Impacts of Indian Ocean SST biases on the Indian Monsoon: as simulated in a global coupled model
In this study, the impact of the ocean–atmosphere coupling on the atmospheric mean state over the Indian Ocean and the Indian Summer Monsoon (ISM) is examined in the framework of the SINTEX-F2 coupled model through forced and coupled control simulations and several sensitivity coupled experiments. During boreal winter and spring, most of the Indian Ocean biases are common in forced and coupled simulations, suggesting that the errors originate from the atmospheric model, especially a dry islands bias in the Maritime Continent. During boreal summer, the air-sea coupling decreases the ISM rainfall over South India and the monsoon strength to realistic amplitude, but at the expense of important degradations of the rainfall and Sea Surface Temperature (SST) mean states in the Indian Ocean. Strong SST biases of opposite sign are observed over the western (WIO) and eastern (EIO) tropical Indian Ocean. Rainfall amounts over the ocean (land) are systematically higher (lower) in the northern hemisphere and the south equatorial Indian Ocean rainfall band is missing in the control coupled simulation. During boreal fall, positive dipole-like errors emerge in the mean state of the coupled model, with warm and wet (cold and dry) biases in the WIO (EIO), suggesting again a significant impact of the SST errors. The exact contributions and the distinct roles of these SST errors in the seasonal mean atmospheric state of the coupled model have been further assessed with two sensitivity coupled experiments, in which the SST biases are replaced by observed climatology either in the WIO (warm bias) or EIO (cold bias). The correction of the WIO warm bias leads to a global decrease of rainfall in the monsoon region, which confirms that the WIO is an important source of moisture for the ISM. On the other hand, the correction of the EIO cold bias leads to a global improvement of precipitation and circulation mean state during summer and fall. Nevertheless, all these improvements due to SST corrections seem drastically limited by the atmosphere intrinsic biases, including prominently the unimodal oceanic position of the ITCZ (Inter Tropical Convergence Zone) during summer and the enhanced westward wind stress along the equator during fall.
KeywordsIndian Monsoon Coupled climate model Model systematic errors Ocean atmosphere interaction Arabian Sea
The Asian Summer Monsoon is one of the most dominant tropical atmospheric circulations, and the economies and livelihood of the populations of India and Southeast Asia depend heavily on its rainfall (Wang 2006). Because of the dynamically interactive nature of the tropical Indo-Pacific ocean–atmosphere system, one of the best tools to study Indian Summer Monsoon (ISM) variability is a global Coupled General Circulation Model (CGCM). In order to provide reliable seasonal predictions and climate projections of monsoon rainfall, it is nevertheless essential that CGCMs are able to produce a reasonable simulation of the mean ISM circulation and rainfall distribution. Unfortunately, coupled climate modeling is still an area under rapid development, and CGCMs are still at a relatively early stage with most of the CGCMs exhibiting pervasive problems and deficiencies (Shukla et al. 2009). As an illustration, simulation of ISM and its variability still remains a significant challenge for many state-of-the-art CGCMs (Annamalai et al. 2007; Kripalani et al. 2007; Terray et al., 2005, 2012; Bollasina and Nigam 2009; Levine and Turner 2012). Out of the 22 CGCMs submitted to the World Climate Research Program’s (WCRP) Coupled Model Intercomparison Project (CMIP) phase 3, Annamalai et al. (2007) and Kripalani et al. (2007) found, respectively, only six and seven models with a realistic ISM rainfall climatology. The large spread and difficulties of the CMIP3 CGCMs in simulating even the mean ISM rainfall during the 20th century add further doubts about the quality of the ISM rainfall projections by current CGCMs.
Even though current CGCMs simulate a substantially more realistic distribution of ISM rainfall compared to atmosphere-only models due to the importance of air–sea coupling in the monsoon variability (Wang et al. 2004, 2005, 2008; Kumar et al. 2005; Wu and Kirtman 2005), a major limiting factor for current CGCMs comes from model deficiencies in capturing the Sea Surface Temperature (SST) over the Indian Ocean, especially during boreal summer and fall. There is no doubt that SST variations over the Indian Ocean have significant impacts on ISM circulation and rainfall (Izumo et al. 2008; Annamalai 2010; Boschat et al. 2011, 2012), even though this role remains controversial (Annamalai and Murtugudde 2004). Therefore, important SST biases in coupled models drastically limit our understanding of the physical processes involved in the climate fluctuations, especially those associated with the ISM and Indian Ocean Dipole (IOD) (Bollasina and Nigam 2009; Fischer et al. 2005; Terray et al. 2012; Levine and Turner 2012).
