1. Introduction
a. The northward-propagating intraseasonal oscillation
While the Indian summer monsoon remains one of the most consistent and stable features of the global climate system on interannual and interdecadal time scales, the monsoon’s intraseasonal variability (ISV) is far less predictable. Intraseasonal oscillations in the monsoon’s strength are dominated by organized convective events that form in the equatorial Indian Ocean (EqIO) and propagate north to the Indian subcontinent (e.g., Annamalai et al. 1999; Annamalai and Slingo 2001). This northward-propagating intraseasonal oscillation (NPISO) has a period of 30–50 days and a speed of approximately 1° latitude day−1 (e.g., Yasunari 1979; Krishnamurti and Subrahmanyam 1982; Gadgil 1990; Lawrence and Webster 2002). Convection over India is to some extent anticorrelated with the eastern EqIO (EEqIO), such that “active” periods of enhanced rainfall over the subcontinent are associated with “break” periods of suppressed convection in the EEqIO (Hartmann et al. 1992; Annamalai and Sperber 2005; Klingaman et al. 2008). Waliser et al. (1999a) demonstrated that these recurring active and break events give the monsoon an ISV greater than its interannual variability. Successful long-range prediction of these events would be a great boon to Indian agriculture, primarily for flood and drought mitigation (Webster and Hoyos 2004). Detailed descriptions of the structure of the NPISO can be found in Kemball-Cook and Wang (2001), Annamalai and Slingo (2001), and Klingaman et al. (2008), among others.
Recent studies have made some progress toward understanding the physical mechanisms underlying the NPISO, although competing hypotheses persist. Using observations and simple numerical experiments, Wang and Xie (1997) suggested that the northward propagation was an artifact of moist Rossby waves excited by an eastward-moving Kelvin–Rossby wave packet; the moist Rossby waves propagated northwestward, giving the entire packet a westward tilt with latitude. Similar efforts have examined reanalysis data and described the meridional propagation as Rossby waves emanating from equatorial convection, with the majority of northward-propagating events also demonstrating eastward propagation along the equator (Annamalai and Slingo 2001; Kemball-Cook and Wang 2001; Lawrence and Webster 2002). The 30–50-day period and equatorial propagation of the summer intraseasonal oscillation strongly suggests that it is connected to—if not a manifestation of—the winter intraseasonal oscillation, the Madden–Julian oscillation (MJO; Madden and Julian 1971, 1972). Waliser (2006) concluded that the MJO and the NPISO can be considered as two instances of the same underlying intraseasonal phenomenon, modified by the seasonal background state.
b. The effect of air–sea coupling on NPISO simulations
Atmosphere–ocean coupled processes have recently gained support as potential mechanisms for driving the NPISO. Waliser et al. (2003) concluded that all 10 atmosphere-only GCMs (AGCMs) from the Climate Variability and Predictability program/Asian–Australian monsoon intercomparison project (Kang et al. 2002) substantially underestimated the monsoon ISV, particularly near the equatorial Indian Ocean, despite displaying reasonable seasonal-mean rainfall. An earlier intercomparison study found that many AGCMs exhibited poor MJO-like variability in northern winter (Slingo et al. 1996). Similar studies with individual AGCMs have confirmed this deficiency (e.g., Rajendran et al. 2002). Still, the fact that AGCMs generated intraseasonal events, albeit weak ones, suggests that the NPISO is an intrinsically atmospheric mode; the oscillation likely arises from internal atmospheric variability, not coupled processes.
Following many studies that showed that air–sea coupling improved simulations of the MJO (e.g., Flatau et al. 1997; Waliser et al. 1999b; Kemball-Cook et al. 2002; Inness and Slingo 2003; Woolnough et al. 2007), recent efforts have focused on the impact of an interactive ocean on the NPISO. These studies were justified by observations of Indian Ocean sea surface temperatures (SSTs) from the Bay of Bengal (BoB) Monsoon Experiment (Bhat et al. 2001) and the Joint Air–Sea Monsoon Interaction Experiment (Webster et al. 2002) field campaigns, which showed the passage of NPISO events to be associated with SST variations of more than 1°C. As in the MJO (Woolnough et al. 2000), anomalous SSTs were in quadrature with anomalous convection, with warm (cool) SSTs leading enhanced (suppressed) convection by 10–15 days. Anomalies in SST and convection were linked by a negative feedback involving surface heat fluxes, boundary layer stability, low-level winds, evaporation, and moisture convergence (e.g., Kemball-Cook and Wang 2001; Klingaman et al. 2008).
Fu et al. (2003) compared a hybrid coupled GCM to its atmosphere-only counterpart and found that the coupled model produced a strong NPISO with correct phase relationships (compared to observations) between SSTs, convection, and surface heat fluxes. The AGCM produced a weaker NPISO and collocated intense convection with the warmest SSTs, whereas in observations, warm SSTs coincided with subsidence, clear skies, and strong insolation (Fu et al. 2002). This error is common in AGCM simulations of intraseasonal behavior in the Indian Ocean and in the West Pacific warm pool. Using the same model, Fu and Wang (2004) demonstrated that coupling substantially improved the two- and three-dimensional structure of the NPISO over an AGCM integration when validated against observations and European Centre for Medium-Range Weather Forecasts (ECMWF) analyses. The authors concluded that AGCMs had little hope of representing observed monsoon ISV because of their fundamental inability to create and modify the requisite intraseasonal SST anomalies via convective feedbacks.
Rajendran and Kitoh (2006) obtained similar results with the Meteorological Research Institute (MRI) coupled GCM. They found that large-scale atmospheric dynamics could account for the existence of the MJO and the NPISO but that atmosphere-to-ocean feedbacks amplified the intensity and corrected the phase speed of propagating convection. The MRI-coupled GCM reasonably replicated the observed phase relationships between rainfall, SST, net surface heat flux, latent heat flux, and surface wind stress, while phase relationships in the AGCM were weaker and temporally distorted. This agrees with Zheng et al. (2004), who employed the coupled model from the Geophysical Fluid Dynamics Laboratory (GFDL). Fu et al. (2007) demonstrated that a coupled GCM could extend NPISO predictability—as measured by the ratio of signal-to-forecast error and by the spatial correlation of anomalies in time-filtered rainfall—by about a week when compared to the same model without an interactive ocean. Maloney and Sobel (2005) compared simulations of the MJO in the National Center for Atmospheric Research (NCAR) Community Atmospheric Model, version 2 (CAM2) coupled to a slab ocean model of globally uniform depth. That study concluded that a 20-m depth was optimal for simulating the wind-induced surface heat exchange (WISHE) mechanism that was critical to the representation of the MJO in CAM2. While shallower depths improved intraseasonal SST variations due to lower thermal inertia, the shallower mixed layer also reduced latent heat variations, weakening the feedback between convection and surface fluxes.
Air–sea coupling does not rectify all errors associated with the NPISO. As for the MJO, errors in the mean state can influence simulations of the NPISO. Inness et al. (2003) showed that reducing systematic mean-state errors in the third Hadley Centre Coupled Ocean–Atmosphere GCM (HadCM3) could substantially improve the MJO. Sperber (2004) also suggested that mean-state errors could project onto intraseasonal behavior and so degrade simulations of the MJO and the NPISO. Further, Fu and Wang (2004) noted that model representations of cumulus clouds, cloud-radiation interactions, boundary layer processes, and land surface feedbacks could dramatically affect simulated ISV.
c. Motivation
Although many studies have found a weak and inconsistent NPISO in AGCMs, the SST forcing in these studies has often exhibited marked deficiencies. Typically, AGCMs have been forced by SSTs from the coupled-model integrations against which the AGCM is to be compared. Fu and Wang (2004), for example, forced their AGCM with daily and climatological monthly mean SSTs from a 16-yr coupled simulation. That study concluded that while daily SST forcing improved the ISV of rainfall over monthly mean SSTs, the ISV in the daily SST AGCM run was still substantially weaker than in the coupled-model run (their Fig. 4). Where observed SSTs have been used (e.g., Waliser et al. 2003; Liess and Bengtsson 2004), they have typically been taken from the weekly optimally interpolated SST analyses (OISSTs) from the National Centers for Environmental Prediction (NCEP; Reynolds and Smith 1994). When compared to observations, both coupled-model and NCEP OISSTs suffer from weak intraseasonal anomalies (Harrison and Vecchi 2001; Senan et al. 2001; Sengupta and Ravichandran 2001; Sengupta et al. 2001). The NCEP OISST product assimilates readings from satellite-based infrared sounders, which are affected by contamination by clouds, which often leads to rejection of cloudy pixels, a critical failing when estimating SSTs during the monsoon season. By contrast, the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI) suffers from far less cloud contamination and so provides an improved SST analysis in tropical regions. Harrison and Vecchi (2001) concluded that the TMI SSTs were more accurate than the NCEP analyses when both were compared to moored observations in the Arabian Sea and the BoB. The same study performed a spectral analysis and found that the TMI SSTs contained as much as 4 times more power on intraseasonal time scales than the NCEP OISSTs. Similar conclusions on the improved accuracy and intraseasonal power of the TMI SSTs were reached by Senan et al. (2001) in a comparison with buoy observations around the Indian coastline. Bhat et al. (2004) found that the NCEP OISSTs had virtually no ISV in the BoB. Our study uses a new SST analysis that assimilates TMI data, among other high-resolution satellite-derived products, as well as ship and buoy observations (section 2b).
