Boreal summer intraseasonal oscillation in a superparameterized GCM: effects of air-sea coupling and ocean mean state

The effect of air-sea coupling on the simulated boreal summer intraseasonal oscillation (BSISO) is examined using atmosphere—ocean-mixed-layer coupled (SPCAM3-KPP) and uncoupled configurations of the Super-Parameterized (SP) Community Atmospheric Model, version 3 (SPCAM3). The coupled configuration is constrained to either the observed ocean mean state or the mean state from the SP coupled configuration with a dynamic ocean (SPCCSM3), to understand the effect of mean state biases on the BSISO in the latter. All configurations overestimate summer mean subtropical rainfall and 15 its intraseasonal variance. All configurations simulate realistic BSISO northward propagation over the Indian Ocean and western Pacific, in common with other SP configurations. Constraining SPCAM3-KPP to the SPCCSM3 mean state reduces the overestimated BSISO variability, but also weakens BSISO propagation. Using the SPCCSM3 mean state also introduces a one-month delay to the BSISO seasonal cycle compared to SPCAM3-KPP with the observed ocean mean state, which matches well with the reanalysis. The phase 20 relationship between intraseasonal rainfall and sea surface temperature (SST) is captured by all coupled models, but with a shorter delay between suppressed convection and warm SST relative to the reanalysis. Prescribing the 31-day smoothed SSTs from the SPCAM3-KPP simulations in SPCAM3 worsens the overestimated BSISO variance. This suggests that airsea coupling improves the amplitude of the simulated BSISO. Based on a Taylor diagram, SPCCSM3 mean state SST biases and air-sea coupling both lead to higher simulated BSISO fidelity, largely due to their ability to suppress the overestimated 25 subtropical BSISO variance.


Introduction
The intraseasonal oscillation (ISO) is the most vigorous sub-seasonal signal in the tropics (Zhang, 2005). It interacts closely with tropical climate and weather systems, such as the El Niño-Southern Oscillation (ENSO) and tropical cyclones (Kessler et al., 1995;Zhang and Gottschalck, 2002;Mcphaden, 2004;Wu et al., 2007), and even mid-latitude systems (Ding and 30 Wang, 2007;Moon et al., 2013). Compared to the boreal winter ISO (i.e., the Madden and Julian Oscillation;MJO;Madden andJulian, 1971, 1972), the boreal summer ISO (BSISO) shifts away from the equator to the Asian summer monsoon (ASM) region Lau and Waliser, 2012). Thus, the BSISO is connected strongly to the onset, active and break phases of ASM (Yasunari, 1979;Annamalai and Slingo, 2001;Lau and Waliser, 2012). The frequency of extreme events over the ASM region is also highly related to BSISO activity (Ren et al., 2013;Li et al., 2015;Hsu et al., 2016Hsu et al., , 2017. 35 Realistic representations of the BSISO in climate models remain a challenge, although some improvements have been achieved in recent decades. Current state-of-art general circulation models (GCMs) still have difficulty to simulate properly the BSISO spatial pattern (Sperber and Annamalai, 2008;Sperber et al., 2013;Hu et al., 2017) and its northwest-southeast tilted rain band structure (Lin et al., 2006;Sabeerali et al., 2013). In contrast, the northward propagation, which is the most significant feature of BSISO, is captured by more models. Fidelity for northward propagation improved in models that 40 contributed to the Coupled Model Intercomparison Project (CMIP) phase 5, relative to the CMIP phase 3 (Sabeerali et al., 2013;Sperber et al., 2013). Most models with reasonable northward propagation also simulated a good eastward propagation along the equatorial Indian Ocean.
The representation of convection is largely responsible for the ability of GCMs to simulate BSISO characteristics (Maloney and Hartmann, 2001;Randall et al., 2007;Jiang et al., 2016). Using the Hadley Centre atmospheric GCM 45 (AGCM), Klingaman and Woolnough (2014) found that increasing the convective entrainment and detrainment rates considerably improved deficient MJO-like variability in the Indian and Pacific Oceans. In recent years, studies have shown that "superparameterized" GCMs have high fidelity for simulating ISO variability (Benedict and Randall, 2009;Jiang et al., 2015;Neena et al., 2017). In superparameterized GCMs, the traditional cumulus parameterization is replaced by a twodimensional cloud-resolving model in each grid column to calculate the cloud and radiation physics tendencies 50 (Khairoutdinov and Randall, 2003;Khairoutdinov et al., 2005). By comparing different versions of the National Center for Atmospheric Research (NCAR) Community Atmospheric Model (CAM), DeMott et al. (2014) showed that the superparameterized CAM (SPCAM3) demonstrated better BSISO characteristics than CAM with the standard convective parameterization.
