Abstract

The Madden–Julian oscillation (MJO) is arguably the most important intraseasonal mode of climate variability, given its significant modulation of global climate variations and attendant societal impacts. Advancing the current understanding and simulation of the MJO using state-of-the-art climate data and modeling systems is thus a necessary goal for improving MJO prediction capability. MJO variability is assessed in NOAA/NCEP reanalyses and two versions of the Climate Forecast System (CFS), CFS version 1 (CFSv1) and its update version 2 (CFSv2). The analysis leans on a variety of diagnostic procedures and includes MJO sensitivity to varying El Niño–Southern Oscillation (ENSO) phases. It is found that significant improvements have been realized in the representation of MJO variations in the new NCEP Climate Forecast System reanalysis (CFSR) as evidenced by outgoing longwave radiation (OLR) power spectral analysis and more coherent propagation characteristics of precipitation and 850-hPa zonal winds over the Eastern Hemisphere in CFSR-only depictions. Conversely, while modest improvements are realized in the CFSv2 as compared to CFSv1, in general the simulation of the MJO continues to be a challenge. Both versions produce strong eastward propagating variance of convection and wind fields in the intraseasonal frequency band. However, the simulated MJO propagates slower than the observed with difficulties traversing the Maritime Continent into the western Pacific, as noted in many previous modeling studies. The CFS shows robust intraseasonal simulations over the west Pacific during El Niño years with diminished simulation capability over the Indian Ocean during La Niña years. This is likely a manifestation of the preference for La Niña MJO activity to occur over the Indian Ocean and the simulation challenges over that domain.

1. Introduction

The recent focus on climate prediction strategies to bridge the gap between two-week deterministic weather forecasts and probabilistic seasonal climate prediction necessitates improved understanding and simulation of intraseasonal climate variability. One such subseasonal climate variability mode offering enhanced prospects for climate prediction is the Madden–Julian oscillation (MJO). The MJO is an eastward-propagating intraseasonal (i.e., ~30–90 days) oscillation in tropical winds and convection, which is strongest during the boreal cold season months of November–April (Madden and Julian 1994). MJO influences are many and include impacts on global tropical cyclogenesis (Maloney and Hartmann 2000; Vitart 2009), North American and Asian monsoon systems (Zhang and Dong 2004), midlatitude storm track variations (Lin and Brunet 2009), and the evolution of El Niño–Southern Oscillation (ENSO) variability (Pohl and Matthews 2007; Zhang and Gottschalck 2002; Zhang 2005).

Despite the significant impacts of the MJO on the global climate system and the benefits afforded by improved forecasts of MJO events, the simulation of MJO variability continues to be an onerous burden for some climate models (Kim et al. 2009; Lin et al. 2006). Furthermore, given the potential for varied mechanisms to produce MJO variations and subsequent evolution, there is no universal indication of what specific processes need to be improved in climate models to advance MJO simulation. Notwithstanding drawbacks in our current understanding of the physical mechanisms giving rise to MJO variability, including simulation deficiencies, there have been notable improvements in some modeling systems in recent years (Kim et al. 2009). Since these systems were not tuned for the improved representation of the MJO, results suggest that the simulated MJO may depend on gradual advancements in multiple model physics components, rather than improvements in one particular parameterization scheme.

In this study, we conduct a comprehensive assessment of MJO variability in different National Centers for Environmental Prediction (NCEP) reanalyses and simulations from the coupled atmosphere–ocean Climate Forecast System (CFS). The target of the analysis is on the simulation of MJO variations as opposed to prediction skill of near-term MJO indices. A main difference from previous MJO studies using long-term free integrations of coupled models is that our analysis for the CFS is based on 9-month seasonal forecasts initialized from observations. Accordingly, while including air–sea coupling, which has been shown to be an important process for the MJO (Inness et al. 2003a), the model maintains a relatively realistic mean state and interannual variability (i.e., ENSO) compared to those in free simulations. The assessment of the initial version [CFS version 1 (CFSv1)] and its upgrade (CFSv2) is critical for understanding its performance in climate predictions and simulations.

In addition to an overall diagnosis of MJO variability for the 25-yr period from 1982 to 2006 in NCEP reanalyses and CFS simulations, an important aspect of the analysis is a focus on differences between MJO features in the Indian and Pacific Oceans as a function of ENSO phase. This is in an effort to better understand physical mechanisms and ENSO-related interannual modulation of MJO activity, especially given that the amplitude and eastward propagation of the MJO vary with the tropical Pacific ENSO variability (Kessler 2001) and that the MJO in turn also influences the evolution of ENSO. This two-way ENSO–MJO relationship is becoming increasingly recognized as an important factor for both subseasonal and seasonal global climate predictions, given the phase dependence of the midlatitude response to the tropical intraseasonal convective activities (Pohl and Matthews 2007; Roundy et al. 2010), and that ENSO-related SST variability is the underpinning for seasonal climate predictions. It is therefore important to understand the variation of MJO features in conjunction with ENSO phases and whether the NCEP coupled models (i.e., the CFS suite) can reproduce aspects of this behavior, as the CFS is used for extended range (weeks 2–4) and seasonal (up to 9 months) prediction efforts that rely critically on the skill of ENSO prediction.

Accordingly, this study uses seasonal retrospective forecasts from two systems: CFSv1 (Saha et al. 2006) and the next generation CFSv2 (S. Saha et al. 2011, personal communication). The use of reforecasts as opposed to the Coupled Model Intercomparison Project (CMIP) style free model runs in the simulation diagnosis facilitates a more robust investigation of the critical interactions of the MJO with ENSO during the periods of observed ENSO events, as investigated here, provided that the model exhibits skill in maintaining ENSO variability during the target forecast period.

Section 2 will describe the datasets and diagnostics used in this study. Section 3 will detail the large-scale tropical climate. Section 4 will describe MJO features over the full 25-yr record, with section 5 stratifying the analysis as a function of ENSO phase. Section 6 is left for the summary and conclusions.

2. Data and methodology

To assess the observed features of MJO variability we employ the NCEP–Department of Energy (DOE) reanalysis-2 (R2; Kanamitsu et al. 2002) and the NCEP CFS reanalysis (CFSR; Saha et al. 2010). Both reanalyses provide representations of the climate on a global scale. The CFSR is the most recent NCEP reanalysis and has numerous upgrades over R2. In addition to the various improvements in the model physics, the CFSR is produced based on a coupled atmosphere–ocean–land guess forecast with a much higher atmospheric horizontal resolution (T382) and includes direct assimilation of radiance data (Saha et al. 2010). The reliance on two NCEP reanalysis systems is twofold: to assess potential differences in MJO features between the R2 and CFSR and to provide the proper benchmark data for the CFSv1 and the updated CFSv2 simulations, as these reforecasts are initialized by the R2 and CFSR, respectively. We also employ the use of observed outgoing longwave radiation (OLR) data from the NOAA Advanced Very High Resolution Radiometer (AVHRR; Liebmann and Smith 1996) and Global Precipitation Climatology Project (GPCP) precipitation data (Xie et al. 2003), which have been interpolated to daily resolution for the period 1982–2006.

