Rectification of surface chlorophyll (Chl) concentration by the atmospheric intraseasonal variability is detected in a numerical biophysical ocean model when it is forced by composite Madden-Julian Oscillation (MJO) events. In addition to the shoaled mixed layer depth (MLD) previously reported, it is found that increased mean Chl by MJO forcing mostly co-occurs with shoaled isothermal depth (ITD) / nutrient isopleths and reduced barrier layer thickness (BLT). Case studies reveal that MJO forcing increases MLD and ITD variations, which enhances vertical mixing and brings nutrients to the surface layer thereby increasing Chl concentration. The shoaled MLD and ITD in the seasonal / annual mean are due to asymmetric responses to MJO wind forcing; i.e., shoaling by weaker wind is more sensitive than deepening by stronger wind. Reduced mean BLT is because ITD shoaling is larger than MLD shoaling. As an exception, it is detected that both the mean Chl and BLT are increased by MJO forcing in the southern Bay of Bengal. Here, the ITD is climatologically deep in the active MJO season (boreal summer), and different phase between the MLD and ITD variations results in temporarily large BLT. However, this barrier layer does not effectively isolate the surface layer from the nutrient-rich deeper ocean. Lastly, observations support our findings in limited regions and seasons, though further investigation is necessary to confirm the effect of atmospheric intraseasonal variability on the mean surface Chl.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 It has been argued that the atmospheric intraseasonal variability (ISV) imparts a low-frequency rectification to oceanic variability [Han et al., 2004; Kessler and Kleeman, 2000; Waliser et al., 2003, 2004]. Kessler and Kleeman  imposed intraseasonal surface wind forcing into an ocean general circulation model (OGCM) in the western equatorial Pacific, and found that sea surface temperature (SST) in the tropical Pacific basin is rectified by this intraseasonal forcing. The authors further examined effects of this rectified SST on El Niño-Southern Oscillation, and concluded that the ISV can constructively influence the El Niño-Southern Oscillation cycle. Waliser et al. [2003, 2004] constructed composite atmospheric forcing from observed Madden-Julian Oscillation (MJO) [Madden and Julian, 1971, 1994] events, and used them to force a reduced gravity, sigma coordinate OGCM having an explicit mixed-layer physics. In their experiments, low-frequency rectification by MJO forcing was detected and analyzed in SST, mixed layer depth (MLD), and surface zonal current (u). The authors concluded that most of the SST rectification to a warmer temperature was caused by a nonzero mean of MJO forcing itself (e.g., shortwave radiation) while most of the MLD rectification to a shallower depth is associated with nonlinear mixed layer processes. Specifically, the shoaling of MLD by reduced wind is larger in absolute magnitude than the MLD deepening by increased wind. The rectification of the zonal current was thought of as a net effect of the shallower MLDs associated with weak winds and deeper MLDs associated with strong winds. Han et al.  focused on the low-frequency rectification of surface zonal current in the tropical Indian Ocean (TIO) in a different model with a different forcing compared to Waliser et al. [2003, 2004], but drew similar conclusions, namely that the asymmetric behavior of MLD responding to intraseasonal wind variations is the major nonlinear process that renders a low-frequency rectification in zonal current.
Jin et al. [2012a, hereafter J12] reported that the atmospheric intraseasonal forcing results in higher mean states of surface Chlorophyll (Chl) concentration in the TIO. When their OGCM was forced by climatological forcing only (control simulation), the mean state of surface Chl concentration in the TIO was too low, by more than an order of magnitude in some regions, compared to observations. When anomalous MJO forcing was added to the climatological forcing, the bias was noticeably reduced although not removed completely. J12 suggested that consecutive entrainment blooms by MJO forcing contribute to a higher mean Chl concentration.
 We note that accurately simulating the Chl concentration is important because (1) primary production, which is intimately related to Chl studied here serves as the base of the marine food web, and (2) Chl concentration affects the vertical profile of penetrating solar radiation by absorbing it, and ultimately can impact the physical climate system [e.g., Ballabrera-Poy et al., 2007; Gnanadesikan et al., 2010; Jochum et al., 2010; Lengaigne et al., 2007; Manizza et al., 2008; Marzeion et al., 2005; Murtugudde et al., 2002; Sweeney et al., 2005; Wetzel et al., 2006]. J12 reported an improved Chl mean state by the intraseasonal forcing, but the effect of the intraseasonal forcing was likely not fully realized in J12. This is because of their experimental design in which the period of each simulation was just 200 days. Additional experiments are clearly needed in which MJO forcing is prescribed for a longer period, i.e., several years or decades, to obtain a stable biophysical ocean state. In this study, a multidecadal simulation with intraseasonal forcing is performed, and it is found that the surface Chl mean state is even closer to the observation than that reported in J12 (Figure 1; details of this result are described in section 3.1.1). The objective of this study is to investigate the mechanism how the intraseasonal forcing increases the surface Chl mean state.
 The rest of this paper is organized as follows. After briefly describing the model, experimental setup and data in section 2, in section 3, we examine the annual and seasonal mean differences between the MJO and control simulations in a number of fields. Based on this result, we select two regions and perform additional sensitivity experiments: starting from the same initial condition, but with and without the anomalous MJO forcing. We also provide observational support for the model processes. A summary with conclusions is presented in section 4.