Besides the convective parameterization scheme, the effect of air-sea interaction on simulated ISO variability has also 55 received growing attention. Many model studies report a substantially improved BSISO representations in coupled GCMs (CGCMs) compared to the corresponding AGCMs (Fu and Wang, 2004;Rajendran and Kitoh, 2006;Klingaman et al., 2008;DeMott et al., 2014). In contrast, Ajayamohan et al. (2011) found no changes in the BSISO properties with coupling. Peatman and Klingaman (2018) investigated the effect of air-sea coupling on the simulated Indian summer monsoon in the UK Met Office Unified Model Global Ocean Mixed layer configuration (MetUM-GOML2.0), a similar model framework 60 that used in this study. Coupling degraded the atmospheric basic state, but slightly improved the BSISO northward propagation.
Incorporating air-sea interaction in CGCMs inevitably introduces atmospheric and oceanic mean state biases. When studies compare CGCMs to AGCMs prescribed with observed SSTs (Kemball-Cook et al., 2002;Zhang et al., 2006;DeMott et al., 2011;Roxy et al., 2013), the differences between simulations may largely come from mean-state differences, rather than air-sea coupling. Due to the strong control on low-level moisture and convergence anomalies, mean state biases may degrade simulated intraseasonal variability (Hendon, 2000;Inness et al., 2003;Sperber et al., 2005;Bollasina and Ming, 2013). Using the National Centers for Environmental Prediction (NCEP) coupled Climate Forecast System (CFS) model, Seo et al. (2007) showed that BSISO development and propagation were largely improved when the CGCM cold SST bias was removed via flux corrections. They suggested that the role of air-sea interactions would be more accurate under an ocean 70 mean state with smaller SST biases. To reduce the mean-state differences between CGCMs and AGCMs, time-varying SST from CGCMs should be used to drive AGCMs (Fu and Wang, 2004;Seo et al., 2007;Levine and Turner, 2012;DeMott et al., 2015). However, this quantifies the role of air-sea coupling only under the biased CGCM mean state.
Through flux correction techniques, CGCMs can be constrained to a given climatological ocean state. Such techniques are more effective in CGCMs with simple ocean models, which lack ocean dynamics, as dynamics may interfere with the 75 prescribed flux correction. Simple ocean models also lack coupled modes of variability, such as ENSO, feedbacks from which can influence the perceived effects of coupling on sub-seasonal variability (Klingaman and DeMott, 2020). CGCMs with simplified oceans are a useful tool to investigate the roles of air-sea coupling and mean-state biases in the simulation of BSISO. In this paper, we examine the role of air-sea interactions and mean state biases in the simulated BSISO using a configuration of SPCAM3 coupled to a mixed-layer ocean model, constrained to observed ocean mean state and mean state 80 from the superparameterized Community Climate System Model (CCSM), version 3 (SPCCSM3; Stan et al. 2010). The model experiments, validation data, and analysis methods are described in section 2. Section 3 analyses the influence of airsea coupling and mean state biases in the spatial pattern and propagation of simulated BSISO. Results are discussed and summarized in sections 4 and 5, respectively.

Models
SPCAM3 (Khairoutdinov et al., 2005) is employed in this study, due to its high fidelity for the simulated ISO (Jiang et al., 2015;Neena et al., 2017). It has a horizontal resolution of T42 and a vertical resolution of 30 levels. The embedded 2-D cloud-process resolving model, used to calculate convective and radiative temperature and moisture tendencies, has 32 columns with a fine resolution (4km) (Khairoutdinov and Randall, 2003). To understand the role of air-sea interaction in the 90 BSISO simulation, SPCAM3 is coupled to the Multi-Column K Profile Parameterization (MC-KPP) mixed-layer ocean model (referred to as "SPCAM3-KPP"). MC-KPP uses the KPP vertical mixing scheme of Large et al. (1994). It consists of many independent 1-D columns, with one column under each AGCM grid point to which it is coupled. Therefore, the horizontal resolution of MC-KPP is the same as SPCAM3. MC-KPP has a fine vertical resolution, with 100 points in a 1000m water column, 70 of which are in the top 300m, and a ~1m resolution in the upper 20m. Further details on SPCAM3-95 KPP can be found in Klingaman and DeMott (2020). https://doi.org/10.5194/gmd-2020-63 Preprint. Discussion started: 24 April 2020 c Author(s) 2020. CC BY 4.0 License.