Using the SINTEX-F2 coupled model, Joseph et al. (2012) hypothesized that the poor representation and weakness of ISM in this particular CGCM are due to a warm bias over the tropical Indian Ocean and the inherent weak meridional temperature gradient in the Indian region that drives the large scale monsoon flow (Chung and Ramanathan 2006). On the other hand, using the HadGEM3 CGCM, Levine and Turner (2012) suggest that cold SST biases over the Arabian Sea significantly reduce the ISM rainfall and circulation. This cold SST bias in the western Indian Ocean is pervasive in current CGCMs and may also drastically affect the simulated monsoon. The relationship between concurrent tropical Indian Ocean SSTs and ISM is thus controversial both in observations and current coupled simulations, partly due to the competing effects of SST on the evaporation and the meridional temperature gradient in the Indian region (Chung and Ramanathan 2006). These results warrant the need for improved monsoon and Indian Ocean SST simulations with current CGCMs.
These earlier studies provide the motivation for examining the roles of Indian Ocean SST biases on the ISM in a state-of-the-art CGCM, the SINTEX-F2 coupled model (Masson et al. 2012, Terray et al. 2012). As a first step toward the accurate coupled simulation of monsoon rainfall and Indian Ocean SST annual cycle, we have therefore examined the ability of this model to simulate the monsoon climate in both coupled and atmosphere-only configurations. Moreover, the present study includes the results of a group of sensitivity experiments with the CGCM, in order to unravel the specific roles of the Indian Ocean SST errors in the ISM rainfall and circulation simulation.
The paper is organized as follows. The model, the sensitivity experiments designed to study the specific role of SST biases in the western and eastern tropical Indian Ocean in the coupled simulation and the validation datasets used in this study are described in Sect. 2. In Sect. 3, we present the performance of the new version of the coupled model in simulating the monsoon climate in atmosphere-only and coupled configurations with a special emphasis on ISM and IOD. In Sect. 4, the results of the sensitivity experiments are analysed. The final section summarizes and discusses the main results of the present work.
2 Model and data description
2.1 SINTEX-F2 model
We have used the standard configuration of SINTEX-F2 model (Masson et al. 2012). It is the upgraded version of SINTEX-F1 CGCM (Guilyardi et al. 2003; Gualdi et al. 2003a; Luo et al. 2003, 2005). The oceanic component is NEMO (Madec 2008; Madec et al. 1998), using the ORCA05 horizontal resolution (0.5°), 31 vertical levels and including the LIM2 ice model (Timmermann et al. 2005). The atmospheric component is ECHAM 5.3 (Roeckner et al. 2003, 2004) with the T106 (1.125°) horizontal resolution and 31 hybrid sigma-pressure levels. A mass flux scheme (Tiedtke 1989) is applied for cumulus convection with modifications for penetrative convection according to Nordeng (1994). The coupling information, without any flux corrections, is exchanged every 2 h by means of the OASIS 3 coupler (Valcke 2006). See Masson et al. (2012) and Terray et al. (2012) for more details.
We run a 110 year control experiment (named CTL hereafter) with the coupled configuration of SINTEX-F2. At the same time, we have run an AGCM experiment (named FOR hereafter) with the atmospheric-only configuration of SINTEX-F2, forced by Advanced Very High Resolution Radiometer (AVHRR) daily SST from 1982 to 2010, in order to assess the impact of the ocean–atmosphere coupling on the simulated Indo-Pacific climatology.
2.2 Design of the sensitivity experiments
For the FTW experiment, the SST damping is applied toward a daily climatology computed from the, AVHRR only, daily Optimum Interpolation SST (OISST) version 2 dataset for the 1982–2010 period (Reynolds et al. 2007). In the FTE experiment, we added a +0.5 °C offset value to this daily SST climatology. The choice to add a constant 0.5° offset factor comes from the uniform differences between the AVHRR OISST climatology and Tropical rainfall measuring mission Microwave Imager (TMI) version 4 3-day mean climatology (see http://www.ssmi.com/tmi/tmi_description.html, Wentz et al. 2000). This difference could be explained by the inability of AVHRR’s satellite to see through clouds in this active convection area or to some coastal “pollution” in TMI data. Being doubtful about the exact reason of this difference, we choose to keep IOSST data for our experiments in order to have a longer time series for computing the SST climatology, but add to it a 0.5 °C constant factor in order to maximize the perturbation introduced in the FTE experiment.