2 Model, Experimental Setup, and Data
 The OGCM used in this study is the reduced gravity, primitive equation, sigma coordinate model of Gent and Cane , which is identical to that used in Waliser et al. [2003, 2004]. In this OGCM, surface heat fluxes are computed by an advective atmospheric mixed layer model (AMLM) [Seager et al., 1995]. Murtugudde et al.  reported that this OGCM coupled with AMLM simulates realistic ocean dynamics and thermodynamic features compared to the observations. This OGCM is forced by surface wind speeds and wind stress, shortwave radiation, cloud fraction, and precipitation. Surface air temperature and humidity are prescribed over land only as boundary conditions for the AMLM but computed over the ocean interactively with model predicted SSTs. In addition to this physical OGCM, a nine-component ecosystem model is coupled to produce bio-physical feedbacks within the ocean [Christian et al., 2001a, 2001b]. Details of the model and prescribed forcings are identical to J12, with the latter described briefly below.
 MJO forcing is constructed by producing a composite MJO from the anomalies of observed MJO events, and this MJO composite is added to the climatological forcing. The MJO events are selected by using the Real-time Multivariate MJO series [Wheeler and Hendon, 2004] which provides daily strength and phase of the MJO. As in J12, a composite for each of the 8 MJO phases is constructed by averaging daily band-pass filtered (30–120 days) data if the Real-time Multivariate MJO magnitude is greater than 1.0. In J12, the winter (Nov–Apr) and summer (May–Oct) MJO composite anomalies are separately assembled, and each composite is used to build one cycle of a 48 day MJO forcing (e.g., 8 phases and assuming each phase is 6 days long). Then, the same (winter or summer) MJO anomaly was repeated 4 times for 6 months. The details of how MJO forcing is constructed are described in J12. In this study, we concatenate the winter and summer MJO forcings to build a continuous MJO forcing for a whole year. For the transition from winter MJO to summer MJO or vice versa, we applied a linear filter for smooth transition. For example, on 30 April and 1 May when the summer MJO succeeds the winter MJO, we let the winter MJO anomaly linearly decrease from 25 April (coefficient is 1) to 6 May (coefficient is 0). At the same time, we let the summer MJO anomaly linearly increase from 25 April (coefficient is 0) to 6 May (coefficient is 1). The examples of continuous MJO forcing are displayed in the later case studies (Figures 4a and 5a). We note that this composite based strategy allows for a temporal average of forcing anomalies over the MJO life-cycle to be nonzero, but the deviations from zero are small.
 In J12, it is noted that the only difference between the MJO and control experiments was the existence of MJO forcing anomalies, which is obtained by multiplying the MJO composite anomalies by a scale factor of 2.5. The reason for using this scale factor was in order to provide MJO forcing comparable to the magnitude of band-pass filtered variability of notable MJO events in nature (see Figure 1 in J12). These details are also true in this study, and we refer to the simulation with MJO forcing anomalies as the “MJO” while the counterpart without MJO anomalies is referred to as the “Control.” Sensitivity to this scale factor will be addressed later in the section 3.1.1. The control and MJO simulations start from the same initial condition that is provided by the 270 years spin-up run, and the control and MJO simulations are performed for 80 and 90 years, respectively, to reach their respective statistical equilibrium. In this study, the last 15 years of each simulation are used for analyses.
 In this study, we present observational MLD, isothermal depth (ITD), and barrier layer thickness (BLT) based on temperature and salinity data from the World Ocean Atlas 2009 (WOA09) [Antonov et al., 2010; Locarnini et al., 2010]. Following Sprintall and Tomczak , we calculated MLD with the variable sigma-t criterion corresponding to 0.5 °C temperature difference, and ITD as the depth where water temperature is 0.5 °C below the SST. Then, BLT is defined as the ITD minus the MLD (see also Murtugudde and Busalacchi ). In addition, we also used observational surface winds and Chl concentration to support our model results. The wind data comes from the cross-calibrated, multiplatform ocean surface wind data set [Atlas et al., 1996]. The Chl data is the level-3 standard mapped image product of Moderate Resolution Imaging Spectroradiometer (MODIS) Aqua [Esaias et al., 1998] and Sea-viewing Wide Field-of-view Sensor (SeaWiFS) [McClain et al., 2004]. All Chl values from observations and model results are log 10-transformed before any analysis or averaging [Campbell, 1995; Jin et al., 2012b]. Similarly, nitrate concentrations from the model results are also log 10-transformed, and we refer to the log-10 transformed Chl and NO3 as “Log10Chl,” and “Log10NO3”, respectively.
3.1 Annual and Seasonal Means
3.1.1 Annual Mean of Chl and Nitrate
 The impact of MJO forcing on the rectified annual means is shown in Figures 1 and 2. In Figure 1, annual mean surface Chl concentrations from the model with and without MJO forcing in the TIO basin are compared with the observations. It is seen that the annual mean surface Chl in the MJO simulation (Figure 1c) is much improved compared to the control simulation (Figure 1b), and is indeed comparable to the observations (Figure 1a) especially in the eastern and southern TIO. In Figure 1d, which shows the difference between the MJO and control simulations, the effect of anomalous MJO forcing is prominent in the central-to-eastern equatorial Indian Ocean with concentration differences over 2 in log10 scale. When MJO forcing with a scale factor of 1.0 is used, it is found that the annual mean of surface Chl concentration in the TIO basin ranges over 0.5–1 higher in the log10 scale (about 3 to 10 times) compared to the annual mean of the control simulation, which is smaller than that of the MJO simulation with scale factor of 2.5, but with conclusions that remain the same.