In a study focused on the West Pacific, Bernie et al. (2005) compared a one-dimensional mixed layer ocean model to observations from the intensive observation period (IOP) of the Tropical Ocean Global Atmosphere Coupled Ocean–Atmosphere Response Experiment (TOGA COARE). The study concluded that to capture the intraseasonal SST variability required the ocean model to be forced by diurnally varying heat fluxes and to have a low thermal inertia in the mixed layer, equivalent to a fine vertical resolution. Bernie et al. (2005) suggested that a vertical resolution of 1 m and 3-hourly coupling were required to capture of order 90% of the observed intraseasonal SST variability. As past studies have used coupled models with a top ocean layer on the order of 10-m thick and coupled only once per day, the intraseasonal variability of the SSTs that were used to force the corresponding AGCM experiments in those studies were too small.
The large errors in the ISV of previous observed and coupled-model SSTs have left unresolved the underlying question of the ability of an AGCM to simulate the NPISO given reasonably accurate high-frequency SST forcing. While it is questionable whether an AGCM would simulate the observed phase relationship between anomalous SSTs and convection even with “perfect” SSTs, it may be that realistic intraseasonal SST anomalies could improve the poor ISV in rainfall seen in previous studies. Further, an observed SST dataset of high temporal resolution and realistic ISV would allow an improved examination of the effect of high-frequency (e.g., daily or weekly) SST anomalies on the NPISO. The tropical sea surface evolves quickly and is tightly coupled to convection and rainfall, so an assessment of the multiscale interactions of SST and rainfall would be valuable. This type of experiment can be conducted with an AGCM, in which SSTs are prescribed and their temporal variability can be controlled. Sensitivity experiments with high-frequency SSTs have been previously conducted using coupled-model SSTs (Fu and Wang 2004) and the NCEP OISSTs (Liess and Bengtsson 2004), and these have generally shown that high-frequency SST variability improved the intraseasonal variability of monsoon rainfall. Given the issues with these SST forcings noted above, a new experiment using observed SSTs with realistic intraseasonal variability is certainly warranted.
In this study, we conduct three ensembles of simulations with the Hadley Centre Atmosphere Model (HadAM3) at a high horizontal resolution. These ensembles are forced by a new SST dataset with fine spatial and temporal resolutions. Our primary interest is to determine the extent to which high-frequency SST variability influences NPISO-like behavior. To that end, we have forced one HadAM3 ensemble with daily SSTs, a second with monthly means, and a third with 5-day means. By comparing simulated NPISO events against observations, this experimental design also allows an investigation of the AGCM’s representation of the NPISO when forced by some of the most accurate SSTs available. The model and the three ensembles are described in section 2; we examine the ensemble-mean response to high-frequency SST forcing in section 3a and compare individual ensemble members to observations in section 3b. We discuss the implications of our results concerning high-frequency SSTs, particularly the implications for coupled models and experiments, in section 4, and we summarize our key findings in section 5.
2. Model and methods
a. The Hadley Centre Atmosphere Model
All experiments were performed with a high-resolution version of the Hadley Centre Atmosphere Model, HadAM3. The model is configured as in Pope et al. (2000), except that the spatial, vertical, and temporal resolutions have been increased to 1.25° longitude by 0.83° latitude, 30 levels, and 10 min, respectively; the spatial resolution is therefore N144L30. The sensitivity of the NPISO to atmospheric horizontal resolution has not been thoroughly examined in literature; studies that have considered the effect of resolution on tropical ISV in general have found mixed results. Liess and Bengtsson (2004) found that increasing the resolution of the Hamburg Atmospheric Model (ECHAM4) from T42 to T106 degraded the simulation of the MJO and the NPISO. Meehl et al. (2006) found little improvement in the low NPISO-like variability in the Community Climate System Model, version 3 (CCSM3) when increasing the resolution from T42 to T85. By contrast, Navarra et al. (2008) found that many aspects of tropical variability were improved in the Scale Interaction Experiment-Frontier (SINTEX-F) model at T106 over a simulation at T30. That study focused mainly on interannual variability (e.g., El Niño), however. Stratton (1999) studied the effect of the N144L30 resolution on the previous version of HadAM (HadAM2b) and concluded that it improved atmospheric variability, bringing HadAM2b closer to the 40-yr ECMWF Reanalysis (ERA-40), which uses a similar spatial resolution. With respect to vertical resolution, Inness et al. (2001) demonstrated that increasing HadAM3 from 19 to 30 levels benefited the ISV of convection. Similarly, Liess and Bengtsson (2004) found that increasing ECHAM4 from 19 to 39 levels improved the phase speed of the intraseasonal oscillation and suggested that finer vertical resolution would be particularly important at high horizontal resolutions.
b. Sea surface temperature forcing
This study is among the first to make use of SST analyses resulting from the Global Ocean Data Assimilation Experiment (GODAE) High-Resolution Sea Surface Temperature (GHRSST) pilot project (Donlon et al. 2007). GHRSST provides a common data format for the dissemination of satellite SST data from a global network of sources in near real time (Stark et al. 2007). This study uses the Operational Sea surface Temperature and Sea Ice Analysis (OSTIA) product from the United Kingdom National Centre for Ocean Forecasting. OSTIA assimilates daily SST products from several microwave and infrared satellite-based sounders (Stark et al. 2007), including the TMI data that has been previously shown to be more accurate and have higher ISV than other optimally interpolated SST products (section 1c). The analysis is available daily at a global spatial resolution of 1/20°, or approximately 6 km. (Data may be obtained online at http://ghrsst-pp.org.) OSTIA is produced through a persistence-based optimal interpolation system (Lorenc 1981) using all data sources available to the GHRSST project (Donlon et al. 2007). Preliminary comparisons against buoy data indicate that the OSTIA analyses have a root-mean-square error of 0.5°C and a cold bias of 0.15°C at a single point. A detailed description of GHRSST can be found in Donlon et al. (2007); initial results from the OSTIA project are available in Stark et al. (2007). OSTIA’s high spatial and temporal resolutions, combined with its use of in situ data and microwave sounders that can estimate SSTs under clouds more accurately than infrared instruments, make the OSTIA analyses far more accurate than other OISST datasets previously used to force AGCMs.
Due to the novelty of this product, SSTs were available for only one year: February 2005–January 2006.
c. Experiment design
We conducted three 30-member ensembles of HadAM3 forced by the OSTIA SST analyses. The first ensemble was forced by the daily OSTIA dataset and so included high-frequency SST variability; this will be referred to as the “daily ensemble.” A second ensemble was forced by monthly mean OSTIA SSTs and will be referred to as the “monthly ensemble.” A third, “5-day ensemble” forced by 5-day means of OSTIA SSTs was conducted to determine the sensitivity of the daily ensemble to the highest-frequency components of the OSTIA SSTs. OSTIA SSTs were area averaged to the HadAM3 spatial resolution; the masking value for sea ice—used to estimate sea ice coverage—was altered to account for the area averaging. For the monthly (5 day) ensemble, time-mean SSTs were calculated using the Atmospheric Model Intercomparison Project (AMIP II) method so that the monthly mean (5-day mean) SSTs in the daily and monthly (5-day) ensembles were equal at every grid point (Taylor et al. 2000). Time-mean SSTs were linearly interpolated to daily values between meaning periods. In all three ensembles, SSTs and sea ice were updated once per day.
The SSTs used to force the daily ensemble have substantial 30–50-day ISV during the monsoon season [June–September (JJAS)], particularly in the BoB and the Arabian Sea, where values of the standard deviation approach 0.5°C (Fig. 1a). This 30–50-day variability was calculated using a bandpass filter, which rejects constituent signals of the original time series with periods shorter than 30 days or longer than 50 days; variability in the 30–50-day band passes through the filter. In the eastern BoB, the 30–50-day variability accounts for more than 50% of the total variability. Equatorial variability is much weaker, but Klingaman et al. (2008) found weak equatorial SST anomalies associated with the NPISO in the TMI SST analyses. The large values off the African coast are due to the spinup of the Somali jet at the beginning of the monsoon season, which leads to coastal upwelling and evaporative sea surface cooling in June (Webster et al. 1998; Schott and McCreary 2001) that projects onto the 30–50-day ISV. The monthly mean OSTIA SSTs used to force the monthly ensemble have far less ISV, as little as one-sixth of that in the daily OSTIA SSTs (Fig. 1b). The 30–50-day variability in the 5-day mean SSTs used to force the 5-day ensemble (not shown) is broadly similar to that of the daily SSTs.