The assessment of MJO simulations is based on four-member ensemble reforecast datasets from the NCEP coupled climate models, CFSv1 and CFSv2, which are initialized in early October for the years 1982–2006. The CFSv1 is initialized once per day during 9–12 October from R2, while the CFSv2 is initialized four times per day on 8 October from CFSR. For the simulation analysis the target dates are 1 November–30 April, effectively eliminating the influence of the initial conditions on the results and the possible influence of the varying dates of initialization between the two systems. Table 1 shows some of the differences in model characteristics between the CFSv1 and CFSv2 and includes spatial and temporal resolution changes, upgrades to both the ocean and land models, and differences in the initial condition datasets. A more extensive presentation of model upgrades can be found in Table 1 of Saha et al. 2010. For the simulation diagnostics the four members from each model version are treated independently, with the displayed results representing the post-diagnostic ensemble average.

Table 1.

Comparison of primary differences between the CFSv1 and CFSv2 modeling systems.

Comparison of primary differences between the CFSv1 and CFSv2 modeling systems.
Comparison of primary differences between the CFSv1 and CFSv2 modeling systems.

One difficulty in earlier assessments of MJO simulations was the lack of consistent performance metrics for the models, as different statistics were used in the various studies. In an effort to capitalize on recent progress in MJO studies and chart a path forward, a U.S. Climate Variability and Predictability (CLIVAR) MJO working group was established to address many of the issues related to our understanding and simulation of MJO variability (Gottschalck et al. 2010; Kim et al. 2009; Waliser et al. 2009). Chief among the outcomes of this effort was the development of a standardized set of diagnostics that modeling groups can use to facilitate more robust model/observational and intermodel comparisons (Waliser et al. 2009).

We employ the MJO Working Group recommendations here to facilitate physical understanding while elucidating the fidelity of the CFSv1 and CFSv2 in depicting MJO variability. The diagnostics are applied to observations, reanalyses, and individual model members. Typical fields include daily-averaged raw and 20–100-day bandpass-filtered OLR (Kaylor 1977), 850-hPa zonal winds, 200-hPa velocity potential, and precipitation anomalies. The lens through which we will view the various MJO characteristics includes an assessment of spatial mean state and variance diagnoses, lag correlation, power spectral, wavenumber frequency, and EOF analysis.

3. Large-scale structure

a. Mean State

MJO simulations tend to be more robust in climate models having a realistic mean state structure of precipitation and low-level zonal winds (Inness et al. 2003a,b; Zhang 2005; Kim et al. 2009). Figure 1 shows the mean state precipitation (shaded) and 850-hPa zonal winds (contoured) for the observations (GPCP/CFSR, top panel), CFSv1 (middle panel), and CFSv2 (bottom panel) for November–April from 1982 to 2006. The solid red line denotes the zero contour. Both versions of the model simulate the distribution of precipitation with fidelity when compared to their observational counterpart, although the amplitude of precipitation in the tropical Indian and Pacific Oceans are somewhat elevated in the simulations. The Pacific intertropical convergence zone (ITCZ) amplitude is better represented by the CFSv2 with the CFSv1 being somewhat stronger, while the South Pacific convergence zone (SPCZ) amplitude is elevated in the simulations as compared to observations. It has been noted that the models that simulate a strong mean SPCZ tend to have more robust MJO simulations (Kim et al. 2009).

Fig. 1.

November–April observed mean precipitation (shaded) and 850-hPa zonal winds (contoured) for 1982–2006 in (top) CFSR and GPCP, (middle) CFSv1, and (bottom) CFSv2. Precipitation is shaded in 2 mm day−1 intervals, and 850-hPa zonal winds are contoured every 3 m s−1 with positive (negative) values indicated by the solid (dashed) line. The red line denotes the zero contour.

Fig. 1.

November–April observed mean precipitation (shaded) and 850-hPa zonal winds (contoured) for 1982–2006 in (top) CFSR and GPCP, (middle) CFSv1, and (bottom) CFSv2. Precipitation is shaded in 2 mm day−1 intervals, and 850-hPa zonal winds are contoured every 3 m s−1 with positive (negative) values indicated by the solid (dashed) line. The red line denotes the zero contour.

Similarly, a robust depiction of the distribution of the low-level zonal wind field is also apparent in both simulations. Mean westerlies extend from the western equatorial Indian Ocean (slightly weaker in simulations) through the Maritime Continent and into the western Pacific, with tropical easterlies entrenched over the eastern Pacific to the date line. Previous studies suggest that the presence of mean westerlies over the equatorial Eastern Hemisphere is a favorable background condition for the formation and maintenance of MJO activity (Zhang 2005), as seen here.

b. Intraseasonal variability

In addition to the mean state, it is necessary to evaluate the spatial structure of intraseasonal variance on the MJO time scale. Shown in Fig. 2 is the 20–100-day filtered variance of precipitation and 850-hPa zonal winds over 1982–2006, plotted as in Fig. 1. The strongest intraseasonal variance in precipitation is located over the western Pacific and Indian Oceans in observations. The simulations generally capture these areas; however, their amplitude is significantly stronger than that from observations. Both versions of the CFS extend their precipitation footprint into the eastern equatorial Pacific (not seen in observations), however with a weaker amplitude in the CFSv2.

Fig. 2.

November–April 20–100-day filtered variance in precipitation (shaded) and 850-hPa zonal winds (contoured) for 1982–2006 in (top) CFSR and GPCP, (middle) CFSv1, and (bottom) CFSv2. Precipitation is shaded in mm2 day−2 (see color bar scale for intervals), and zonal winds are contoured every 3 m2 s−2. The red line denotes the >9 m2 s−2 contour and is arbitrary.

Fig. 2.

November–April 20–100-day filtered variance in precipitation (shaded) and 850-hPa zonal winds (contoured) for 1982–2006 in (top) CFSR and GPCP, (middle) CFSv1, and (bottom) CFSv2. Precipitation is shaded in mm2 day−2 (see color bar scale for intervals), and zonal winds are contoured every 3 m2 s−2. The red line denotes the >9 m2 s−2 contour and is arbitrary.