 To examine the oceanic vertical structure associated with this increased mean Chl concentration at the surface, we present latitude – depth cross-sections based on a meridional average from 4°S to 0°N (Figures 2a and 2b). Here, NO3 concentration is superimposed onto Chl concentration as black contours because nitrate is the most important nutrient for Chl growth in the TIO region. As noted in J12, Log10NO3 increases with depth while a subsurface Log10Chl maximum occurs roughly in the range of 50–100 m. This subsurface Log10Chl maximum is generally thought of as due to the combined effect of nutrient and light availability.
 In the control simulation, a local minimum of Log10Chl as well as Log10NO3 occurs in the surface layer between 60°E and 75°E. A local maximum of Log10Chl in the deeper ocean is also detected in the same longitudinal range at about 90 m depth. One interesting point here is that the nutrient concentration at this depth does not vary much longitudinally (Figure 2a; the black contour corresponds to 0.5) compared to Log10Chl variation. J12 noted that low concentration of Chl at the surface, which means a larger amount of light below the surface, can lead to enhanced Chl concentration in the deeper ocean. This process is also in play for this local Chl maximum in Figure 2a, although nutrient and light are both affected by the level of Chl. Other than this central region, the vertical distribution of Log10Chl and Log10NO3 is generally similar between the eastern and western TIO regions, each exhibiting strong vertical gradients of Chl and nitrate.
 In the MJO simulation, Log10Chl and Log10NO3 concentrations increase to the east at most depths. In the eastern part of the basin where MJO amplitude is generally larger, the vertical gradient of these quantities is diminished considerably and a higher concentration of nutrients is found closer to the surface. The increased nutrients in the deeper ocean are also seen in the MJO simulation. For example, comparing the depth of Log10NO3 contour in the 80°E to 90°E band, 0.5 isopleth is just below 80 m in the control simulation, while the values are nearly double at the same depth in the MJO simulation.
 As mentioned in the Introduction (section 1), it has been reported in earlier studies that the MLD is rectified by ISV forcing to a shallower long-term average. In this study, the shoaled annual mean of MLD is prominent in the eastern equatorial Indian Ocean east of 75°E (Figures 2a and 2b, purple solid line). In this region, the added MJO forcing leads to an approximately 10 m shoaling of the MLD annual mean. In addition, we also examine the ITD and BLT for further analysis. The barrier layer inhibits the interaction between the surface mixed layer and deeper ocean [Lukas and Lindstrom, 1991], so it can affect the surface Chl variability by reducing nutrient supply. In general, it is known that heavy precipitation associated with the MJO lowers surface salinity and tends to enhance the barrier layer, while the strong surface wind favors vertical mixing and weakens the barrier layer; a signature of intraseasonal variability on the BLT is incontrovertible (e.g., Schiller and Godfrey ; Zhou and Murtugudde ; see also Duvel ; Hendon ; Kessler ). Figures 2a and 2b show the annual mean of the equatorial barrier layer in the model simulations; between MLD (purple solid line) and ITD (purple long dash line). In the control simulation, the equatorial barrier layer is prominent around 90°E, approximately from 40 to 70 m in depth (Figure 2a). In the MJO simulation, the vertical structure is lifted and the BLT decreases to a range of 35 to 55 m. It means that when the MLD is rectified to be shallower due to MJO forcing, the ITD is also rectified to be shallower, even more than the MLD. The reduced BLT is of particular interest in light of the previously reported correlation between SSTs, BLTs, and precipitation in this region [Murtugudde and Busalacchi, 1999] because it indicates a potential positive feedback between the MJO, SSTs, and atmospheric convection in this region.
 On the other hand, the annual mean of the 24 °C isothermal depth (hereafter D24C; purple short dash line in Figures 2a and 2b) along the equator does not exhibit large differences between the control and MJO simulations, distinct from the MLD or ITD. D24C in the eastern equatorial Indian Ocean represents the upper thermocline [e.g., Annamalai et al., 2005]. This relatively small difference in the annual mean of D24C implies that the rectification of vertical structure by the MJO is mostly confined to the shallow upper layer. However, despite similar annual means, the standard deviation of D24C in the MJO simulation is larger than that in the control simulation (not shown) because of Kelvin wave and surface divergence / convergence activity by the MJO [e.g., Zhang, 2005; Matthews et al., 2010; Zhou and Murtugudde, 2010; Jin et al., 2012b; Kessler, 2012]. This variability of the D24C influences the nutrient supply and thus affects the Chl variability in the water column.