Initial conditions for all three ensembles were obtained from a previous 17-yr AMIP II integration of HadAM3 at N144L30. That integration was forced by the combination of observed and reanalysis SSTs used for the AMIP II project (Fiornio 1997). Members were initialized from consecutive days in February from the AMIP II integration, with the validity date of each initial condition set to 1 February. All members were integrated for the year corresponding to the OSTIA SSTs: 1 February 2005 through 31 January 2006. As the monsoon season begins in June, initializing the members in February provides ample spinup time for the model to adjust to the new SST forcing and for the individual members to diverge.
d. Wavelet transforms
Wavelet transforms are alternatives to Fourier transforms and fast Fourier transforms (FFTs) that analyze a one-dimensional discrete time series (e.g., rainfall) as a diffuse, two-dimensional time-frequency image (Torrence and Compo 1998). Wavelets thus provide the decided advantage of identifying not only at what periods a time series exhibits power but at what temporal points in the time series that power occurs. The result of a wavelet transform can be expressed as a line contour plot of frequency (or, inversely, period) against time, with power as the contoured value. Statistical significance of a wavelet transform can be determined by assuming a background spectrum of either white noise—a constant power at each period, giving a flat Fourier spectrum—or red noise—increasing power with increasing period—and performing a chi-squared test (Torrence and Compo 1998). Wavelet transforms have been extensively used in atmospheric and oceanic sciences in the last two decades, including to analyze SSTs, wind stress, and sea level off the California coast (Breaker et al. 2001), observed and proxy records of East Asian and Indian monsoon rainfall (Hu and Nitta 1996; Yadava and Ramesh 2007), forecast error covariances in variational data assimilation methods (Bannister 2007), and the El Niño-Southern Oscillation (Kestin et al. 1998; Chang et al. 2004), among many other applications. Torrence and Compo (1998) provide an excellent introduction to wavelet transforms, while further technical details can be found in Daubechies (1990, 1992).
In this study, we make use of wavelet transforms to show when and at what periods monsoon-season time series of rainfall and SST display statistically significant power. All wavelet transforms use a “Morlet” basis function, which is essentially a plane wave modulated by a Gaussian (Goupillaud et al. 1984; Torrence and Compo 1998). Statistical significance was calculated by assuming a red-noise background spectrum.
e. Calculation of ensemble-mean quantities
Where we use ensemble-mean metrics, they are calculated by first calculating the metric on each ensemble member and then taking the mean of the metric across the ensemble. For example, to calculate the ensemble-mean standard deviation of 30–50-day bandpass-filtered JJAS rainfall, we first take the standard deviation of 30–50-day bandpass-filtered JJAS rainfall from each member and then compute the ensemble mean of that variance.
3. Impacts of high-frequency SSTs on intraseasonal variability
As comparisons of the daily and monthly ensembles provide the clearest case for the importance of high-frequency SST variability in simulations of intraseasonal behavior, most of our results and discussion focus on those ensembles. We include comparisons of the daily and 5-day ensembles to demonstrate the impact of removing only the highest-frequency SST variability in the OSTIA dataset.
a. Ensemble-mean diagnostics
When forced by daily OSTIA SSTs, HadAM3 reproduces well the JJAS-mean rainfall across the monsoon domain, with local maxima in the northern BoB, along the hilly western coast of India, and in the Indian Ocean south of the equator (Fig. 2a). The model’s spatial distribution of rainfall across India resembles to a high degree the 1° × 1° gridded 1951–2004 climatology of JJAS rainfall compiled by the Indian Meteorological Department (Fig. 2b; Rajeevan et al. 2005). The model underestimates the rainfall near the southwest tip of India; produces a weaker rain-shadow effect from the western mountains, resulting in wet biases over central southern India; and overestimates rainfall in the northeastern and northern subcontinent.
The difference between the daily and monthly ensembles gives the impact of including submonthly SST variability in the forcing SSTs. Submonthly SST variability causes small but statistically significant changes in ensemble-mean, JJAS-mean rainfall. The daily ensemble shows higher seasonal-mean rainfall over the northern BoB, the Arabian Sea, and the South China Sea, and less rainfall to the south of the peninsula (Fig. 2c). These differences occur even though the SST forcing for both ensembles has the same monthly—and hence seasonal—mean at each grid point. Furthermore, the differences in the seasonal-mean rainfall are negative as well as positive. This behavior suggests that the differences are caused not only by the nonlinear response of precipitation to SSTs (i.e., the Clausius–Clapeyron relationship) but also by circulation changes. While the statistical significance of the changes to the ensemble-mean, JJAS-mean rainfall is a noteworthy result, the low magnitude of these changes—with respect to the ensemble mean (Fig. 2a)—implies that including submonthly SST variability does not substantially alter the model climatology. Differences in ensemble-mean, JJAS-mean rainfall between the daily and 5-day ensembles (not shown) are less than ±0.25 mm day−1 and are statistically significant at the 5% level only over the ocean immediately around the Indian subcontinent and in the South China Sea.
The daily ensemble shows a high variance in intraseasonal rainfall north of 5°N, measured as the ensemble mean of the standard deviation in 30–50-day bandpass-filtered JJAS rainfall (Fig. 2d). This variability decreases toward the equator, particularly in the eastern Indian Ocean, because HadAM3 produces little rainfall in this region (Fig. 2a). Still, the standard deviation is about 30% of the mean, indicating that the model can substantially vary equatorial rainfall on intraseasonal time scales. When forced by daily OSTIA SSTs, HadAM3 is capable of producing substantial 30–50-day ISV in monsoon rainfall.
The daily ensemble demonstrates far greater intraseasonal rainfall variability than the monthly ensemble across the vast majority of the monsoon domain (Fig. 2e). These 10%–30% increases in the standard deviation are mostly confined to the ocean, although some scattered significant values exist across northwestern and central India. The increased ISV across the eastern Indian Ocean and northward into the BoB is particularly encouraging for the NPISO, as past studies have indicated that the oscillation was more clearly observable in the eastern basin (Lawrence and Webster 2002; Klingaman et al. 2008). The daily ensemble shows far smaller statistically significant increases in ISV when compared to the 5-day ensemble (Fig. 2f); these are limited to the eastern equatorial Indian Ocean. For the regions of interest to this study, the 5-day and daily ensembles have essentially the same ISV in monsoon rainfall.
Substantial differences occur between the daily and monthly ensemble means in the time series of India area-averaged rainfall (red lines in Fig. 3). Both ensembles reasonably simulate the evolution of the daily climatology from the all-India rainfall dataset provided by the Indian Institute for Tropical Meteorology, although with a persistent wet bias of 3–5 mm day−1. On qualitative inspection, however, only the daily ensemble mean displays clear ISV, particularly after 1 July (Fig. 3a), which we will confirm quantitatively with wavelet analysis. One active-break-active cycle begins with an active event in late July, followed by break conditions during August, and a return to active conditions in early September. Most of the individual ensemble members (black lines in Fig. 3) mimic the ensemble mean through this oscillation. However, no member in either ensemble reliably simulates the observed all-India rainfall for 2005 (cyan line in Fig. 3), the same year as the SST forcing. That season was characterized by MJO-like equatorial convection in May and a delayed onset of monsoon rains across India; during June the observed rainfall lies outside the range of both ensembles. This suggests that forced SSTs cannot drive this behavior in this AGCM.
Further evidence for the ability of submonthly SSTs to affect intraseasonal monsoon behavior comes from the ensemble-mean wavelet transform of rainfall in the northern BoB, the region that showed one of the largest increases in 30–50-day rainfall in the daily ensemble (Fig. 2e). The ensemble-mean wavelet transform from the ensemble confirms that including daily SST variability induces additional variance in intraseasonal rainfall, particularly at 20–50-day periods (Fig. 4a). The NPISO has a characteristic period of 30–50 days, and so this peak in power suggests that the oscillation might be better resolved in the daily ensemble. In this band, the daily ensemble has power exceeding 95% confidence beginning in early August, with power exceeding 90% confidence extending to mid-July. The wavelet transforms of the forcing SSTs in the BoB indicate that the daily ensemble SST forcing contained statistically significant power in the 30–50-day band during late July, August, and September (Fig. 4b), a period which broadly corresponds to the occurrence of statistically significant power in rainfall in Fig. 4a.
In striking contrast, the monthly ensemble-mean wavelet transform has no power at intraseasonal periods exceeding any reasonable confidence level at any time during the monsoon season (Fig. 4c). In taking the monthly means of the OSTIA SSTs, all power has been removed at periods shorter than 30 days and variability has been damped down to at least 60-day periods (Fig. 4d). The association between the ISV in SSTs and rainfall demonstrates the potential for a strong link between SSTs and rainfall in this region. This inference is further supported by the 5-day ensemble-mean wavelet transform, which shows statistically significant power in both rainfall (Fig. 4e) and SST (Fig. 4f) on intraseasonal time scales. The removal of power in SST at periods shorter than 5 days has altered the wavelet transform of rainfall in the 5-day ensemble. Compared to the daily ensemble, the 5-day ensemble exhibits statistically significant power in rainfall for a smaller fraction of the monsoon season and only within the 40–50-day band; there is no statistically significant power in July and no peak in the 20–40-day band.
