Quantitative Analysis of the Feedback Induced by the Freshwater Flux in the Tropical Pacific Using CMIP5

Freshwater exchange at the atmosphere-ocean interface is one of the most important forcing conditions for both oceanic general circulation and the climate system. In particular, the freshwater flux (FWF), i.e., precipitation minus evaporation, is a key factor in controlling the salinity distribution and modulating thermohaline circulation (Huang et al., 2005). Forced by atmospheric fields, the oceanic changes can produce feedback to the atmosphere in which the important oceanic field affected is sea surface temperature (SST). In addition, different from ocean heat flux forcing, FWF forcing acts to drive a change in SST whereas the heat flux represents a passive response to SST change (Zhang and Busalacchi, 2009). The significance and implications of these relationships need to be investigated more thoroughly to understand the physical characteristics and to improve model simulations of the interannual variability in the tropical Pacific, particularly that associated with ENSO and its diversity (Zhang et al., 2013; Zheng et al., 2014).

Previous research has mainly focused on the effects of atmospheric forcing components of surface heat flux and winds in the tropical Pacific (e.g., Bjerknes, 1969; Meehl et al., 2001; Yu and Boer, 2002; Kim et al., 2007; Yu and Weller, 2007; Zhang et al., 2014). Recently, however, FWF forcing and its related salinity effects on climate variability have attracted great attention, and there has been significant progress in understanding the physical characteristics and modeling the roles of FWF forcing in climate variability (e.g., Zhang and Busalacchi, 2009; Zhang et al., 2010; Hackert et al., 2011; Ham et al., 2012; Wu et al., 2010; Zheng and Zhang, 2012; Zheng et al., 2014). These studies have demonstrated that FWF and its related salinity fields first affect the oceanic density gradients and stratification, and then have the ability to maintain the mean climate and its variability (e.g., Maes, 2000; Lagerloef, 2002; Curry et al., 2003; Boyer et al., 2005; Huang et al., 2005; Levitus et al., 2005; Ballabrera-Poy et al., 2007; Cravatte et al., 2009; Collins et al., 2010). For instance, sensitivity analyses following an introduction of FWF perturbation have revealed obvious effects on SSTs (Levitus, 1989; Delcroix and Hénin, 1991; Manabe and Stouffer, 1995; Béthoux et al., 1998; Maes, 1998; Wong et al., 1999, 2001; Dickson et al., 2002; Jacobs et al., 2002; Fedorov et al., 2004; Huang et al., 2005; Huang and Mehta, 2005; Ma et al., 2013). In considering the coupling between the ocean and atmosphere, some additional processes start to act because the FWF forcing-induced changes in SSTs can produce additional feedback to the atmosphere. Additionally, FWF affects the depth of the mixed layer through its direct contributions to the buoyancy flux (Q _B), which further causes the entrainment of subsurface water into the mixed layer.

The effect of FWF on ocean salinity and the transport of freshwater in the ocean are not thoroughly understood, and there are many unresolved questions remaining with regard to the tropical Pacific. For example, previous studies have tended to analyze the relationship of FWF feedback with related physical fields through diagnosing, instead of quantifying, the effects among these relationships. Additionally, model performances are strikingly different in representing the related feedbacks, with large uncertainty and biases. The released simulation results of phase 5 of the Coupled Model Intercomparison Project (CMIP5) experiments (Taylor et al., 2012) have provided a good opportunity to further explore FWF feedback to climate variability. Preliminary analyses indicate that interannual variability of FWF in the tropical Pacific bears a close relationship with SST associated with the evolution of ENSO (Zhang et al., 2012; Zheng and Zhang, 2012). Thus, interannual variabilities of FWF display a nonlocal positive relationship with SST during ENSO evolution. Accordingly, we can evaluate the FWF modulation of SST in coupled models to not only understand interannual anomalous signal mechanisms, but also to analyze FWF bias in the interannual variability simulated by the CMIP5 multi-model through assessing the feedback in the tropical Pacific. These CMIP5-based analyses support the view that FWF forcing and its related feedback should receive adequate attention due to its strong interannual anomalies related to ENSO.

