1. Introduction
General circulation models (GCMs) often fail to capture the finescale structures that affect regional climate due to their coarse resolutions of typically 100–300 km. This partially accounts for the deficiencies shown by GCMs in reproducing important aspects of mean convection and climate over the tropics. One way to represent regional climate information in GCMs is to use nested regional climate models with fine resolution. Alternatively, the GCM resolution can be enhanced considerably either globally or in a particular region by employing variable resolution. In a global GCM with very high resolution the regional–global scale interactions can be incorporated comprehensively. It can also provide regional climate information such as the land–sea distribution, lakes, soil characteristics, and orography at a finer spatial scale, and such simulations will be useful for local impact assessments.
Continuing advances in computing power enabled the use of GCMs with high resolution to address the impact of increase in horizontal resolution (e.g., Boer and Lazare 1988; Boyle 1993; Phillips et al. 1995; Williamson et al. 1995; Gualdi et al. 1997; Brankovic and Gregory 2001). However, subsequently, the vertical resolution also was found to play an important role (Lindzen and Fox-Rabinowitz 1989). Recently, Roeckner et al. (2006), performing a systematic horizontal and vertical resolution sensitivity study using seven horizontal resolutions up to T159 and with two sets of vertical resolutions (L19 and L31), suggested that only at the higher vertical resolution of L31 simulations showed gradual convergence and systematic improvement in seasonal climate simulations as the horizontal resolution is increased. Tompkins and Emanuel (2000) showed that the vertical distribution of water vapor in GCMs can be sensitive to the vertical resolution of a model. Pope et al. (2001) also addressed the impact of increased vertical resolution on the distribution of water vapor. Inness et al. (2001) studied the behavior of tropical convection in an aquaplanet GCM with the vertical resolution doubled in the free troposphere and around the tropopause, and discussed the implications of the results for Madden–Julian oscillation (MJO) (Madden and Julian 1971, 1994) simulation. These studies highlighted the need to use high vertical resolution along with high horizontal resolution. However, none of the studies, so far, addressed the impact of very high resolution on the simulation of mean convection and climate over the tropics. With this objective, we analyze the simulation of the Meteorological Research Institute/Japan Meteorological Agency (MRI/JMA) global atmospheric GCM (AGCM) at TL959L60 resolution, which corresponds approximately to a 20-km mesh horizontal grid with 60 vertical levels. The motivation for this study also comes from our intention to use a very high resolution coupled GCM for tropical climate studies. To have a realistic high-resolution coupled simulation of tropical convection and its variability, its atmospheric component must have the fidelity to represent the mean convection (Hendon 2000; Inness et al. 2003; Zhang et al. 2006).
Variability in convection over the equatorial belt plays a decisive role in the tropical mean convection and climate. Several studies suggest that equatorial convection does not occur randomly but is organized on selected spatial and intraseasonal time scales corresponding to convectively coupled large-scale equatorial waves such as Kelvin, equatorial Rossby (ER), mixed Rossby–gravity (MRG), and eastward and westward propagating inertio–gravity (EIG and WIG, respectively) waves (e.g., Nakazawa 1988; Takayabu 1994; Numaguti 1995; Wheeler and Kiladis 1999). In addition, over the near-equatorial regions, mean convection is significantly modulated by the MJO at 30–60-day time scale. Various studies have found that the MJO and equatorial waves are difficult to reproduce with GCMs (e.g., Slingo et al. 1996; Lin et al. 2006). The major shortcomings of MJO simulation in a majority of GCMs are found to be their inability to represent the dominance of 30–60-day time scales and to partition this variance relative to higher frequency variations and between eastward and westward propagating modes. The coherent eastward propagation and the phase relationship with the low-level moisture convergence are generally not well represented in models (e.g., Sperber 2004).
However, a reasonable representation of the MJO and equatorial waves, which is a measure of the fidelity of a model in representing the interaction between convection and large-scale dynamics, is important for a realistic simulation of mean climate over the tropics. Therefore, in addition to assessing the simulation of tropical mean convection in TL959L60, this study analyses its skill in simulating the organization of tropical convection on different spatiotemporal scales. Factors found to be important for the simulation of tropical convection are model physics, resolution, and air–sea coupling. Although the MJO simulation was reported to improve with resolution (e.g., Hayashi and Golder 1986; Kuma 1994; Gualdi et al. 1997; Inness et al. 2001; Duffy et al. 2003), its sensitivity to very high resolutions has not been addressed. The effect of resolution on the simulation of convectively coupled equatorial waves has, so far, not been investigated. Hence, in addition to the verification of the TL959L60 simulation against recent observation datasets, we also compared it with simulations at two lower resolutions of TL159L40 and TL95L40, so as to understand the effect of very high resolution on the simulation of tropical mean convection and its variability.
In section 2, we briefly discuss the details of the model, integrations, and validation datasets. The simulated mean convection and climatological characteristics over the tropics are presented in section 3. The effect of very high resolution on large-scale mean circulation over the tropics is discussed in section 4. Simulation of intraseasonal variability, convectively coupled equatorial waves, and the MJO are analyzed in section 5. In section 6, we summarize the results.
2. Model, integrations, and datasets
A global climate model with horizontal grid size of ∼20 km and 60 vertical levels, jointly developed by JMA and MRI (Mizuta et al. 2006), is used. The availability of the Earth Simulator, a massively parallel vector supercomputer, provided a unique opportunity to perform long (10 years or more) climate simulations at very high resolution, which conserve mass and are still based on the primitive equation system in hydrostatic approximation. The interaction among planetary-, synoptic-, and mesoscale phenomena and the topographic modulation of mesoscale precipitation can be explicitly represented in these very high resolution global models. This model became the operational global model for short-term numerical weather prediction at JMA starting November 2007.
