiation between the MJO and non-MJO composites.
As early as 20–25 days prior to its convective initiation, eastward-propagating signals of the MJO start over the central-western Indian Ocean in negative anomalies of 850-hPa zonal wind (Fig. 5a) and 400-hPa temperature (Fig. 5e). No such early eastward-propagating signal exists with statistical significance in the non-MJO composites (Figs. 5b,f). Interestingly, there is westward propagation in 400 hPa temperatures of the non-MJO 10 days before its convective initiation (Fig. 5f). There are also obvious westward propagation signals in 500-hPa PV for both the MJO and non-MJO immediately after their convective initiation (Figs. 5g,h), suggesting the presence of equatorial Rossby waves. SST appears to be higher over a broader area of the Indian Ocean prior to convective initiation of the MJO than non-MJO, but signals with statistical significance are limited (Fig. 5i). There are significant negative anomalies of SST near 100°E in the non-MJO composite from 10 days before to 10 days after the convective initiation. There is no obvious difference between the MJO and non-MJO in PV at 500 hPa and moisture convergence at 850 hPa (Figs. 5c,d) or in the lower troposphere (850–500 hPa, not shown) before their convective initiation.
a. Low-level wind
The eastward propagation of low-level easterly anomalies emerging over the tropical Indian Ocean 20 days before convective initiation of the MJO is one of the features that clearly distinguishes the MJO from non-MJO convective events (Figs. 5a,b). This low-level easterly signal has been observed by Straub (2013). It can also be seen in the composite of 850-hPa wind based on the RMM index [phases 7 and 8 in Fig. 12 of Waliser et al. (2009)], where it coincides with eastward-propagating negative anomalies in precipitation of the MJO from the Indian Ocean to the Maritime Continent. However, the RMM composite to a large degree describes the successive rather than primary MJO.
Composites of horizontal distributions of anomalies in precipitation and 850-hPa wind vectors for the (primary) MJO show easterlies over the equatorial western Indian Ocean on day −15 when most of the tropical Indian Ocean and Maritime Continent are covered by negative anomalies in precipitation (Fig. 6a). As time progresses, the easterlies extend northward as well as eastward (Figs. 6c,e). By day −6, they have reached the Maritime Continent (Fig. 6g). Up to this point, there is hardly any significant positive anomaly in precipitation over the equatorial Indian Ocean, and there is hardly eastward propagation of the negative anomalies in precipitation, in contrast to the RMM composite [Fig. 12 of Waliser et al. (2009)]. A positive anomaly in precipitation then quickly develops over the central equatorial Indian Ocean (day 0, Fig. 6k), which is accompanied by strengthened and widened easterlies over the Maritime Continent and emergence of equatorial westerlies over the western Indian Ocean (Fig. 6m). Both are part of the MJO circulation pattern associated with MJO convection as depicted by Madden and Julian (1972). In the non-MJO composite (Fig. 6, right column), significant low-level easterlies are absent over the tropical Indian Ocean until positive anomalies in precipitation appear near convective initiation (day −3).
The difference in zonal winds between the MJO and non-MJO is not only in the timing of their easterlies (much earlier for the MJO), but also in their latitudinal and vertical extent (Fig. 7). Easterlies extend from the surface to the midtroposphere and from 15°S to 10°N for the MJO. They are more confined in space for the non-MJO events.
b. Surface pressure
There are two possible reasons for the low-level easterly anomalies discussed in the previous subsection. One is anomalies in the zonal pressure gradient. The other is anomalous momentum transport from extratropics, upper levels, and/or flows of other scales. Only the first possibility is discussed here. Intraseasonal anomalies in pressure (geopotential height) do not change sign vertically from the surface to 5 km (from 1000 to 500 hPa) in the tropics. Their signals are strongest near the surface. We use surface pressure (SP) anomalies as a proxy of lower-tropospheric pressure anomalies.
For the MJO, there is a zonal wavenumber-1 structure in equatorial SP anomalies (Fig. 8, left column). About 20 days before the convective initiation, negative SP anomalies are over the longitudes of South America, the Atlantic Ocean, and Africa, and positive SP anomalies are over the rest of the tropics. There is an equatorial low-pressure surge penetrating from Africa eastward into the Indian Ocean (Fig. 8c). This low-pressure surge quickly covers the entire equatorial Indian Ocean (Figs. 8e,g,i) and advances farther to the Maritime Continent (Figs. 8k,m) and the western Pacific (Fig. 8o). By the time of the convective initiation over the Indian Ocean, this low-pressure surge has reached the eastern Pacific (Fig. 8q). The average eastward propagation speed of the low-pressure surge is around 5 m s−1 over the Indian Ocean and Maritime Continent before the convective initiation (days −21 to −6) and accelerates to 27.5 m s−1 after convection emerges over the Indian Ocean (days −6 to 0). Meanwhile, the eastern edge of the region with positive SP anomalies over the western coast of South America moves eastward from day −12 and reaches the eastern coast of Africa at the convective initiation day, with an average propagation speed of around 11 m s−1. Such zonal wavenumber-1 structure in SP anomalies, the equatorial low-pressure surge, and the eastward propagation of SP anomalies are all missing in the non-MJO composite (Fig. 8, right column), where SP anomalies appear to change sign locally.