 The annual means of the variables in the equatorial TIO region discussed thus far reveal that the anomalous MJO forcing results in lifting nutrients to a shallower depth as well as shoaling MLD and reducing BLT, while there is no significant change below 100 m depth. Considering that the shoaled MLD in the annual average translates to the increased variability of MLD responding to MJO forcing, we hypothesize that the shoaled ITD and corresponding reduced BLT in the annual average is also a result of the increased variability of ITD by MJO forcing. In this sense, we suggest that the MJO events enhance vertical mixing of upper ocean (including not only the MLD, but also the ITD or deeper layers), and lead to an increased nutrient supply and higher Chl concentration in the surface layer. The effect of MJO forcing on the mean barrier layer is examined further below.
3.1.2 Seasonal Mean of BLT
 In Figure 2, it is seen that the annual mean BLT in the eastern equatorial Indian Ocean becomes thinner by MJO forcing. In this subsection, we further examine the rectified BLT based on seasonal means. Figure 3 shows the boreal winter (Nov-Apr) and summer (May-Oct) seasonal mean BLTs. In boreal winter, the observed BLT is negligible in the southwestern region, but is significant in the northeastern Indian Ocean (Figure 3a). The MJO run also simulates the smaller western and larger eastern BLTs along the equator though large BLTs in the Bay of Bengal are not simulated (Figure 3c color shading). In the eastern TIO, the BLT in the control simulation exists broadly from 8°S to 8°N but the southern part of this BLT is carved out in the MJO simulation. Thus the BLT pattern in the MJO simulation compares more favorably to the observations. In this region, just south of the equator, (1) a large increase of Log10Chl in Figure 1d is seen, and (2) MJO variability is known to be prominent particularly in boreal winter. In addition, focusing on the western TIO, the MJO anomaly forcing results in reduced BLT compared to the control simulation (e.g., between 0° and 8°N, and 8°S and 16°S), bringing the model's winter mean BLT closer to the observations. The large BLT observed in the Bay of Bengal is not captured in either simulations, likely due to insufficient vertical resolution to capture the complex vertical structure and also due to the dependence of BLT on temperature inversions in the region [see Howden and Murtugudde, 2001; Thadathil, 2008]. The equatorial and the southeastern TIO are regions of Bjerknes feedback and play an important role in regional climate variability [Annamalai et al., 2003; Annamalai and Murtugudde, 2004]. Simulating accurate BLTs in these regions is not only essential for improving climate predictions but may also be important for ocean-atmospheric feedbacks during MJO events [e.g., Duvel, 2012].
 In the boreal summer season, the observed BLT in the TIO is large (over 30 m) in the eastern equatorial basin and the southern Arabian Sea (Figure 3b). The control run simulates similar patterns as the observations, but 1) the eastern equatorial BLT is confined to narrower latitudes, and is locally too thick (> 40 m), and 2) the longitude of the local maximum BLT in the southern Arabian Sea is shifted westward approximately by 5–10° in longitude. In the MJO simulation, the simulation of BLT is improved compared to the control simulation (Figure 3d). For example, MJO forcing reduces excessively large values of BLT in the eastern equatorial region that occur in the control simulation, and expands the pattern of thicker BLT to the north of the equator by increasing the BLT there. Off the Somali coast, MJO forcing removes the local BLT maximum that occurs in the control simulation. Actually in this region, the MJO composite of Log10Chl simulated in J12 was too weak compared to the observations, while that in the MJO simulation is more consistent with nature, particularly in boreal winter (not shown). This improvement of MJO composite seems to be related to the decreased BLT in this region. The role of the diurnal cycle of the mixed layer has been invoked in the Arabian Sea for seasonal to interannual variability of Chl [e.g., Wiggert et al., 2002 and references therein], but the diurnal cycle is not resolved in our simulations. The role of the diurnal cycle of MLDs at intraseasonal time scales needs further investigation.
 In summary, we found that MJO forcing modifies the vertical structure of the upper layer of the ocean. The rectifying effect of MJO forcing results in the reduced mean BLT in several regions. Figures 3e and f indicate that, in both boreal winter and summer, the pattern of increased seasonal mean surface Chl concentration by MJO forcing is generally consistent with the pattern of decreased seasonal mean BLT. This consistency particularly appears in the central to eastern equatorial Indian Ocean just south of equator, with one exception: both the BLT and surface Chl increase by MJO forcing in boreal summer in the southern Bay of Bengal.
 To investigate the relationship between the increased or decreased mean BLT and increased mean surface Chl by MJO forcing, we perform case studies with an additional experiment: repeat the control simulation (no MJO forcing anomalies) but with the initial condition provided by the MJO simulation after spin-up (hereafter “No_MJO” simulation). We expect that, by removing MJO forcing, the surface mean state of the “high Chl concentration”, which is maintained in the MJO simulation, might be quickly restored back to the “low Chl concentration” mean state, and the difference between the MJO and No_MJO simulation would reveal the detailed role of MJO forcing.