The dynamical framework of the model consists of a full primitive equation system (Kanamitsu et al. 1983). At highest resolution, the model corresponds to triangular truncation 959 with a linear Gaussian grid (TL959) in the horizontal (the transform grid uses 1920 × 960 grid cells and grid size ∼20 km). The linear grid adopted in the model using a semi-Lagrangian scheme can suppress the aliasing effect arising from the quadratic Eulerian advection terms in addition to reducing the number of grid points (Hortal 2002). The model has 60 layers in the vertical with the model top at 0.1 hPa. The simulation at this resolution is hereafter referred to as TL959L60. At all resolutions, an Arakawa–Schubert scheme with prognostic closure similar to that of Randall and Pan (1993) is used for cumulus parameterization. The effect of entrainment and detrainment between the cloud top and cloud base in convective downdrafts, instead of reevaporation of convective precipitation (Nakagawa and Shimpo 2004), is also included. To improve the organization of tropical cyclones in high-resolution simulations, the effect of convective-scale pressure gradient forces (Wu and Yanai 1994; Gregory et al. 1997) is parameterized by reducing the vertical transport of horizontal momentum in the convection scheme. The time integration was accelerated by introducing a two-time-level semi-Lagrangian scheme (Yoshimura and Matsumura 2005).
The large resolution increase in TL959L60 was found to cause biases in resolution-dependent characteristics such as an increase in global-average precipitation and tropical upper-tropospheric temperature and a decrease in cloud amount. To reduce these biases, the model was tuned by changing the parameters in the evaporation process, cloud water content diagnosis, and gravity wave drag. As unresolved variance of water vapor within a model grid cell decreases with finer resolution, the subgrid-scale variance was reduced by 10% to slightly reduce the overestimation of the global-mean precipitation. Evaporation increases as inhomogeneity of the variables in a fixed larger domain increases with increased resolution. Hence, its estimation was reduced by 10%, which was found to reduce the evaporation by only a couple of percent of its long-term mean as negative feedback works against the modification. Cloud water detrainment at the top of cumulus convection and its speed of transformation to precipitation was reduced by 20%, consistent with the observed radiation balance, to increase the clouds. The gravity wave drag coefficient for short waves (λ < 100 km) was increased to reduce the excessive development of extratropical cyclones. Though this tuning helped to reduce biases, the basic parameters such as precipitation, temperature, and cloud amounts did not show any significant change with respect to their long-term means due to the complex feedbacks involving them.
The model was forced with monthly mean climatological SST and sea ice concentration [Reynolds and Smith (1994), averaged from 1982 to 1993]. A 10-yr integration was carried out after a spinup of 5.5 years. To study the relative role of model resolution, two additional 10-yr simulations were made at two lower resolutions of TL95 (∼180 km grid size) and TL159 (∼120 km grid size) with 40 vertical layers using the same boundary forcing (hereafter referred to as TL95L40 and TL159L40, respectively). Over the tropics, the highest simulation skill is realized during the boreal winter season when the global circulation is effective in communicating the impact of boundary anomalies to the tropical atmosphere (without interference from complex regional anomalies associated with the dominant Asian–African monsoon system). Further, MJO, the dominant mode of tropical convection variability, due to its pronounced seasonality in amplitude, is observed to be stronger, more coherent, and more frequent over the equatorial Eastern Hemisphere during November–April. Hence, our analysis focuses on the extended boreal winter season, November–April (hereafter referred to as the winter season).
Daily precipitation interpolated from Global Precipitation Climatology Project (GPCP) pentad data on a 2.5° × 2.5° grid for the period 1979–2002 are used for validating simulated daily mean precipitation. Daily mean circulation fields for 1979–2002 are estimated from 6-hourly 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40) data [1958–2002; Uppala et al. (2005)]. The primary reference and a full report documenting the dataset for the reanalysis can be found online (http://www.ecmwf.int/research/era/). Daily and twice-daily interpolated outgoing longwave radiation (OLR) from the National Oceanic and Atmospheric Administration for the period 1979–2005 are also used. The Tropical Rainfall Measuring Mission (TRMM) based latent heating profiles estimated using the Goddard profiling (GPROF) convective–stratiform heating (CSH) algorithm of Tao et al. (1994) for 1998–2006 have been used. In addition, Tropical Ocean Global Atmosphere Couple Ocean (TOGA) intensive field array (IFA) 6-hourly profiles of apparent heat sources for the period December 1992–February 1993 (Ciesielski et al. 2003) have also been used.
3. Mean convection over the tropics
Mean outgoing longwave radiation overlaid with corresponding 850-hPa winds for the winter season from the observations and simulations are shown in Fig. 1. The observed OLR indicates convection over the equatorial region stretching from the Indian Ocean to the date line including the South Pacific convergence zone (SPCZ), roughly from 15°S to 10°N. In TL959L60, simulated low OLR is realistic over the equatorial regions of the Indian Ocean and SPCZ but slightly overestimated over the western Arabian Sea. However, in terms of intensity, location, and distribution of convective activity, TL959L60 tends to give an improved simulation over the observed seasonal convective centers. For example, in TL959L60 the spatial extent of the region of OLR less than 220 W m−2, indicating sustained deep convection associated with the Australian–Indonesian monsoon, compares better with observations. Whereas in TL159L40 and TL95L40, the zonal extent of simulated OLR less than 240 W m−2 is much smaller and is mainly restricted to the northern parts of Australia, Indonesia, and the Maritime Continent of the western Pacific. Consistently, in TL959L60, the strength of the low-level circulation and the location of the convergence over major convective centers (equatorial regions of the Indian Ocean and the SPCZ) are closer to observations compared to TL159L40 and TL95L40. Although TL159L40 and TL95L40 capture the convection and associated low-level circulation with weaker magnitude over these regions, the convection over the equatorial Indian Ocean and SPCZ is significantly restricted. The resolution sensitivity is thus most pronounced over the equatorial Indian Ocean, the Maritime Continent, and the SPCZ.