The 850-hPa zonal wind intraseasonal variance is robust over southern portions of the Maritime Continent and the tropical Indian and west Pacific Oceans, as in precipitation; however, the maxima in observations is located over the Maritime Continent and the western equatorial Pacific, as indicated by the red contour (9 m2 s−2). The simulations capture the maxima in 850-hPa zonal wind over the west Pacific with the CFSv2 being significantly stronger.

4. MJO variability

a. Filtered lag correlations

An appraisal of the basic features of MJO variability is a necessary first step in understanding physical characteristics and model assessment. Indeed, a climate model that does not exhibit variability consistent with the observed subseasonal perturbations will most likely be ineffective at capturing the onset and evolution of individual MJO events.

Quasi-observed MJO variability is examined in Fig. 3 via lagged correlations of the reanalysis and observed 20–100-day bandpass-filtered daily anomalies (i.e., MJO time scale) of precipitation and 850-hPa zonal winds against an Indian Ocean precipitation index, representing the area-averaged daily precipitation anomalies over the domain 10°S–10°N, 70°–100°E(Waliser et al. 2009).1 The analysis is carried out over the years 1982–2006 for the months November–April. Figures 3a and 3b display the lag correlations using the observed GPCP precipitation with reanalysis 850-hPa zonal winds from the R2 and CFSR, respectively. The 850-hPa zonal wind propagation from R2 and CFSR compare quite nicely when used in conjunction with the same observed precipitation and, similarly, capture known features of the MJO evolution—for example, robust precipitation propagation to the date line and faster 850-hPa zonal wind propagation in the Western Hemisphere.

Fig. 3.

Lag correlations of 20–100-day filtered 10°S–10°N averaged precipitation (shaded) and 850-hPa zonal winds (contoured) with respect to the Indian Ocean precipitation index: (a) R2 850-hPa zonal winds and GPCP precipitation, (b) CFSR 850-hPa zonal winds and GPCP precipitation, (c) R2 850-hPa zonal winds and R2 precipitation, and (d) CFSR 850-hPa zonal winds and CFSR precipitation. Warm (cold) colors indicate positive (negative) correlations for precipitation, while solid (dashed) lines indicate positive (negative) correlations for 850-hPa zonal winds and are contoured at 0.1 intervals.

Fig. 3.

Lag correlations of 20–100-day filtered 10°S–10°N averaged precipitation (shaded) and 850-hPa zonal winds (contoured) with respect to the Indian Ocean precipitation index: (a) R2 850-hPa zonal winds and GPCP precipitation, (b) CFSR 850-hPa zonal winds and GPCP precipitation, (c) R2 850-hPa zonal winds and R2 precipitation, and (d) CFSR 850-hPa zonal winds and CFSR precipitation. Warm (cold) colors indicate positive (negative) correlations for precipitation, while solid (dashed) lines indicate positive (negative) correlations for 850-hPa zonal winds and are contoured at 0.1 intervals.

Significant differences emerge when the lag correlations are computed using internally consistent reanalysis fields, that is, reanalysis-derived winds and precipitation. Figures 3c and 3d highlight this feature as evidenced by the weak propagation of both precipitation and 850-hPa zonal winds in the R2-only representation (Fig. 3c) as compared to the more robust propagating features depicted in the CFSR-only representation (Fig. 3d). Particularly noteworthy in the R2-only depiction is the lack of any coherent eastward propagation of an 850-hPa zonal wind anomaly at any lead time prior to t minus 10 days. While both R2-only and CFSR-only lag correlations exhibit some degree of westward propagation over equatorial Africa and the western Indian Ocean, not seen in the more observationally constrained representation (i.e., Figs. 3a and 3b), it is quite weak in the CFSR-only realization. This feature combined with the eastward propagation of precipitation to the date line (albeit weaker) and stronger 850-hPa zonal wind propagation into the Western Hemisphere in the CFSR-only analysis demonstrates the improvement in the depiction of tropical intraseasonal precipitation variability in this new reanalysis system. Since the R2 and CFSR 850-hPa zonal winds show similar lag correlations with the GPCP rainfall (Figs. 3a and 3b), the improved CFSR-only (Fig. 3d) correlations compared to R2 only (Fig. 3c) indicate a better rainfall simulation. Due to the mutual interaction between convection and large-scale circulation, the improved rainfall in the CFSR implies an enhancement of model physics or rainfall-related dynamical processes such as the low-level moisture convergence or both. It would be interesting to ascertain if this representation in the pure CFSR extends to improved simulation of MJO variability in the CFSv2, given that the CFSv2 and CFSR use the same model formulation.

Figure 4 shows the lagged correlations for precipitation and 850-hPa zonal winds in the CFSv1 (top) and CFSv2 (bottom), as in Fig. 3. The structure and propagation of 850-hPa zonal winds is similar in the two CFS model versions, with the CFSv1 exhibiting slightly more coherent eastward propagation of the zonal wind anomaly there, although both models capture the faster zonal wind propagation in the Western Hemisphere. The slow eastward propagation over the Eastern Hemisphere has been previously noted in the CFSv1 (Seo and Wang 2010) and appears to persist in the CFSv2. The CFSv1 does have significantly strong westward propagation of low-level zonal winds over the equatorial Indian Ocean and African continent, not seen in the observed (cf. Fig. 3) or CFSv2 depictions, a minor but notable improvement in the new version of the CFS. The similarity in the lag correlations in Fig. 4 between the CFSv1 and CFSv2 is interesting given that the CFSv1 is initialized with the R2 reanalysis. One may expect an equally poor simulation as shown in the R2-only representation (Fig. 3c). However, there were model upgrades between the production time of the R2 and that of the CFSv1. Apparently these model improvements are manifest in the CFSv1 (Fig. 4, top panel) via the improved large-scale circulation–convection relationship when compared to R2 only (Fig. 3c).

Fig. 4.

As in Fig. 3 but for the (top) CFSv1 and (bottom) CFSv2.

Fig. 4.

As in Fig. 3 but for the (top) CFSv1 and (bottom) CFSv2.

b. Multivariate EOFs

While lag correlations nicely capture the propagation features of precipitation and low-level winds on the MJO time scale, a more robust indicator of MJO variability resides in a model”s ability to depict the multivariate EOFs of combined u850, u200, and OLR and consequently the characteristic pattern of the three-dimensional MJO structure (Kim et al. 2009; Wheeler and Hendon 2004). Figure 5 shows the structure of the first two multivariate EOFs of equatorially averaged (15°S–15°N) daily anomalies of OLR, u200, and u850.

Fig. 5.