We qualitatively examined the wavelet transforms from each ensemble member (not shown) and found that the majority of the daily-ensemble (5-day ensemble) members contained significant power above 95% confidence in the 30–50-day (40–50 day) band during the season, while the majority of the monthly-ensemble members did not. The limited temporal extent of the statistically significant power at periods shorter than 10 days in all three ensembles suggests that this power is probably associated with sharp changes in the seasonal cycle of rainfall rather than with persistent variability. Comparing the dates of this high-frequency power with the time series of all-India rainfall (Fig. 3) supports this hypothesis, as the ensemble-mean rainfall decreases rapidly in late July and increases again in early September. These changes likely project onto short-frequency wavelets and so appear in the wavelet analysis as statistically significant power.
To explore the quantitative difference in intraseasonal activity across all ensemble members, we construct a metric that considers the power on those time scales. For each member, we consider the area-averaged rainfall for each day in the same region used to construct Fig. 4 and perform a one-dimensional wavelet transform on the resulting time series. Because we are most interested in the intraseasonal periods where the daily ensemble shows statistically significant power, we consider only periods between 30 and 50 days; periods between 20 and 30 days are excluded so as to clearly separate the 30–50-day NPISO from 10–20-day variability along the monsoon trough (Annamalai and Slingo 2001). We normalize the power at each period in the 30–50-day range by the 95% confidence level for that particular period. The normalization ensures that longer periods—which have greater power but also a greater confidence threshold from our red-noise background spectrum (section 2d)—do not carry greater weight than shorter periods. All normalized values greater than unity (i.e., for which the power exceeded 95% confidence) are summed across JJAS. This “intraseasonal-power metric” provides a measure of both the frequency of occurrence and the intensity of power in the 30–50-day band in each ensemble member.
We calculated the intraseasonal-power metric on each member in all three ensembles and for each year between 1997 and 2006 in the l° × l° gridded, daily precipitation analyses from the Global Precipitation Climatology Project (GPCP). The high spatial and temporal resolutions of the GPCP data nearly match the resolutions of the HadAM3 simulations. We also consider the 2005 GPCP data separately, because the 2005 OSTIA SSTs were used to force the HadAM3 simulations. The probability density function (PDF) of the intraseasonal-power metric values confirms that the vast majority of members in the daily ensemble had greater 30–50-day power in rainfall than their counterparts in the monthly ensemble (Fig. 5). By this metric, most of the members in the monthly ensemble have very little statistically significant intraseasonal power in BoB rainfall compared to the GPCP analysis. On the other hand, the shape of the PDF of the daily ensemble approximates the GPCP PDF well. (The daily ensemble’s lower probability value at its peak is likely due to the larger sample size: the daily ensemble has 30 members while there are only 10 yr of GPCP data.) If one were to choose an ensemble member at random from the daily and monthly ensembles, the daily ensemble member would be about twice as likely to have ISV close to the 2005 GPCP data.
Most members of the 5-day ensemble have values of intraseasonal power close to the peak of the GPCP PDF (Fig. 5). When compared to the daily ensemble, however, the 5-day ensemble shows a decided shift toward lower values of statistically significant intraseasonal power; the majority of the 5-day members have less ISV than the majority of the daily-ensemble members. This is consistent with the rainfall wavelet transform (Fig. 4e), which demonstrated that the 5-day ensemble showed significant 30–50-day power in rainfall for a smaller fraction of the monsoon season. In the analysis and in the intraseasonal-power metric, the 5-day ensemble lies between the daily and monthly ensembles, emphasizing that removing even just the highest-frequency components of the OSTIA SSTs results in a degradation to the simulation of ISV.
b. Model NPISO events compared with GPCP analysis
Having determined that the daily ensemble contained greater ISV and possibly a stronger NPISO than either the monthly or 5-day ensembles, we must now consider whether the modeled NPISO in the daily ensemble more closely resembles the observed NPISO. First, it is important to note that providing daily, observed SST forcing does not correct a common error in AGCM simulations of intraseasonal monsoon variability: the phase relationship between SST anomalies and atmospheric deep convection. We computed the lead–lag correlations between the linear trend in longitude-averaged (80°–90°E) rainfall and SST over JJAS across all ensemble members; the trend was calculated using a centered 11-day window at each day. This band is chosen because it contains little land south of 25°N; the NPISO is strongest in the eastern one-third of the Indian Ocean; and all three ensembles had their highest frequencies of occurrence of intraseasonal power in this band. The linear trend in rainfall is a more useful diagnostic than the raw rainfall time series because of the strong latitudinal gradient in climatological rainfall between the equator and the Indian subcontinent in JJAS. The 11-day trend is ideal as it is short enough to retain sufficient data points to calculate a useful diagnostic, yet long enough to limit the mixing of intraseasonal modes. The ensemble-mean lead–lag correlation coefficients are calculated by taking the mean of the lead–lag covariances from the individual members, then dividing by the product of the standard deviations of the time series of rainfall and SST created by concatenating data from all 30 ensemble members. A similar method is used for the 10 yr of GPCP analysis. To account for the serial correlations in rainfall within each ensemble member, degrees of freedom are estimated using the method of Livezey and Chen (1983).
Correlating the trends in GPCP rainfall and TMI SSTs over 1998–2006 demonstrates that in these satellite-derived analyses, SSTs warm most strongly 8–10 days before the maximum increase in rainfall, then cool most strongly 8–10 days after (Fig. 6a). This suggests a shorter time scale for atmosphere–ocean interactions than Fu and Wang (2004). In contrast to the present study, Fu and Wang (2004) used different observed datasets, a wider band for longitude averaging, and a 20–70-day bandpass filter. In the daily ensemble and north of the equator, SSTs are most strongly correlated with rainfall within 2 days of zero lag (Fig. 6b), consistent with many previous AGCM studies (e.g., Fu and Wang 2004; Rajendran and Kitoh 2006). Results are similar for the 5-day ensemble, although with slightly weaker correlations at all leads and lags (Fig. 6c). The monthly ensemble uses daily SSTs from a linear interpolation between monthly means, and so correlating the 11-day trends in rainfall and SST produces no signal (Fig. 6d).
Even without the correct phase relationship between SSTs and deep convection, an AGCM may still be able to simulate an NPISO with reasonable intensity and propagation speed if the atmosphere responds to warm SST anomalies by generating organized convection. If the SST anomalies have a realistic magnitude and propagate northward—as they do in observations of the NPISO (e.g., Klingaman et al. 2008)—they may be able to induce convection to follow. To diagnose this behavior, we examine the same 11-day linear trend in rainfall that was used to compute the correlations in Fig. 6. Because the mean of the GPCP record or the ensembles could be distorted by variability in the timing of intraseasonal events—if events occur at different times in each year or ensemble member, the mean might falsely indicate that no events occurred—we select 3 yr from the GPCP record and three members from each ensemble to examine closely. We select the three ensemble members (years) with values of the intraseasonal-power metric closest to the most probable value, determined from the PDF for that ensemble (the GPCP record). In other words, we select the members (years) that lie closest to the peak of the corresponding curve in Fig. 5. We refer to these as the “typical” members (years).
Two or three northward-propagating active-break cycles occur in each of the typical years of the GPCP analysis (Figs. 7a–c). In Fig. 7, we have drawn by hand black (pink) lines to trace the northward propagation of active (break) events. As we are using the trend in rainfall to diagnose these events, a persistent active (break) event will appear as white space following the initial red (blue) contours. Thus, the extended break event in July 2002 appears as mostly white space with some scattered synoptic-scale events (Fig. 7a). Most active and break events have a propagation speed of approximately 1° latitude day−1, reaching Indian latitudes 15–20 days after forming in the equatorial Indian Ocean. This is consistent with past observational studies (e.g., Yasunari 1979; Krishnamurti and Subrahmanyam 1982; Gadgil 1990; Lawrence and Webster 2002). Organized ISV does not occur continuously through the season; events can sometimes be as much as 2 months apart (Fig. 7b). Significant variation also occurs in propagation speed and origin, as in the events in 2006 (Fig. 7c). Also, it is worth mentioning that short-time-scale southward propagation can occur, particularly between active-break cycles over the Indian subcontinent and the northern BoB (15°–25°N). Several of these events can be seen around 20°N during early August in Fig. 7b. While these events are notable and appear in each year of the GPCP analysis, they are outside the scope of the present study, which is focused on northward-propagating events on longer, 30–50-day time scales.