Comparing the vertical moist static stability [VMS, the difference between gross moist static (GMS) enthalpy of the lower troposphere (1000–400 hPa) and that of the upper troposphere (400–50 hPa)], we note that the winter distributions of VMS from the observations and simulations (Fig. 2a) have the same pattern as for the large-scale distribution of the corresponding winter-mean organized convection (Fig. 1). This can be understood with the simple model suggested by Neelin and Held (1987) based on a moist static energy budget. Rajendran et al. (2002) used this simple model to demonstrate the association between variations in VMS and precipitation/convection. According to this model, increased convection is associated with increased atmospheric instability and hence a decrease in VMS. The zonally oriented region of maximum instability along 10°S is confined to the Maritime Continent in lower-resolution simulations, while it is stronger and more extensive across the Asian Pacific and compares better with observations in TL959L60. The change in VMS can be understood in terms of changes in low-level moisture convergence.
Observed and simulated winter-mean moisture convergence integrated over 1000–600 hPa are shown in Fig. 2b. The regions with strong convection are associated with strong large-scale moisture convergence in the lower troposphere. In TL959L60, as in observations, a zonally oriented belt of strong moisture convergence with relative maxima on either side of the equator extends from the Indian Ocean to the Pacific. Corresponding to the overestimated convection over the western Indian Ocean along the east African coast (Fig. 1), TL959L60 overestimates moisture convergence over the narrow longitudinal band along the coast. This is associated with overestimated low-level wind convergence. Similar bias is not evident in the mean VMS distribution (Fig. 2a). The regions of decreasing (increasing) VMS are associated with regions of stronger (weaker) moisture convergence in all simulations. Enhanced moisture convergence in one region increases the moisture in the air column above it, reducing its static stability (Neelin and Held 1987). Associated with this reduced stability is the enhanced convective activity, stronger ascent, and precipitation over that region. This in turn changes the circulation pattern, resulting in subsidence over surrounding areas where weaker ascent (or stronger descent) causes the atmosphere to be more stable, thus inhibiting organized convective activity and precipitation. Although, the three simulations are similar in representing the large-scale features of strong moisture convergence over the tropics, there is significant lack of moisture convergence over the equatorial Indian Ocean at lower resolution. In addition, over the SPCZ, the low-level moisture convergence is weaker compared to the observations and TL959L60. Correspondingly, the simulations at lower resolutions show weaker convection and low-level circulation (Fig. 1) over these regions.
The Taylor diagram (Taylor et al. 2000) shown in Fig. 3 summarizes the validation of the climatological mean distribution of OLR, precipitation (PR), precipitable water content (PWAT), mean sea level pressure (SLP), surface temperature (TA), and zonal wind at 200 hPa (U2) and 850 hPa (U8) over the global tropics (30°S–30°N, 0°–360°). This quantifies the degree to which the correlation with the observations increases, errors decrease, and the model converges at different resolutions. The basic fields of SLP, and PWAT, TA, and PR that are relevant for radiative balance do not vary significantly in root-mean-square error among the simulations. This indicates that the tropical mean winter patterns are robust features of the model. But, these statistical results indicate that, when the horizontal and vertical resolution is increased in TL959L60, the climate simulation is improved either by slightly reducing the rms error (RMSE) and/or by improving the correlations.
4. Large-scale circulation
Equatorially averaged zonal wind overlaid with corresponding (u, ω) vectors from TL959L60, TL159L40 and the difference between them are compared with observations in Fig. 4. The simulated winter-mean Walker circulation along the equatorial region shows that the location, zonal extent, and vertical depth of the descending limbs over the Western Hemisphere are realistic in TL959L60. The difference between TL959L60 and TL159L40 significant at 95% (Fig. 4d) shows the strengthening and widening of the core of the ascending limb longitudinally and vertically in TL159L40. Compared to observations, in the Eastern Hemisphere where the first baroclinic structure and the strength of the ascending limb tends to be overestimated in TL159L40, it is slightly underestimated in TL959L60. In contrast, the upper-level circulation in the descending limb is slightly overestimated (underestimated) in TL959L60 (TL159L40).
The vertical structures of the simulated Walker circulation with the two lower resolutions of TL159L40 and TL95L40 (not shown) are generally similar, suggesting that the model captures the Walker circulation reasonably well at all resolutions and a change in resolution alone does not lead to any significant improvement in the simulated Walker circulation. The relatively small difference between the simulated vertical structures of the Walker circulationfor TL159L40 and TL95L40 (not shown) suggests that the model captures the Walker circulation reasonably well at all resolutions and a change in resolution alone does not lead to any remarkable change in its simulation.
Another factor having significant impact on tropical circulation is the spatial distribution and vertical structure of latent heating from organized deep convection. For example, the spatial gradient of diabatic heating is primarily responsible for the thermally direct tropical circulation cells (Rind and Rossow 1984). Longitude–height profiles of the December–February (DJF) mean convective heating from (i) TRMM and spatially averaged profiles from TOGA IFA, as well as (ii) TL959L60 and (iii) TL159L40, and the difference between them, are shown in Fig. 5. In spite of the difference in the magnitude of heating, the vertical variation and height of the heating maxima are similar in TRMM and TOGA. It is to be noted that the TRMM-based estimates are derived using the GPROF algorithm. In comparison, the vertical structure of the simulated moist convective heating, which is one of the two main components of the total diabatic heating associated with the Walker circulation, is found to differ from the two observations not only in magnitude but in the location and extent of the upper-level heating maximum. In both models, two relative heat source maxima are located over the convective centers of the intertropical convergence zone. In TL959L60, maximum heating above the boundary layer is between 500 and 400 hPa. The heating rate decreases with depth below 700 hPa and thereafter increases in the boundary layer with a relative maximum around 850 hPa. In comparison, TL159L40 has weaker heating at all levels (in turn associated with less active deep convection, as seen in Fig. 1) with shallower heating maxima located at slightly lower levels above 700 hPa. The entire Eastern Hemisphere to the Maritime Continent (60°–140°E) appears to have weaker heating profiles where significant differences occur between the two simulations (Fig. 5c). The significant differences between the two simulations indicate the impact of increased resolution. However, the most noticeable feature is the large difference between the simulated heating structure and the observations, particularly in the height and extent of the upper-level heating maximum.