Since ocean dynamics are absent in MC-KPP, SPCAM3-KPP must be constrained to a reference ocean climatology, which can be taken from an observation-based dataset, an ocean model or a CGCM simulation. This makes MC-KPP an attractive option for investigating the role of air-sea coupling under different ocean mean states. To represent mean ocean dynamics and correct for biases in atmospheric surface fluxes, prescribed vertical profiles of heat and salt corrections are 100 applied at each grid point and each timestep. These are computed from a 10-year SPCAM3-KPP "relaxation" simulation with a 15-day relaxation timescale towards the reference seasonal cycle of oceanic temperature and salinity. The daily climatology of corrections is then computed and imposed on a "free-running" coupled integration with no relaxation, which displays only small SST biases against the reference climatology ( Fig. 1). Further details on the correction method can be found in Hirons et al. (2015), Peatman and Klingaman (2018) and Klingaman and DeMott (2020). 105

Experiments and validation data
Two pairs of atmospheric-only (SPCAM3) and coupled (SPCAM3-KPP) models are analysed to examine the role of air-sea coupling and mean state biases in the simulated BSISO (Table 1). These simulations are also analysed for the MJO in Klingaman and DeMott (2020). For brevity, we write SPCAM3 and SPCAM3-KPP as "SPA" and "SPK", respectively. All simulations are 50-year integrations. Next, we explain the setup of each experiment in detail. 110 To study the effect of mean state biases on the simulated BSISO, SPK is constrained to two ocean mean states: 1) the 1980-2009 climatology from the Smith and Murphy (2007) 3D ocean analysis; and 2) the climatology from a 20-year simulation of SPCCSM3 ( "SPC" for short; Stan et al., 2010;DeMott et al., 2011DeMott et al., , 2014. The former is considered as the observed ocean state (Fig. 1c), against which SPC shows large cold SST biases throughout the Indo-Pacific in boreal summer (Fig. 1f). These two coupled simulations are referred to as "SPK-OBS" and "SPK-SPC", respectively. To 115 investigate the effect of air-sea interaction on the simulated BSISO under the observed ocean mean state, SPCAM3 is forced by the 31-day smoothed SST from SPK-OBS (hereafter "SPA-KOBS"). Similarly, we prescribe the 31-day smoothed SST from SPK-SPC to SPCAM3 to understand the air-sea coupling effect under the SPC mean state ("SPA-KSPC"). Table 1 summarises all four experiments.
All validation data in this study are from the European Centre for Medium-range Weather Forecasts (ECMWF) Interim 120 (ERAI) reanalysis dataset (Dee et al., 2011). The daily variables used include: precipitation, outgoing longwave radiation (OLR), 850-hPa wind, surface variables related to latent heat and sensible heat (LH and SH) fluxes, and 3-D variables associated with moist static energy (MSE; Maloney, 2009). All the variables have the horizontal resolution of 1.5°×1.5°; we analyse the period of 1986-2016 for compatibility with Gao et al. (2019).

Methods 125
The BSISO convective intensity is represented by 20-100-day-filtered variability of boreal summer (May-October) precipitation, OLR or MSE, depending on the process-oriented diagnostic applied. To reveal the overall propagation features of BSISO convection, lead-lag regression analysis is performed on area-averaged filtered precipitation. Since the Indian and https://doi.org/10.5194/gmd-2020-63 Preprint. Discussion started: 24 April 2020 c Author(s) 2020. CC BY 4.0 License.
Pacific basins exhibit different basic states (Lau and Waliser, 2012), we compute propagation over one region in each basin: the eastern Indian Ocean (80°E-90°E, 0°-10°N) and western Pacific (130°E-140°E, 0°-10°N). 130 DeMott et al. (2016) and Gao et al. (2019) provide a useful diagnostic method to quantitatively assess the contribution of intraseasonal SST variability to the MSE budget of ISO in boreal winter and boreal summer, respectively. We employ the same approach to understand oceanic feedbacks to the simulated BSISO, and compare these feedbacks to those in reanalysis data. Intraseasonal SST can affect atmospheric convection through modifying surface turbulent fluxes (LH and SH) and hence MSE, via the near-surface gradients in specific humidity and temperature. Based on decomposition of surface bulk 135 formulae (Weare et al., 1981), SST modulation of surface fluxes is represented as the difference between fluxes calculated using the full and 61-day smoothed SST. These SST-modulated fluxes are then projected onto column-integrated MSE and its tendency (〈m〉 and ∂〈m〉/ ∂t) to reveal the oceanic feedback to the maintenance and propagation of BSISO convection, respectively. Further details can be found in DeMott et al. (2016) and Gao et al. (2019).