The typical ensemble members confirm that the daily ensemble not only has greater ISV in rainfall but also far greater organization to its convective events than the monthly ensemble. The three daily typical members (Figs. 7d–f) demonstrate consistent, northward-propagating increases and decreases in rainfall throughout the monsoon season. Most of these active and break events originate between the equator and 10°N, and many require about 15 days to propagate to India, consistent with the GPCP events. There are some events that propagate more quickly, such that the oscillation appears to have a period of 20–30 days rather than 30–50 days, such as the July events in Fig. 7d. These are consistent with the statistically significant power at periods of 20–30 days in the wavelet analysis (Fig. 4a). On the whole, however, the typical members from the daily ensemble show consistent northward propagation of organized convection on intraseasonal time scales.
Intraseasonal events in the typical members from the 5-day ensemble (Fig. 7g–i) generally have a faster propagation speed than those in the daily ensemble or in GPCP. One of the members shows decidedly weaker events with less-consistent propagation (Fig. 7i). Still, most of the events in the 5-day ensemble are of similar magnitude to those in GPCP, form at near-equatorial latitudes, and show decided northward propagation. The monthly typical members have weaker variability (Figs. 7j–l); most events propagate far too quickly, often appearing at all latitudes between 5° and 20°N simultaneously.
The daily typical members produce intraseasonal events at approximately the same times as the 2005 GPCP data (Fig. 7i)—beginning from the equator near 1 July and 1 September—which suggests that these events are connected to the OSTIA SSTs. The fact that the events are shifted earlier by 8–10 days in the simulations supports this hypothesis, because this was the discrepancy between the model and observations in the phase relationship between SSTs and rainfall (Figs. 6a,b). Thus, the warming (cooling) trend in SSTs that has been observed to precede the active (break) event is coincident with an active (break) event in HadAM3; the model is one-quarter of an NPISO cycle out of phase with observations.
It is interesting to note the southward propagation from the equator that is seen in the September events in the 2005 and 2006 GPCP data (Figs. 7b,c, respectively) and occasionally in the model simulations (particularly Figs. 7f,g). Several previous studies (e.g., Lawrence and Webster 2002) have suggested that equatorial convection could generate equatorially symmetric Rossby cells, of which the Northern Hemisphere cell would be amplified by the northern summer basic state. While the weaker Southern Hemisphere cell was not consistently detectable in our ensembles or the GPCP analysis, the equatorial symmetry seen in some events in Fig. 7 implies that such a mechanism might exist in those simulations.
To further examine the northward propagation of organized convective events in each ensemble, we computed the lead-lag correlations of 30–50-day bandpass-filtered, longitude-averaged (80–90°E) rainfall during the monsoon season. Longitude-averaged rainfall at each latitude point was correlated with the time series of longitude-averaged rainfall at 20°N. This northern latitude was chosen for the base point because while some ensemble members have limited northward propagation, almost all members in the daily ensemble demonstrate some 30–50 day variability in rainfall in the northern BoB, as shown for the ensemble mean in Fig. 2d. Selecting a northerly latitude for the base point therefore provides the best opportunity to detect any latitudinal propagation in the intraseasonal events. For brevity, in Fig. 8 we show the results for the mean correlation over the ensemble or the GPCP analysis record, as well as the leftmost two “typical” members or years shown in Fig. 7.
Substantial interannual variability occurs in the GPCP lead–lag correlations. The mean correlation for the 1997–2006 record shows that intraseasonal equatorial rainfall leads rainfall in the northern BoB by 20–25 days (Fig. 8a). The opposite phase of the oscillation occurs 15–20 days later, although the correlations are weaker and sometimes not statistically significant. In the first typical year (2002; Fig. 8b), however, the propagation speed is much faster, giving a period of approximately 30 days for the oscillation. The second typical year (2005; Fig. 8c) has a somewhat longer period of 30–40 days.
The daily ensemble–mean lead–lag correlation is remarkably similar to the GPCP mean correlation, showing statistically significant northward propagation from 5°N at a lead of about 20 days, with the opposite phase of the oscillation occurring 20–25 days later (Fig. 8d). The daily typical members demonstrate northward propagation from 10°N (Figs. 8e,f), with Fig. 8f also showing southward propagation from the equator. These members and the ensemble mean show a propagation speed of about 1° day−1 from 10°N, in line with past observations and the GPCP analysis. Results from the 5-day ensemble mean (Fig. 8g) and typical members (Figs. 8h,i) are similar to the daily ensemble. By contrast, the monthly ensemble mean shows no propagation of intraseasonal rainfall in any direction (Fig. 8j). The two typical members also display no signal of any kind (Figs. 8k,l).
When combined with the results of Fig. 7 described above, these correlations suggest that in some cases an atmosphere-only model can respond to daily SST variability to produce an NPISO that resembles observations in intensity, speed, and timing. On the other hand, the monthly typical members exhibit no clear northward propagation; they completely fail to produce any signal resembling an organized intraseasonal convective event.
While the typical members from each ensemble represent the most probable amounts of ISV one could obtain from that ensemble, the typical members from the daily ensemble contain substantially more ISV than the typical members from the monthly ensemble. Comparing the typical members from each ensemble has shown that the daily ensemble was far more likely to produce organized, northward-propagating intraseasonal events, but it did not prove that the monthly ensemble was incapable of doing so. The monthly ensemble does contain some members with values of the intraseasonal-power metric equal to or higher than members from the daily ensemble (Fig. 5), and these monthly-ensemble members may show an organized NPISO similar to their daily-ensemble counterparts.
To test this hypothesis, we compare the three members from each ensemble with a value of the intraseasonal-power metric closest to the value for the 2005 GPCP analysis (the dashed line in Fig. 5); these will be referred to as the “observational” members. Because the peak of the daily-ensemble PDF occurs close to the 2005 GPCP value, we exclude the daily typical members from the selection of the daily observational members to prevent repetition. The selection of these members creates a like-with-like comparison not only between the three ensembles in terms of intraseasonal power but also between the ensemble members and the 2005 GPCP data. We repeat the diagnostics of the 11-day centered linear trend in rainfall (Fig. 9) and the lead–lag correlations of 30–50-day rainfall with a base point at 20°N (Fig. 10). As before, the members from the daily ensemble and the 5-day ensemble display more intense and more coherent NPISO-like events than the corresponding members from the monthly ensemble. The daily observational members show several strong events in July and September (Figs. 9a–c) with timing similar to the GPCP analyses (Figs. 7a–c), as for the typical members from the daily ensemble. The observational members from the 5-day ensemble again show one member with weak intraseasonal events (Fig. 9d), but two members with consistent intraseasonal events with magnitude similar to those in GPCP (Figs. 9e,f). Intraseasonal events in the monthly observational members are scattered at best, propagate too quickly, and rarely form south of 10°–15°N (Figs. 9g–i). The lead–lag correlations also demonstrate that the daily and 5-day observational members display consistent movement of convection from the equatorial Indian Ocean to the subcontinent (Figs. 10a–f), while none of the monthly ensemble members showed any latitudinal propagation in their intraseasonal events (Figs. 10g–i).
The results of Figs. 9, 10 are particularly remarkable because they compare members from each ensemble that have similar amounts of intraseasonal power. This indicates that even when the monthly ensemble members manage to generate substantial power in 30– 50-day rainfall, the spatiotemporal pattern of rainfall does not resemble the northward-propagating oscillation. While the monthly ensemble could not generate NPISO-like events even in members with high (for that ensemble) amounts of intraseasonal power, the observational members from the daily ensemble produce intense northward-propagating events that in some cases very closely resembled observations. This speaks to the ability of high-frequency SSTs with realistic ISV to not only generate greater ISV in rainfall but to organize convective events and encourage their northward propagation toward India. We have previously demonstrated that the OSTIA SSTs have more accurate ISV than coupled-model SSTs or the NCEP OISSTs used in previous studies (sections 1c and 2b). Thus, the fact that an AGCM driven by daily OSTIA SSTs can produce an NPISO similar to observations strengthens the case for high-frequency SST variations as a critical component of monsoon intraseasonal variability.
In the lead–lag correlations for the daily and 5-day ensembles, as well for the GPCP analyses, the signal of an intraseasonal oscillation is often seen first in the north (i.e., over India) followed by the opposite phase forming in the south (i.e., over the eastern equatorial Indian Ocean). This holds for the mean correlations (Fig. 8) as well for the typical (Fig. 8) and observational members (Fig. 10). Such a pattern implies that intraseasonal monsoon rainfall cannot be predicted solely from events propagating from the south; intraseasonal rainfall over India displays the greatest predictability once either an active or a break event has reached the subcontinent itself. Klingaman et al. (2008) reached a similar conclusion by examining lead–lag correlations of NOAA OLR observations between the equatorial Indian Ocean and the Indian subcontinent. Confirmation from a GCM of the dominance of the Indian terminus of the oscillation for predictability lends further credibility to this theory.