5. Intraseasonal variability of tropical convection
Standard deviations of 6-hourly winter precipitation on time scales ≤1 day from TL959L60 and TL159L40 are shown in Fig. 6. We have used precipitation due to its binary nature while the corresponding diurnal variance for the OLR shows similar features. There are distinct differences in diurnal variance between land and ocean. The diurnal variance is largest over land regions, as a result of the large response to daytime solar heating. Significant variance also occurs over seasonal convective regions over land (e.g., the Maritime Continent), consistent with the convective forcing by the diurnal cycle in land surface heating, which spreads to adjacent oceans and is indicative of complex land–sea-breeze effects (Arakawa and Kitoh 2005). Over the oceans there is smaller variance associated with the SPCZ and the ITCZ. Both simulations are broadly consistent and correspond well with the observed diurnal variance suggested by previous studies (Janowiak et al. 1994; Chang et al. 1995; Yang and Slingo 2001). Since the diurnal variance is important for the fluctuations in surface fluxes and the initiation of convection over the tropics, its realistic representation is indicative of a reasonable interaction among the various physics schemes in the model (Arakawa and Kitoh 2005).
In TL959L60, the amplitude is strengthened over Australia and the Maritime Continent. Over the equatorial Indian Ocean, the diurnal variance seen in TL959L60 is absent in TL159L40 (which is deficient in simulating low-level moisture convergence and convection, as shown in Figs. 1 and 2b). The impact of resolution is also marginally evident in the coherent diurnal variation linked to the response of the surface atmosphere system to the diurnal cycle in the solar radiation, which is computed as the seasonal mean for four times of the day (from 6-hourly data). In TL959L60, the diurnal variation over these regions (averaged over two boxes in Fig. 6) is only marginally amplified compared to TL159L40.
The standard deviation of daily winter OLR from the observations, TL959L60, and TL159L40 (Fig. 7a) shows that regions of largest OLR variability in the observations are coincident with regions of deep convection. But, the models tend to have the largest variability along the periphery of the mean convective maxima (Fig. 1). The percentage of total OLR variance explained by the 20–90-day time scale during the winter season from observations and the two simulations are shown in Fig. 7b. The maximum intraseasonal variance is located over major seasonal convective centers and over regions with large total variance. While TL159L40 appears to be closer to observations in simulating the total variance, the percentage of the intraseasonal variance is strengthened and compares better with the observations in TL959L60, especially over the Indian Ocean and the monsoonal areas of northern Australia and Indonesia. However, both simulations tends to underestimate intraseasonal variance.
a. Convectively coupled equatorial waves
Other modes that are important for producing prominent intraseasonal fluctuations over the tropics include the convectively coupled pure shallow-water-like equatorial waves (e.g., Takayabu 1994; Numaguti 1995; Wheeler and Kiladis 1999). A realistic representation of these waves is essential for the prediction of variability at all time scales. Failure of GCMs to capture these waves is largely due to inadequacies in cumulus parameterization schemes and representation of the interaction between the physics and dynamics, arguably the most fundamental errors.
Figure 8 shows the symmetric and antisymmetric OLR (defined with respect to the equator) power spectra from the observations, TL959L60, and TL159L40 as a ratio of OLR power to the background power, to emphasize the dominant spatial and time scales. The space–time spectra of OLR are computed in the same manner as in Wheeler and Kiladis (1999). Theoretical dispersion curves for equatorial wave modes (Matsuno 1966) are drawn for equivalent depths of 8, 12, 25, 50, and 90 m. Signals of Kelvin and ER waves and the MJO are seen in the observed symmetric spectrum (Fig. 8a) and MRG and EIG waves in the antisymmetric spectrum (Fig. 8b). In TL959L60 (Figs. 8c and 8d), the organization of the OLR spectral variance shows a resemblance to Kelvin, ER, EIG, and MJO. However, the amplitude for MJO is much weaker than for observations and does not show any relative peak at wavenumbers 1–2. Also, it appears more as a standing oscillation where the westward power is dominant in the corresponding low-frequency band. For the ER dispersion curve, the westward power is stronger and broader, extending to other equivalent depths as well. The model fails to simulate the MRG mode, which is important for cyclogenesis. However, for Kelvin waves, the variance matches with observations, suggesting that the model captures the wave–heating feedback associated with it and, hence, its phase speed.
In comparison, Kelvin and EIG waves in TL159L40 (Figs. 8e and 8f) are not as comparable to the observations in terms of their variance and scaling to corresponding equivalent depths. For example, in TL959L60 EIG power is larger and that corresponding to the Kelvin waves (and its equivalent depth) matches better with observations. Significant improvement occurs for Kelvin waves along with their scaling to realistic equivalent depths (with slightly slowing speed) in TL959L60. One factor that slows the Kelvin wave is moist convection, which reduces the effective static stability of the atmosphere. Hence, slower Kelvin waves and reduced static stability (Fig. 2b) indicates that moist convection is better represented in TL959L60. Alternately, the mountain ranges of South/Central America and East Africa, which were better resolved in TL959L60, can also slow Kelvin waves from a fast “dry” phase speed to a speed closer to observations, as suggested by Matthews (2000). These aspects and their role in the simulation of equatorial waves are being investigated in a separate study. The MJO and ER modes in TL159L40 grossly remain as in TL959L60. Thus, as in the case of the seasonal mean, increasing the resolution only contributes to improving some aspects of equatorial waves.