Summary of all experiments
Type of experiment
No correction applied
No correction applied
Western Indian Ocean (WIO):
10°S–30°N (see Fig. 1)
Eastern Indian Ocean (EIO): (see Fig. 1)
AVHRR + 0.5 °C
2.3 Reference datasets
For rainfall comparison between observations and the model outputs, we used the Tropical Rainfall Measuring Mission (TRMM) observations, specifically the 0.25° by 0.25° horizontal resolution merged 3B43 dataset, which is available from 1998 to 2010 (Kummerow et al. 2001; Huffman et al. 1997). For wind, wind stress and atmospheric temperature, we used the ERA-Interim reanalysis from 1989 to 2009, the latest global atmospheric reanalysis produced by the European Centre for Medium-Range Weather Forecasts (Dee et al. 2011). For SST, we used the AVHRR infrared satellite SST product from 1982 to 2010 (Reynolds et al. 2007). We also used TRMM Microwave Imager (TMI) from 1998 to 2008 (Wentz et al. 2000). This product has a much better spatiotemporal sampling of observations over the cloudy areas of the tropical Indian Ocean, especially the eastern tropical Indian Ocean, than older products, because this sensor is nearly free of cloud interferences, but this product has a shorter time series than AVHRR. Finally, the Simple Ocean Data Assimilation (SODA) version 2.2.4 is used for diagnosing the 20° isotherm depth (used as a proxy for the thermocline depth) in the Indian Ocean (Carton and Giese 2008). SODA 2.2.4 covers the long period from 1871 to 2008 and uses winds from the new 20Crv2 atmospheric reanalysis (Compo et al. 2006), but we used only data from 1978 to 2008.
3 The monsoon annual cycle in observations, atmosphere-only and coupled runs
In this section, we compare the performances of the atmosphere-only (FOR) and coupled (CTL) configurations of SINTEX-F2 in simulating the annual cycle in the Indian areas. The focus here is to disentangle the biases, which are due specifically to the atmospheric model from those due to the ocean–atmosphere coupling in the coupled simulation of the monsoon cycle. More precisely, our aim is to identify the ≪atmospheric model errors≫, which are the errors in the atmospheric simulated fields, which can be detected in both the FOR and CTL simulations.
3.1 Boreal winter
The coupled and forced experiments are successful in locating the mean ITCZ position during boreal winter (Fig. 2b, c). However, a striking feature is that both experiments share many errors. First, a dry bias is found over the islands of the Maritime Continent and, to a lesser extent, over Madagascar and North Australia in both experiments (Fig. 2e, f). Furthermore, both the coupled and forced runs show a wet bias over oceanic regions of the Maritime Continent, with unrealistically sharp land–sea contrasts. This suggests a significant underestimation of convective anomalies over the Maritime Continent, probably related to a misrepresentation of the strong diurnal cycle of convection over land in this area (related to small-scale phenomena like the land-sea breeze) and, also, to the very crude land model used in ECHAM (Alessandri et al. 2007; Neale and Slingo 2003). This large decrease of rainfall over the land of the Maritime Continent is associated with a pair of low-level anticyclonic anomalies straddling the equator in the central Indian Ocean and a consistently too weak westerly flow on and south of the equator during December-February (Fig. 2e, f). All these features can be attributed to Rossby wave responses to decreased diabatic heating over the Maritime Continent associated with the dry bias (Gill 1980). Another striking similarity between the two runs is a decreased low-level northeasterly flow over the Arabian Sea, suggesting a too weak winter monsoon. This is again consistent with a Gill atmospheric response to the decrease of precipitation over the Maritime Continent.
Focusing now on the differences between CTL and FOR, we note that the Indian Ocean ITCZ is too zonal and more active in CTL, especially over the western Indian Ocean and between Australia and Java (Fig. 2d, e). On the other hand, in FOR, a reduction of rainfall is seen over the eastern Indian Ocean with an opposite excess of precipitation from the African coast to the central south Indian Ocean (Fig. 2d).
These distinct rainfall patterns, over the South West Indian Ocean, in FOR and CTL simulations, are consistent with the simulated wind pattern (Fig. 2e, f). As an illustration, the erroneous low-level wind pattern simulated in the southern tropical Indian Ocean (induced by the heating errors over the Maritime Continent) implies that the minimum of the wind stress curl is significantly shifted eastward in the simulations compared to the observations (not shown). Consistent with the work of Yokoi et al. (2008), this leads to an eastward shift in the location of the thermocline dome in the South tropical Indian Ocean in the coupled run (Fig. 2h, i). Finally, this feature may, in turn, explain the warm SST bias over the southwest tropical Indian Ocean and provides an explanation for the excessive precipitation in the western Indian Ocean during this season in CTL (Fig. 2d, i).
Over the far south-eastern Indian Ocean, Fig. 2i, f show an excess of precipitation and a warmer than observed SST. This suggests a driving role of the ocean in this region. Koch-Larrouy et al. (2007) show that the misrepresentation of the tidal mixing in the OPA8.2 model is partly responsible of the warm SST bias in this region. The same mechanism could explain the warm bias in the south-eastern Indian Ocean, which, by an error compensation, attenuates the dry bias observed in this region in FOR.