4. Discussion
Studies that have investigated the ability of atmosphere-only models to simulate the intraseasonal variability of the Indian monsoon have suggested, without exception, that AGCMs contain NPISO-like variability but cannot reproduce either the strength or propagation speed of the oscillation (e.g., Fu et al. 2003; Waliser et al. 2003; Fu and Wang 2004; Rajendran and Kitoh 2006). Those studies that have also employed coupled models have found that air–sea coupling improved representations of the NPISO. The first result led to the conclusion that the NPISO is an internal atmospheric mode, while the second implied that atmosphere–ocean feedbacks were essential to generate a strong NPISO with the appropriate meridional velocity. Our results support the first conclusion, as some of the monthly-ensemble members contain intraseasonal power that agrees with the GPCP analyses (Fig. 5). These studies considered the key failing of AGCMs to be their inability to represent the near-quadrature phase relationship between sea surface temperatures and convection. We noted in section 1c, however, that all previous studies to simulate the NPISO with an AGCM have employed SST forcing that substantially underestimates the intraseasonal SST anomalies associated with individual active and break events.
Here, we have demonstrated that an AGCM can reproduce NPISO-like variability with greater fidelity if forced by SSTs with more realistic intraseasonal variability, even though the phase relationship between rainfall and SST remains incorrect (Fig. 6). It is reasonable to conclude, therefore, that it is not necessarily the zero-lag phase relationship between the SSTs and convection that resulted in weak NPISO-like variability in past AGCM experiments, but the low magnitude of daily, submonthly, and intraseasonal SST variability in the forcing SSTs. Using coupled-model SSTs Fu and Wang (2004) found improvement in ISV when an AGCM was forced by daily SSTs instead of monthly means, but they reported that the ISV in a simulation with daily SSTs was still 20%–25% less than the corresponding coupled-model integration (their Fig. 4). By contrast, we have shown that our daily ensemble has variability that agrees with observations (Fig. 5). This suggests that not only is the frequency of the SST anomalies critical for an accurate simulation of the NPISO but also the magnitude of the SST anomalies on submonthly time scales. It could be argued that without atmosphere-to-ocean feedbacks our modeled NPISO events may not have been driven by entirely the correct physical mechanisms, but the very existence of NPISO events speaks to the ability of an AGCM to respond to spatially coherent and realistic high-frequency SST anomalies with organized intraseasonal convection.
When we compared the daily and 5-day ensembles, we found that while the 5-day ensemble contained somewhat less power at intraseasonal periods (Figs. 4e, 5), the simulation of intraseasonal events was still far more similar to those in the GPCP analyses than the monthly ensemble. Thus, we can conclude that SST variability on periods of less than 5 days makes some contribution to an accurate representation of intraseasonal variability but that one can probably neglect these shortest periods without completely destroying the fidelity of the ISV. This conclusion is contingent upon the retention of substantial ISV in the 5-day means of SST. We noted in section 2b that the 5-day means of the OSTIA SST dataset had only slightly less ISV than the daily OSTIA product; a comparison of Figs. 4b,d confirms this.
It is also interesting to note that the 5-day ensemble appears in all metrics to be a slight degeneration of the daily ensemble toward the monthly ensemble. This can be seen most clearly in the wavelet transforms of rainfall (Fig. 4, left column) and in the PDF of the intraseasonal-power metric (Fig. 5). Without conducting any further ensembles, it is impossible to say whether this degeneration would continue in a linear fashion if successively longer means (10, 15, 20 days, and so on) were taken. The prospect that the simulation of the ISV might smoothly degrade with longer means of SST is intriguing for its possible implications for the study of multiscale atmosphere–ocean interactions.
The results of this study have clear implications for future ACGM and coupled-model experiments. Simulations that remove or underrepresent high-frequency SST anomalies severely limit the ability of the model to reproduce an NPISO that resembles observations in frequency, intensity, and propagation speed. This is equally true for coupled models with dynamic SSTs as for atmosphere-only simulations. As previously mentioned in section 1c, many coupled models substantially underestimate SST variability due to excessively high thermal inertia, a consequence of coarse vertical resolution in the upper ocean. Our results support the claim of Bernie et al. (2007) that this vertical resolution must be improved, because it will improve daily and intraseasonal SST variability and hence the predictability of the NPISO. Similarly, atmosphere-only simulations that continue to use monthly mean SST forcing are imposing an unnecessary constraint on the ability of the model to generate organized convection in response to high-frequency SST anomalies.
Further, our experiments suggest that including high-frequency, realistic SST anomalies is necessary for capturing the initiation and propagation of NPISO events in an AGCM. Thus, the predictability of active and break events in such a model must be in part a function of the accuracy of these anomalies. Fu et al. (2007) recently showed an increase in the NPISO predictability time scale of nearly one week in a coupled model over an AGCM. Many short- and medium-range weather prediction models, however, are AGCMs with persisted SST anomalies. Without either an interactive ocean or a technique for adding high-frequency SST perturbations, these models will likely fail to predict the frequency and intensity of NPISO events. In this case, an interactive ocean need not be more than a mixed layer scheme (e.g., Woolnough et al. 2007), as the response of the Indian Ocean on intraseasonal time scales is likely to be dominated by thermodynamic processes. Future work will examine the impact of adding a mixed layer scheme to an AGCM on NPISO predictability.
5. Summary and conclusions
Our ensembles of HadAM3 simulations forced with OSTIA SSTs demonstrated that an AGCM could respond to accurate, high-frequency SST forcing to organize convection and generate NPISO-like variability. We noted in section 3a that the members of the daily and 5-day ensembles contained substantially more power in 30–50-day rainfall than the members from the monthly ensemble. Figure 5 demonstrated this conclusively; it also showed that the intraseasonal variability in the daily ensemble matched that of the GPCP analysis remarkably well. Not all members of the daily and 5-day ensembles contained appropriate amounts of intraseasonal power. The inclusion of high-frequency SST anomalies from an SST analysis with more realistic ISV, however, allowed more ensemble members to contain this power than those in the monthly ensemble. The key result here, then, is that while HadAM3 did not always respond to daily SST forcing by organizing NPISO-like variability in convection, the daily SST forcing significantly improved the chances that an individual ensemble member would produce such variability (over monthly mean SST forcing).
Not only did the members of the daily ensemble twice as frequently reproduce intraseasonal rainfall variability in line with observations than the monthly-ensemble members, but those that did showed an organization and propagation to the convection that resembled the NPISO in the GPCP analyses. Even when an integration forced with monthly mean SSTs produced substantial intraseasonal variability in rainfall, this was not organized into a coherent, northward-propagating intraseasonal event. Taken together, these results and those from the previous paragraph indicate that daily SST forcing can induce an atmosphere-only model to produce accurate amounts of 30–50-day power in rainfall far more frequently than monthly mean SST forcing. Daily SST forcing can also aid the organization of that variability into coherent convective events that move northward with a phase speed that agrees with the NPISO.
Furthermore, the results of the 5-day ensemble indicated that 5-day means of OSTIA SSTs still produced a reasonable simulation of NPISO-like events. The 5-day ensemble contained slightly less intraseasonal power than the daily ensemble, which indicated that the highest-frequency SST variability in the OSTIA analyses plays some role in improving the ISV of convection. Individual NPISO-like events in the 5-day ensemble were mostly of reasonable magnitude and propagation speed when compared to GPCP, however, which indicated that the most critical frequencies of SST for the NPISO are likely 5 days or longer.
In our lead–lag correlations of intraseasonal rainfall (Figs. 8, 10), we concluded that the signal of the NPISO was frequently found first over India, before the opposite phase of the oscillation forms over the equatorial Indian Ocean. This correlation pattern was found for the daily and 5-day ensembles, as well as for the GPCP analyses, in both the mean correlation and in correlations from individual ensemble members or years. Klingaman et al. (2008) reached similar conclusions of NPISO development using TMI SSTs and ECMWF reanalysis; our GCM results have strengthened those findings.
Even with accurate, daily SST forcing, our AGCM still collocated the heaviest rainfall too readily over the warmest sea surface temperatures (Fig. 6b). This was in direct contrast to observations, which have consistently shown the strongest convection to be associated with cooling SSTs; past studies have found that warm SSTs are associated with NPISO break events and suppressed convection (e.g., Fu and Wang 2004; Klingaman et al. 2008). We determined the incorrect phase relationship to likely be due to the lack of feedbacks between convection and the ocean surface, an intrinsic failure of atmosphere-only models when simulating tropical intraseasonal variability that cannot be resolved by improving the SST forcing. This failure could also be due to the HadAM3 convection scheme responding too readily to the warm SSTs. Our results indicated that a coupled model must be able to accurately represent the high-frequency SST anomalies that are so critical to the intraseasonal variability of convection and rainfall. High-frequency SSTs play a key role in strengthening and maintaining the NPISO and so cannot be neglected, regardless of whether they are being forced in an atmosphere-only model or simulated in a coupled model.