The MJO comprises hierarchical substructures of eastward-moving super cloud clusters (SCCs with horizontal scale of several thousand kilometers) and westward-moving cloud clusters (CCs of several hundred kilometers) [Nakazawa (1988); Wheeler and Kiladis (1999)]. To investigate the hierarchical substructures within the MJO and their associated wave interrelationship, MJO, Kelvin, and ER waves are isolated by filtering OLR for specific regions of the wavenumber–frequency diagram (Fig. 8). Hovmöller diagrams of 7.5°S–7.5°N averaged filtered waves and the unfiltered OLR for a representative 6-month period from the observations, TL959L60, and TL159L40 are shown in Fig. 9. Here MJO envelopes are delineated usinge a threshold of −10 W m−2 (−5 W m−2) in the observations (simulations) as the models underestimate the MJO considerably (Fig. 8). In the observations, the MJO envelopes occasionally encompass successive Kelvin waves and intruding ER waves, which flares the active convection (Masunaga et al. 2006). The isolated eastward MJO in TL959L60 shows reasonable wave interaction among the three waves. The pattern resembles the observed hierarchical movement structure of cloud clusters. In TL159L40 the wave association is not as comparable to the observations as in TL959L60. This is mainly because Kelvin waves appear to have slightly faster speed and different spatial scales than both observations and TL959L60. Consequently, the superposed unfiltered OLR anomaly is also weaker in TL159L40. However, the unfiltered OLR anomalies in the simulations clearly show the fundamental deficiency in both models in underestimating the MJO variance, its reduced spatial extent, and coherent eastward propagation.
b. Structure and propagation characteristics of MJO
To investigate the structure of the simulated MJO, space–time spectral analysis (Hayashi 1982) is applied on detrended (by removing the first three harmonics of the long-term annual cycle) OLR anomalies for the winter season. The spectra at each grid between 10°S and 10°N for each year are estimated and then averaged over the entire grid-box ensemble for 10 years. Figure 10 shows the spectra from the observations and two simulations. The observed spectrum shows an eastward propagating signal at intraseasonal (30–90 days) and planetary scales (wavenumber 1–2), with much higher eastward power than the corresponding westward power. In the simulations, intraseasonal spectral peaks exist but not exactly at observed periodicities. In TL959L60, the broad spectral maximum extends from wavenumbers 1–2 within an eastward time scale of about 45–90 days. But the low-frequency intraseasonal power is dominant at westward frequencies and wavenumbers 1–2. However, in TL159L40, the space–time distribution of intraseasonal power is slightly better at reducing the westward power and also extending the eastward power to higher frequencies for wavenumber 1. This picture gives additional insight into assessing the MJO simulations using the two versions. For basic characteristics of the MJO, the increase in resolution does not contribute to any improvement whereas TL159L40 compares slightly better with the observations.
Furthermore, the dominant structure associated with the MJO was derived based on extended empirical orthogonal function (EEOF) analysis (Weare and Nasstrom 1982; Waliser et al. 2003) of velocity potential (χ) at 200 hPa (hereafter χ200) using temporal lags of 15 pentads (from −7 to +7 pentads) during the winter season over the region 30°S–30°N, 0°–180°. A 241-point Lanczos bandpass filter (Duchon 1979) is applied to the datasets after detrending by removing the mean annual cycle and the first three harmonics to separate the 20–90-day time scale. This region during the winter season is associated with the maximum amplitude of MJO variability. The upper-level χ was chosen to study the MJO because the oscillation has a strong signature in this field, and hence it is advantageous for the simulations for which much weaker MJO variability is compared to the observations. In addition, χ is not a locally measurable quantity; it is an inverse Laplacian of the divergence field and can be regarded as a overly smoothed divergence field that overemphasizes the larger scales. The χ field provides a first-order description of the gross repeatable features of the oscillation, indicating the large-scale divergent circulation features associated with convective disturbances. Therefore it is a good choice for the comparison of the basic characteristics of the oscillation obtained from different simulations.
Using the composite MJO event, it is possible to make a more critical assessment of the model’s ability to simulate the MJO. Dominant MJO events for compositing were then chosen from the amplitude of the EEOF-1 time series. The events in the observations (simulations) with peak amplitude smaller than −1 (−0.75) standard deviation were selected as the dominant events. According to this criteria, there were 16 events for the observations, 14 events for TL959L60, 10 events for TL159L40, and 9 events for TL95L40. This indicates a slightly higher occurrence of eastward propagating events in TL959L60. For each selected event, the corresponding 15-pentad filtered anomalies of χ200 and OLR are extracted. Finally, a composite MJO cycle of 15-pentad duration was then obtained by averaging over all events.
The composite maps of anomalies of χ200 overlaid with the corresponding OLR anomalies for lags −5 to +5 pentads from the observations and TL959L60 are shown in Figs. 11 and 12, respectively. The observed EEOF pattern exhibits a predominantly zonal κ = 1 structure covering the entire tropics. Negative anomalies of χ200 and OLR associated with active convection in the observations (TL959L60) are centered over the Indian Ocean at about 75°E at lag −4 (−3), while a relative χ200 maximum with maximum OLR is found lying across the date line, shifted about 10°S over the SPCZ (beyond the date line). In the active phase (over the Indian Ocean) the anomalies of χ200 and OLR are in phase, whereas in the suppressed phase (near the date line) they have their maxima shifted some 30° longitude. By the next pentad the active convection region has moved eastward to about 90°E (80°E) and then, by another pentad, has moved to about 120°E. The sequence of composite patterns allows us to roughly estimate the velocity of propagation of the MJO. Between lag −3 and 0 the minimum of χ200 and the convective anomaly in the observations moves from ∼90°E to 170°E; therefore, they have moved with a propagation velocity of ∼7 m s−1.