The earliest appearance of the low-level easterly anomalies over the western equatorial Indian Ocean is across the boundary between negative SP anomalies to the west and positive anomalies to the east about 18–15 days before the convective initiation (Figs. 8e,g). As the low-pressure surge reaches the Maritime Continent, the low-level easterly anomalies are enhanced there (day −6). The strengthening and widening of the equatorial low-level easterly anomalies (days −3 and 0) is more directly related to the developing convection over the Indian Ocean than the low-pressure surge. It should be pointed out that signals of the low-pressure surge are not always statistically significant, but they are robust in a sense that they become significant in a composite based on data of a longer record (see section 5).
Matthews (2008) showed in his composite of the primary MJO negative temperature anomalies in the midtroposphere propagating both westward and eastward from the eastern Indian Ocean and Maritime Continent 15 days prior to convective initiation. He interpreted the westward-propagating negative temperature anomalies as a destabilizing factor for convective initiation of the primary MJO. In the present composite of the (primary) MJO, there are also signals of both westward and eastward propagation of negative temperature anomalies at 400 hPa over the Indian Ocean starting 20 days prior to convective initiation, but only the eastward-propagating signals are significant (Fig. 5e). Significant westward propagation in negative temperature anomalies is found in the non-MJO composite starting 10 days before its convective initiation (Fig. 5f). The presence and absence of the eastward-propagating signals in negative temperature anomalies prior to convective initiation are another main distinction between the MJO and non-MJO events.
The negative temperature anomalies start at 400 hPa over the central Indian Ocean and gradually expand vertically and zonally as they move eastward (Fig. 9, left column). Their eastward propagation speed is about 5 m s−1 over the Indian Ocean and slightly faster over the Maritime Continent and the western Pacific. The average speed is about 6.5 m s−1 (determined by the moving speed of the trailing or western edge of the significant part of the anomalies). If clear-sky radiative cooling caused negative temperature anomalies, it could hardly explain their vertical and zonal expansion and eastward propagation. The eastward-propagating negative temperature anomalies before convection initiation of the non-MJO composite (Fig. 9, right column) are ambiguous, if they exist at all.
In the MJO composite, there are anomalous westerlies and subsidence coinciding with the negative temperature anomalies. Negative precipitation anomalies are found underneath the negative temperature anomalies. Positive precipitation anomalies start on day −6 at longitudes (~60°E) without obvious tropospheric temperature anomalies. This suggests that the eastward-propagating negative temperature anomalies do not trigger MJO convective initiation by destabilizing the atmosphere.
Once MJO convection is initiated, the tropospheric temperature responds quickly with positive anomalies centered at 300 hPa, flanked by negative anomalies above and to the west. This pattern is the same as observed by Kiladis et al. (2005) for the mature MJO. Meanwhile, the eastward-propagating negative temperature anomalies continue moving eastward, albeit losing much of their significance.
Horizontally, the negative temperature anomalies in the MJO composite are mostly confined to the tropics (Fig. 10, left column). They start over the tropical Indian Ocean, without much sign of being induced by extratropical or upstream perturbations. As positive precipitation anomalies emerge over the equatorial Indian Ocean immediately prior to convective initiation (day −6), the negative temperature anomalies exhibit a swallowtail pattern, with equatorial signals to the east and off-equatorial signals to the west. This is the same pattern that precipitation anomalies will develop into at a later time (Zhang and Ling 2012). No such swallowtail pattern in temperature anomalies exists in the non-MJO composite (Fig. 10, right column).