Acknowledgments
NPK was supported by a scholarship from the Marshall Aid Commemoration Commission. PMI, HW, and JMS were supported by and are members of the National Centre for Atmospheric Science, a Natural Environment Research Council Collaborative Centre. The authors thank Dr. George Kiladis and three anonymous reviewers for their comments, which resulted in substantial improvements to a previous version of this manuscript. The authors also thank Mr. John Stark of the Hadley Centre at the Met Office for his assistance in obtaining the OSTIA analyses. The Indian Meteorological Department provided the gridded dataset of climatological monsoon rainfall shown in Fig. 2b. The Indian Institute for Tropical Meteorology provided the all-India rainfall dataset used in Fig. 3. The wavelet transforms were computed using software provided by C. Torrence and G. Compo (available online at http://atoc.colorado.edu/research/wavelets.
REFERENCES
Annamalai, H., and J. M. Slingo, 2001: Active/break cycles: Diagnosis of the intraseasonal variability of the Asian Summer Monsoon. Climate Dyn., 18 , 85–102.
Annamalai, H., and K. R. Sperber, 2005: Regional heat sources and the active and break phases of boreal summer intraseasonal (30–50 day) variability. J. Atmos. Sci., 62 , 2726–2748.
Annamalai, H., J. M. Slingo, K. R. Sperber, and K. Hodges, 1999: The mean evolution and variability of the Asian summer monsoon: Comparison of ECMWF and NCEP–NCAR Reanalyses. Mon. Wea. Rev., 127 , 1157–1186.
Bannister, R. N., 2007: Can wavelets improve the representation of forecast error covariances in variational data assimilation? Mon. Wea. Rev., 135 , 387–408.
Bernie, D. J., S. J. Woolnough, and J. M. Slingo, 2005: Modeling diurnal and intraseasonal variability of the ocean mixed layer. J. Climate, 18 , 1190–1202.
Bernie, D. J., E. Guilyardi, G. Madec, J. M. Slingo, and S. J. Woolnough, 2007: Impact of resolving the diurnal cycle in an ocean-atmosphere GCM. Part 1: A diurnally forced GCM. Climate Dyn., 29 , 575–590.
Bhat, G. S., and Coauthors, 2001: BOBMEX: The Bay of Bengal Monsoon Experiment. Bull. Amer. Meteor. Soc., 82 , 2217–2243.
Bhat, G. S., G. A. Vecchi, and S. Gadgil, 2004: Sea surface temperature of the Bay of Bengal derived from the TRMM Microwave Imager. J. Atmos. Oceanic Technol., 21 , 1283–1290.
Breaker, L. C., P. C. Liu, and C. Torrence, 2001: Intraseasonal oscillations in sea surface temperature, wind stress, and sea level off the central California coast. Cont. Shelf Res., 21 , 727–750.
Chang, Y. S., D. Jeon, H. Lee, H. S. An, J. W. Seo, and Y. H. Youn, 2004: Interannual variability and lagged correlation during strong El Nino events in the Pacific Ocean. Climate Res., 27 , 51–58.
Daubechies, I., 1990: The wavelet transform time-frequency localization and signal analysis. IEEE Trans. Inf. Theory, 36 , 961–1004.
Daubechies, I., 1992: Ten Lectures on Wavelets. Society for Industrial and Applied Mathematics, 357 pp.
Donlon, C., and Coauthors, 2007: The Global Ocean Data Assimilation Experiment High-Resolution Sea Surface Temperature Pilot Project. Bull. Amer. Meteor. Soc., 88 , 1197–1213.
Fiornio, M., cited. 1997: AMIP II sea surface temperature and sea ice concentration observations. [Available online at http://www-pcmdi.llnl.gov/projects/amip/AMIP2EXPDSN/BCS_OBS/amip2_bcs.htm.].
Flatau, M., P. J. Flatau, P. Phoebus, and P. P. Niler, 1997: The feedback between equatorial convection and local radiative and evaporative processes: The implications for intraseasonal oscillations. J. Atmos. Sci., 54 , 2373–2386.
Fu, X., and B. Wang, 2004: The boreal-summer intraseasonal oscillations simulated in a hybrid coupled atmosphere—ocean model. Mon. Wea. Rev., 132 , 2628–2649.
Fu, X., B. Wang, and T. Li, 2002: Impacts of air–sea coupling on the simulation of mean Asian summer monsoon in ECHAM4 model. Mon. Wea. Rev., 130 , 2889–2904.
Fu, X., B. Wang, T. Li, and J. P. McCreary, 2003: Coupling between northward-propagating, intraseasonal oscillations and sea surface temperature in the Indian Ocean. J. Atmos. Sci., 60 , 1733–1753.
Fu, X., B. Wang, D. E. Waliser, and L. Tao, 2007: Impact of atmosphere–ocean coupling on the predictability of monsoon intraseasonal oscillations. J. Atmos. Sci., 64 , 157–174.
Gadgil, S., 1990: Poleward propagation of the ITCZ: Observations and theory. Mausam, 41 , 285–290.
Goupillaud, P., A. Grossman, and J. Morlet, 1984: Cycle-octave and related transforms in seismic signal analysis. Geoexploration, 23 , 85–102.
Harrison, D. E., and G. A. Vecchi, 2001: January 1999 Indian Ocean cooling event. Geophys. Res. Lett., 28 , 3717–3720.
Hartmann, D. L., M. L. Michelsen, and S. A. Klein, 1992: Seasonal variations of tropical intraseasonal oscillations: A 20–25-day oscillation in the western Pacific. J. Atmos. Sci., 49 , 1277–1289.
Hu, Z. Z., and T. Nitta, 1996: Wavelet analysis of summer rainfall over North China and India and SOI using 1891–1992 data. J. Meteor. Soc. Japan, 74 , 833–844.
Inness, P. M., and J. M. Slingo, 2003: Simulation of the Madden–Julian oscillation in a coupled general circulation model. Part I: Comparison with observations and an atmosphere-only GCM. J. Climate, 16 , 345–364.
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.
Kang, I., and Coauthors, 2002: Intercomparisons of the climatological variations of Asian summer monsoon precipitation simulation by 10 GCMs. Climate Dyn., 19 , 383–395.
Kemball-Cook, S., and B. Wang, 2001: Equatorial waves and air–sea interaction in the boreal summer intraseasonal oscillation. J. Climate, 14 , 2923–2942.
Kemball-Cook, S., B. Wang, and X. Fu, 2002: Simulation of the ISO in the ECHAM4 model: The impact of coupling with an ocean model. J. Atmos. Sci., 59 , 1433–1453.
Kestin, T. S., D. J. Karoly, J. I. Yang, and N. A. Raynor, 1998: Time-frequency variability of ENSO and stochastic simulations. J. Climate, 11 , 2258–2272.
Klingaman, N. P., H. Weller, J. M. Slingo, and P. M. Inness, 2008: The intraseasonal variability of the Indian summer monsoon using TMI sea surface temperatures and ECMWF reanalysis. J. Climate, 21 , 2519–2539.
Krishnamurti, T. N., and D. Subrahmanyam, 1982: The 30–50 day mode at 850 mb during MONEX. J. Atmos. Sci., 39 , 2088–2095.
Lawrence, D. M., and P. J. Webster, 2002: The boreal intraseasonal oscillation: Relationship between northward and eastward movement of convection. J. Atmos. Sci., 59 , 1593–1606.
Liess, S., and S. Bengtsson, 2004: The intraseasonal oscillation in ECHAM4. Part II: Sensitivity studies. Climate Dyn., 22 , 671–688.
Livezey, R. E., and W. E. Chen, 1983: Statistical field significance and its determination by Monte Carlo techniques. Mon. Wea. Rev., 111 , 46–59.
Lorenc, A. C., 1981: A global three-dimensional multivariate statistical interpolation scheme. Mon. Wea. Rev., 109 , 701–721.
Madden, R. A., and P. R. Julian, 1971: Detection 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, 1972: Description of global-scale circulation cells in the tropics with a 40–50 day period. J. Atmos. Sci., 29 , 1109–1123.
Maloney, E. D., and A. H. Sobel, 2005: Surface fluxes and ocean coupling in the tropical intraseasonal oscillation. J. Climate, 17 , 4368–4386.
Meehl, G. A., J. M. Arblaster, and D. M. Lawrence, 2006: Monsoon regimes in the CCSM3. J. Climate, 19 , 2482–2495.
Pope, V. D., M. L. Gallani, P. R. Rowntree, and R. A. Stratton, 2000: The impact of the new physical parameterizations in the Hadley Centre climate model. Climate Dyn., 16 , 123–146.
Rajeevan, M., J. Bhate, J. D. Kale, and B. Lal, 2005: High resolution daily gridded rainfall data the Indian region: Analysis of break and active monsoon spells. Curr. Sci. India, 91 , 296–306.
Rajendran, K., and A. Kitoh, 2006: Modulation of tropical intraseasonal oscillations by atmosphere–ocean coupling. J. Climate, 19 , 366–391.
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.
Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate, 7 , 929–948.
Schott, F. A., and J. P. McCreary, 2001: The monsoon circulation of the Indian Ocean. Prog. Oceanogr., 51 , 1–123.