The model composite MJO shows only certain characteristics in terms of space–time variability of MJO-induced convection. This includes the locations of the maxima in variability, coincidence of negative anomalies of χ200 with that of OLR, and their eastward advancement. Also, the MJO variability is more pronounced in the Eastern Hemisphere. However, there are significant shortcomings as well in the model, namely, the much weaker amplitude of the MJO variability, incoherent eastward advancement with pentads, much faster speed, and the smaller spatial scale of convection over convective centers of the equatorial Indian Ocean and the western Pacific. Thus, the dominant EEOF structure that successfully separates the strongest eastward propagating events conclusively implies the need for further improvement in capturing the observed MJO structure and propagation.
6. Summary and discussion
This study examines the mean convection and organization of convection on different intraseasonal time scales over the tropics in a very high resolution simulation of the Meteorological Research Institute/Japan Meteorological Agency (MRI/JMA) global atmospheric general circulation model (AGCM). The model, TL959L60, has a horizontal resolution that corresponds approximately to a 20-km mesh grid with 60 vertical levels. Verification of the simulation against recent observations shows its ability to capture certain major aspects of the seasonal mean convection over the tropics. Comparison with simulations at two lower resolutions of TL159L40 and TL95L40 reveals the effect of very high resolution in improving the simulation of seasonal-mean convection and climatological parameters over the tropics and moist stability over the Indian Ocean and SPCZ.
However, increasing the resolution does not impact the simulation of large-scale circulation features such as the vertical structure of the Walker and Hadley circulations and the location and intensity of the subtropical westerly jet stream. Although TL959L60 shows marginal improvement in simulating the vertical structure of heating and the height of the upper-tropospheric heating maximum, there exists a large bias in the simulated heating structure at all resolutions. This implies the need for further refinement in the cumulus parameterization scheme to improve the simulation of heating and the resultant large-scale circulation in the tropics. Many studies on resolution sensitivity suggest that increased spatial resolution may lead to improvement in some aspects of the simulations and degradation in others. But this analysis reveals that concurrent high vertical resolution helps alleviate this drawback, as reported by Roeckner et al. (2006). In conjunction with a vertical resolution of L60, the TL959L60 simulation does not show any serious degradations such as intensification and the poleward shifting of subtropical jet streams as found in previous modeling studies (e.g., Boyle 1993; Williamson et al. 1995).
The impact of resolution on intraseasonal time scales and on simulating the spatiotemporal variability of tropical convection have been investigated in detail. The effect of resolution on the amplitude of the diurnal cycle and total intraseasonal variance was found to be marginal. Marked improvement occurs in important features of convectively coupled Kelvin waves in TL959L60, but the model fails to simulate the MJO amplitude, which is significantly underestimated, and its partition with respect to low-frequency westward power. The composite MJO life cycle constructed based on EEOF analysis reveals the shortcomings in the simulation of the structure and propagation characteristics of the MJO, which are common to all resolutions. The absence of pronounced resolution sensitivity is consistent with the findings of previous studies (Duffy et al. 2003; Lin et al. 2006). However, when the eastward MJO signals and ER and Kelvin waves are isolated in TL959L60, the hierarchical structure of cloud clusters associated with MJO convection and the wave interrelationship among MJO, ER, and Kelvin waves are slightly better represented than in lower resolution versions. The improvement in simulating the convectively coupled Kelvin waves and the gross organization of tropical convection at different spatiotemporal scales suggests a slightly better representation of the interaction between convection and the large-scale dynamics at very high resolution in TL959L60. At the same time, the results strongly suggest that even the highest resolution simulation still shows significant deficiencies in the simulation of the most important aspects of the observed MJO and convectively coupled waves such as ER and MRG. This implies the need for further improvement in the simulation of the MJO and equatorial waves in this GCM. Notably, recent analysis of Lin et al. (2006) suggests that models with moist-convergence-type closure tend to produce more realistic MJO signals than schemes with CAPE-type closure, which is used in the present model. Thus, the MJO should be improved through continued refinements in the cumulus parameterization of the model.
Acknowledgments
K. Rajendran was supported by the Cooperative System for Supporting Priority Research, Japan Science and Technology Agency, and thanks Prof. J. Srinivasan for discussions and useful suggestions. We are grateful to Dr. A. D. Del Genio, the editor, and one anonymous reviewer for careful and insightful suggestions that greatly improved the original manuscript. This work was done within the framework of the Kyosei Project and the KAKUSHIN program funded by MEXT. The 20-km mesh model calculations were made on the Earth Simulator.
REFERENCES
Arakawa, O., and A. Kitoh, 2005: Rainfall diurnal cycle variation over the Indonesian Maritime Continent simulated by 20km mesh GCM. Sci. Online Lett. Atmos., 1 , 109–112.
Boer, G. J., and M. Lazare, 1988: Some results concerning the effect of horizontal resolution and gravity wave drag on simulated climate. J. Climate, 1 , 789–806.
Boyle, J. S., 1993: Sensitivity of dynamical quantities to horizontal resolution for a climate simulation using the ECMWF (cycle 33) model. J. Climate, 6 , 796–815.
Brankovic, C., and D. Gregory, 2001: Impact of horizontal resolution on seasonal integrations. Climate Dyn., 18 , 123–143.
Chang, A. T-C., L. S. Chiu, and G. Yang, 1995: Diurnal cycle of oceanic precipitation from SSM/I data. Mon. Wea. Rev., 123 , 3371–3380.
Ciesielski, P. E., R. H. Johnson, P. T. Haertal, and J. Wang, 2003: Corrected TOGA COARE sounding humidity data: Impact on diagnosed properties of convection and climate over the warm pool. J. Climate, 16 , 2370–2384.
Duchon, C. E., 1979: Lanczos filtering in one and two dimensions. J. Appl. Meteor., 18 , 1016–1022.
Duffy, P. B., B. Govindasamy, J. P. Iorio, J. Milovich, K. R. Sperber, K. E. Taylor, M. F. Wehner, and S. L. Thompson, 2003: High-resolution simulation of global climate, part 1: Present climate. Climate Dyn., 21 , 371–390.