3.2 Case Study 1: Central-Eastern Equatorial Indian Ocean
 We choose the central-eastern equatorial Indian Ocean (83°E–85°E, 2°S–4°S) as the first case study region (Figure 4). This region is selected because Figure 1d indicates that here the Log10Chl annual mean difference is large in the open ocean, and because Figure 3e indicates that decreased BLT by MJO forcing in boreal winter is also prominent here. As a seasonal characteristic in this region, the red line in Figure 4a shows that surface wind variability associated with the MJO is much larger in boreal winter than in boreal summer. The temporal evolution of vertical structure responding to this MJO forcing is shown in Figure 4b, which is in the same format as Figure 2 except x-axis (longitude in Figure 2 vs. time in Figure 4b). Comparing the boreal winter and summer seasons, the MLD variability in winter is also larger than that in summer, with a shallower mean state. Log10Chl in the surface layer is also higher in boreal winter than in summer. Similar to the results in J12, the entrainment process due to oscillating wind is prominent in the winter season, and higher nutrient availability at shallower depths contributes to a more effective entrainment bloom of Chl.
 Conversely, in the No_MJO simulation, Figure 4c shows that surface Chl concentration is quickly lowered, and the vertical gradient of Log10NO3 steepens. For example to the right of the gray line marked as “1Nov83”, the depth of zero line of Log10NO3 (the first solid black line from the top, hereafter D0_Log10NO3) in Figure 4c is below 60 m while that in Figure 4b is about 40 m. In Figure 4c, Log10NO3 around 40 m depth is as low as -3. Hence, the combined effect of weak entrainment and lower concentration of nitrate around the MLD results in a significant drop in surface Chl concentration. In Figure 4d, which shows the surface Log10Chl difference between the MJO and No_MJO simulations, the lowered Chl concentration in the No_MJO simulation is as large as 3 in log10 scale (1/1000).
 The relationship between nutrient distribution and surface Chl concentration is clearly seen in Figures 4d and 4e, which show differences in Log10Chl at the surface and depth of the NO3 isopleths (i.e., D0_Log10NO3 and “Log10NO3 = 1”; hereafter D1_Log10NO3), respectively, between the MJO and No_MJO simulations. Comparing the two figures, particularly the variability of D0_Log10NO3 difference (Figure 4e black solid line; orange line indicates low-pass filtered (48 day running mean; see J12 for details) difference) appears to be closely related to the surface Log10Chl difference. For example from October to December, when the negative difference of D0_Log10NO3 becomes larger, which means D0_Log10NO3 in the No_MJO simulation is deeper than that in the MJO simulation, the positive Chl difference becomes larger (less Chl concentration in the No_MJO simulation). On the other hand, the variability of D1_Log10NO3 difference (Figure 4e black dash line; green line indicates low-pass filtered difference) is quite small compared to the D0_Log10NO3 difference variability. This result is consistent with the negligible difference in the annual means of D24C, as stated earlier; i.e., the effect of MJO forcing is weak around or below 100 m depth. However, the decreasing trend of D1_Log10NO3 difference (deepening in the No_MJO simulation) is obvious in Figure 4e.
 Figure 4f shows the difference in MLD (black solid line; orange line indicates low-pass filtered difference) and ITD (black dash line; green line indicates low-pass filtered difference) between the MJO and No_MJO simulations. Consistent with previous studies noted in the Introduction (section 1), shoaled MLD by MJO forcing appears in Figure 4f as negative differences in the boreal winter season. In this strong MJO season, variability of MLD responding to each MJO event is large (black lines), and temporally smoothed MLD difference indicates approximately a 10 m shoaling by MJO forcing (orange line). In boreal summer, MLD variability itself is small, and low-frequency MLD differences are close to zero. On the other hand, the seasonal behavior of ITD difference between the two simulations is quite similar to the MLD difference. In boreal summer when the MJO activity is weak, the ITD difference is negligible, while the low-frequency difference in boreal winter is similarly negative as the MLD, but with a bigger amplitude than that of the MLD (comparing green and orange lines, approximately -20 m vs. –10 m). In both the MLD and ITD cases, the negative difference means that the MLD and ITD shoal (deepen) with (without) MJO forcing anomalies. Because the BLT is defined as the difference between the MLD and ITD in this study, in the No_MJO run, a larger deepening of ITD than MLD means an increased BLT in the bsence of MJO forcing. During the 3 year period shown in Figure 4, the ITD in the No_MJO simulation generally deepens with time (Figure 4c; purple long dash line) more quickly than the MLD, which translates to an increasing BLT with time. This trend of increasing BLT is consistent with that of deepening nutrient isopleths mentioned above (Figure 4e).
 In terms of the temporal order in the No_MJO simulation, the lowered Chl concentration at surface as well as deepening D0_Log10NO3 occur nearly simultaneously with the increasing BLT. For example, in Figure 4d, the Log10Chl difference abruptly increases in September (between vertical dash and solid lines in boreal summer), and this happens every year. Wind speed at this time is weakening from the annual peak in August (blue line in Figure 4a). At the same time, in Figure 4c, the MLD starts to shoal while the ITD stays flat or deepens a little. This different behavior between the MLD and ITD leads to noticeably thick barrier layer (over 20 m) in the No_MJO simulation. During this process, D0_Log10NO3 stays just below the ITD; it stays flat in the No_MJO simulation, while it shoals in the MJO simulation. As a result, in the No_MJO simulation, the barrier layer isolates the surface layer from the high nutrient layer. This is the reason why the seasonally low surface Chl concentration appears from November to January in the No_MJO simulation.