To conclude about this season, the common errors in CTL and FOR suggest that most of the rainfall and circulation biases in the coupled run are driven by errors in the atmospheric model, especially over the Maritime Continent. The warmer SST in the south-eastern Indian Ocean in CTL (Fig. 2i) corrects the dry bias observed in FOR over this region thanks to an error compensation. However, the coupling also increases some biases: the erroneous low-level wind pattern in the south tropical Indian Ocean causes a warmer SST in the south-western Indian Ocean consistent with a stronger excess of precipitation over this region (Fig. 2i).
3.2 Boreal spring
One hypothesis is that the delayed ISM onset in the coupled run is mostly attributable to the SST, thermocline, rainfall and low-level wind errors simulated in the western Indian Ocean during the pre-onset phase, which, in turn, are associated with the atmospheric circulation errors described for the preceding season (Joseph et al. 2003; Annamalai et al. 2005; Sijikumar and Rajeev 2012). This hypothesis will be tested more thoroughly with the help of dedicated sensitivity experiments in the next section.
To conclude about this transition season, many of the simulated rainfall and low-level wind errors may be again attributable to errors in the atmospheric model, especially the misrepresentation of convection over Maritime Continent islands and the associated Gill response to the lack of precipitation over this region.
3.3 Boreal summer
During the boreal summer (June–August), the pair-wise differences between observations, CTL and FOR are significantly enhanced (Fig. 4d, f) suggesting that, during ISM, the biases found in CTL are not solely due to errors in the atmospheric model. The dry islands bias in the Maritime Continent is still found during boreal summer, but is no longer the main rainfall errors in both simulations. This reinforces the hypothesis of a driving role of SST errors or the amplification of atmospheric errors by positive ocean–atmosphere feedbacks during this season.
Concerning the low level winds, both FOR and CTL are able to simulate a reasonable monsoon circulation, including a realistic monsoon trough (Fig. 5b, c). However, Fig. 5e, f demonstrate that the Somali jet is not confined near the African coast and the southern trade winds cross the equator from 40° to 100°E in both simulations. This bias is consistent with excessive/less precipitation just north/south of the equator in both simulations (Fig. 5e, f). This common error is however more pronounced in CTL than FOR (Fig. 5d), in addition, the ISM circulation is weaker in CTL (see also Fig. 4a), which is consistent with the lack of precipitation over central and north India in CTL compared to FOR (Figs. 4, 5d). The weaker monsoon strength in CTL, in response to weakened diabatic heating in the monsoon trough, is a clear improvement of mean wind circulation (see Fig. 4a) and highlights the impacts of the ocean–atmosphere coupling on the simulation of ISM (Wang et al. 2005; Kumar et al. 2005; Cherchi and Navarra 2007).
Since the coupling induces important changes in the simulated atmospheric and rainfall mean state, it is interesting to investigate the possible links between Indian Ocean SST biases and ISM circulation in the coupled simulation (Fig. 5g, i). The clear dipole structure of SST biases with a warm bias in the Western Indian Ocean (WIO, see Fig. 1) and a cold bias in the Eastern Indian Ocean (EIO, see Fig. 1) drives us to analyse these two regions separately.
In the WIO, the part of the warm SST bias located south of the Equator is consistent with the SST errors noted during the previous seasons and can be explained by the same mechanisms. North of the Equator, the poor representation of the Somalia upwelling (Schott et al. 1997; Schott and McCreary 2001) is very likely the cause of the warm SST bias, as documented in previous versions of SINTEX (Gualdi et al. 2003b). The reasons of the poor representation of the upwelling are not clear, the low resolution of the oceanic component could explain a part of the error, but the inability of the CTL run to reproduce a correct wind stress profile along the coast, especially the tangent component of the wind stress, is also a possible reason. Furthermore, the delayed ISM onset in CTL could also contribute to the reinforcement of this warm bias in the whole WIO during the first part of ISM. The SST bias in WIO appears to dominate the SST east–west gradient error in the equatorial Indian Ocean. In this respect, several recent studies have suggested a possible impact of SST in the western Arabian Sea on precipitation over India (Izumo et al. 2008; Levine and Turner 2012; Joseph et al. 2012). Izumo et al. (2008) and Levine and Turner (2012) show that cold (warm) SST biases in the WIO could decrease (enhance) monsoon rainfall in their coupled model. On the other hand, Joseph et al. (2012) suggest that the warm SST biases in the WIO could explain the weakness of the monsoon, north of 20°N, in SINTEX-F2, by setting up an erroneous meridional tropospheric temperature gradient in the Indian region (Chung and Ramanathan 2006). In other words, it is rather difficult to relate excessive precipitation over the Indian Ocean (just north of the equator) and the lack of precipitation over north India during ISM with the WIO SST errors in CTL. This question will be investigated in Sect. 4, with the help of the FTW experiment.