Gregory, D., R. Kershaw, and P. M. Innes, 1997: Parameterization of momentum transport by convection. II: Tests in single-column and general circulation models. Quart. J. Roy. Meteor. Soc., 123 , 1153–1183.
Gualdi, S., A. Navarra, and H. von Starch, 1997: Tropical intraseasonal oscillation appearing in operational analyses and in a family of general circulation models. J. Atmos. Sci., 54 , 1185–1202.
Hayashi, Y., 1982: Space–time spectral analysis and its application to atmospheric waves. J. Meteor. Soc. Japan, 60 , 156–171.
Hayashi, Y., and D. G. Golder, 1986: Tropical intraseasonal oscillation appearing in the GFDL general circulation model and FGGE data. Part I: Phase propagation. J. Atmos. Sci., 43 , 3058–3067.
Hendon, H. H., 2000: Impact of air–sea coupling on the Madden–Julian oscillation in a general circulation model. J. Atmos. Sci., 57 , 3939–3952.
Hortal, M., 2002: The development and testing of a new two-time-level semi-Lagrangian scheme (SETTLS) in the ECMWF forecast model. Quart. J. Roy. Meteor. Soc., 128 , 1671–1687.
Inness, P. M., J. M. Slingo, S. J. Woolnough, R. B. Neale, and V. D. Pope, 2001: Organization of tropical convection in a GCM with varying vertical resolution; implications for the simulation of the Madden–Julian oscillation. Climate Dyn., 17 , 777–793.
Inness, P. M., J. M. Slingo, E. Guilyardi, and J. Cole, 2003: 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.
Janowiak, J., P. A. Arkin, and M. Morrissey, 1994: An examination of the diurnal cycle in oceanic tropical rainfall using satellite and in situ data. Mon. Wea. Rev., 122 , 2296–2311.
Kanamitsu, T., K. Tada, T. Kudo, N. Sato, and S. Isa, 1983: Description of the JMA operational spectral model. J. Meteor. Soc. Japan, 61 , 812–828.
Kuma, K. I., 1994: The Madden and Julian and tropical disturbances in an aqua-planet version of JMA global model with T63 and T159 resolution. J. Meteor. Soc. Japan, 72 , 147–172.
Lin, J., and Coauthors, 2006: Tropical intraseasonal variability in 14 IPCC AR4 climate models. Part I: Convective signals. J. Climate, 19 , 2665–2690.
Lindzen, R. S., and M. Fox-Rabinowitz, 1989: Consistent vertical and horizontal resolution. Mon. Wea. Rev., 117 , 2575–2583.
Madden, R. A., and P. R. Julian, 1971: Description of a 40–50 day oscillation in the zonal wind in the tropical Pacific. J. Atmos. Sci., 28 , 702–708.
Madden, R. A., and P. R. Julian, 1994: Observations of the 40–50-day tropical oscillation—A review. Mon. Wea. Rev., 122 , 814–837.
Masunaga, H., T. S. L’Ecuyer, and C. D. Kummerow, 2006: The Madden–Julian oscillation recorded in early observations from the Tropical Rainfall Measuring Mission (TRMM). J. Atmos. Sci., 63 , 2777–2794.
Matsuno, T., 1966: Quasi-geostrophic motions in the equatorial area. J. Meteor. Soc. Japan, 44 , 25–43.
Matthews, A. J., 2000: Propagation mechanisms for the Madden–Julian oscillation. Quart. J. Roy. Meteor. Soc., 126 , 2637–2652.
Mizuta, R., and Coauthors, 2006: 20-km-mesh global climate simulations using JMA–GSM model—Mean climate states. J. Meteor. Soc. Japan, 84 , 165–185.
Nakagawa, M., and A. Shimpo, 2004: Development of a cumulus parameterization scheme for the operational global model at JMA. RSMC Tokyo-Typhoon Center Tech. Review 7, Japan Meteorological Agency, 10–15.
Nakazawa, T., 1988: Tropical super clusters within intraseasonal variations over the western Pacific. J. Meteor. Soc. Japan, 66 , 823–839.
Neelin, J. D., and I. M. Held, 1987: Modeling tropical convergence based on the moist static energy budget. Mon. Wea. Rev., 115 , 3–12.
Numaguti, A., 1995: Characteristics of 4–20 day period disturbances observed in the equatorial Pacific during the TOGA COARE IOP. J. Meteor. Soc. Japan, 73 , 353–377.
Phillips, T. J., L. C. Corsetti, and S. L. Grotch, 1995: The impact of horizontal resolution on moist processes in the ECMWF model. Climate Dyn., 11 , 85–102.
Pope, V. D., A. Pamment, D. R. Jackson, and A. Slingo, 2001: The representation of water vapor and its dependence on vertical resolution in the Hadley Centre Model. J. Climate, 14 , 3065–3085.
Rajendran, K., R. S. Nanjundiah, and J. Srinivasan, 2002: The impact of surface hydrology on the simulation of tropical intraseasonal oscillation in NCAR (CCM2) atmospheric GCM. J. Meteor. Soc. Japan, 80 , 1357–1381.
Randall, D., and D. M. Pan, 1993: Implementation of the Arakawa–Schubert cumulus parameterization with a prognostic closure. The Representation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 145–150.
Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analysis using optimum interpolation. J. Climate, 7 , 929–948.
Rind, D., and W. B. Rossow, 1984: The effects of physical processes on the Hadley circulation. J. Atmos. Sci., 41 , 479–507.
Roeckner, E., and Coauthors, 2006: Sensitivity of simulated climate to horizontal and vertical resolution in the ECHAM5 atmosphere model. J. Climate, 19 , 3771–3791.
Slingo, J. M., and Coauthors, 1996: Intraseasonal oscillation in 15 atmospheric general circulation models. Results from an AMIP diagnostic subproject. Climate Dyn., 12 , 325–357.