We employ the BSISO indices of Lee et al. (2013) to investigate the simulated convective features in each BSISO phase. 140 Firstly, daily anomalies are computed as the departure from the climatological annual cycle (annual mean and first three harmonics) and a 120-day running mean. Multi-variate Empirical Orthogonal Function (MV-EOF) is then performed on the combination of OLR and u850 anomalies, each normalized by their respective area-mean standard deviation over the ASM region (40°E-160°E, 10°S-40°N) during boreal summer. The first two leading principal components (PC1 and PC2) define the BSISO indices, which separate the BSISO life cycle into eight phases. For model results, normalized OLR and u850 145 anomalies are projected onto the observed EOF modes to obtain the model PCs. Model simulated BSISO phase composites are constructed based on these PCs (Sperber et al. 2013). Consistent with Lee et al. (2013), only strong BSISO events (√PC1 -+ PC2 -> 1.5) are selected for phase composites.

Mean state and intraseasonal variability 150
Previous studies indicate that ISO variability is closely related to the climatological state of convection (Wheeler and Kiladis, 1999;Sperber et al., 2000;Waliser et al., 2003). Thus, we firstly examine the summer mean precipitation and 850-hPa wind ( Fig. 2). In ERAI, substantial rainfall appears over the ASM region where the average exceeds 8 mm/day (Fig. 2a). SPK-OBS overestimates subtropical rainfall (10°N-20°N), particularly in the western North Pacific (Figs. 2b and e), but underestimates rainfall over the southern Indian Ocean, Maritime Continent and India north to 20°N. These precipitation 155 biases are associated with low-level wind biases. Simulated mean westerlies extend into the western North Pacific, increasing convergence across the band of subtropical enhanced precipitation. In SPK-SPC (Fig. 2c), which is constrained to the cold climatological SPCCSM3 SST (Fig. 1d), rainfall reduces compared to that in SPK-OBS: wet biases are reduced while dry biases become more severe (Figs. 2e and f).  where mean rainfall is also overestimated (Fig. 2). This reinforces that model BSISO biases are largely associated with incorrect representations of mean rainfall (Sperber and Annamalai, 2008;Sabeerali et al., 2013;Hu et al., 2017). The overestimation of intraseasonal rainfall in SPK-SPC (Fig. 3h) is smaller than that in SPK-OBS (Fig. 3g), consistent with the lower mean rainfall in SPK-SPC. SPA-KOBS (Fig. 3i) and SPA-KSPC (Fig. 3j) show increases in intraseasonal rainfall relative to SPK-OBS and SPK-SPC, respectively. This indicates that both cold SST biases and air-sea coupling exert 170 negative feedbacks to BSISO convection, and improve the simulated BSISO amplitude.

Northward propagation of BSISO
Lag regression analysis is employed to understand the simulated BSISO propagation in each experiment. Over the Indian Ocean, 80°E-90°E averaged intraseasonal precipitation is regressed onto the (80°E-90°E, 0°-10°N) averaged intraseasonal precipitation time series (Fig. 4). In ERAI, the convective anomaly originates south of the equator and propagates southward 175 to 10°S and northward into the Bay of Bengal (Fig. 4a). All simulations capture the northward propagation, confirming the high fidelity of SPCAM3 for simulating BSISO northward propagation (DeMott et al., 2014;Neena et al., 2017). However, the simulations fail to capture the southward branch. And the simulated strongest convection shifts northward (10°N) relative to ERAI (5°N). A weaker propagating signal is observed under the SPC mean state than under the observed mean state for both SPK and SPA, especially over the subtropical regions from Lag 0 to Lag 20 (marked with black box; Figs. 4b and d; 180 Figs. 4c and e). This is consistent with weaker BSISO variability under the SPC mean state (Figs. 3g-j), suggesting that cold tropical mean SST biases are unfavourable for BSISO variability and northward propagation. In SPA, the amplitude of convective maxima increases relative to SPK, but with weaker propagation (Figs. 4b-e). Air-sea coupling plays a similar role in BSISO convection under both observed and cold mean states: it damps BSISO amplitude but supports northward propagation over the eastern Indian Ocean. 185 The BSISO northward propagation over the western Pacific is shown in Fig. 5. In ERAI, the convection also originates south of equator, but propagates only north, into the western North Pacific (Fig. 5a). These characteristics are properly simulated by all experiments (Figs. 5b-e). As in the Indian Ocean, due to the large overestimation of subtropical BSISO variability, the simulated strongest convection moves north compared to ERAI. It is notable that considerable suppressed offequatorial convection leads the strongest convection by ~1 week, which degrades model performance. Over the western 190 Pacific, mean-state biases also reduce the northward propagating signal in SPK-SPC (SPA-KSPC) relative to in SPK-OBS (SPA-KOBS). Air-sea interaction plays little role in the BSISO northward propagation over the western Pacific, as there is no substantial difference between SPK and SPA under either ocean mean state (Figs. 5b-c; Figs. 5d-e).