 In the No_MJO simulation, the deeper ITD is maintained from August to January (Figure 4c; between two gray dashed lines; 50–60 m depth). However, the added MJO forcing changes the seasonal behavior of the ITD. In the MJO simulation (Figure 4b), the deepest ITD is observed in August to September (around 50 m depth), and in the following boreal winter, the ITD hovers persistently around 40 m depth. The modified seasonal behavior by MJO forcing is also detected in the surface Log10Chl. In the MJO simulation, the mean Log10Chl in boreal summer is lower than that in boreal winter, while the lowest surface Chl concentration in the No_MJO simulation is around January. Considered together, these results suggest that the seasonal character of MJO forcing is strong enough to modify the seasonal cycle of ITD to be shallower in boreal winter. The shallower ITD contributes to increased nutrient concentration at a given depth around the mixed layer bottom because the ITD and D0_Log10NO3 occur at a similar depth all season long. Thus, more nutrients available at the mixed layer interface aid the entrainment process to work effectively and increase surface Chl concentration. This case study clearly shows that the removal of MJO anomaly from the atmospheric forcing results in increased BLT and lowered surface Chl concentration in as little as a year.
3.3 Case Study 2: Southern Bay of Bengal
 The second case study is performed in the southern Bay of Bengal (90°E–92°E, 4°N–6°N). This location is selected because the local maximum concentration of annual mean surface Chl in the MJO simulation is quite evident (Figure 1c), while the increased BLT is also notable in this region in the boreal summer season (Figure 3d), in contrast to the previous case study. At this location, the climatological wind speed as well as MJO activity is stronger in boreal summer than in boreal winter (Figure 5a). The climatologically high wind speed results in a relatively deep MLD. Figure 5c shows that during July to August when the surface wind speed is at its peak, the MLD in the No_MJO simulation is as deep as 70 m. After August, as the MLD shoals, surface Chl concentration decreases, and its minima occurs around November. This characteristic of the No_MJO simulation is totally different from that of the MJO simulation. Figure 5b shows that during August to November, the oscillating wind associated with the MJO leads the entrainment, and high concentration of nutrients around the bottom of the mixed layer helps to maintain high concentration of surface Chl.
 In the No_MJO simulation, D0_Log10NO3 is just below the MLD, so D0_Log10NO3 is quite deep (below 70 m) in boreal summer. At the same time, the surface value of Log10NO3 in the MJO simulation is even higher than 0, particularly in June and July. Thus, a significant difference in D0_Log10NO3 (the MJO – No_MJO; black solid and orange line) is evident in Figure 5e. Here, the orange line shows a notable trend; e.g., the large differences during July to August become more negative year by year. In the case of D1_Log10NO3, as in the previous case, the difference (black dash and green line in Figure 5e) is much smaller than D0_Log10NO3 difference (note again that D0_Log10NO3 and D1_Log10NO3 correspond to the isopleths depth of Log10NO3 = 0 and Log10NO3 = 1, respectively).
 In Figure 5f, the variation of the MLD difference (the MJO – No_MJO runs; black solid and orange line) in this case is similar to the previous case; a shallower mean MLD occurs during the active MJO season, and a smaller difference between the MJO and No_MJO simulations is seen during the calm MJO season. On the other hand, the variation of the ITD difference (black dash and green line) is not similar to the previous case, unlike the MLD difference. In the former case shown in Figure 4f, the shoaled ITD in the MJO simulation was significant in boreal winter, which is an active MJO season at that location, while the ITD difference was negligible in boreal summer. In Figure 5f, the minimum difference of ITD is detected in boreal summer as in the previous case in spite of the strong MJO activity in this season at this location. In both the cases, the peak of climatological wind forcing occurs during July to August, and the ITD and MLD deepen from May to August. Hence, we can infer that the rectifying effect of MJOs on the ITD is limited during the period when the ITD is quite deep. This result is consistent with J12 in terms of the importance of background seasonal cycle.
 Figure 5f indicates that in boreal summer, the MLD shoaling is greater than the ITD shoaling in the MJO simulation in terms of the low-frequency signal (orange and green lines), so the mean BLT in the MJO simulation increases in the active MJO season. It seems that the different timing of the shallowest depth of the local MLD and ITD, which is a response to the calm-wind MJO phase, contributes to the temporally large BLT in this season. For example, between the two solid gray lines marked as “1May83” and “1Nov83” in Figure 5b, the timing of local trough of the purple solid and dash lines are the same, but the timing of local peak of the purple solid line (MLD) precedes that of the purple dash line (ITD) by about 10 days. This occurs because, in the case of MLD, the shallowing is faster than the deepening (15–20 days vs. 30–35 days). It appears that the shallowest MLD occurs nearly simultaneously with the weakest wind forcing, while the deepest MLD occurs a few days after the strongest wind forcing. In the case of ITD, both peaks and troughs are lagged by a few days with respect to the phase of wind forcing.