Sperber, K. R., 2004: Madden–Julian variability in NCAR CAM2.0 and CCSM2.0. Climate Dyn., 23 , 259–278.
Takayabu, Y. N., 1994: Large-scale cloud disturbances associated with equatorial waves. Part I: Spectral features of the cloud disturbances. J. Meteor. Soc. Japan, 72 , 433–448.
Tao, W-K., and Coauthors, 1994: Retrieved vertical profiles of latent heat release using TRMM rainfall products for February 1998. J. Appl. Meteor., 40 , 957–982.
Taylor, K. E., D. Williamson, and F. Zwiers, 2000: The sea surface temperature and sea-ice concentration boundary conditions for AMIP II simulations. PCMDI Tech. Rep. 60, 24 pp.
Tompkins, A. M., and K. A. Emanuel, 2000: The vertical resolution sensitivity of simulated equilibrium tropical temperature and water vapour profiles. Quart. J. Roy. Meteor. Soc., 126 , 1219–1238.
Uppala, S. M., and Coauthors, 2005: The ERA-40 reanalysis. Quart. J. Roy. Meteor. Soc., 131 , 2961–3012.
Waliser, D. E., W. Stern, S. Schubert, and K. M. Lau, 2003: Dynamic predictability of intraseasonal variability associated with the Asian summer monsoon. Quart. J. Roy. Meteor. Soc., 129 , 2897–2925.
Weare, B. C., and J. S. Nasstrom, 1982: Examples of extended empirical orthogonal function analyses. Mon. Wea. Rev., 110 , 481–485.
Wheeler, M., and G. N. Kiladis, 1999: Convectively coupled equatorial waves: Analysis of clouds and temperature in the wavenumber–frequency domain. J. Atmos. Sci., 56 , 374–399.
Williamson, D. L., J. T. Kiehl, and J. J. Hack, 1995: Climate sensitivity of the NCAR Community Climate Model (CCM2) to horizontal resolution. Climate Dyn., 11 , 377–397.
Wu, X., and M. Yanai, 1994: Effects of vertical wind shear on the cumulus transport of momentum: Observations and parameterization. J. Atmos. Sci., 51 , 1640–1660.
Yang, G. Y., and J. M. Slingo, 2001: The diurnal cycle in the Tropics. Mon. Wea. Rev., 129 , 784–801.
Yoshimura, H., and T. Matsumura, 2005: A two-time-level vertically-conservative semi-Lagrangian semi-implicit double Fourier series AGCM. CAS/JSC WGNE Research Activities in Atmosphere and Ocean Modeling, No. 35, WMO, 3.27–3.28.
Zhang, C., and Coauthors, 2006: Simulations of the Madden–Julian oscillation in four pairs of coupled and uncoupled global models. Climate Dyn., 27 , 573–592.
Climatological winter mean OLR and 850-hPa winds from (a) observations, (b) TL959L60, (c) TL159L40, and (d) TL95L40.
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
Climatological winter mean (a) vertical moist stability (VMS) defined as GMS(400–50hPa) − GMS(1000–400hPa) and (b) moisture convergence integrated over 1000–600 hPa from the observations and three simulations.
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
Taylor diagram of normalized standard deviation and correlation of climatological OLR, PR, PWAT, SLP, TA, U2, and U8 with respect to the corresponding observations over the global tropics (30°S–30°N, 0°–360°) for TL159L40 and TL959L60.
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
Longitude–height cross section of winter-mean zonal wind overlaid with (u, ω) vectors averaged over 10°S–10°N from the (a) observations, (b) TL959L60, and (c) TL159L40 and (d) the difference between TL959L60 and TL159L40. Differences significant at the 95% level are stippled in (d).
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
Longitude–height cross section of DJF mean convective heating averaged over 10°S–10°N from (a) TRMM, (b) TL959L60, and (c) TL159L40 and (d) the difference between TL959L60 and TL159L40. The differences above the 95% significance level are stippled in (c). Solid line in (a) represents the vertical structure of the DJF mean apparent heat source from TOGA IFA data.
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
Standard deviation of 6-hourly winter precipitation on time scales ≤1 day from TL959L60 and TL159L40. Coherent diurnal variation averaged over the two boxes (solid line for left box and dashed line for the right box; time in UTC) is shown in insets.
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
(a) Standard deviation of daily OLR total variance and (b) percentage of total variance of OLR explained by the 20–90-day time scale (average from all November–April periods) from the observations, TL959L60, and TL159L40.
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
Wavenumber–frequency spectra of (a), (b) observed, (c), (d) TL959L60, and (e), (f) TL159L40 OLR. The (left) symmetric and (right) antisymmetric components around the equator are divided by the background power, averaged over 15°S–15°N. Superimposed thin lines are the dispersion curves of equatorial Kelvin, n = 1 Rossby (ER), n = 1 westward inertio–gravity (WIG), mixed Rossby–gravity, and n = 0 EIG waves for equivalent depths h = 8, 12, 25, 50, and 90 m. Thick boxes indicate the filtering regions to retrieve MJO, Kelvin, and ER waves.
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
Time–longitude diagram of OLR (shaded) averaged between 7.5°S and 7.5°N for a representative year from the observations, TL959L60, and TL159L40. The MJO filtered anomaly less than 10 (5) W m−2 for the observations (models) is represented by thick dashed contours. The Kelvin-filtered anomaly and ER-filtered anomaly, of less than 10 W m−2, are represented by thin dashed and solid contours, respectively.
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
Wavenumber–frequency power spectra of OLR anomalies from the observations, TL959L60, TL159L40, and TL95L40.
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
Composite MJO life cycle based on 200-hPa velocity potential anomalies (shaded) overlaid with correspond-ing OLR anomalies (contours where the thick contour corresponds to the zero anomaly) from the observations.
Citation: Journal of Climate 21, 15; 10.1175/2008JCLI1950.1