Quantitative contributions of SST fluctuations to BSISO
The temporal and spatial evolutions of BSISO and SST variability over the Indian Ocean and western Pacific are shown in 195 Figs. 6 and 7, respectively. Over the Indian Ocean, in ERAI, warm SSTs always lead enhanced convection, with maxima north/northeast of the convective center (Figs. 6a, f and k). The SST-rainfall phase relationship is properly represented in SPK and misrepresented in SPA. However, in more detail, SPK shows a shorter delay between suppressed convection and warm SST relative to ERAI. This suggests the ocean responds more quickly to BSISO convection in SPK than in ERAI, which might be caused by the lack of ocean dynamics in MC-KPP. Alternatively, the high coupling frequency (15 minutes)  200 and fine oceanic vertical resolution (~1m) in SPK may make the mixed-layer depth too sensitive to atmospheric convection.
SST anomalies in SPA are small due to the 31-day smoothing applied (Figs. 6d,e,i,j,n,and o). Compared to SPK-OBS, the convective region becomes smaller under the SPC mean state or without air-sea coupling (SPK-SPC, SPA-KOBS and SPA-KSPC). To some degree, it reflects the negative (positive) role of mean state biases (air-sea interaction) in supporting the propagating BSISO signal, in agreement with Fig. 4. 205 Over the western Pacific, SPK simulates well the ERAI phase relationship between the convection and SST anomalies: a near-quadrature relationship over the equatorial regions (Figs. 7a-c), but nearly out-of-phase over the off-equatorial regions (Figs. 7f-h and k-m). However, SPK produces substantial suppressed convection north of the active convection over the western North Pacific, consistent with Fig. 5. As a result, simulated warm SST appears north of the convective center, while the ERAI SST maximum is always located northwest of the convective center. Imposing the SPC mean state weakens 210 convection. In SPA, the intraseasonal SST is very small by design.
As revealed by Gao et al. (2019), intraseasonal SST can affect the BSISO by modulating surface LH and SH fluxes. The temporal and spatial distributions of SST-modulated fluxes are highly coherent with those of SST anomalies themselves (Gao et al., 2019;their Figs. 14 and 15). Therefore, the evolutions of SST anomalies shown in Figs. 6 and 7 can also roughly reveal the evolutions of SST-modulated surface fluxes. By regressing the SST-modulated flux anomalies onto the BSISO 215 MSE and the time rate of change of MSE (〈m〉 and ∂〈m〉/ ∂t), Figure 8 illustrates the "SST effect" on the maintenance and propagation of convection, respectively. In ERAI, intraseasonal SST supports 〈 〉 in the central Indian Ocean and Bay of Bengal but destroys 〈 〉 in the western North Pacific (Fig. 8a). The inter-basin differences are linked to the differences in SST-convection phase relationships (Gao et al. 2019). However, in SPK, simulated SST anomalies largely damp the convection over both the Bay of Bengal and western North Pacific (Figs. 8b and c), because of the quicker response of the 220 SPK ocean to BSISO convection, combined with the erroneously strong suppressed convection just before the active phase.
In SPK, the negative SST effect on 〈 〉 agrees with the stronger BSISO variability in SPA ( Fig. 3i; Fig. 3j) than in SPK ( Fig.   3g; Fig. 3h) over the subtropics. SST anomalies have similar effects on BSISO propagation under the observed and SPC mean states, with some amplitude difference. The simulated SST anomalies in SPK favor convection propagation (Figs. 8g https://doi.org/10.5194/gmd-2020-63 Preprint. Discussion started: 24 April 2020 c Author(s) 2020. CC BY 4.0 License. and h), but with a smaller contribution (8-12%) than in ERAI (12-20%). This is caused by the incorrect phase relationship 225 between the simulated BSISO and SST anomalies (Figs 6 and 7). A weak SST effect on the BSISO can still be found in SPA (Figs. 8d, e, i, and j), as these simulations retain some intraseasonal SST variability (Fig. 3). In summary, SST variability enhances the simulated amplitude of BSISO variability by suppressing the excessive subtropical convection and contributes to the propagation of convection over the entire BSISO region.