Multivariate EOFs (left) 1 and (right) 2 of 15°S–15°N averaged 20–100-day filtered (top) 850-hPa zonal winds, (middle) 200-hPa zonal winds, and (bottom) OLR, for observations (black), CFSv1 (green), and CFSv2 (red).

Fig. 5.

Multivariate EOFs (left) 1 and (right) 2 of 15°S–15°N averaged 20–100-day filtered (top) 850-hPa zonal winds, (middle) 200-hPa zonal winds, and (bottom) OLR, for observations (black), CFSv1 (green), and CFSv2 (red).

The observed structures of modes 1 and 2 (black lines) depict a baroclinic circulation pattern highlighted by out-of-phase lower-level and upper-level winds and corresponding OLR anomalies. Mode 1 (left column) representing 17.26% of the observed variance indicates positive (negative) u850 (u200) wind anomalies over the western and central Pacific with a coincident negative OLR anomaly, signifying enhanced convection in this region. Conversely, the reverse situation exists over portions of the Indian Ocean, indicative of suppressed convection there. Mode 2 with an observed percentage of explained variance of 16.57% is in quadrature with mode 1 and features a baroclinic circulation in the Indian Ocean and collocated negative OLR anomalies with a maximum over the Maritime Continent. For the simulations the CFSv2 shows a much improved MJO EOF representation for both modes, as tracked by the similarities in the black and red lines. To be sure, there are some problem areas for CFSv2, most notably in the central Pacific for mode 1 and the Maritime Continent for mode 2, however, only in the OLR field. This suggests a continued deficiency of climate models in resolving MJO features when interacting with the Maritime Continent (Vintzileos and Pan 2007).

c. Spectral analysis

Another useful diagnostic to assess the MJO is to compare the power spectra of selected fields from observations and climate simulations. Figure 6 shows the power spectra of unfiltered daily anomalies of 850-hPa zonal winds and OLR in the various observed and simulated datasets for the November–April period of 1982–2006 over the Indian Ocean (a and b) and western Pacific (c and d) regions. The area averages for the anomalies encompasses the latitude bands of 10°S–5°N for OLR and 15°S–0° for 850-hPa zonal winds, while the longitude bands are taken as 70°–100°E (160°–185°E) for the Indian Ocean (western Pacific), as suggested by inspection of the maximum of variance in precipitation and 850-hPa zonal wind in Fig. 2 and the MJO working group.

Fig. 6.

Power spectra of 850-hPa zonal winds and OLR for the (top) Indian Ocean and (bottom) western Pacific for the CFSv2 (black), CFSv1 (blue), NCEP R2 (red), NCEP CFSR (purple), and NOAA AVHRR (cyan dashed). The units for 850-hPa zonal winds are m2 s−2 day while for OLR is W2 m−4 day. (a),(c) 850-hPa zonal winds are area averaged over the domain of 15°S–0°, 70°–100°E (15°S–0°, 160°–185°E), while (b),(d) OLR is area-averaged 10°S–5°N, 70°–100°E (20°S–5°S, 160°–185°E).

Fig. 6.

Power spectra of 850-hPa zonal winds and OLR for the (top) Indian Ocean and (bottom) western Pacific for the CFSv2 (black), CFSv1 (blue), NCEP R2 (red), NCEP CFSR (purple), and NOAA AVHRR (cyan dashed). The units for 850-hPa zonal winds are m2 s−2 day while for OLR is W2 m−4 day. (a),(c) 850-hPa zonal winds are area averaged over the domain of 15°S–0°, 70°–100°E (15°S–0°, 160°–185°E), while (b),(d) OLR is area-averaged 10°S–5°N, 70°–100°E (20°S–5°S, 160°–185°E).

Figures 6a and 6c show the 850-hPa zonal wind power spectra in the NCEP R2 (red), CFSR (purple), CFSv1 (blue), and CFSv2 (black). Over the Indian Ocean (Fig. 6a), the extended winter daily anomalies exhibit a maximum in spectral power in the 30–90-day band, representative of the MJO time scale. Both the CFSv1 and CFSv2 simulate a similar peak in the power spectra of 850-hPa zonal winds when compared to reanalysis data; however, their amplitude is much stronger, although the CFSv2 shows some improvement over its predecessor. The western Pacific 850-hPa zonal winds similarly agree in structure among the various reanalyses and simulations, although the relative amplitude between the simulations and reanalysis datasets is different, with the R2 and CFS producing a similar amplitude and the CFSR and CFSv2 nearly identical in the intraseasonal band.

Figures 6b and 6d show the power spectra of unfiltered daily OLR in the NOAA AVHRR (cyan dashed), NCEP R2 (red), CFSR (purple), CFSv1 (blue), and CFSv2 (black). Over the Indian Ocean the observed power spectra from the NOAA satellite data exhibit a broadened spectrum in the 30–90-day band. While both reanalysis products show less power in the intraseasonal band, the structure of the CFSR representation follows that from the observations more closely, that is, the broadened spectrum. The NCEP R2, CFSv1, and CFSv2 all produce sharp intraseasonal peaks at lower frequencies, as compared to observations, and the simulations have greater amplitude. Although the CFSv2 peaks in the lower frequency end of the intraseasonal spectrum, the amplitude is closer to the observed depiction. The OLR depiction over the western Pacific shows much more agreement among the reanalysis, observations, and CFS simulations than over the Indian Ocean, although this time the CFSR exhibits the strongest amplitude peak in the intraseasonal band.

A limitation of power spectral analysis is that it does not provide information about the propagation features of intraseasonal anomalies. A useful diagnostic to understand the propagation characteristics of MJO-related fields is the wavenumber–frequency diagram, which compactly provides information regarding the spectral power for both eastward and westward propagating components, including their wavenumber characteristics.

Figure 7 shows the wavenumber–frequency diagrams for equatorially averaged (10°S–10°N) OLR (left column) and 850-hPa zonal winds (right column) for the observations from NOAA satellites and CFSR (top panels), CFSv1 (middle panels), and CFSv2 (bottom panels). Both observed equatorial OLR and 850-hPa zonal winds exhibit dominance in their eastward-propagating power spectra maximized at wavenumber 1, although significant eastward propagation also occurs at wavenumber 2. The OLR and 850-hPa zonal winds depiction from CFSv1 does not compare well to its observed counterpart. While the dominance of eastward vis-à-vis westward propagating components is captured, the CFSv1 again places significant power for both fields at low frequencies in relation to observations, corresponding to an overall slow eastward propagation.

Fig. 7.