Senan, R., D. S. Anitha, and D. Sengupta, 2001: Validation of SST and wind speed from TRMM using North Indian Ocean moored buoy observations. CAOS Tech. Rep. 2001AS1, Centre for Atmospheric and Oceanic Sciences, Indian Institute of Science, Bangalore, India, 31 pp.
Sengupta, D., and M. Ravichandran, 2001: Oscillations of the Bay of Bengal sea surface temperature during the 1998 summer monsoon. Geophys. Res. Lett., 28 , 2033–2036.
Sengupta, D., B. N. Goswami, and R. Senan, 2001: Coherent intraseasonal oscillations of ocean and atmosphere during the Asian summer monsoon. Geophys. Res. Lett., 20 , 4127–4130.
Slingo, J. M., and Coauthors, 1996: Intraseasonal oscillations 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.
Stark, J. D., C. J. Donlon, M. J. Martin, and M. E. McCulloch, 2007: OSTIA: An operational, high resolution, real time, global sea surface temperature analysis system. Proc. Conf. on Oceans ’07 IEEE Aberdeen, Rep. 061214-029, Aberdeen, United Kingdom, IEEE, 4 pp. [Available online at http://ghrsst-pp.metoffice.com/pages/latest_analysis/ostia.html.].
Stratton, R. S., 1999: A high resolution AMIP integration using the Hadley Centre model HadAM2b. Climate Dyn., 15 , 9–28.
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, UCRL-MI-125597, Program for Climate Model Diagnosis and Intercomparison, Lawrence Livermore National Laboratory, Livermore, CA, 28 pp.
Torrence, C., and G. P. Compo, 1998: A practical wavelet analysis. Bull. Amer. Meteor. Soc., 13 , 61–78.
Waliser, D. E., 2006: Intraseasonal variability. The Asian Monsoon, B. Wang, Ed., Praxis, 203–257.
Waliser, D. E., C. Jones, J. K. Schemm, and N. E. Graham, 1999a: A statistical extended-range tropical forecast model based on the slow evolution of the Madden–Julian oscillation. J. Climate, 12 , 1918–1939.
Waliser, D. E., K. M. Lau, and J. H. Kim, 1999b: The influence of coupled sea surface temperatures on the Madden–Julian oscillation: A model perturbation experiment. J. Atmos. Sci., 56 , 333–358.
Waliser, D. E., and Coauthors, 2003: AGCM simulations of intraseasonal monsoon variability associated with the Asian summer monsoon. Climate Dyn., 21 , 423–446.
Wang, B., and X. Xie, 1997: A model for boreal summer intraseasonal oscillation. J. Atmos. Sci., 54 , 72–86.
Webster, P. J., and C. Hoyos, 2004: Prediction of monsoon rainfall and river discharge on 15–30-day time scales. Bull. Amer. Meteor. Soc., 85 , 1745–1765.
Webster, P. J., V. O. Magaña, T. N. Palmer, J. Shukla, R. A. Tomas, M. Yanai, and T. Yasunari, 1998: Monsoons: Processes, predictability, and the prospects for prediction. J. Geophys. Res., 103 , 14451–14510.
Webster, P. J., and Coauthors, 2002: The JASMINE pilot study. Bull. Amer. Meteor. Soc., 83 , 1603–1630.
Woolnough, S. J., J. M. Slingo, and B. J. Hoskins, 2000: The relationship between convection and sea surface temperatures on intraseasonal timescales. J. Climate, 13 , 2086–2104.
Woolnough, S. J., F. Vitart, and M. A. Balmaseda, 2007: The role of the ocean in the Madden–Julian oscillation: Implications for MJO prediction. Quart. J. Roy. Meteor. Soc., 133 , 117–128.
Yadava, M. G., and R. Ramesh, 2007: Significant longer-term periodicities in the proxy record of the Indian monsoon rainfall. New Astron., 12 , 544–555.
Yasunari, T., 1979: Cloudiness fluctuations associated with the Northern Hemisphere summer monsoon. J. Meteor. Soc. Japan, 57 , 227–242.
Zheng, Y., D. E. Waliser, W. F. Stern, and C. Jones, 2004: The role of coupled sea surface temperatures in the simulation of the tropical intraseasonal oscillation. J. Climate, 17 , 4109–4134.
(a) The std dev in JJAS 30–50-day bandpass-filtered daily OSTIA SSTs (°C) and (b) the std dev in JJAS 30–50-day bandpass-filtered daily OSTIA SSTs, divided by the std dev in JJAS 30–50-day bandpass-filtered daily OSTIA SSTs. For (b), the monthly mean SSTs were first linearly interpolated to daily values. In (a), black line contours show the percentage of the total variability for which the 30–50-day band accounts, with contours at 25%, 50%, and 75%.
Citation: Journal of Climate 21, 23; 10.1175/2008JCLI2329.1
(a) The ensemble-mean JJAS-mean rainfall rate (mm day−1) from the daily ensemble; (b) the climatological JJAS-mean rainfall rate (mm day−1) from the IMD gridded data; (c) the difference in the ensemble-mean JJAS-mean rainfall rate (mm day−1) for the daily minus the monthly ensemble; (d) the ensemble-mean std dev of 30–50-day bandpass-filtered JJAS rainfall from the daily ensemble; (e) the ratio of the ensemble-mean std dev of 30–50-day bandpass-filtered JJAS rainfall for the daily divided by the monthly ensemble; (f) same as (e), but for the daily divided by the 5-day ensemble. Gray (black) dots indicate statistical significance at the 5% (10%) level using (c) a two-tailed Student’s t test and (e),(f) a two-tailed F test.
Citation: Journal of Climate 21, 23; 10.1175/2008JCLI2329.1
The area-averaged rainfall rate (mm day−1) taken over all Indian land points (10°–30°N, 70°–90°E) for (a) the daily ensemble and (b) the monthly ensemble for (black) the individual ensemble members and (red) the ensemble mean. The yellow line gives the daily climatology from the IITM all-India rainfall dataset (1901–2005); the cyan line gives the all-India rainfall for 2005.
Citation: Journal of Climate 21, 23; 10.1175/2008JCLI2329.1
The ensemble-mean one-dimensional wavelet transform of (left column) area-averaged rainfall and (right column) area-averaged SSTs in the northern BoB (15°–20°N, 85°–90°E), for (top row) the daily ensemble, (middle row) the monthly ensemble, and (bottom row) the 5-day ensemble. The solid black contours indicate the 90% and 95% confidence intervals against red noise, while the dashed black contour indicates the region outside of which edge effects distort the results.
Citation: Journal of Climate 21, 23; 10.1175/2008JCLI2329.1
PDFs of the intraseasonal-power metric for (long dash) the daily ensemble, (short dash) the monthly ensemble, (dashed–dotted) the 5-day ensemble, and (solid) the 1997–2006 1° × 1° gridded, daily precipitation analyses from GPCP. The value for GPCP in 2005—the year of the SST forcing—is shown as a vertical, dotted line.
Citation: Journal of Climate 21, 23; 10.1175/2008JCLI2329.1
Lead–lag correlations between the linear trend (11-day window) in longitude-averaged (80°–90°E) rainfall and SST over JJAS for (a) GPCP rainfall and TMI SSTs from 1998–2006, (b) the daily ensemble, (c) the 5-day ensemble, and (d) the monthly ensemble. Gray shading indicates statistical significance at the 5% level.
Citation: Journal of Climate 21, 23; 10.1175/2008JCLI2329.1
The linear trend (11-day window) in longitude-averaged rainfall (80°–90°E; mm day−2) for (a)–(c) three typical years from the GPCP analysis and three typical members from the (d)–(f) daily ensemble, (g)–(i) 5-day ensemble, and (j)–(l) monthly ensemble with values of the intraseasonal-power metric closest to the most probable value for that ensemble. The hand-drawn black (pink) lines trace the northward propagation of active (break) events in each ensemble. The bottom color bar applies to all panels.
Citation: Journal of Climate 21, 23; 10.1175/2008JCLI2329.1
Lead–lag correlations of 30–50-day bandpass-filtered, longitude-averaged JJAS rainfall (80°–90°E) with the base point at 20°N for (top row) the GPCP analysis, (second row) the daily ensemble, (third row) the 5-day ensemble, and (bottom row) the monthly ensemble. (left column) The mean correlation over the entire 1997–2006 GPCP analysis and each ensemble; (right columns) typical years from the GPCP analysis and two of the typical ensemble members from each ensemble. Gray shading indicates statistical significance at the 5% level.
Citation: Journal of Climate 21, 23; 10.1175/2008JCLI2329.1
Same as Fig. 7, but for the three observational members from (a)–(c) the daily ensemble, (d)–(f) the 5-day ensemble, and (g)–(i) the monthly ensemble, with a value of the intraseasonal-power metric closest to the value from the 2005 GPCP analysis.
Citation: Journal of Climate 21, 23; 10.1175/2008JCLI2329.1
Same as Fig. 8, but for the three observational ensemble members shown in Fig. 9.
Citation: Journal of Climate 21, 23; 10.1175/2008JCLI2329.1