As mentioned earlier, there is no obvious difference in low-level moisture convergence between the MJO and non-MJO events prior to their convection initiation in the Hovmöller diagrams (Figs. 5c,d). There is a westward tilt in specific humidity q starting 6 days before and persisting through convective initiation of the MJO (Fig. 11, left column). This low-level moisture ahead (east of) the MJO convection center has been observed by many previous studies (e.g., Johnson et al. 1999; Kemball-Cook et al. 2002; Kiladis et al. 2005). It appears to be related to the moisture flux by the low-level easterly anomalies. It also exists in the non-MJO composite (Fig. 11, right column), if less obviously.
e. Diabatic heating
One interesting feature of diabatic heating associated with the MJO is its westward tilt (Lin et al. 2004; Kiladis et al. 2005; Jiang et al. 2011; Ling and Zhang 2011). In the MJO composite (Fig. 12, left column), the westward tilt exists only in the lower troposphere. In the upper troposphere, there is an eastward tilt, which emerges 3 days before the convective initiation. These tilts make the diabatic heating of the MJO into a boomerang shape. To a lesser degree, this boomerang heating structure also exists in the non-MJO composite (Fig. 12, right column). The only obvious difference in diabatic heating between the MJO and non-MJO composites is the zonal scale of the heating, which is larger for the MJO than the non-MJO.
f. Potential vorticity
Composite Hovmöller diagrams of PV at 500 hPa, the peak level of tropospheric MJO PV, show tendencies of westward propagation of positive PV anomalies (Figs. 5g,h) starting on day 0 for both the MJO and non-MJO. This suggests the presence of equatorial Rossby waves (Wheeler and Kiladis 1999). There is no obvious difference over the Indian Ocean in PV between the MJO and non-MJO prior to day 0.
The only PV signal that distinguishes the MJO and non-MJO is the cyclonic PV pair straddling the equator, a unique PV signature of the MJO (Zhang and Ling 2012). This PV pair is well established at convective initiation of the MJO and further strengthens afterward (Fig. 13, left column). It is generated by diabatic heating associated with the MJO in the tropics, with obvious connections to neither higher latitudes nor the stratosphere. A similar PV pair also emerges in the non-MJO composite (Fig. 13, right column), but its Southern Hemispheric part is weak and appears to connect with the extratropics.
g. Gross moist stability
The concept of gross moist stability (GMS) was introduced by Neelin and Held (1987) to measure the relationship between large-scale atmospheric flows and moist convection. It has been used to diagnose the MJO in model simulations (Sobel and Maloney 2013) as well as in observations (Haertel et al. 2008). Raymond et al. (2007) defined normalized GMS (NGMS) as the ratio of vertical integrated total divergence of moist entropy to vertical integrated horizontal moisture divergence.
Total NGMS and its horizontal and vertical components for the MJO and non-MJO as well as the corresponding precipitation anomalies, all averaged over 10°S–10°N, 60°–90°E, are compared in Fig. 14. Negative NGMS leading to convective initiation of the MJO has been documented (Raymond and Fuchs 2009; Hannah and Maloney 2011) but not found in this study. NGMS remains positive before and after convective initiation. It is larger than its October–April mean 10–15 days before convection initiation, starts to decrease as precipitation anomalies increase, reaches its minimum 4–5 days before the convective initiation but is still positive, and gradually increases until after the time of precipitation maximum. The horizontal component is larger than the vertical one before day −5. The behavior of NGMS is of no difference for the MJO and non-MJO. This suggests that NGMS might be a useful diagnostic tool for convective development in the tropics, but not unique for the MJO.
5. Summary and discussion
The purpose of this study is to seek precursors in large-scale fields that may distinguish events of convective initiation over the tropical Indian Ocean that lead to the MJO from those that do not. Such precursors may help us understand and predict convective initiation of the MJO. During October–April of 1998–2009, 36 large-scale convective events over the Indian Ocean were identified, out of which 13 primary MJO events and 13 non-MJO events were compared. Three very obvious and interesting precursor signals were found that are unique to the primary MJO. They all first emerge 10–20 days before convective initiation of the MJO and move eastward at a speed close to that of the MJO, while being absent or weak in non-MJO events. These signals are as follows.
Lower-tropospheric (surface to 500 hPa) easterly anomalies. They start over the equatorial western Indian Ocean at low levels (850 hPa), extend upward into the entire lower troposphere and eastward to the rest of the Indian Ocean and the Maritime Continent, and expand over most tropical latitudes (Fig. 5a and the left columns of Figs. 6–9). These easterly anomalies after the convective initiation become part of the canonical MJO circulation that is closely related to the eastward-propagating convection center of the MJO as portrayed in the MJO schematic by Madden and Julian (1972). Their possible role in MJO initiation over the Indian Ocean might be westward moisture transport from the Maritime Continent and generation of the low-level moisture convergence east of the developing convective center of the MJO (Fig. 11, left column). These eastward-propagating low-level easterly anomalies before MJO initiation have been observed by Straub (2013).
Surface pressure (SP) anomalies. They represent pressure anomalies in the lower troposphere (surface–5 km). They exhibit a zonal structure of wavenumber 1. Negative SP anomalies first cover the longitudes from South America to Africa, and positive anomalies cover the rest of the tropics (Fig. 8, left column). An equatorial low-pressure surge penetrates from Africa eastward through the Indian Ocean and Maritime Continent. Positive SP anomalies over the eastern Pacific also move eastward.