Wavenumber–frequency spectra for (left) OLR and (right) 850-hPa zonal winds in (top) observations, (middle) CFSv1, and (bottom) CFSv2. The unit is W2 m−4 day for OLR and m2 s−2 day for 850-hPa zonal winds.

Fig. 7.

Wavenumber–frequency spectra for (left) OLR and (right) 850-hPa zonal winds in (top) observations, (middle) CFSv1, and (bottom) CFSv2. The unit is W2 m−4 day for OLR and m2 s−2 day for 850-hPa zonal winds.

For OLR the CFSv2 does not fare much better save for a reduced amplitude; however, for 850-hPa zonal winds there is significant improvement over its predecessor in depicting the structure and amplitude of propagating components in 850-hPa zonal winds, as evidenced by a significant reduction in amplitude at wavenumber 1 (as compared to CFSv1) and a shift toward higher frequency. This is further supported by the spatial representation of enhanced tropical intraseasonal variability in 850-hPa zonal winds in the CFSv2 (Fig. 2, bottom panel).

5. Interannual variability

MJO variability has the potential to be significantly modulated by interannual variations of the climate system and, in particular, by ENSO given the colocation of the MJO and ENSO climate variability modes over the Indo-Pacific region. This modulation evidently occurs as a result of modification of the background state of convection, winds, and temperature due to ENSO through which the MJO propagates (Roundy et al. 2010). Observed characteristics of the MJO–ENSO interaction are also highlighted by variations in propagation speed (Pohl and Matthews 2007), zonal extensions and retractions of the MJO envelope (Kessler 2001), and implications for ENSO prediction (Zhang and Gottschalck 2002). Given that the coupled models analyzed here are used for the extended range (weeks 2–4) and seasonal (up to 9 months) prediction efforts that rely critically on the skill of ENSO prediction, it is necessary to understand the characteristics of MJO variations during the varying phases of ENSO and the fidelity of the CFSv1 and CFSv2 in capturing such behavior.

To separate the impacts of ENSO on the structure of MJO variations we first stratify the daily anomaly fields by positive and negative ENSO phases. The classification is gleaned from the Climate Prediction Center (CPC) historical Niño-3.4 (5°N–5°S, 170°–120°W) indices from 1982 to 2006 (http://www.cpc.ncep.noaa.gov/products/precip/CWlink/MJO/enso.shtml#history). The index is calculated as a 3-month running mean. To be considered an ENSO winter the Niño-3.4 index is required to be larger than ±0.5 for the running mean seasons from October–December (OND) through February–April (FMA). According to this definition there were six positive ENSO winters—1982/83, 1986/87, 1991/92, 1994/95, 1997/98, and 2002/03—and five negative ENSO winters—1984/85, 1988/89, 1995/96, 1998/99, and 1999/2000.

Given the focus in this section on the characteristics of the MJO simulation as a function of ENSO phase, it is necessary to examine the degree to which both versions of the CFS can predict the evolution of ENSO events. Shown in Fig. 8 is the composite Niño-3.4 index for the chosen El Niño and as predicted by the four-member ensemble mean for the October starts during El Niño and La Niña events from the observations (black), CFSv1 (red), and CFSv2 (blue) as a function of monthly lead time. Positive (negative) values denote an El Niño (La Niña). Both versions of the CFS show skill in predicting the Niño-3.4 index at a several month lead time with the CFSv2 tracking the Niño-3.4 index more closely.

Fig. 8.

Composite of Niño-3.4 SST index (K) for the CFSv1 (red), CFSv2 (blue), and observations (black). The Niño-3.4 index is calculated using a four-member ensemble average for the defined ENSO events. Forecasts are from 9 to 12 Oct for CFSv1 and from 8 Oct for CFSv2.

Fig. 8.

Composite of Niño-3.4 SST index (K) for the CFSv1 (red), CFSv2 (blue), and observations (black). The Niño-3.4 index is calculated using a four-member ensemble average for the defined ENSO events. Forecasts are from 9 to 12 Oct for CFSv1 and from 8 Oct for CFSv2.

a. ENSO mean state

Given the strong dependence of MJO simulation on the mean state system (Zhang 2005; Waliser et al. 2009) in Figs. 9 and 10 we first assess the mean features of the low-level winds and precipitation during El Niño (Fig. 9) and La Niña (Fig. 10) years. This provides context for the following analysis with regard to potential differences in background state as a function of ENSO phase.

Fig. 9.

As in Fig. 1 but for El Niño years.

Fig. 9.

As in Fig. 1 but for El Niño years.

Fig. 10.

As in Fig. 1 but for La Niña years.

Fig. 10.

As in Fig. 1 but for La Niña years.

The mean background state during El Niño exhibits stronger precipitation over the central Pacific, not surprising given the warm SST anomaly there during a positive ENSO event. Additionally, the fetch of mean low-level zonal winds exhibits considerable eastward extension [Fig. 1 provides the context for this assessment; also see Kessler and Kleeman (2000)]. Both versions of the CFS generally represent well the mean features during El Niño, save for elevated precipitation amplitude (as noted in Fig. 1), and a more eastward-positioned low-level wind as depicted by the extension of the zero line (i.e., the extent of the low-level zonal westerlies) by ~20° of longitude.

The La Niña depiction (Fig. 10) is quite different, characterized by a retraction of the mean low-level westerly winds into the Indian Ocean with a more horseshoelike precipitation footprint. The low-level westerlies over the Indian Ocean are strong and extend to just east of the Maritime Continent. The terminus of the mean westerlies is some 30°–40° of longitude to the west of the El Niño state. The CFS model simulations again capture the general spatial representation of the low-level winds and precipitation, as in El Niño years, albeit with slightly weaker amplitude over the Maritime Continent.

b. ENSO-filtered lag correlations

MJO propagation characteristics as a function of ENSO phase are examined in this section via lagged correlations of the 20–100-day bandpass filtered daily anomalies of precipitation and 850-hPa zonal winds, as in Figs. 3 and 4. Given our focus on ENSO modulation, the lag correlations are performed against a western Pacific precipitation index, calculated as the area-averaged precipitation in the 10°S–10°N, 160°–185°E domain.2

Figures 11 and 12 show the observed representation (top panel) and the CFSv1 (middle) and CFSv2 (bottom) during El Niño and La Niña years, respectively. Striking differences in the propagation features are noted between El Niño and La Niña years. During El Niño the intraseasonal anomalies of precipitation and 850-hPa zonal winds propagate coherently through the Indian Ocean, across the Maritime Continent, and into the western Pacific, separating as they enter the eastern Pacific domain. During La Niña years the propagation of the precipitation and 850-hPa zonal wind anomalies appears to be thwarted as they enter the western Pacific, perhaps a consequence of the negative feedback via coupling of the atmosphere to a cold ENSO SST anomaly.