 In the previous case, D0_Log10NO3 was of similar depth and variability as the ITD, and the barrier layer isolated the surface layer from the deeper nutrient-rich layer particularly in boreal summer (Figure 4c). However, in this location in boreal summer, the impact of the barrier layer is unclear because the entrainment bloom clearly occurs in association with the summer MJO. Here, the possibility of the same entrainment process during the deepening ITD as in the case of the deepening MLD is suggested. Both Figures 5b and 5c indicate that when the ITD abruptly deepens from May to June (e.g., to the right of the solid gray line marked as “1May83”), the depth difference between the ITD (purple long dash line) and the solid black line that indicates D0_Log10NO3 decreases close to 0, and nutrient concentration in the barrier layer increases. The increased nutrients in the barrier layer finally contribute to the effective entrainment bloom at the surface when the MLD deepens.
 In summary, for the southern Bay of Bengal, it is shown that oscillating wind forcing results in fluctuations of the MLD and ITD, and the deepening of both MLD and ITD might trigger the entrainment process. The entrainment induced by the deepening MLD and ITD cooperate with each other to supply nutrients to the surface layer. Hence, despite the large BLT values in the low-frequency signal, intraseasonal variations of MLD and ITD are actually large, and vigorous mixings occur by MJO forcing. It is also worth noting that at this location, wind anomalies in boreal winter are quite weak compared to those in summer, but even these relatively weak wind anomalies drive perturbations of oceanic vertical structure in the MJO simulation.
3.4 Observational Evidence
 Until now, our numerical experiments strongly suggest that the atmospheric ISV forcing can increase surface Chl mean concentration. Can this hypothesis be supported by observational evidence? To answer this, we calculate local correlation coefficients between seasonal ISV strength and seasonal mean of surface Chl concentration. The seasonal ISV strength is defined as a standard deviation of 30–120 day band-pass filtered wind speed anomaly for six months (boreal winter or summer). The seasonal mean Chl concentration is prepared for both the MODIS Aqua and SeaWiFS datasets by averaging monthly data from July to October (for boreal summer) or from January to April (for boreal winter). The reason why we average Chl concentration for 4 months (e.g., Jul to Oct) instead of 6 months (e.g., May to Oct) is because we assume that the initiation of seasonal ISV (i.e., starting from May or Nov) does not result in immediate increase of surface Chl concentration; some time is needed for breaking down the vertical stratification, pulling up nutrients to the surface layer, and phytoplankton growth taking place. Lastly, we applied horizontal smoothing with a 5° × 5° box averaging to both wind and Chl data sets in order to focus on large-scale variability. Considering the data span of MODIS Aqua (2002–2010) and SeaWiFS (1998–2010), we calculated the 95% significance level based on 9 and 13 degrees of freedom, respectively, and display the boreal summer results in Figures 6a and 6b. Boreal winter results are shown in Appendix A rather than shown here because the correlation coefficients are less significant, and less consistent between MODIS Aqua and SeaWiFS data.
 Figures 6a and 6b show significant positive correlations in the northwestern tropical Pacific, and several local points in the tropical Indian Ocean. Among those regions, the eastern equatorial Indian Ocean overlaps with the region that is: (1) on or around the route of MJO convections, and (2) reported as a region with a significant barrier layer. The northwestern tropical Pacific is also known as the region affected by boreal summer MJO convections [e.g., Waliser et al., 2004, 2009; Jin et al., 2012b].
 We can say that the positive correlation in the open ocean is consistent with our suggested mechanism, but of course this correlation analysis is not sufficient to support the dynamical linkage implied in the hypothesis – at least in part because a higher standard deviation of intraseasonal wind variability over a season is not necessarily due to strong episodic forcing like the MJO, and it may be influenced by other coupled aspects of the system that are not controlled for relative to the model experiments presented here. Time series of the vertical profile of the oceanic upper layer temperature, salinity, nitrate, etc. would be ideal to analyze this further, but the data for performing such an analysis is not available at present for sufficiently long periods. Hence, one approach is to examine the WOA09 climatology. In Figure 6c, boreal summer season mean BLT is displayed as color shading, and the 95% significance level of correlation is overlaid as contour lines (black for the MODIS Aqua, and gray for the SeaWiFS Chl). Here, the broad positive correlation signals in the northwestern tropical Pacific as well as those in the eastern equatorial Indian Ocean coincide with the region of climatological BLTs of 20–40 m. This suggests that seasonally persistent ISV can mix through this range of BLT, and draw upon nutrients from a deeper layer to drive surface Chl blooms. If the BLT is thinner, even a weaker ISV can contribute to the surface Chl bloom similar to a stronger or more persistent ISV.
 In Figures 6a and 6b, significant negative correlations are notable northwest of Australia (Figure 6a), and in the Bay of Bengal (Figure 6b). For the former, we found significant positive correlations between seasonal mean SST and ISV strength (not shown). Hence we hypothesize that there is a relationship between the ISV and coastal Ekman pumping. In the Bay of Bengal case, it is reported that the effect of surface wind on Chl variability is limited in this region due to severe vertical stratification [Prasanna Kumar et al., 2002]. Monsoon related heavy rainfall and corresponding river discharge in boreal summer also act to enhance the vertical stratification in this basin [e.g., Howden and Murtugudde, 2001]. Hence, the negative correlation can be explained, at least in part, as the effect of intraseasonal rainfall variability, though there is larger observational uncertainty in ocean color data in the Bay of Bengal in boreal summer due to persistent clouds associated with the monsoon. Overall, the observations support, albeit tentatively, the hypothesis of increased surface Chl concentration by the atmospheric ISV, although the effects are limited to a few regions and periods through the year.