Focusing now on the EIO, the cold SST bias is clearly consistent with the lack of precipitation over this region in CTL and has also been a long-standing problem in the SINTEX coupled model (see Fig. 5i; also Fischer et al. 2005; Terray et al. 2012). But, again, it is not obvious to relate this SST bias with the precipitation and wind patterns observed to the north of the equator in the coupled model. Annamalai (2010) shows with a forced linear baroclinic model that the equatorial EIO SST could have an important impact on precipitations over north of India and Bay of Bengal. We will also go back to this problem with the FTE experiment in Sect. 4.
In summary, FOR and CTL share many biases mainly linked to an incorrect position of the ITCZ during boreal summer. On the other hand, the coupling induces important changes with a significant reduction of the ISM strength, but also a severe deficit of precipitation off Sumatra and Java and excessive rainfall over the WIO. The possible impact of the SST errors on the ISM rainfall and circulation in the Northern Hemisphere is not clear in CTL and will be addressed in the following section.
3.4 Boreal fall
In FOR, the southward shift of the ITCZ is significantly delayed as illustrated by the positive (negative) rainfall biases in the Bay of Bengal and China Sea (in the southeastern Indian Ocean and Maritime Continent) in Fig. 6f. This is consistent with the persistence of the ISM circulation in the northern hemisphere (Fig. 6f) and the positive values of the IMDI in September–October (Fig. 4a) in this simulation. Consistent with delayed southward migration of the ITCZ and the dry islands bias in the Maritime Continent, FOR is affected by a very strong easterly wind bias over the equatorial Indian Ocean during boreal fall.
The importance of IOD-like air-sea interactions during this season hints an important and retroactive forcing of SST onto the atmospheric circulation in CTL. Figure 7 shows that the enhancement of the warm bias in the WIO, in the early summer, occurs simultaneously with the degradation of the equatorial wind stress in CTL and is closely followed by the formation of the cold bias in the EIO. This suggests that the WIO SST bias may be responsible of the emergence of the EIO SST bias during boreal summer and fall, for example via equatorial wind stress forcing, as in some positive dipole events (during El Niño years) in which the warming of the WIO is the major feature and governs the evolution of the event (Loschnigg et al. 2003; Drbohlav et al. 2007; Boschat et al. 2012). We will also address this question in more details in the next section.
To conclude this section, the comparison of seasonal cycles in FOR, CTL and observations suggests that, during boreal winter and spring the main source of errors is the atmospheric model, including prominently land surface and coastal processes (e.g. land-sea breeze) over the islands of the Maritime Continent. During summer and fall, the SST biases seem to have a more significant impact on the oceanic and atmospheric circulation errors. We will, in the next section, investigate more thoroughly these impacts with two dedicated sensitivity coupled experiments.
4 Sensitivity experiments
What is the exact contribution of WIO and EIO SST biases to the various deficient features affecting the ISM rainfall and circulation, including the delayed ISM onset and the wrong position of the ITCZ, during boreal summer in the CGCM?
What are the exact contributions of WIO and EIO SST biases in the emergence of the positive dipole-like structure in the mean-state of the CGCM during boreal fall?
To simplify the presentation of the results, we will first discuss the results related to the ISM rainfall and circulation and then address the problem of the emergence of the dipole-like structure in the boreal fall mean-state of the CGCM.
4.1 Impact of SST biases on ISM onset, rainfall and circulation
4.1.1 ISM onset
The monsoon onset is often represented by various indices defined from rainfall as well as dynamical parameters. Following the work of Xavier et al. (2007), we use a thermodynamical index based on the tropospheric temperature gradient (hereafter referred as Tropospheric Temperature Gradient (TTG) Index) to find out the onset of monsoon season. More precisely, the TTG index is defined by the difference in the tropospheric temperature (TT; defined as the temperature averaged between 600 and 200 hPa) between a northern box (40°–100°E; 5°–35°N) and a southern box (40°–100°E; 15°S–5°N). The onset of the monsoon is then defined when the value of TTG becomes positive from negative. The monsoon onsets in Julian days are 148, 142 and 158 in observations (ERA interim), FOR and CTL simulations, respectively. In other words, the monsoon onset is significantly delayed in CTL compared to observations and FOR, as suggested in Sect. 3.
4.1.2 ISM rainfall pattern and mean position of the ITCZ
We now focus on the ISM rainfall distribution and circulation errors during boreal summer in CTL and their possible relationships with the warm (cold) SST bias in the WIO (EIO). Figure 8b illustrates again the inability of the model, both in FOR and CTL, to reproduce the two observed locations of convection in the Indian region during boreal summer—the first one over the Indian subcontinent and the Bay of Bengal, the second over the eastern equatorial Indian Ocean (black curve in Fig 8b). Instead of these two precipitation maxima, all experiments simulate only one unique rainfall maximum, around 10°N in FOR and over 5°N in CTL and both sensitivity experiments.