Fig. 11.

Lag correlations of 20–100-day filtered 10°S–10°N averaged precipitation (shaded) and 850-hPa zonal winds (contoured) with respect to the west Pacific precipitation index for El Niño years. (top) CFSR 850-hPa zonal winds and GPCP precipitation, (middle) CFSv1 850-hPa zonal winds and precipitation, and (bottom) CFSv2 850-hPa zonal winds and precipitation. Warm (cold) colors indicate positive (negative) correlations for precipitation, while solid (dashed) lines indicate positive (negative) correlations for 850-hPa zonal winds and are contoured at 0.1 intervals.

Fig. 11.

Lag correlations of 20–100-day filtered 10°S–10°N averaged precipitation (shaded) and 850-hPa zonal winds (contoured) with respect to the west Pacific precipitation index for El Niño years. (top) CFSR 850-hPa zonal winds and GPCP precipitation, (middle) CFSv1 850-hPa zonal winds and precipitation, and (bottom) CFSv2 850-hPa zonal winds and precipitation. Warm (cold) colors indicate positive (negative) correlations for precipitation, while solid (dashed) lines indicate positive (negative) correlations for 850-hPa zonal winds and are contoured at 0.1 intervals.

Fig. 12.

As in Fig. 11 but for La Niña years.

Fig. 12.

As in Fig. 11 but for La Niña years.

The model responses are mixed. Both versions of the CFS capture quite well the propagation features over the Eastern Hemisphere during El Niño—however, tend to speed up the propagation of 850-hPa zonal wind anomalies upon crossing the date line. The situation during La Niña is not as clear. Neither version of the CFS seems to coherently propagate the 850-hPa zonal wind and precipitation anomalies over the Indian Ocean as in the observed depiction. This indicates that the simulated intraseasonal activities in the western Pacific are less connected to the variability in the Indian Ocean than in the observed, a manifestation of the model’s failure to propagate across the Maritime Continent. The simulated features east of the date line do exhibit some degree of fidelity, mostly in the wind field. The interruption to the eastward-propagating intraseasonal anomalies near the date line is captured to some degree, more so in the CFSv2.

c. Regional power spectra

Figure 13 shows the power spectra over the Indian Ocean (as in Fig. 5) for 850-hPa zonal winds (top row) and OLR (bottom row) for observations (left), CFSv1 (middle), and CFSv2 (right) as a function of ENSO phase. ENSO positive and negative phases are shown by the red and blue lines, respectively. Only minor differences emerge in the observed power spectra as a function of ENSO phase over the Indian Ocean. In general, the spectral density in 850-hPa zonal winds is quite similar regardless of ENSO phase; however, during La Niña the OLR spectra reaches a stronger intraseasonal peak and at lower frequencies than during an El Niño winter.

Fig. 13.

Indian Ocean power spectra of (top) 850-hPa zonal winds and (bottom) OLR as a function of El Niño (red) and La Niña (blue) phases. The 850-hPa zonal winds are area averaged over the domain of 15°S–0°, 70°–100°E while OLR is area averaged over 10°S–5°N, 70°–100°E for the (left) observations, (middle) CFSv1, and (right) CFSv2. The unit is W2 m−4 day for OLR and m2 s−2 day for 850-hPa zonal winds.

Fig. 13.

Indian Ocean power spectra of (top) 850-hPa zonal winds and (bottom) OLR as a function of El Niño (red) and La Niña (blue) phases. The 850-hPa zonal winds are area averaged over the domain of 15°S–0°, 70°–100°E while OLR is area averaged over 10°S–5°N, 70°–100°E for the (left) observations, (middle) CFSv1, and (right) CFSv2. The unit is W2 m−4 day for OLR and m2 s−2 day for 850-hPa zonal winds.

Aside from the relative amplitude during the various ENSO phases, both CFSv1 and CFSv2 do not faithfully represent the ENSO phase dependence, with both versions of the model placing too much emphasis on the lower portion of the intraseasonal frequency band regardless of ENSO phase, similar to the full 25-yr record depiction (Fig. 5). This feature of a preference for lower frequency in the CFS has been previously noted (Seo and Wang 2010). To be sure, some modest improvement can be claimed in the CFSv2 representation, as evidenced by a systematic reduction of amplitude in the power spectra of 850-hPa zonal winds and OLR when comparing CFSv1 and CFSv2, especially during La Niña winters.

The observed power spectra of 850-hPa zonal winds (15°S–0°, 165°–190°E) and OLR (20°–5°S, 160°–185°E) over the western Pacific in Fig. 14 shows much more significant variability as a function of ENSO phase, with El Niño winters exhibiting the strongest spectral density and, in the case of 850-hPa zonal winds, a peak at slightly lower frequency than that of La Niña phases. This is in general opposite to that occurring over the Indian Ocean. Furthermore, during El Niño the OLR and 850-hPa zonal winds do not have the same spectral peak, with 850-hPa zonal winds maximizing at a lower frequency than that of OLR. This characteristic is absent during La Niña. Another interesting feature when comparing the ENSO phase power spectra in the Indian Ocean (Fig. 13) and the western Pacific (Fig. 14) is the stronger amplitude of 850-hPa zonal winds in the higher frequency band (i.e., Ω > 30 days) during El Niño years in the west Pacific (Fig. 13). This is perhaps a manifestation of the increased Kelvin wave activity over the western Pacific observed during ENSO years, which has been shown to impact the development characteristics of El Niño (Zhang and Gottschalck 2002). As in the Indian Ocean representation for OLR, the relative amplitude of the power spectra as a function of ENSO phase is captured by both versions of the CFS for both OLR and 850-hPa zonal winds.

Fig. 14.

As in Fig. 13 but over the western Pacific domain of 15°S–equator, 160°–185°E for 850-hPa zonal winds and 20°–5°S, 160°–185°E for OLR.

Fig. 14.

As in Fig. 13 but over the western Pacific domain of 15°S–equator, 160°–185°E for 850-hPa zonal winds and 20°–5°S, 160°–185°E for OLR.