BSISO indices 230
Propagating BSISO characteristics can also be examined using the BSISO indices. Figure 9 shows the annual cycle of variance of each of the first four PCs from ERAI and model simulations. Since EOF analysis is only applied to May-October, PC time series for November-April are obtained by projecting OLR and u850 anomalies onto the same EOF patterns.
In ERAI, PCs have a similar seasonal cycle as those from Lee et al. (2013). PC1 has the largest variance throughout boreal summer (Fig. 9a). It increases sharply from late April, maximising in August, followed by PC2 with about half a month 235 delay. In contrast to PC1 and PC2, PC3 exhibits most of its variance in early summer (May-June). PC4 also has a broad peak, but its amplitude is much smaller than the other PCs. SPK-OBS produces a similar seasonal cycle of PC1 as ERAI, except for deficient variance in August-September, which largely reduces the amplitude difference between the first two PCs (Fig. 9b). Simulated PC1 also increases abruptly from late April, with a half-month delay between PC1 and PC2, which matches ERAI well. However, PC1 in SPK-SPC begins to increase in late May (Fig. 9c), about one month later than ERAI 240 and SPK-OBS. The variance of the first two PCs in SPK-SPC has a much narrower peak and tails off from early July, without a second peak in October. This suggests that cold SST error leads to a late onset of BSISO activity, and also strongly degrades the annual cycle of BSISO activity. In both SPK simulations, the amplitude of PC3 is smaller than that of PC4, which indicates that the order of those simulated PCs is reversed. All model simulations show a similarly poor ability to simulate PC3 and PC4. Therefore, hereafter we only focus on PC1 and PC2. Removing air-sea coupling in SPA increases 245 significantly the strength of PC1 relative to SPK. This suggests that the role of air-sea coupling in suppressing BSISO variability is mainly expressed through damping variability in PC1.
Based on the PC1 and PC2 time series, the BSISO life cycle is separated into eight phases, which represents the canonical northward propagating BSISO (Annamalai and Sperber, 2005;Wang et al., 2005). Figure 10 shows the phase composite of anomalous precipitation and 850-hPa wind from ERAI and SPK-OBS. SPK-OBS displays a similar evolution of the 250 structure of convection and circulation as in ERAI, but with significant amplitude differences. The simulated circulation also matches well with the simulated convection: easterlies (westerlies) occur to the north (south) of the ascent; cyclonic (anticyclonic) circulations coincide with the active (suppressed) off-equatorial convection. However, the northwest-southeast titled rainfall band in SPK-OBS is more zonal than in ERAI ( Fig. 10e and m), largely due to the poor representation of BSISO variability over the western Pacific. In ERAI, most strong BSISO events occur in phases 2, 5 and 7. In contrast, in 255 SPK-OBS, a stronger preference for simulated strong BSISO events appear in off-equatorial regions (phases 3, 4, 7, and 8) https://doi.org/10.5194/gmd-2020-63 Preprint. Discussion started: 24 April 2020 c Author(s) 2020. CC BY 4.0 License. rather than in near-equatorial regions (phases 1, 2, 5, and 6). It implies that the BSISO variability is largely determined by the propagating BSISO signal (Sperber et al., 2013;Neena et al., 2017).
We further compute the frequency of strong BSISO events in each phase from each dataset (Fig. 11). Overall, there are fewer strong BSISO events in the simulations (25-30%) than in ERAI (~33%), although the amplitude of simulated 260 precipitation associated with these events is much larger. Consistent with Fig. 10, in ERAI, phases 2, 5 and 7 show the highest frequencies, while in SPK-OBS, most strong simulated BSISO events occur in phases 3, 4, 7, and 8. Comparing SPK-SPC and SPK-OBS shows that the SPC ocean mean state has little consistent effect on the frequency of BSISO events.