Joseph et al. (2012) have suggested that the warm bias in the WIO is responsible of the biased TT gradient, which, in turn, is responsible of the wrong ITCZ position in the CTL simulation. In order to test this hypothesis, the TTG index, defined in Sect. 4.2.1 has been cumulated during the whole monsoon season, in order to measure the strength of ISM. The ISM strength remains constant in CTL (218 K) and FTW (219 K), which is still far from the observed value of 260 K. These results demonstrate that the WIO SST has also only a weak impact on the biased representation of the meridional TT gradient in the CGCM. Note that this result does not contradict the hypothesis that the overly weak meridional TT gradient in CTL is responsible of the wrong position of the ITCZ (Joseph et al. 2012).
However, Fig. 8a, b show that there is a significant impact of the correction of SST biases in both WIO and EIO regions on precipitation amplitude over Peninsular India and neighboring oceanic regions without changing the time evolution (e.g. Fig. 8a) or the shape of the meridional rainfall distribution (e.g. Fig. 8b) in the CGCM. In both sensitivity experiments, there is a significant decrease of precipitation in the northern hemisphere that is stronger in FTW than in FTE. In FTE, we observe also a reduction of the dry bias in the coupled run off Sumatra, around 5°S (see below).
4.1.3 Mechanisms in FTW
4.1.4 Mechanisms in FTE
Interestingly, Fig. 10a illustrates also a non-local response to the SST changes in the EIO with a significant decrease of precipitation north of the equator over the oceanic regions surrounding the Indian subcontinent (nearly one-third of the excess of ISM precipitation) and, to a much lesser extent, over India compared to CTL. Furthermore, these rainfall pattern modifications are associated with significant improvements of the inter-hemispheric monsoon flux in the central and eastern equatorial Indian Ocean, which are one of the major biases of the ISM circulation in CTL (see Fig. 10a and the previous section). Finally, these improvements extend and amplify during boreal fall with a partial correction of the rainfall and low-level wind components of the dipole-like pattern, seen in CTL (Fig. 10b).
The increased precipitation over the EIO in FTE is associated with more ascent over this region compensated by subsidence over the ocean regions surrounding peninsular India and an increased northward flux at 200 hPa (Fig. 10c, d). The non-local rainfall response in FTE is thus associated with a reorganization of the monsoon Hadley cell, induced solely by the changes of the EIO SSTs applied in FTE. This result, consistent with the works of Annamalai (2010) and Krishnan et al. (2000), highlights that the monsoon is partly thermally driven. However, it is important to keep in mind the key role played by the atmospheric biases (or alternatively the absence of coupling over the EIO in FTE) as the reduction of the EIO SSTs errors is (once again) not sufficient to restore the observed bimodal latitudinal structure of the monsoon rainfall (Fig. 8b).
To summarize our results about question (1), the corrections of the WIO and EIO SST biases do not impact the large-scale circulation errors in CTL, like the erroneous ITCZ location over the ocean near 5°N, nor the delayed ISM onset. The WIO area seems to be an important source of moisture for the ISM. The SST decrease in this region corrects an important part of the excess of precipitation over the ocean, but gives birth to new biases such as a lack of precipitation over south India and a too weak monsoon flux. On the other hand, the correction of the much smaller, but seasonally evolving cold SST bias in the EIO consistently restores a reasonable amount of precipitation in the EIO region, reduces significantly the excess of precipitation over the ocean in the northern hemisphere and improves the equatorial low-level wind mean state. All these changes are interrelated through a modulation of the Hadley cell during boreal summer and fall, but the amplitude of the improvements seems to be limited by the atmospheric biases, the absence of coupling over the EIO (in the FTE experiment) or SST errors in other key oceanic regions for the monsoon (for a comprehensive review, see Wang 2006 or Gimeno et al. 2010).
4.2 Formation of dipole-like SST structure and associated circulation errors
The similarity between the seasonal evolution of the WIO and EIO SST biases in CTL (Fig. 7) and the formation of a positive IOD event (Chang et al. 2006; Drbohlav et al. 2007, their Fig. 4), suggests a comparable formation mechanism.