The purpose of this study is to evaluate CMIP5 multi-model simulation performance with regard to the interannual variability of oceanic physical fields in the equatorial Pacific. In addition, FWF feedback to SST in the tropical Pacific is explored and FWF feedback to SST and related oceanic fields is quantified. A description of the multi-model, the observational and reanalysis data, and the definition and calculation used for Q _B are presented in section 2. In section 3, the positions and intensities in the interannual spatial distribution are compared. In section 4, multi-model simulations and observations of physical fields are compared, and the difference in the correlation between FWF and SST within the

CMIP5 multi-model is explained by analyzing oceanic processes. The linear relationship induced for FWF and SST is qualitatively analyzed based on the differences within the CMIP5 models and observation in section 5. A summary and conclusion are provided in section 6.

Data from the CMIP5 multi-model archive are used, which are available in the Program for Climate Model Diagnosis and Inter-comparison (PCMDI) Earth System Grid (ESG) (http://pcmdi3.llnl.gov/esgcet/). In the current study, we perform analyses using the 23 models listed in Table 1. We make use of Pre-industrial control (Pi-control) scenario simulations (Taylor et al., 2012) to analyze the following variables in the tropical Pacific: SST, precipitation, evaporation, and SSS. The first ensemble member is used where multiple ensemble members are available for any model. All CMIP5 data are interpolated to a 1^°× 1^° resolution global grid to compare the data. In this paper, data from the last 100 years for the 23 Pi-control models are selected to focus on climate characteristics and variability. Unless stated otherwise, Pi-control simulations are compared in the period 1909-2008 (100 yr).

To compare with CMIP5 multi-model simulations, we use the following datasets. The observed precipitation data are from version 2 of the Global Precipitation Climatology Project (GPCP) dataset, covering 1979-2013, with a 2.5^°× 2.5^° horizontal resolution (Adler et al., 2003). Evaporation is derived from the Objectively Analyzed Air-Sea Fluxes (OAFlux; Yu and Weller, 2007). Net heat flux data of monthly and long-term climatology fields at the sea surface (latent and sensible heat fluxes and flux-related surface meteorology) are derived from the OAFlux data from 1958 to present (1^° gridded) and the surface radiation data from the International Satellite Cloud Climatology Project from 1983 to 2009 (resolution: 2.5^°) (Schiffer and Rossow, 1985). Monthly mixed layer depth (MLD) data and its climatological fields are directly available from the International Pacific Research Center/Asia-Pacific Data-Research Center Argo products, which cover the period from 2005 to 2013. SST data are from version 3b of the Extended Reconstructed Sea Surface Temperature (ERSST) dataset, which includes SST data from January 1854 to the present day (Smith et al., 2008). SSS data are from the quality-controlled subsurface ocean temperature and salinity data of the Met Office Hadley Centre observation datasets, which are available from 1950 to the present day and provide separate files for each month (Ingleby and Huddleston, 2007). Trends for all of the above data are removed to eliminate the response to global warming and the data are interpolated to a 1^°× 1^° resolution global grid.

Figure 1.  STD of FWF calculated from the results of the CMIP5 multi-model simulations and based on observations (right of the bottom panel) in the tropical Pacific. Units: mm d^-1.

The Q _B field, together with wind and heat flux, controls the formation of MLD, which affects the entrainment of subsurface cold water into the upper surface in the equatorial Pacific. The Q _B at the sea surface can be defined as (Zhang et al., 2010) \begin{equation} \label{eq1} Q_{\rm B}=\dfrac{\alpha {\rm HF}}{(\rho c_{p})}+\beta S_0{\rm FWF}=Q_{\rm T}+Q_{\rm S} , \end{equation} where HF is the net heat flux at the sea surface (positive when the ocean is receiving heat flux); FWF (=P-E, where P is precipitation and E is evaporation) is the net freshwater flux (when the ocean is gaining net freshwater, FWF is positive); α is the thermal expansion coefficient; β the haline contraction coefficient; S₀ the reference surface salinity; c_p the heat capacity of seawater, and ρ the density of seawater. The surface Q _B is the net contribution of the HF part (Q _T) and the FWF part (Q _S).

(Zhang and Busalacchi, 2009) indicated that SST anomalies generated by ENSO induce large nonlocal anomalous FWF variability over the western and central Pacific, which directly affects SSS, leading to changes in oceanic processes, and enhances the SST anomalies. Figure 1 shows the FWF interannual standard deviation (STD) of the CMIP5 multi-model and observations in the tropical Pacific. It can be seen that a large FWF variability (larger than 2.5 mm d^-1) is located in the west-central equatorial Pacific and in the South Pacific Convergence Zone (SPCZ), which extends southeastward in the South Pacific Ocean, with a maximum FWF variability of more than 4.0 mm d^-1 from 160^°-180^°W in the west-central equatorial Pacific. The distribution of FWF interannual variability indicates that FWF displacement is similar to the precipitation pattern in tropical regions, such as the intertropical convergence zone (ITCZ). Hence, precipitation interannual variability is a primary factor leading to FWF anomalies in the equatorial Pacific.

Figure 2.  STD of SSS calculated from the results of the CMIP5 multi-model simulations and based on observations (right of the bottom panel) in the tropical Pacific. Units: psu.

Figure 3.  STD of SST calculated from the results of the CMIP5 multi-model simulations and based on ERSST (right of the bottom panels) in the tropical Pacific. Units: ^°C.

FWF variability caused by a precipitation anomaly in the equatorial Pacific leads to a response in the local anomalous ocean physics, in which fresh water gained and lost in the ocean can directly lead to a change in salinity at the surface layer. The spatial distribution of SSS interannual anomalous STDs simulated by the multi-model and from observations is shown in Fig. 2. A large SSS variability (larger than 2.5 psu), covarying with the large FWF variability, is found in the western tropical Pacific based on observations. High salinity is located in the subtropical convergence zone of the Northern Hemisphere and extends eastward across the date line, with a maximum FWF variability of more than 0.5 psu between 160^°W and 180^°W in the west-central equatorial Pacific. It is confirmed that FWF is the main factor responsible for SSS change in the equatorial Pacific, where the evaluation of the sensitivity of precipitation was also conducted to assess the possible impact of salinity on the tropical Pacific climate variability.

Most models (21 out of the 23) agree well with observations in terms of the distribution of interannual FWF (Fig. 1) and SSS (Fig. 2) variability. As shown in Fig. 1, most of the CMIP5 multi-models (all except two: IPSL-CM5A-LR and IPSL-CM5A-MR) show large FWF variability in the west-central Pacific. However, 10 of the 21/23 models that have a large variability in the equatorial region illustrate weaker interannual variability, with a large variability located farther west than observed, and west of the date line. The amplitude of variation for most models (17 of 23), 5.0 mm d^-1 on average, is larger than that of observations.

Not only can the multi-model simulate the interannual FWF feature observed in the tropical Pacific, but it can also reflect the FWF sensitivity to precipitation contributions, which is the primary controlling factor of FWF interannual variability. Moreover, precipitation is the primary factor causing the difference of FWF anomalous patterns simulated by coupled models (Kang et al., 2014).

Furthermore, the corresponding SSS distribution simulated by the CMIP5 multi-model shows the observed feature of a large SSS interannual variability [large than 3.5 psu (Practical salinity units: \textperthousand)] located in the west-central equatorial Pacific. The amplitudes of the interannual variabilities of most CMIP5 multi-models are greater than that in observations and the areas of the large variabilities are commonly greater than that observed. In addition, a difference among CMIP5 multi-models exists in the locations of large variabilities. For example, some models (9 of 23) extend across the date line, but others (14 of 23) are west of the date line. With the eastern front of a large SSS variability simulated from the multi-model moving near the date line, the eastern borders of the large variabilities of most models (16 of 23) move eastward across the date line. SSS variabilities of some models (7 of 23) are weaker and are located in the west equatorial Pacific, where the displacement of large variabilities are farther west than observed. For example, the weaker FWF interannual variability of IPSL-CM5A-MR leads to a smaller SSS anomaly. By contrast, the stronger FWF interannual change of CMCC-CESM corresponds to a larger variability of SSS.

FWF directly affects SSS and changes the upper ocean density so that the ocean physical fields have a positive feedback to SST anomalies in the equatorial Pacific. Figure 3 shows the STD distribution of an SST anomaly in the equatorial Pacific, which is the same as the feature of the observed ENSO pattern. A large SST variability (larger than 0.8^°C) is located in the eastern tropical Pacific, and the area of large variability extends westward along the equator and is near the date line.

Figure 3 shows that the large SST variabilities simulated by most models are located in the east-central Pacific, but the interannual SST variabilities simulated by the multi-model show the difference in the location and intensity of the larger variability. The large SST variabilities of some models (13 of 23) are located westward, across the date line, with a stronger SST variability than that observed.