The stronger observed intraseasonal OLR during La Niña without a similar signature in 850-hPa winds in the Indian Ocean is an interesting finding (Fig. 13). One possible explanation is that in observations the moisture convergence that leads to convection is due to the convergence of moisture anomalies by the mean zonal winds (not the anomalies), which are stronger during La Niña over the Indian Ocean (Fig. 10), and that the feedback of the convection on to the anomalous wind field is marginal. This potential mechanism does not occur in the CFSv1 as intraseasonal anomalies of both 850-hPa zonal winds and OLR are both stronger during La Niña. The coherence of these fields over the western Pacific Ocean between the observations and simulations (Fig. 14) with respect to the relative amplitudes of 850-hPa zonal winds and OLR suggest that in this region the moisture convergence is potentially more related to the convergence of mean moisture by wind anomalies (not the mean).

d. Discussion

The analysis presented in the previous section brings to light an interesting situation. It appears that both versions of the CFS seem to simulate MJO features with greater fidelity over the western Pacific as opposed to the Indian Ocean. The propagation of anomalies with respect to an Indian Ocean reference point in Fig. 4 clearly shows the slow propagation over the Indian Ocean and MJO termination over the Maritime Continent in the simulations for the entire 1982–2006 time period. When we switch to the ENSO phase dependence (i.e., Figs. 11 and 12) as a function of a western Pacific reference point, we again see the propagation termination near the Maritime Continent, especially in the La Niña case. Such a difficulty for MJO events initiated over the Indian Ocean to traverse the Maritime Continent barrier has been noted in various simulation studies (Inness et al. 2003a; Inness and Slingo 2006; Vitart et al. 2007; Vintzileos and Pan 2007; Seo and Wang 2010). One possible reason for this difficulty is errors in modeling the MJO air–sea interaction over the Indian Ocean (Zhang et al. 2006). Although the simulated propagation in positive ENSO years is more realistic in general (Fig. 11), one could argue that the much weaker Indian Ocean correlations at negative lags in the El Niño years also hint at the propagation problem through the Indian Ocean and Maritime Continent.

While it is outside the scope of this examination to conduct model sensitivity tests to understand the simulation differences between El Niño and La Niña events, those differences, nonetheless, are intriguing given the relatively sound representation in the simulations of the mean state through which the MJO propagates. It appears that it is not modification of the background state during El Niño or La Niña (Figs. 9 and 10) that directly impacts the simulated propagating features of the MJO—however, it is more likely modeling deficiencies with regard to MJO propagation through the Indian Ocean and Maritime Continent. Inness and Slingo (2006) proposed that the unrealistic orographic effect in the models is responsible for the discontinuity of the propagation across the Maritime Continent, while Seo and Wang (2010) showed that the convection parameterization could be the reason. Perhaps the intensive observations over the Indian Ocean gathered from the upcoming Dynamics of the Madden-Julian oscillation (DYNAMO) campaign will help to shed further light on these issues.

6. Summary and conclusions

Wintertime MJO variability is assessed in NCEP reanalyses (R2 and CFSR) and simulations from two versions of the Climate Forecast System, CFSv1 and CFSv2. The assessment is important for both furthering our understanding of the physical mechanisms that give rise to MJO fluctuations and the ability of current climate prediction systems at NCEP in simulating aspects of its behavior. The MJO has significant impacts on the climate system in nature, influencing ENSO, tropical cyclogenesis, storm-track variations, and monsoon systems, to name a few. Given the interactions of the MJO with other components of the climate system (i.e., ENSO), the MJO has the potential to impact global weather features via modulation of tropical teleconnections and potential enhancement or disruption of the climate prediction signal coming from the interannual variability. As such, it is a vital component to the intraseasonal-to-interannual climate prediction strategy at NCEP.

The new NCEP CFSR shows improvement over its predecessor (R2) in representing aspects of the MJO, including proper phasing and propagation features of intraseasonal variability. This is apparently a reflection of the numerous model upgrades that have occurred since the last NCEP reanalysis. Some modest improvements in the new version of the CFS have also been noted, especially in representing the multivariate EOF structure of the MJO. To be sure, some aspects of the MJO simulation continue to be challenging for this class of climate models, especially with regard to spectral OLR components and propagation through the Indian Ocean during La Niña events.

However, through inspection of the differences of MJO variations as a function of ENSO phase both versions of the CFS appear to simulate the intraseasonal variability quite well over the Pacific region, including the much stronger intraseasonal amplitude over the west Pacific and more coherent propagation features during El Niño years. An investigation of the El Niño and La Niña mean states reveals that the difficulty in simulating MJO behavior during La Niña is likely a reflection of the preference for MJO variability to occur over the Indian Ocean in a La Niña event, where the strong mean state resides, as opposed to some intrinsic interaction of La Niña with the other features of the climate system (i.e., a tropical teleconnection). Air–sea coupling over the Indian Ocean and difficulty for models to propagate the MJO from the Indian Ocean over the Maritime Continent are a known bias for many climate models, including the NCEP CFS, and are continuing to be the likely culprit.

It is not possible in this analysis to specifically determine the root causes of the continued deficiency in MJO simulations over the Indian Ocean, or alternatively their robust simulation during El Niño in the west Pacific, given the numerous changes between modeling systems. Future research will target further understanding of these particular aspects including sensitivity to intraseasonal SST anomalies over the Indo-Pacific warm pool, air–sea interactions, the formulation of convective parameterization schemes, and issues with the Maritime Continent barrier.

Despite only modest advancements in the CFS MJO simulations, the notable gains in representing the MJO in the new NCEP–CFSR reanalysis system offer the prospect for significant improvement in near-term (i.e., ~30 days) prediction of the MJO given the noticeable progress in MJO-related propagation characteristics in the CFSR-only depiction as compared to that from R2 (Fig. 3), coupled with the prospects for improved initial conditions offered by the CFSR. The prediction skill of the MJO in the 1–4-week timeframe in the two versions of the NCEP coupled climate models is currently under active investigation and will be reported on in a forthcoming paper.

Acknowledgments

The authors wish to thank Drs. Emily Becker and Hui Wang for reviewing an early version of the manuscript, two anonymous reviewers for their insightful comments, and Dr. Shang-Ping Xie for his editorial guidance.