4 Summary and Conclusion
 In this study, we examined how intraseasonal forcing enhances surface Chl. Contrasting two multidecadal simulations, with the only difference being MJO forcing anomalies, we showed that the mean surface Chl concentration in the MJO simulation is significantly higher than that in the control simulation. The Chl concentration in the MJO-forced case is closer to the observations, while that in the control run is significantly lower than observations. The examination of vertical structure along the equator suggested that the added MJO forcing enhances vertical mixing, and reduces the vertical gradient of nutrients and Chl concentration in the upper layer, while there are only minor differences in the depth of the 24 °C isotherm (around 100 m depth). The seasonal mean of BLT is also improved in several locations by the rectifying effect of MJO forcing; the mean BLT mostly decreases with one exception which was examined in the second case study. It is also found that the improved BLTs in the seasonal mean co-occur with increased mean Chl concentration at the surface (Figures 3e and 3f).
 The first case study was performed in the central-eastern equatorial Indian Ocean. In this region, the effect of boreal winter MJO is larger than that of summer MJO. In boreal winter, MJO forcing enhances oceanic vertical mixing by increasing the MLD and ITD variability. This enhanced vertical mixing lifts nutrients to the upper ocean, and results in increased surface Chl concentration. As for the low-frequency signal, the ITD shoaling by MJO forcing is larger than the MLD shoaling, which translates to a thinner mean barrier layer.
 The second case study was for the southern Bay of Bengal in which the effect of boreal summer MJO is much larger than that of the winter MJO. In boreal summer, the ITD is relatively deep, and the rectifying effect of MJO forcing on the ITD is small. Because the MLD rectification to shallower depths by MJO forcing is larger than that of ITD in this region, the seasonal mean of the BLT becomes thicker in the MJO simulation (Figure 3d). However, intraseasonal variation of the ITD responding to MJO forcing is still large, and the stratification weakens. In this case, the barrier layer is temporally variable in its response to MJO forcing, and is not effective in isolating the surface layer from the deeper layer in terms of Chl and nutrient distribution. It appears that the entrainment process via the deepening ITD allows nutrients to increase in the barrier layer, resulting in an entrainment bloom in the surface layer. The nonlocal processes of barrier layer formation and nutrient supply need to be considered in the future.
 Our experiments with a numerical coupled biophysical ocean model clearly show that the atmospheric ISV can increase mean surface Chl concentration by enhancing vertical mixing and providing more nutrients to the surface if the ISV is strong and persistent enough. This result is also supported by the observational data. With a simple correlation analysis between seasonal ISV and seasonal mean surface Chl concentration, we found a statistically significant positive relationship in a few limited regions particularly in the boreal summer season. This observational evidence is supportive of our conclusions, but not sufficient because there are many uncertainties. Previously, Waliser et al.  suggested and Jin et al. [2012b] found consistent results that surface wind forcing is the primary source for the MJO-related Chl variability. However, there are several other factors which can affect surface Chl concentration other than the wind associated with the MJO; e.g., wind stress curl and Ekman pumping, dust deposition, local storms and large-scale monsoon variability or global scale climate perturbations like El Niño/ La Niña. For further study to clarify the relationship, it would be helpful to analyze oceanic vertical profiles of salinity, temperature, Chl, etc., at a number of regions for a multiyear period.
 Lastly, related to the previously noted opposing effects of surface winds and precipitation on the BLT, our model results are biased to the diminishing effect of winds on the BLT rather than the enhancing effects of precipitation. Similar to the experimental setup used in the case study (the No_MJO simulation), when we include the anomalous MJO forcing only into the winds (u, v, wspd, utau, vtau) while other forcings are climatological, the results are fairly similar to those in the MJO simulation (not shown). It is possible that the sensitivity of our OGCM (and maybe of other OGCMs also) to the anomalous precipitation is weaker than that in nature. We expect that an improved numerical model which can simulate the BLT variability by more accurately responding to both winds and precipitation would be necessary for improving not only the physical features but also biological phenomena like the Chl modulation by the ISV.
 The same correlation coefficients shown in Figure 6, but for the boreal winter season (Nov–Apr) are displayed in Figure 7. Comparing Figures A1a and A1b, the locations of positive correlation are somewhat inconsistent between the two Chl datasets. The MODIS Aqua data shows positive correlation coefficients around the Maritime Continents, while those are less significant in Figure Alb, with the exception in the Arabian Sea. Negative correlations in the equatorial western Pacific are significant in both panels. Figure A1c indicates that this region has a climatologically deep BLT around 40–60 m. Hence, it seems that winds associated with the MJO are not strong enough to break this thick BLT. We hypothesize that the negative correlation is due to heavy precipitation associated with the MJO which enhances the BLT, although it is not supported in this study because the effect of precipitation on the BLT appears too weak in the model.
 D.J., R.M., and D.W. acknowledge support from NASA PO grant NNX09AF43G. R.M. also acknowledges the ONR DYNAMO grant. D.W.’s contribution to this study was carried out on behalf of the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration.