1. Introduction
Collectively, the current suite of coupled general circulation models (CGCMs) is not able to simulate the Asian monsoon mean state (climatological annual cycle) realistically. Biases include weak monsoonal winds, a rainfall deficit in the Bay of Bengal, excess rainfall in the western equatorial Indian Ocean (WEIO), and an easterly bias along the equator (Annamalai et al. 2017). Despite considerable effort, the causes of these errors are still not known. Some researchers have sought to relate them to biases in the atmospheric component of the coupled models (e.g., Martin et al. 2010; Ma et al. 2014). Others have focused on the ocean component, considering the causes and impacts of surface (e.g., Han et al. 2012; Levine et al. 2013; Sandeep and Ajayamohan 2014) and subsurface (e.g., Chowdary et al. 2016) temperature biases and of Bay of Bengal salinity biases (e.g., Seo et al. 2009). Recently, Annamalai et al. (2017) argued that the cause of poor monsoon simulations is the models’ near-equatorial coupled processes (Bjerknes feedback) being too strong, thereby suggesting that both oceanic and atmospheric errors are involved.
In this study, we investigate the causes of errors in the depth of the 20°C isotherm (D20), a measure of thermocline depth in tropical oceans. Its misrepresentation can lead to SST errors in upwelling regions where it rises near to the surface. Moreover, its accurate simulation has been shown to be essential for the successful prediction of climate modes, such as for El Niño (e.g., Ji and Leetmaa 1997) and the Indian Ocean dipole (e.g., Luo et al. 2007).
a. Background
Figure 1 provides maps of biases in annual-mean D20 (ΔD20; Fig. 1a) and the mixed-layer thickness (MLT) in January (ΔMLT; Fig. 1b). In both panels, errors are defined by differences between observations and the multimodel-mean (MMM) fields from a suite of coupled ocean–atmosphere models (see section 2 for details). We show ΔMLT during January because that is when the mixed layer is thickest in response to wintertime cooling and, hence, impacts the upper-ocean stratification most strongly (see below). The most striking biases occur in the northern Arabian Sea (NAS), where ΔD20 and ΔMLT attain maxima of 182 and 126 m, respectively, in the northwest corner of the basin.
Maps of (a) annual-mean D20 and (b) January MLT bias. The bias is defined as the difference between the multimodel mean of 31 coupled models and observations (the former minus the latter).
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
A possible cause of ΔD20, suggested by the property that it is largest in the northwest corner, is error in the models’ formation of Persian Gulf water (PGW). In the annual mean, Arabian Sea water enters the Gulf near the surface, and PGW exits near the bottom, with an overall exchange of 0.1–0.3 Sv (1 Sv ≡ 106 m3 s−1; e.g., Schott and McCreary 2001; Kämpf and Sadrinasab 2006; Yao 2008; Yao and Johns 2010). When PGW enters the Arabian Sea its core salinity and temperature are
Another possible cause of
A number of processes might force
Near-surface salinity is another variable with a large bias in the NAS. The Arabian Sea is a region of net evaporation. As such, the local production of salty water must be balanced by the advection of freshwater into the basin. There are two freshwater pathways into the NAS: the West India Coastal Current (WICC) and summertime Somali Current, with the former bringing freshwater into the Arabian Sea from the Bay of Bengal and the latter from the south Indian Ocean (Shetye et al. 1994; Prasad and Ikeda 2002b; Rao and Sivakumar 2003). Shankar et al. (2016) explicitly demonstrated the strong impact of salinity advection by the WICC on upper-ocean stratification in the NAS, tending to lower near-surface density there. Thus, errors in these freshwater sources could also impact
b. Present research
In this study, we seek to understand the causes of the NAS mean-state biases that occur in the ocean components of a suite of 31 coupled ocean–atmosphere models. For this purpose, we describe the biases that occur in their MMM fields and, to illustrate differences among the solutions, also in four model subgroups (defined in section 3a) and for several individual models. To identify the processes that cause the biases, we then report a series of scatterplots of data points from pairs of model variables. In some of the plots, the points are highly correlated, an indication that variations in biases among the models occur systematically, despite large differences in process parameterizations across the models. Of course, high correlations between variables x and y do not necessarily indicate causality:
Key results are the following. Large D20 biases result from the wintertime MLT near the northern boundary of the NAS being too thick, which generates ASHSW that is too deep. For the models in which PGW spreads into the NAS,
The paper is organized as follows. Section 2 summarizes our data sources and analysis techniques. Section 3 describes the model biases, and section 4 identifies the processes that determine them. Section 5 provides a summary and discussion of our results, the latter including their implications for the improvement of climate models.
2. Data and analyses
a. Observational and model data
The observational data we use include oceanic variables from the World Ocean Atlas 2013 (WOA13; Locarnini et al. 2013; Zweng et al. 2013), rainfall from the Global Precipitation Climatology Project (GPCP; Huffman et al. 1997), and surface heat and evaporative fluxes from OAFlux (Yu et al. 2008). Surface winds are obtained from the European Centre for Medium-Range Weather Forecast (ECMWF) interim reanalysis (ERA-Interim; Dee et al. 2011). Horizontal grid intervals are 1° × 1° for WOA13 and OAFlux, 2.5° × 2.5° for GPCP rain, and 1.5° × 1.5° for the ERA-Interim reanalysis. The time series we use for each dataset extend from 1979 to 2015 for GPCP rainfall and the ERA-Interim reanalysis, 1984 to 2009 for OAFlux surface heat flux, and 1985 to 2014 for OAFlux evaporative flux. All the datasets are provided as monthly averages.
The coupled models we analyze are listed alphabetically in appendix A. The suite consists of solutions from 31 models from phase 5 of the Coupled Model Intercomparison Project (CMIP5); see Sperber et al. (2013) and Nagura et al. (2013) for details. It also includes a solution from the Coupled Model For the Earth Simulator (CFES) developed at the Japan Agency for Marine-Earth Science and Technology (JAMSTEC). In our analyses, we use output from the models’ historical runs, that is, from their solutions forced by observed atmospheric composition changes and solar radiation. Table 1 and Fig. 2 (below) list each model’s acronym, symbol, and basic properties relevant to the present study.
List of model properties. For each model, the table columns list its acronym (Model), group (Group), Persian Gulf representation (PG), mixed-layer model (ML), and horizontal (Hor) and vertical resolution (Ver). The table is divided into four blocks, which, from top to bottom, correspond to groups 1A, 1B, 2A, and 2B defined in section 3a. The Persian Gulf representations, discussed in section 2a, are as follows: no Persian Gulf (N); and with a Persian Gulf in which exchange with the NAS is directly determined (D), parameterized using the Griffies et al. (2004) cross-land mixing scheme (C), or externally prescribed (P). The symbol [D] indicates that the exchange is likely direct, but we are unable to confirm that property. The symbol N′ indicates that the model has a Persian Gulf but no across-strait transport is allowed. Acronyms for the mixed-layer models are defined at the end of section 2a. Column Hor lists the longitudinal and latitudinal (lon × lat) resolution of each model in degrees. Column Ver denotes whether a model uses level (Z), density (D), or hybrid (H) coordinates, and provides the number of vertical divisions.
Values of (a) annual-average
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
The CGCMs are all global in extent. In their ocean components, the number of vertical levels varies from 30 to 50 and their horizontal resolution is typically 1°. Given their low horizontal resolution, almost all the models cannot resolve mesoscale eddies. With regard to the NAS, misrepresentation of eddy activity is potentially serious, as eddies are known to be vigorous in the Gulf of Oman and to contribute to the spreading of PGW (L’Hégaret et al. 2013, 2015, 2016; Vic et al. 2015). On the other hand, PGW does spread throughout the NAS in many of the CGCMs, and so must occur by a large-scale process rather than by eddies [see section 3b(1)(iv)].
Not surprisingly, the models specify the flow of PGW into the Arabian Sea differently. The PG column in Table 1 separates the models into four general categories: models without any Persian Gulf (N); with a Persian Gulf that freely exchanges water with the Arabian Sea (D); and with an isolated Persian Gulf region in which the exchange is either prescribed (P) or parameterized by enhanced cross-land mixing (C; Griffies et al. 2004). To our knowledge, none of the models without a Persian Gulf prescribed the impact of PGW within the NAS (e.g., by restoring temperature and salinity toward observed values in the northwest corner of the basin); rather, the NAS stratification was entirely determined by processes within the basin. The strait that connects the Persian Gulf to the NAS is only about 50 km wide, too narrow to be resolved in most of the models; in the models that have a Persian Gulf, then, the strait is unrealistically wide. A result of these model differences is that the depth, salinity, and temperature of PGW vary considerably among the models (see the discussion of Fig. 2 below).
The models also differ in their mixed-layer parameterizations, using six different types (see the ML column of Table 1): 1) K-profile parameterization (Large et al. 1994; KPP) or its updated version (Danabasoglu et al. 2006); 2) the Kraus-Turner (1967) scheme (KT); 3) the Noh and Kim (1999) method (NK) or its updated version (Noh et al. 2005); 4) the Blanke and Delecluse (1993) method (BD) or its updated version (Madec et al. 2008); 5) the Pacanowski and Philander (1981) scheme (PP); and 6) Oberhuber’s (1993) parameterization (OB).
b. Analyses
Monthly climatologies are computed for the observations using the complete time series for each dataset and for the models using the last 50 years of model output. We define MLT, which is not included in the model datasets, to be the depth where density increases by
For a particular variable









Assuming the data points are scattered with a Gaussian distribution, there is a correlation value above which we can say that the correlation exists (i.e.,
In each scatterplot, we also plot lines,
3. Model biases
In this section, we discuss the stratification biases that occur in the coupled models. We first note that their magnitude varies considerably and that they divide into four distinct groups (section 3a). Then, we describe the horizontal, temporal, and vertical structures for the complete set of models [section 3b(1)] and for four subgroups and several individual models [section 3b(2)].
Throughout the text, we define
Maps of (a) annual-mean D20, (b) annual-mean salinity at 200-m depth,
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
a. Bias differences and model groups
There is a large variation in magnitude of the NAS biases among the models. To illustrate, for all the models Fig. 2 lists from left to right
For our analyses, we found it useful to divide the models into four groups, indicated by the different colored bars in Fig. 2. Two major groups are defined by whether the model lacks (group 1; black and gray bars) or includes a Persian Gulf (group 2; blue and red bars). Within these two groups, D20 values either increase to unrealistic values larger than 200 m (groups 1A and 2A; black and blue bars) or remain smaller than 200 m (groups 1B and 2B; gray and red bars). For convenience, throughout the text we shorten the group names to Gn, GX, and GnX, where n = 1, 2 and X = A, B (e.g., group 2a is G2A).
There are obvious model–observation differences for all the variables. Most notably, observed
It is visually apparent that variations between some pairs of variables are correlated. The models are arranged in Fig. 2 in order of increasing
b. Bias structures
Although biases vary considerably among the models, many of them tend to have a common structure. To illustrate these commonalities, we first describe basic features of the MMM stratification and forcing fields [section 3b(1)]. We then discuss differences that occur within the four groups [section 3b(2)] and for a few individual models [section 3b(2)(iii)].
1) Common features
(i) D20, S, and MLT biases
Figure 3 shows the horizontal structure of the stratification errors, providing maps of annual-mean D20 (left column), annual-mean salinity S at a depth of 200 m (
The curves in the bottom panels of Fig. 3 illustrate the time dependence of the biases, plotting areal averages of D20,
Figure 4 illustrates the vertical structure of the stratification errors, showing sections along 65°E of (left to right) annual-mean, temperature T, salinity S, and density
Sections along 65°E, showing (left) annual-mean temperature, (middle) salinity, and (right) potential density for (top) observations and (bottom) differences between the MMM fields of the coupled models and observations. The black curves are the January mixed-layer thicknesses along 65°E (top) from observations and (bottom) from the MMM fields.
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
The stratification within the wintertime mixed layer (above the black curves in Fig. 4) is influenced by surface buoyancy fluxes. The negative
(top) Maps of MMM fields, showing (a) OND surface heat flux Q (positive means warming to the ocean), (b) OND evaporation minus precipitation, E − P, and (c) annual-mean Ekman pumping velocity,
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
(a),(b) Maps of depth-integrated current and pressure anomalies from annual-mean MMM fields and the Sverdrup flow driven by ECMWF-Interim winds, (c),(d) velocity sections, and (e),(f) maps of currents averaged from 100 m to the surface during JJA and OND. In (a) and (b), values of
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
(ii) Density bias
Because the
(iii) Forcing
Figure 5 illustrates the horizontal structure and time dependence of the MMM surface forcings in the NAS. The top panels show
The
The E − P map shows strong evaporation during OND in the NAS (Fig. 5b), with a maximum of 7 mm day−1 near the northern boundary due to the influence of dry continental air. The E − P bias in the NAS (difference between curves in Fig. 5e) is weak throughout most of the year, being negative only during March–May and positive during November–December.
The
(iv) Circulation
The NAS circulation is complex, with a prominent annual cycle driven by the annually reversing monsoon winds. Given this complexity, a thorough discussion of the circulation and dynamics of the region is beyond the scope of our study (see, e.g., Prasad and Ikeda 2002a,b). Here, then, we note two aspects of the flow field relevant to our results: the existence of mean, subsurface, southward flow away from the northern boundary of the basin; and of seasonal, northward, near-surface currents into the boundary.
(a) Annual mean southward flow
Because the intense winds of the southwest monsoon dominate the annual cycle, the NAS has a significant mean circulation. Figures 6a and 6b illustrate its horizontal structure, showing annual-mean, depth-integrated current vectors and pressure anomalies, both for the MMM,
The integrations to determine




















Sverdrup (1947) derived solution (2) from depth-integrated equations, and in so doing lost all information about the vertical structure of the flow. In a model that allows for vertical structure, baroclinic adjustments (namely, the radiation of baroclinic Rossby waves across the basin) tend to trap the Sverdrup flow in the upper ocean (e.g., Anderson and Gill 1975; McCreary 1981a,b). It is reasonable, then, to compare the surface-trapped, annual-mean circulations in the coupled models to that of a Sverdrup flow. Indeed, the striking similarity of the interior circulations in Figs. 6a and 6b indicates that Sverdrup theory does capture the basic dynamics of the models’ annual-mean circulation. One prominent difference in the two flow fields is the presence of a western boundary current in the MMM, a flow that is not included in (2). Another obvious difference is the jump in
Figures 6c and 6d illustrate the vertical structure of the annual-mean MMM flow, showing meridional and zonal sections of across-section (color shading) and along-section (contours) currents, respectively. Although the strongest flows are surface trapped, the flow field is not confined above the wintertime MLT (green line); moreover, currents are not unidirectional but can reverse vertically. These properties are evident in the structure of the υ field (contours) in Fig. 6c: Near the northern boundary, υ is southward very near the surface and below 175 m, and is positive in between; farther to the south, it is southward well below the mixed layer. We interpret these features to result from the overturning circulations associated with PGW and ASHSW, which introduce baroclinicity and deepen the overall Sverdrup response. The southward current below the mixed layer likely contributes to the southward spreading of the saline warm bias shown in Fig. 4.
(b) Seasonal northward currents
Figures 6e and 6f provide maps of the average currents in the upper 100 m during summer (JJA) and fall (OND), respectively. Meridional and zonal sections like those in Figs. 6c and 6d show that both circulations are generally confined above 200 m (not shown). During the fall, there is northward flow in the WICC that extends from the southern tip of India into the NAS (Fig. 6f). During the summer, there is northward flow in the western Arabian Sea via the Somali and Omani Currents, but it does not extend to the northern boundary (Fig. 6e). We checked the summertime flow in each of the models, and in almost all of them the Omani Current separates from the coast before reaching the northern boundary. These results demonstrate that the WICC and the summertime Somali and Omani Currents are realistically simulated in the coupled models, and that they are capable of advecting freshwater into the NAS.
2) Group and individual differences
(i) D20, S, and MLT biases
Figures 7 and 8 plot annual-mean stratification biases for the four groups, showing maps of annual-mean ΔD20 and
(top) Maps of annual-mean
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
Maps of (top) annual-mean error in 200-m depth salinity,
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
The G1B (panels b and f) and G2B (panels d and h) biases differ most strikingly from the MMM fields in that they have no subsurface-intensified biases, a consequence of their wintertime mixed layers being relatively (realistically) thin (Fig. 9). Note also that
Maps of January
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
Figure 9 plots maps of January ΔMLT for each group. As for ΔD20 and
The above properties are also apparent in the MLT temporal curves in the bottom panels of Fig. 3. For
To investigate possible circulation differences among the groups, we obtained figures similar to Fig. 6 for each of them (not shown). The horizontal structures of the annual-mean and seasonal circulations are very similar to those for the MMM (panels a, b, e, and f), a consequence of the wind forcing being similar in each of the groups. The circulations, however, differ considerably in their depth, extending to greater depths in the GA models but remaining shallow in the GB models. The deep southward flow likely advects S and T bias to the south in the GA models (Figs. 7e,g and 8e,g).
(ii) SSS biases
As discussed in section 4e, an important influence on the NAS biases is SSS biases that are advected into the region. Figure 10 presents maps of (top) annual-mean ΔSSS and (bottom) rainfall bias for the four groups. Significant
Maps of (top) annual-mean
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
We attribute the various ΔSSS patterns in Fig. 10 to the relative strengths of these two sources. In G2B, bias 1 is smaller and bias 2 is larger in magnitude compared to G2A; as a result, the NAS is erroneously fresh (salty) in the G2B (G2A) models. Similar differences exist for the G1A and G1B models, except that
There are two pathways by which remotely generated salinity biases reach the NAS. During the fall and winter, anomalously salty water due to bias 1 flows southward along the east coast of India, around Sri Lanka, and northward along the west coast of India within the WICC (Fig. 6f; Han et al. 2001; Shankar et al. 2016). The impact of this pathway is evident by the tongues of salty water that extend into the NAS (Figs. 10a,c,d). During the summer, freshwater generated in the WEIO due to bias 2 flows northward in the western Arabian Sea in all the models (Fig. 6e), but only appears to reach the NAS in the G2B models for which bias 1 is weak (Fig. 10d). In the G1B models, salinity is less fresh in the WEIO than elsewhere, due to the excessive rainfall outside the WEIO and river runoff into the northern Bay; thus, the freshest water appears to be carried into the NAS by the WICC rather than by western Arabian Sea currents.
(iii) PGW biases
PGW spreads into the interior of the NAS in some, but not all, of the models. To illustrate, Fig. 11 plots sections of
Zonal sections near the northern boundary, showing density (contours) and
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
For all the G2B models, the mixed layer is thin and PGW appears as a distinct salinity maximum within the upper pycnocline. For most of them (9 of the 11), PGW is trapped near the western boundary, and the
In contrast, for most of the G2A models the
The above properties suggest that PGW spreads weakly or not at all into the NAS for the G2B models but spreads efficiently for the G2A models. The dynamical cause of this difference is not clear. One possibility is that PGW is deeper than the NAS MLT for the G2B models (
4. Processes
In section 3, we identified key properties of the model biases in the NAS. Here, we organize those properties into an interrelated chain of processes. We begin with an overview of our analysis approach (section 4a). In the rest of the section, we fill out the chain, discussing one bias and process at a time (sections 4b to 4e).
a. Overview
The discussion in sections 4b to 4e is centered on analyses of a series of scatterplots between model variables. Figure 12 shows six members from the series, ordered so that the identification of one process leads logically to another. Model data points in Fig. 12 are colored by their group as in Table 1: G1A (black), G1B (gray), G2A (blue), and G2B (red). The straight lines are best-fit lines either for all the models (solid black) or separately for the G1 (black dashed) and G2 (blue dashed) models (section 2b). The G1 and G2 lines are plotted (Figs. 12a,e) if the difference in their slopes is significant at the 95% level (section 2b); otherwise, the data points do not tend to separate for the two groups, so a single best-fit line is included (Figs. 12b,c,d,f). Given that PGW is present in the G2 but not in the G1 models, the existence of slope differences in Figs. 12a and 12e suggests that only those linkages are impacted by PGW.
A sequence of scatterplots that highlights the processes that determine
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
As noted in section 1b, an issue is that high correlation between two variables,
Note that, although significant correlations indicate that a systematic process causes biases to vary among the models, they provide no information about why models differ from observations [since Eq. (1) does not involve observations at all]. The green dots in each panel of Fig. 12 indicate observational values. In all but one of the plots (Fig. 12b), some data points lie near the dot, indicating that their models represent the observations better than the others. We infer that the process that causes biases to vary among the models is also represented “better” in those models.
b. D20 bias
Existing knowledge of the formation of ASHSW and ventilated thermocline theory (section 1), as well as results from section 3, suggests that the wintertime MLT along the NAS northern boundary determines D20. Specifically, the thick wintertime mixed layer there generates a warm and saline water mass that subducts during spring and spreads southward (Figs. 7 and 8), leading to a deeper D20. In this view, as MLT increases, the warm bias extends deeper and D20 should increase.
Figure 12a provides a scatterplot of annual-mean
For the G1 models (black and gray points), the nearly one-to-one linkage between the two variables exists because it involves only basic properties of the stratification:
For the G2A models, the correlation between
In the above discussion, we have assumed that
c. MLT bias
What processes, then, cause







According to Eqs. (3a) and (3b),
The correlations of
The weak correlations with the surface fluxes are surprising, particularly for




















The above results suggest that
d. 
bias

What processes determine
In a second step, we determine whether temperature or salinity affects

















Similar to Fig. 12a, the slopes of the best-fit lines between
For both groups, wintertime mixed-layer water in the NAS is a mixture of water advected primarily from the WICC [section 3b(2)(ii); Prasad and Ikeda 2002b; Shankar et al. 2016] and entrained from below. The temperature of the entrained water, however, differs between them: For the G1 and G2B models, it is colder because warmer PGW does not spread into the interior of the NAS, whereas for the G2A models it is warmer because it does. In confirmation of this difference, water 100 m below the mixed layer is colder than that in the mixed layer by 3.5°C for the G1 and G2B models but by only 1.6°C for the G2A models.
The different temperatures of entrained water lead to the opposite signs in the
To summarize, for a given
e. 
bias

What processes determine
To investigate the above processes, we obtained scatterplots of January
For the G2 models, SSS in the Persian Gulf, and hence
It remains to assess the sources and impacts of the
Maps of correlation coefficients between annual-mean SSS at every point in the basin with SSS at (a) 15°N, 55°E and (b) 15°N, 72°E, both points indicated by
Citation: Journal of Climate 31, 5; 10.1175/JCLI-D-17-0417.1
Note also in Fig. 13 the correlations are positive everywhere (red and blue colors show strong and weak positive correlation, respectively). This property appears to contradict the fact that the deficit in rainfall in the Bay of Bengal (excess rainfall in the WEIO) generates a saline (fresh) bias (Fig. 10). The resolution of the contradiction is that, although WEIO salinity and
f. G1B and G2B models
The most striking difference between the GB models (G1B + G2B models; gray and red points) and the GA models (G1A + G2A models; black and blue points) is that
One possible reason is suggested by the separation of GA and GB data points in the other panels of Fig. 12: Although there is some overlap, their mean values are clearly different. In particular, in Fig. 12d note that
Another possibility is that different mixed-layer schemes are used in the GA and GB models: The KPP and BD mixing schemes are used in 12 of the 17 GA models, but not at all in the GB models (Table 1). Further, all the mixed-layer models require parameter choices, and perhaps those for the GB models are just “better.”
5. Summary and discussion
In this study, we investigate biases in the climatological mean state of the northern Arabian Sea (NAS) that are present in 31 coupled ocean–atmosphere models (appendix A and Table 1). Model differences lead to large variations in the biases (Fig. 2). Based on these differences, we divide the models into four groups: two groups depending on whether the model lacks (G1) or includes (G2) a Persian Gulf, and two groups depending on whether D20 values increase to unrealistically large values (GA) or remain small (GB).
Basic bias properties are captured in the multimodel-mean (MMM) fields for all the models (Figs. 1, 3, and 4). In particular, errors in the depth of the 20°C isotherm
Other bias properties, however, point toward
Figure 12 provides scatterplots that sequentially illustrate the chain of processes that connect
The different slopes of the best-fit lines for the G1 and G2 models in Figs. 12a and 12e illustrate the impacts of PGW in the NAS. In Fig. 12a,
The processes that cause the prominent model biases (large D20 and wintertime MLT; misrepresentation of PGW properties) also suggest possible pathways for model improvements. Regarding ΔMLT, it is most sensitive to ΔSSS along the west Indian coast, which is linked to the strength of models’ monsoon-rainfall deficit; thus, model improvements that increase monsoon rainfall (a coupled rather than purely oceanic problem) should lower the SSS advected into the NAS and hence reduce ΔMLT. MLT error could also result from the mixed-layer model used (Table 1) or the parameter values chosen for it; for example, it is possible that the mixed-layer models are tuned to be too sensitive to
In conclusion, we have determined that interactions among a number of ocean processes (mixed-layer physics, near-surface salinity advection, PGW formation and spreading, etc.) lead to the large stratification biases in the NAS. We have not, however, considered either the dynamics of those interactions in detail or the impact of the biases on climate. Regarding circulation dynamics, the Prasad and Ikeda (2002a,b) studies of ASHSW formation and spreading provide a good start, but a number of questions remain unanswered. They include the following: Since PGW is generated west of the NAS, what processes allow PGW to spread eastward along the northern boundary, against the direction of Rossby wave propagation? Why does the eastward spreading of PGW depend on MLT (i.e., require
Acknowledgments
The authors are grateful to Ryo Furue and Fabian Schloesser for helpful discussions. We also thank the reviewers, whose comments and suggestions improved our manuscript. This research was supported by MEXT/JSPS KAKENHI Grant 26800249 and by NSF Grant 1460742, and also by the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) through its sponsorship of research at the International Pacific Research Center. This manuscript is SOEST Contribution No. 10308 and IPRC Contribution No. 1311.
APPENDIX
List of Coupled Models
This appendix provides an alphabetical list of the coupled models used in this study (column 1 of Table A1) and the organization that produced them (column 2). Further expansions of acronyms can be found at http://www.ametsoc.org/PubsAcronymList.
List of coupled models.
REFERENCES
Anderson, D. L. T., and A. E. Gill, 1975: Spin-up of a stratified ocean, with applications to upwelling. Deep-Sea Res., 22, 583–596, https://doi.org/10.1016/0011-7471(75)90046-7.
Annamalai, H., B. Taguchi, J. P. McCreary, M. Nagura, and T. Miyama, 2017: Systematic errors in South Asian monsoon simulation: Importance of equatorial Indian Ocean processes. J. Climate, 30, 8159–8178, https://doi.org/10.1175/JCLI-D-16-0573.1.
Blanke, B., and P. Delecluse, 1993: Variability of the tropical Atlantic Ocean simulated by a general circulation model with two different mixed-layer physics. J. Phys. Oceanogr., 23, 1363–1388, https://doi.org/10.1175/1520-0485(1993)023<1363:VOTTAO>2.0.CO;2.
Chowdary, J. S., A. Parekh, G. Srinivas, C. Gnanaseelan, T. S. Fousiya, R. Khandekar, and M. K. Roxy, 2016: Processes associated with the tropical Indian Ocean subsurface temperature bias in a coupled model. J. Phys. Oceanogr., 46, 2863–2875, https://doi.org/10.1175/JPO-D-15-0245.1.
Danabasoglu, G., W. G. Large, J. J. Tribbia, P. R. Gent, B. P. Briegleb, and J. C. McWilliams, 2006: Diurnal coupling in the tropical oceans of CCSM3. J. Climate, 19, 2347–2365, https://doi.org/10.1175/JCLI3739.1.
de Boyer Montégut, C., G. Madec, A. S. Fischer, A. Lazar, and D. Iudicone, 2004: Mixed layer depth over the global ocean: An examination of profile data and a profile-based climatology. J. Geophys. Res., 109, C12003, https://doi.org/10.1029/2004JC002378.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/10.1002/qj.828.
de Szoeke, R. A., 1987: On the wind-driven circulation of the South Pacific Ocean. J. Phys. Oceanogr., 17, 613–630, https://doi.org/10.1175/1520-0485(1987)017<0613:OTWDCO>2.0.CO;2.
Fathrio, I., S. Iizuka, A. Manda, Y.-M. Kodama, S. Ishida, Q. Moteki, H. Yamada, and Y. Tachibana, 2017: Assessment of western Indian Ocean SST bias of CMIP5 models. J. Geophys. Res. Oceans, 122, 3123–3140, https://doi.org/10.1002/2016JC012443.
Griffies, S. M., M. J. Harrison, R. C. Pacanowski, and A. Rosati, 2004: Cross-land mixing. A technical guide to MOM4, GFDL Ocean Group Tech. Rep. 5, 177–188.
Han, W., J. P. McCreary, and K. E. Kohler, 2001: Influence of precipitation minus evaporation and Bay of Bengal rivers on dynamics, thermodynamics, and mixed layer physics in the upper Indian Ocean. J. Geophys. Res., 106, 6895–6916, https://doi.org/10.1029/2000JC000403.
Han, Z.-Y., T.-J. Zhou, and L.-W. Zou, 2012: Indian Ocean SST biases in a Flexible Regional Ocean Atmosphere Land System (FROALS) model. Atmos. Oceanic Sci. Lett., 5, 273–279, https://doi.org/10.1080/16742834.2012.11447012.
Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78, 5–20, https://doi.org/10.1175/1520-0477(1997)078<0005:TGPCPG>2.0.CO;2.
Ji, M., and A. Leetmaa, 1997: Impact of data assimilation on ocean initialization and El Niño prediction. Mon. Wea. Rev., 125, 742–753, https://doi.org/10.1175/1520-0493(1997)125<0742:IODAOO>2.0.CO;2.
Kämpf, J., and M. Sadrinasab, 2006: The circulation of the Persian Gulf: A numerical study. Ocean Sci., 2, 27–41, https://doi.org/10.5194/os-2-27-2006.
Kraus, E. B., and J. S. Turner, 1967: A one-dimensional model of the seasonal thermocline II. The general theory and its consequences. Tellus, 19 (1), 98–106, https://doi.org/10.3402/tellusa.v19i1.9753.
Kumar, S. P., and T. G. Prasad, 1999: Formation and spreading of Arabian Sea high-salinity water mass. J. Geophys. Res., 104, 1455–1464, https://doi.org/10.1029/1998JC900022.
Large, W. G., J. C. McWilliams, and S. C. Doney, 1994: Oceanic vertical mixing: A review and a model with a nonlocal boundary layer parameterization. Rev. Geophys., 32, 363–403, https://doi.org/10.1029/94RG01872.
Levine, R. C., A. G. Turner, D. Marathayil, and G. M. Martin, 2013: The role of northern Arabian Sea surface temperature biases in CMIP5 model simulations and future projections of Indian summer monsoon rainfall. Climate Dyn., 41, 155–172, https://doi.org/10.1007/s00382-012-1656-x.
L’Hégaret, P., L. Lacour, X. Carton, G. Roullet, R. Baraille, and S. Corréard, 2013: A seasonal dipolar eddy near Ras Al Hamra (Sea of Oman). Ocean Dyn., 63, 633–659, https://doi.org/10.1007/s10236-013-0616-2.
L’Hégaret, P., R. Duarte, X. Carton, C. Vic, D. Ciani, R. Baraille, and S. Corréard, 2015: Mesoscale variability in the Arabian Sea from HYCOM model results and observations: Impact on the Persian Gulf water path. Ocean Sci., 11, 667–693, https://doi.org/10.5194/os-11-667-2015.
L’Hégaret, P., X. Carton, S. Louazel, and G. Boutin, 2016: Mesoscale eddies and submesoscale structures of Persian Gulf water off the Omani coast in spring 2011. Ocean Sci., 12, 687–701, https://doi.org/10.5194/os-12-687-2016.
Locarnini, R. A., and Coauthors, 2013: Temperature. Vol. 1, World Ocean Atlas 2013, NOAA Atlas NESDIS 73, 40 pp., https://data.nodc.noaa.gov/woa/WOA13/DOC/woa13_vol1.pdf.
Luo, J.-J., S. Masson, S. Behera, and T. Yamagata, 2007: Experimental forecasts of the Indian Ocean dipole using a coupled OAGCM. J. Climate, 20, 2178–2190, https://doi.org/10.1175/JCLI4132.1.
Ma, H.-Y., and Coauthors, 2014: On the correspondence between mean forecast errors and climate errors in CMIP5 models. J. Climate, 27, 1781–1798, https://doi.org/10.1175/JCLI-D-13-00474.1.
Madec, G., and Coauthors, 2008: NEMO ocean engine. Note du Pôle de modélisation 1288-1619, Rep. 27, Institut Pierre-Simon Laplace, France, 209 pp.
Marathayil, D., A. G. Turner, L. C. Shaffrey, and R. C. Levine, 2013: Systematic winter sea-surface temperature biases in the northern Arabian Sea in HiGEM and the CMIP3 models. Environ. Res. Lett., 8, 014028, https://doi.org/10.1088/1748-9326/8/1/014028.
Martin, G. M., S. F. Milton, C. A. Senior, M. E. Brooks, S. Ineson, T. Reichler, and J. Kim, 2010: Analysis and reduction of systematic errors through a seamless approach to modeling weather and climate. J. Climate, 23, 5933–5957, https://doi.org/10.1175/2010JCLI3541.1.
McCreary, J. P., 1981a: A linear stratified ocean model of the equatorial undercurrent. Philos. Trans. Roy. Soc. London, 298A, 603–635, https://doi.org/10.1098/rsta.1981.0002.
McCreary, J. P., 1981b: A linear stratified ocean model of the coastal undercurrent. Philos. Trans. Roy. Soc. London, 302A, 385–413, https://doi.org/10.1098/rsta.1981.0176.
Morrison, J. M., 1997: Inter-monsoonal changes in the T-S properties of the near-surface waters of the northern Arabian Sea. Geophys. Res. Lett., 24, 2553–2556, https://doi.org/10.1029/97GL01876.
Nagura, M., W. Sasaki, T. Tozuka, J.-J. Luo, S. K. Behera, and T. Yamagata, 2013: Longitudinal biases in the Seychelles Dome simulated by 35 ocean–atmosphere coupled general circulation models. J. Geophys. Res. Oceans, 118, 831–846, https://doi.org/10.1029/2012JC008352.
Niiler, P. P., and E. B. Kraus, 1977: One-dimensional models of the upper ocean. Modeling and Prediction of the Upper Layers of the Ocean, E. B. Kraus, Ed., Pergamon Press, 143–172.
Noh, Y., and H. J. Kim, 1999: Simulations of temperature and turbulence structure of the oceanic boundary layer with the improved near-surface process. J. Geophys. Res., 104, 15 621–15 634, https://doi.org/10.1029/1999JC900068.
Noh, Y., Y. J. Kang, T. Matsuura, and S. Iizuka, 2005: Effect of the Prandtl number in the parameterization of vertical mixing in an OGCM of the tropical Pacific. Geophys. Res. Lett., 32, L23609, https://doi.org/10.1029/2005GL024540.
Oberhuber, J. M., 1993: Simulation of the Atlantic circulation with a coupled sea ice–mixed layer–isopycnal general circulation model. Part I: Model description. J. Phys. Oceanogr., 23, 808–829, https://doi.org/10.1175/1520-0485(1993)023<0808:SOTACW>2.0.CO;2.
Pacanowski, R. C., and S. G. H. Philander, 1981: Parameterization of vertical mixing in numerical models of tropical oceans. J. Phys. Oceanogr., 11, 1443–1451, https://doi.org/10.1175/1520-0485(1981)011<1443:POVMIN>2.0.CO;2.
Pedlosky, J., 1996a: Sverdrup theory. Ocean Circulation Theory, J. Pedlosky, Ed., Springer, 11–14.
Pedlosky, J., 1996b: Theory of the ventilated thermocline. Ocean Circulation Theory, J. Pedlosky, Ed., Springer, 171–276.
Prasad, T. G., and M. Ikeda, 2002a: The wintertime water mass formation in the northern Arabian Sea: A model study. J. Phys. Oceanogr., 32, 1028–1040, https://doi.org/10.1175/1520-0485(2002)032<1028:TWWMFI>2.0.CO;2.
Prasad, T. G., and M. Ikeda, 2002b: A numerical study of the seasonal variability of Arabian Sea high-salinity water. J. Geophys. Res., 107, 3197, https://doi.org/10.1029/2001JC001139.
Prasad, T. G., M. Ikeda, and S. P. Kumar, 2001: Seasonal spreading of the Persian Gulf water mass in the Arabian Sea. J. Geophys. Res., 106, 17 059–17 071, https://doi.org/10.1029/2000JC000480.
Premchand, K., J. S. Sastry, and C. S. Murty, 1986: Watermass structure in the western Indian Ocean—Part II: The spreading and transformation of the Persian Gulf water. Mausam, 37, 179–186.
Rao, R. R., and R. Sivakumar, 2003: Seasonal variability of sea surface salinity and salt budget of the mixed layer of the north Indian Ocean. J. Geophys. Res., 108, 3009, https://doi.org/10.1029/2001JC000907.
Rochford, D. J., 1964: Salinity maxima in the upper 1000 metres of the north Indian Ocean. Aust. J. Mar. Freshwater Res., 15, 1–24, https://doi.org/10.1071/MF9640001.
Sandeep, S., and R. S. Ajayamohan, 2014: Origin of cold bias over the Arabian Sea in climate models. Sci. Rep., 4, 6403, https://doi.org/10.1038/srep06403.
Sayantani, O., C. Gnanaseelan, J. S. Chowdary, A. Parekh, and S. Rahul, 2016: Arabian Sea SST evolution during spring to summer transition period and the associated processes in coupled climate models. Int. J. Climatol., 36, 2541–2554, https://doi.org/10.1002/joc.4511.
Schott, F. A., and J. P. McCreary, 2001: The monsoon circulation of the Indian Ocean. Prog. Oceanogr., 51, 1–123, https://doi.org/10.1016/S0079-6611(01)00083-0.
Seo, H., S.-P. Xie, R. Murtugudde, M. Jochum, and A. J. Miller, 2009: Seasonal effects of Indian Ocean freshwater forcing in a regional coupled model. J. Climate, 22, 6577–6596, https://doi.org/10.1175/2009JCLI2990.1.
Shankar, D., R. Remya, P. N. Vinayachandran, A. Chatterjee, and A. Behera, 2016: Inhibition of mixed-layer deepening during winter in the northeastern Arabian Sea by the West India Coastal Current. Climate Dyn., 47, 1049–1072, https://doi.org/10.1007/s00382-015-2888-3.
Shenoi, S. S. C., S. R. Shetye, A. D. Gouveia, and G. S. Michael, 1993: Salinity extrema in the Arabian Sea. Monsoon Biogeochemistry, V. Ittekkot and R. R. Nair, Eds., Geologisch-Paläontologischen Institutes der Universität Hamburg, 37–49.
Shetye, S. R., A. D. Gouveia, and S. S. C. Shenoi, 1994: Does winter cooling lead to the subsurface salinity minimum off Saurashtra, India? Oceanography of the Indian Ocean, B. N. Desai, Ed., A. A. Balkema, 617–625.
Sperber, K. R., H. Annamalai, I.-S. Kang, A. Kitoh, A. Moise, A. Turner, B. Wang, and T. Zhou, 2013: The Asian monsoon: An intercomparison of CMIP5 vs. CMIP3 simulations of the late 20th century. Climate Dyn., 41, 2711–2744, https://doi.org/10.1007/s00382-012-1607-6.
Stommel, H., 1979: Determination of water mass properties of water pumped down from the Ekman layer to the geostrophic flow below. Proc. Natl. Acad. Sci. USA, 76, 3051–3055, https://doi.org/10.1073/pnas.76.7.3051.
Sverdrup, H. U., 1947: Wind-driven currents in a baroclinic ocean; with application to the equatorial currents of the eastern Pacific. Proc. Natl. Acad. Sci. USA, 33, 318–326, https://doi.org/10.1073/pnas.33.11.318.
Talley, L. D., 1985: Ventilation of the subtropical North Pacific: The shallow salinity minimum. J. Phys. Oceanogr., 15, 633–649, https://doi.org/10.1175/1520-0485(1985)015<0633:VOTSNP>2.0.CO;2.
Tomczak, M., and J. S. Godfrey, 1994: Regional Oceanography: An Introduction. Elsevier, 422 pp.
Toyoda, T., and Coauthors, 2017: Intercomparison and validation of the mixed layer depth fields of global ocean syntheses. Climate Dyn., 49, 753–773, https://doi.org/10.1007/s00382-015-2637-7.
Vic, C., G. Roullet, X. Capet, X. Carton, M. J. Molemaker, and J. Gula, 2015: Eddy–topography interactions and the fate of the Persian Gulf outflow. J. Geophys. Res. Oceans, 120, 6700–6717, https://doi.org/10.1002/2015JC011033.
Yao, F., 2008: Water mass formation and circulation in the Persian Gulf and water exchange with the Indian Ocean. Ph.D. dissertation, University of Miami, 144 pp.
Yao, F., and W. E. Johns, 2010: A HYCOM modeling study of the Persian Gulf: 1. Model configurations and surface circulation. J. Geophys. Res., 115, C11017, https://doi.org/10.1029/2009JC005781.
Yu, L., X. Jin, and R. A. Weller, 2008: Multidecade global flux datasets from the Objectively Analyzed Air-Sea Fluxes (OAFlux) project: Latent and sensible heat fluxes, ocean evaporation, and related surface meteorological variables. Woods Hole Oceanographic Institute OAFlux Project Tech. Rep. OA-2008-01, 64 pp.
Zweng, M. M., and Coauthors, 2013: Salinity. Vol. 2, World Ocean Atlas 2013, NOAA Atlas NESDIS 74, 39 pp., https://data.nodc.noaa.gov/woa/WOA13/DOC/woa13_vol2.pdf.