Negative temperature anomalies in the middle to upper troposphere (500–200 hPa). They start at 400 hPa over the Indian Ocean and immediately propagate eastward (Fig. 5e). They expand vertically and zonally, but most of the time they are confined to the equatorial region (Figs. 9 and 10, left columns). There is no obvious role of the negative temperature anomalies in convective initiation of the MJO. They were observed previously by Matthews (2008).
These three major signals that distinguish the MJO from non-MJO are confirmed by composites based on OLR data of 1980–2010. Other fields were also examined, but they do not clearly distinguish convective initiation of the MJO from that of non-MJO. Low-level moisture increases before convective initiation (Fig. 11). This precondition for deep convection, previously observed by many (e.g., Johnson et al. 1999; Kemball-Cook et al. 2002; Kiladis et al. 2005), is not unique to the MJO. Low-level diabatic heating starts before deep heating, resulting in a westward tilt that exists in both MJO and non-MJO, only to a different extent (Fig. 12). The distinct structure of PV for the MJO emerges only during and after its convective initiation (Fig. 13). Warmer-than-normal sea surface temperature over the Indian Ocean appears to exist before convective initiation of the MJO, but the significant signal is limited (Figs. 5i,j). For both MJO and non-MJO events, equatorial Rossby waves appear to exist (e.g., in PV) during the convective initiation, and normalized gross moist stability decreases when precipitation increases (Fig. 14) as theoretically predicted (Raymond and Fuchs 2009).
There are several shortcomings in this study. Its results are based on a single global reanalysis product. They need to be confirmed by other reanalysis products and, especially, by observations. We have focused on large-scale precursors between MJO and non-MJO convective initiation. We did not explain why convection would occur in a large area in either case. We are not able to fully explain the three major signals that distinguish the MJO from non-MJO, namely, the low-level easterly anomalies, middle-to-upper-tropospheric negative temperature anomalies, and surface pressure anomalies. Here, we can only offer some preliminary thoughts on them.
Matthews (2008) and Straub (2013) both suggested that a primary MJO event is preceded by global-scale perturbations, not by another MJO event. We agree. The three signals unique to the MJO do not show any obvious sign of extratropical, stratospheric, or upstream influences. We view them as intrinsically local to the tropospheric atmosphere over the tropical Indian Ocean. Although we did not make any effort to seek coherence between these three fields, we found their common features. They all appear 10–20 days before convective initiation of the MJO, and they all propagate eastward at a speed close to that of the MJO before its convective initiation. These common features made us ponder the possibility that there is a dry dynamic mode of the MJO.
The idea of a possible dry dynamic MJO mode is not new. Frederiksen and Frederiksen (1997) suggested that tropical intraseasonal oscillation might result from barotropical and baroclinic instability when the tropics and extratropics are coupled. Lin et al. (2007) extracted tropical intraseasonal, eastward-propagating (16 m s−1) signals in 250-hPa velocity potential and zonal wind from a dry global atmospheric model. They suggested that these signals in the dry model emerge as a response to extratropical influences. Wedi and Smolarkiewicz (2010) analytically and numerically tested a hypothesis that the core dynamic of the MJO is a weak, nonlinear, dry Rossby wave. Under favorable conditions (e.g., a zonal mean meridional shear of the zonal flow or extratropical influences), the nonlinear Rossby wave exhibits a solitary wave structure, persists at planetary scales, and moves slowly eastward (5.7 m s−1).
The eastward-propagating low-level easterly anomalies, surface low-pressure anomalies, and mid-to-upper-tropospheric negative temperature anomalies observed in this study do not provide direct evidence for these previous hypotheses but support their general spirit. If a dry dynamic mode of the MJO exists in any form, these anomalies might be part of its signatures. MJO initiation would then be a process of supplying energy to its dynamic mode. There could be various sources of such energy supply, including but not limited to convection–circulation coupling, upstream and extratropical influences, scale interaction, and stochastic forcing. When active convection occurs in a large area without a proper receptor of an MJO dynamic mode, no MJO would be initiated.
Three anonymous reviewers provided careful comments on the submitted manuscript, which helped improve this article. Jian Ling thanks Anton Beljaars, Peter Bauer, and Frederic Vitart for hosting his visit to ECMWF. This study was supported by NSF Grant AGS-1062202. The online availability of the ERA-Interim products, TRMM rainfall data, ENSO index, SST, and MJO RMM index is highly appreciated.
This article is included in the DYNAMO/CINDY/AMIE/LASP special collection.