In a similar fashion, there is no improvement of the WIO SST warm bias during both summer and autumn in FTE (Fig. 11a). But, Figs. 10b and 11c do show a significant atmospheric response to the EIO SST bias with a more realistic equatorial wind stress in FTE than in CTL. This in turn leads to an oceanic adjustment response, with a typical zonal dipole pattern in thermocline depth (deepening in the east, shallowing in the west), which slightly corrects the biased thermocline in CTL (see Fig. 5i). However, FTE equatorial wind stress is still affected by an easterly bias and remains westward instead of the strong eastward maximum observed during boreal fall (Fig 11c). Despite a slight improvement, the thermocline remains too deep in the west (not shown). This could explain why there is no significant improvement of the WIO warm bias in FTE (Fig. 11a). Interestingly, Fig. 11c illustrates that FOR is also affected by a similar westward equatorial wind stress bias during boreal fall. This suggests that, even during boreal fall, the atmospheric biases may control the evolution of the equatorial wind stress, and the associated biased thermocline in CTL. Further sensitivity experiments with more moderate flux adjustments (e.g. allowing some SST interannual and intraseasonal variability in the constrained region) are required to further confirm this hypothesis.
To conclude this section, the corrections of the SST WIO and EIO biases in the sensitivity experiments lead to improvements of the atmospheric circulation and rainfall distribution during boreal summer and fall, but without changing sufficiently the large-scale pattern of the biased atmospheric circulation or erroneous rainfall pattern in CTL. In other words, these SST biases seem only to induce a reinforcement of circulation biases already existing in FOR in most of the cases. Furthermore, all the improvements due to the SST corrections seem drastically limited by the atmosphere intrinsic biases, including prominently the unimodal oceanic position of the ITCZ and the enhanced westward wind stress along the equator. The correction of the SST in the WIO/EIO does not significantly impact the opposite side of the basin. This result contradicts the theory, suggested previously, that the formation of SST biases is due to a coupled mechanism, similar to the one occurring during the growth of a positive IOD event. On the other hand, the inability of the atmospheric model forced by the observed SSTs to reproduce an eastward wind stress along the equator during fall could explain the formation of these biases. Another hypothesis is that these biases are due to local coupled processes.
5 Conclusion and discussion
In this work, we therefore examine the impacts of SST biases on the ISM rainfall and the Indian Ocean mean state, in the SINTEX-F2 model, with the help of a comparison of forced and coupled control simulations and two sensitivity coupled experiments.
During boreal spring and winter, the coupled and forced simulations share many deficiencies. These common features suggest that, during boreal winter and spring, the main source of errors is coming from the atmospheric model, especially a dry islands bias in the Maritime Continent. The key-role of the Maritime Continent on atmospheric model biases was already highlighted by Neale and Slingo (2003). Conversely, during boreal summer and autumn, FOR and CTL exhibit significant mean state differences, suggesting a significant impact of the SST errors on the mean state of the coupled model. Therefore, in order to highlight the role of the Indian Ocean SST biases on the rainfall and atmospheric circulation, as well as the mutual interactions between these SST biases, two coupled sensitivity experiments have been designed, in which the SST biases are corrected either in the warm western (FTW) or cold eastern (FTE) Indian Ocean. Surprisingly and in spite of the correction of the SST biases, the large-scale pattern of the biased atmospheric circulation and rainfall distribution found in CTL is still evident in the two sensitivity experiments. Moreover, the emergence of these SST biases seem only to induce a reinforcement of circulation biases already existing in the atmosphere-only run, which include prominently a too southward and unimodal position of the ITCZ during boreal summer and a delayed southward migration of this ITCZ during boreal fall. This delayed transition of the ITCZ seems to play a key-role in the enhanced westward wind stress simulated along the equatorial Indian Ocean in all the experiments. These common features suggest again that deficiencies in the atmospheric model (e.g. convection) or missing land-vegetation processes may be responsible for these errors. Alessandri et al. (2007) showed a strong improvement in the monsoon representation, especially over India, when replacing, in the previous version of ECHAM (e.g. ECHAM4), the simple surface scheme by a land surface model. It remains, however, to be seen if such modifications will correct the above coupled model deficiencies.
Nevertheless, FTW experiment shows that the WIO is an important source of moisture for the ISM. The decrease (increase) of WIO SSTs leads to a global decrease (increase) of the ISM rainfall and strength consistent with the conclusions of Levine and Turner (2012), Izumo et al. (2008) and Gimeno et al. (2010). This modification is strongest over the southeastern Arabian Sea where it corrects a large part (and sometime even more) of the excess of oceanic precipitation in the CGCM. Meanwhile, it also decreases precipitation over India and monsoon strength, which was already slightly underestimated in the CGCM. Thus, the WIO SSTs seem to impact ISM via a “local” or regional evaporation effect rather than the modulation of the meridional temperature in the Indian region (Chung and Ramanathan 2006).
On the other hand, the correction of the cold SST bias in the EIO leads to a global improvement of the precipitation mean state via modulations of both the local Hadley and Walker circulations during boreal summer and fall. Despite the relatively small size of this area, it plays an important role on the Indian Ocean circulation, consistent with the work of Annamalai (2010). This cold EIO bias is common in many CGCMs (Lin 2007) and similar experiments with other coupled models could be interesting to confirm this result.