Removing air-sea coupling in SPA further increases the frequency of strong BSISO events in off-equatorial regions (phases 3, 4, 7, and 8), associated with the greater intraseasonal convective variability in SPA relative to SPK. 265 To more quantitatively evaluate the performance of each simulation, we construct Taylor diagrams over the Indian Ocean (50°E-100°E, 10°S-20°N) and western Pacific (110°E-160°E, 0°-30°N), respectively (Fig. 12). These diagrams provide pattern correlation coefficients and root-mean-square-errors of the simulated intraseasonal precipitation against ERAI intraseasonal precipitation in each BSISO phase. Over the Indian Ocean, the fidelity of each simulation varies from phase to phase (Fig. 12a). Generally, higher pattern correlations appear in equatorial phases (phases 1, 2, 5 and 6) than in off-270 equatorial phases (phases 3, 4, 7 and 8). The area-averaged rainfall variability is overestimated in almost all phases, with largest amplitude biases in phases 3 and 6 (1.5-2.25 of the standard deviation of ERAI). Overall, SPK-SPC outperforms SPK-OBS, largely due to its smaller amplitude biases compared to ERAI. Under the observed ocean mean state, air-sea coupling improves the simulation fidelity in phases 1, 3, 7, and 8. Under the SPC ocean mean state, coupling greatly improves the simulated BSISO life cycle over the Indian Ocean, except for phase 2. 275 In almost all simulations, higher pattern correlations are found over the western Pacific (0.7-0.9) than over the Indian Ocean (0.5-0.9), but with much larger overestimations of amplitude (Fig. 12b). The overall simulation performance is ranked as follows: SPK-SPC, SPA-KSPC, SPK-OBS and SPA-KOBS. Simulations with the SPC ocean mean state perform better than those under the observed ocean mean state, in terms of both amplitude and distribution. Since the excessive subtropical rainfall variability is the biggest error in SPCAM3, the underlying cold SSTs largely weaken convection, which 280 helps to reduce the amplitude biases and improve the pattern correlation with ERAI. Air-sea coupling improves model performance under both mean states, particularly in suppressing the overestimated simulated BSISO variability over the western North Pacific.

Discussion
Despite the large correct SST-rainfall phase relationship in SPK simulations, a shorter delay between suppressed convection 285 and warm SST exists relative to the reanalysis. This may be associated with the configuration of the MC-KPP ocean (lack of ocean dynamics, high coupling frequency and fine vertical resolution). As a consequence, it is easy to surmise the effect of air-sea coupling on the amplitude of convection, while its effect on the propagation is not obvious. Overall, in SPK, https://doi.org/10.5194/gmd-2020-63 Preprint. Discussion started: 24 April 2020 c Author(s) 2020. CC BY 4.0 License.
intraseasonal SST anomalies largely damp intraseasonal subtropical convection variability and make a smaller contribution (8-12%) to ∂〈m〉/ ∂t than in ERAI (12-20%). Propagation is slightly enhanced in SPK relative to SPA over the Indian 290 Ocean, however, coupling makes nearly no difference to the propagating signal over the western Pacific. This may be related to the larger underestimation of the contribution of intraseasonal SST to ∂〈m〉/ ∂t over the western Pacific than over the Bay of Bengal (Fig. 8). More importantly, it implies that atmospheric internal processes are essential to the propagation of convection, since SPCAM3 simulates BSISO propagation well even without coupling or sub-seasonal SST variability (Gao et al. 2019). 295 Stan et al. (2010) and Neena et al. (2017) reported that SPCCSM3 showed better ISO simulation skill than SPCAM3 with prescribed observed SST. In our experiments, the BSISO variability is similar in SPK-OBS and SPA-KOBS and in SPK-SPC and SPA-KSPC, which suggests a limited role for air-sea coupling in the simulated BSISO. Comparing SPK-OBS and SPK-SPC strongly suggests that the ocean mean-state biases in SPC improve BSISO amplitude and spatial distribution. This implies that mean-state bias in SPC, not sub-seasonal air-sea coupled feedbacks, may be the primary reason for the improved 300 BSISO in SPCCSM3 relative to SPCAM3. However, while the SPC mean state improves the amplitude and spatial pattern of simulated BSISO convection, it also weakens northward propagation. By examining Figs. 1g-h, these amplitude/spatial pattern and propagating differences between SPK-SPC and SPK-OBS may be related to the stronger horizontal SST gradients in SPK-SPC, since the pattern of mean state column water vapor biases closely follows that of mean SST biases (DeMott et al., 2019). Stronger meridional SST gradients in SPC may largely suppress the overestimated subtropical rainfall 305 than in SPK-OBS. SPC also seems to reduce the zonal SST (moisture) gradient that plays an essential role in ∂〈m〉/ ∂t for both the boreal winter and boreal summer ISO, leading to less propagating BSISO signal (Jiang, 2017;Jiang et al., 2018;Gao et al., 2019). Thus, as suggested by Seo et al. (2007) and Klingaman and Woolnough (2014), it is important to control for the effect of mean-state biases as well as its gradients when evaluating the influence of air-sea coupling on the simulated ISO (DeMott et al., 2019). 310 In CGCMs with dynamical oceans, it is also important to consider the effects of inter-annual SST variability (such as ENSO), due to its strong control on sub-seasonal variability, particularly on the boreal winter MJO (Weaver et al., 2011;Kapur and Zhang, 2012;Klingaman and DeMott, 2020). We note that none of the simulations considered here represent ENSO or other coupled modes of inter-annual variability, as the MC-KPP ocean model lacks the requisite ocean dynamics. By using the same simulations, Klingaman and DeMott (2020) found that the intensity and propagation of the boreal-winter 315 MJO was strongly suppressed in SPCAM3-KPP under the SPC ocean mean state, while here the SPC ocean mean state enhances BSISO amplitude and only slightly weakens propagation. Klingaman and DeMott (2020) found that the including of the SPC ENSO variability on top of the SPC mean state substantially strengthened the MJO, suggesting that the strong MJO in SPCCSM3 arose from an excessively intense response to ENSO. The effect of ENSO is not considered here, which could be a subject for further study. 320

Summary
This study investigates the role of oceanic mean-state biases and air-sea coupling in simulating the BSISO by coupling the Superparameterized Community Atmospheric Model (SPCAM3) to the Multi-Column K Profile Parameterization (MC-KPP) mixed-layer ocean. To diagnose the sensitivity to the ocean mean state, SPCAM3-KPP is constrained to either the observed ocean mean state or the ocean mean state from the coupled configuration of SPCAM3 with a dynamical ocean (SPCCSM3; 325 "SPC" for short). The SPC mean state introduces substantial cold SST biases across the Indo-Pacific. To diagnose the sensitivity to air-sea coupling under each mean state, SPCAM3 is driven by the 31-day running mean SST from each SPCAM3-KPP simulation.
Systematic errors in SPCAM3 result in overestimated subtropical summer mean rainfall and intraseasonal variability. This overestimation is greatly reduced with coupling, or with the SPC ocean mean state. Lag regression composites show that the 330 simulated convection exhibits realistic northward propagation over both the Indian Ocean and western Pacific. However, the strongest convective center shifts north relative to reanalysis. The SPC ocean mean state degrades the propagating BSISO signal in both coupled and uncoupled simulations, relative to simulations that use the observed ocean mean state. Air-sea coupling slightly enhances the strength of the propagating signal under both ocean mean states. The coupled simulations capture a reasonable SST-rainfall phase relationship, but with warm SSTs shifting toward the suppressed convection relative 335 to the reanalysis. Intraseasonal SST variability plays a similar role in the BSISO MSE budget in the coupled simulations, regardless of mean state: it damps subtropical convection and favors BSISO northward propagation, but with a smaller contribution (8-12%) than the reanalysis (12-20%).
The simulation skill of propagating BSISO characteristics is also examined by using BSISO indices. Experiments with the observed ocean mean state produce a realistic annual cycle of BSISO variance, while a one-month delay to the onset of 340 BSISO activity is found in simulations with the SPC mean state. Air-sea coupling weakens convective variability mainly through suppressing the first EOF mode. Different from the reanalysis, all model simulations favour strong BSISO activity in off-equatorial regions, associated with the intense subtropical variability. Simulations under the SPC ocean mean state produce an improved BSISO than those under the observed ocean mean state, in both the amplitude and pattern correlation of anomalous precipitation. Air-sea coupling improves the BSISO fidelity in most phases. The enhancement of simulated 345 BSISO by the ocean mean state and air-sea coupling largely arises from suppressing the erroneously strong subtropical convection.
In our study, air-sea coupling has a similar effect on the simulated BSISO under different oceanic background states, suggesting that the role of coupling may largely depend on the phase relationship between the convection and SST anomalies.
Compared to daily coupling frequency between the atmosphere and ocean components of CGCMs, improved fidelity for the 350 BSISO simulation is shown in CGCMs with sub-daily coupling frequency (Woolnough et al., 2007;Klingaman et al., 2011;Hu et al., 2015). In our study, the ocean tends to respond quickly to the atmospheric convection, which may be associated with the sensitive mixed layer depth due to the absent ocean dynamics in KPP. Therefore, to better understand the role of coupling in the simulated ISO, efforts should aim to simulate a realistic phase relationship between the ISO convection and SST anomalies. 355 Code and data availability. All model outputs are available on the U.K. JASMIN collaborative research analysis facility (http://www.jasmin.ac.uk). Access to data can be obtained by contacting Nicholas Klingaman (nicholas.klingaman@ncas.ac.uk Author contributions. YG and PCH developed the code, conducted the analysis, and wrote the manuscript. NPK and CAD designed and performed SPCAM3-KPP simulations presented in the paper. All authors gave comments and contributed to the development of the manuscript.

365
Competing interests. The authors declare that they have no conflict of interest.