REFERENCES

REFERENCES
Gottschalck
,
J.
, and
Coauthors
,
2010
:
A framework for assessing operational model MJO forecasts: A CLIVAR MJO working group project
.
Bull. Amer. Meteor. Soc.
,
91
,
1247
1258
.
Inness
,
P. M.
, and
J. M.
Slingo
,
2006
:
The interaction of the Madden–Julian oscillation with the maritime continent in a GCM
.
Quart. J. Roy. Meteor. Soc.
,
132
,
1645
1667
.
Inness
,
P. M.
,
J. M.
Slingo
,
E.
Guilyardi
, and
J.
Cole
,
2003a
:
Simulation of the Madden–Julian oscillation in a coupled general circulation model. Part I: Comparisons with observations and an atmosphere-only GCM
.
J. Climate
,
16
,
345
364
.
Inness
,
P. M.
,
J. M.
Slingo
,
E.
Guilyardi
, and
J.
Cole
,
2003b
:
Simulation of the Madden–Julian oscillation in a coupled general circulation model. Part II: The role of the basic state
.
J. Climate
,
16
,
365
382
.
Kanamitsu
,
M.
,
W.
Ebisuzaki
,
J.
Woollen
,
S.-K.
Yang
,
J. J.
Hnilo
,
M.
Fiorino
, and
G. L.
Potter
,
2002
:
NCEP–DOE AMIP-II Reanalysis (R-2)
.
Bull. Amer. Meteor. Soc.
,
83
,
1631
1643
.
Kaylor
,
R. E.
,
1977
:
Filtering and decimation of digital time series
.
Institute of Physical Science Technology Tech. Note BN 850, 42 pp
.
Kessler
,
W. S.
,
2001
:
EOF representations of the Madden–Julian oscillation and its connection with ENSO
.
J. Climate
,
14
,
3055
3061
.
Kessler
,
W. S.
, and
R.
Kleeman
,
2000
:
Rectification of the Madden–Julian oscillation into the ENSO cycle
.
J. Climate
,
13
,
3560
3575
.
Kim
,
D.
, and
Coauthors
,
2009
:
Application of MJO simulation diagnostics to climate models
.
J. Climate
,
22
,
6413
6436
.
Liebmann
,
B.
, and
C. A.
Smith
,
1996
:
Description of a complete (interpolated) outgoing longwave radiation dataset
.
Bull. Amer. Meteor. Soc.
,
77
,
1275
1277
.
Lin
,
H.
, and
G.
Brunet
,
2009
:
The influence of the Madden–Julian oscillation on Canadian wintertime surface air temperature
.
Mon. Wea. Rev.
,
137
,
2250
2262
.
Lin
,
J.-L.
, and
Coauthors
,
2006
:
Tropical intraseasonal variability in 14 IPCC AR4 climate models. Part I: Convective signals
.
J. Climate
,
19
,
2665
2690
.
Madden
,
R. A.
, and
P. R.
Julian
,
1994
:
Observations of the 40–50-day tropical oscillation—A review
.
Mon. Wea. Rev.
,
122
,
814
837
.
Maloney
,
E. D.
, and
D. L.
Hartmann
,
2000
:
Modulation of eastern North Pacific hurricanes by the Madden–Julian oscillation
.
J. Climate
,
13
,
1451
1460
.
Pohl
,
B.
, and
A. J.
Matthews
,
2007
:
Observed changes in the lifetime and amplitude of the Madden–Julian oscillation associated with interannual ENSO sea surface temperature anomalies
.
J. Climate
,
20
,
2659
2674
.
Roundy
,
P.
,
K.
MacRitchie
,
J.
Asuma
, and
T.
Melino
,
2010
:
Modulation of the global atmospheric circulation by combined activity in the Madden–Julian oscillation and the El Niño–Southern Oscillation during boreal winter
.
J. Climate
,
23
,
4045
4059
.
Saha
,
S.
, and
Coauthors
,
2006
:
The NCEP Climate Forecast System
.
J. Climate
,
19
,
3483
3517
.
Saha
,
S.
, and
Coauthors
,
2010
:
The NCEP Climate Forecast System reanalysis
.
Bull. Amer. Meteor. Soc.
,
91
,
1015
1057
.
Seo
,
K.-H.
, and
W.
Wang
,
2010
:
The Madden–Julian oscillation simulated in the NCEP Climate Forecast System: The importance of stratiform heating
.
J. Climate
,
23
,
4770
4793
.
Vintzileos
,
A.
, and
H.-L.
Pan
,
2007
:
On the importance of horizontal resolution and initial conditions to forecasting tropical intraseasonal oscillations: The Maritime Continent prediction barrier
.
Extended Abstracts, NOAA CTB-COLA Joint Seminar Series. [Available online at http://www.nws.noaa.gov/ost/climate/STIP/CTB-COLA/Augustin_091907.htm.]
Vitart
,
F.
,
2009
:
Impact of the Madden–Julian oscillation on tropical storms and risk of landfall in the ECMWF forecast system
.
Geophys. Res. Lett.
,
36
,
L15802
,
doi:10.1029/2009GL039089
.
Vitart
,
F.
,
S.
Woolnough
,
M. A.
Balmaseda
, and
A. M.
Tompkins
,
2007
:
Monthly forecast of the Madden–Julian oscillation using a coupled GCM
.
Mon. Wea. Rev.
,
135
,
2700
2715
.
Waliser
,
D.
, and
Coauthors
,
2009
:
MJO simulation diagnostics
.
J. Climate
,
22
,
3006
3030
.
Wheeler
,
M. C.
, and
H. H.
Hendon
,
2004
:
An all-season real-time multivariate MJO index: Development of an index for monitoring and prediction
.
Mon. Wea. Rev.
,
132
,
1917
1932
.
Xie
,
P.
,
J. E.
Janowiak
,
P. A.
Arkin
,
R.
Adler
,
A.
Gruber
,
R.
Ferraro
,
G. J.
Huffman
, and
S.
Curtis
,
2003
:
GPCP pentad precipitation analyses: An experimental dataset based on gauge observations and satellite estimates
.
J. Climate
,
16
,
2197
2214
.
Zhang
,
C.
,
2005
:
Madden–Julian oscillation
.
Rev. Geophys.
,
43
,
RG2003
,
doi:10.1029/2004RG000158
.
Zhang
,
C.
, and
J.
Gottschalck
,
2002
:
SST anomalies of ENSO and the Madden– Julian oscillation in the equatorial Pacific
.
J. Climate
,
15
,
2429
2445
.
Zhang
,
C.
, and
M.
Dong
,
2004
:
Seasonality in the Madden–Julian oscillation
.
J. Climate
,
17
,
3169
3180
.
Zhang
,
C.
,
M.
Dong
,
S.
Gualdi
,
H. H.
Hendon
,
E. D.
Maloney
,
A.
Marshall
,
K. R.
Sperber
, and
W.
Wang
,
2006
:
Simulations of the Madden–Julian oscillation in four pairs of coupled and uncoupled global models
.
Climate Dyn.
,
27
,
573
592
.

Footnotes

1

The reference to “quasi-observed” reflects the uncertainty in reanalysis-based fields over sparsely observed areas (i.e., Indian Ocean) as they are highly model dependent there.

2

This area sufficiently captures observed maxima in u850 and precipitation variance for both phases of ENSO. The area is also suggested by the U.S. CLIVAR MJO Working Group (Waliser et al. 2009).