During boreal fall, both sensitivity experiments show that the correction of one SST bias does not have any significant impact on the other. This apparent de-connection could be explained by the inability of the atmospheric model to reproduce realistic equatorial winds along the equator during all the year. However, these results also suggest that the formation of these SST biases is mainly due to an oceanic local response to the atmospheric biases or to local coupled processes and not to an IOD-like mechanism in the framework of the SINTEX-F2 coupled model (Li et al. 2003; Spencer et al. 2005).
Nevertheless, several questions remain unanswered. First, the ISM onset date is significantly delayed in all coupled simulations, including the sensitivity experiments, whereas it is quite correct in the atmospheric run. This result is in contradiction with the study of Joseph et al. (2003) and Sijikumar and Rajeev (2012), but in accordance with Soman and Slingo (1997). In this last study, they suggested that the onset is mainly control by the West Pacific SSTs. Joseph et al. (1994), Boschat et al. (2011) and Levine and Turner (2012) have also suggested a remote impact of the Pacific SSTs on the onset. This problem will be investigated in a future work.
Second, the interannual variability has not been examined here. In both sensitivity experiments, the interannual variability has been suppressed in the corrected area. This could lead to important changes in the Indian and Pacific Ocean interannual variability. As an illustration, the sensitivity experiments could provide us some answers about the formation of the IOD and how Indian Ocean variability could impact ENSO variability (Luo et al. 2010, Izumo et al. 2010).
A part of this research was supported by the Japan Science and Technology Agency/Japan International Cooperation Agency through the Science and Technology Research Partnership for Sustainable Development (SATREPS). This work was performed using HPC resources from GENCI-IDRIS (Grant 2012-x2012016895).
- Dee DP et al (2011) The ERA-interim reanalysis: configuration and performance of the data assimilation system. Q J R Meteorol Soc 137:656, 553–597, Part AGoogle Scholar
- Gualdi S, Guilyardi E, Navarra A, Masina S, Delecluse P (2003a) The interannual variability in the tropical Indian Ocean as simulated by a CGCM. Clim Dyn 20:567–582Google Scholar
- Gualdi S, Navarra A, Guilyardi E, Delecluse P (2003b) Assessment of the tropical Indo-Pacific climate in the SINTEX CGCM. Ann Geophys 46:1–26Google Scholar
- Joseph PV, Sooraj KP, Rajan CK (2003) Conditions leading to monsoon onset over Kerala and the associated Hadley cell. Mausam 54(1):155–164Google Scholar
- Madec G (2008) NEMO ocean engine. Note du Pôle de modélisation, Institut Pierre-Simon Laplace (IPSL), France. No 27. ISSN No 1288-1619Google Scholar
- Madec G, Delecluse P, Imbard M, Levy C (1998) OPA 8.1 ocean general circulation model reference manual. Note du Pôle de modélisation, Institut Pierre-Simon Laplace (IPSL), France. No 11, 91 pp. Available at http://www.lodyc.jussieu.fr/NEMO/general/manual/Doc_OPA8.1.pdf
- Masson S, Luo JJ, Madec G, Vialard J, Durand F, Gualdi S, Guilyardi E, Behera S, Delecluse P, et al (2005) Impact of barrier layer on winter-spring variability of the southeastern Arabian Sea. Geophys Res Lett 32: L07703, 1–4Google Scholar
- Nordeng TE (1994) Extended versions of the convective parameterization scheme at ECMWF and their impact on the mean and transient activity of the model in the tropics. ECMWF Research Dept. Tech. Memo. 206, European Centre for Medium-Range Weather Forecasts, Reading, United Kingdom, 41 ppGoogle Scholar
- Roeckner E, Alfred S, Amala Y (1996) The atmospheric general circulation model ECHAM-4: model description and simulation of present-day climate. Max-Planck-Institut für Meteorologie. Report no. 218, 94 pp, HamburgGoogle Scholar
- Roeckner E, Baüml G, Bonaventura L, Brokopf R, Esch M, Giorgetta M, Hagemann S et al (2003) The atmospheric general circulation model ECHAM5: Part 1: model description. Max-Planck-Institut für Meteorologie, MPI-Report 353, HamburgGoogle Scholar
- Roeckner E, Brokopf R, Esch M, Giorgetta M, Hagemann S, Kornblueh L, Manzini E, Schlese U, Schulzweida U (2004) The atmospheric general circulation model ECHAM5 Part II: sensitivity of simulated climate to horizontal and vertical resolution. Max-Planck-Institute for Meteorology, MPI-Report 354, HamburgGoogle Scholar
- Valcke S (2006) OASIS3 user guide (prism_2-5). PRISM support initiative report No 3, 64 ppGoogle Scholar
- Wang B (2006) The Asian Monsoon. Springer/Praxis Publishing, New York, p 787Google Scholar