Based on the above assessment of the spatial distribution, it is evident that most of the CMIP5 multi-models can reproduce the corresponding spatial pattern of FWF, SSS and SST, in which a large FWF-SSS variability is located in the west-central equatorial Pacific and an SST anomaly is located in the east-central equatorial Pacific. It is demonstrated that inherent relationships can be clearly observed among the interannual variabilities of FWF, SSS and SST simulated by the CMIP5 multi-model. Regarding the differences shown by model simulations and observations, the spatial distribution and the magnitude of interannual variabilities are completely inconsistent. For example, in the west-central Pacific, the stronger FWF variability, covarying with the stronger SSS, corresponds to a higher SST interannual anomaly in the east-central Pacific and vice versa for the weaker interannual variabilities of FWF, SSS and SST.

Figure 4.  SST EOF1 spatial pattern and the regressive spatial patterns from SST PC1 in the equatorial Pacific: (a) SST EOF1, (b) FWF, (c) SSS, (d) MLD with observation; (e) SST EOF1, (f) FWF, (g) SSS, (h) MLD simulated by GFDL-ESM2M. The units are ^°C for SST, 1.0× 10^-5 kg m^-2 s^-1 (^°C)^-1 for FWF, psu (^°C)^-1 for SSS, and m (^°C)^-1 for MLD. The colored areas in (b-d) and (f-h) are statistically significant at the 99% confidence level.

The CMIP5 multi-model can satisfactorily represent FWF feedback effects on SST in the tropical Pacific, for which (Zhang et al., 2010) and (Zheng and Zhang, 2012) have carried out detailed analysis and diagnosis. The effect of FWF on this feedback process is further verified by the observed interannual variability and CMIP5 multi-model simulations. FWF affects SSS anomalies directly and then modulates other fields, such as MLD and Q _B. Relationships among these interannual ocean variabilities have been discussed previously (Zhang and Busalacchi, 2009; Zheng and Zhang, 2012), and it has been shown that the effect induced by FWF on the related physical fields associated with ENSO evolution is the intrinsic linkage between FWF and SSS, and MLD and Q _B.

The FWF forcing in the equatorial Pacific tends to indirectly modulate SST conditions in two ways (Zhang et al., 2010). First, during El Niño, FWF into the ocean surface acts to decrease the salinity in the west-central region where the SST is relatively high. The decreased salinity acts to stabilize the upper ocean and suppress the vertical mixing at the base of the mixed layer. Secondly, the FWF into the ocean has a direct effect on Q _B, which in turn exerts an influence on MLD and the entrainment of subsurface water at the base of the mixing layer. Because Q _T and Q _S are negatively correlated during ENSO cycles, their effects on Q _B tend to compensate for one another. Thus, as part of Q _B, the FWF into the ocean acts to compensate for the Q _T, leading to a smaller negative Q _B. The reduced negative Q _B tends to decrease MLD, and FWF into the ocean induces oceanic processes that lead to a less pronounced cooling effect on the surface layers in the west-central basin, which in turn acts to enhance the warming conditions during El Niño, indicating a positive feedback to SST. Therefore, MLD and Q _B are two important physical fields in FWF feedback to SST in the equatorial Pacific.

Figure 5.  The (a-c) observed and (d-f) GFDL-ESM2M simulated spatial patterns in the equatorial Pacific from SST PC1: (a, d) the Q _S part; (b, e) the Q _T part; (c, f) Q _B. The units are 1.0× 10^-5 kg m^-2 s^-1 (^°C)^-1 for Q _S, Q _T and Q _B. The colored areas in (a-f) indicate where values are statistically significant at the 99% confidence level.

To more clearly identify the relationships between interannual anomalies of various oceanic variables and SST, the GFDL-ESM2M model is selected because of its relatively better performance. Then, a regression analysis is conducted along the equator for interannual anomalies of SST, SSS, MLD, and the buoyancy flux components (i.e., Q _B, Q _T and Q _S) during ENSO. The dominant pattern for interannual SST variability is extracted using empirical orthogonal functions (EOF) analysis. Then, SST spatial patterns and corresponding temporal coefficients are obtained from EOF1 (observed variance contribution is approximately 42%; variance contribution simulated by GFDL-ESM2M is approximately 38%), which primarily represent the interannual pattern of SST associated with ENSO. Using the first principal component (PC1) of SSTA, we obtain the spatial distribution of some related fields using the regression analysis, including SSS, MLD and Q _B.

The first spatial pattern of SST extracted using EOF analysis and the relevant regression spatial patterns for various anomalous fields in observations are shown in Fig. 4. For example, during El Niño, a positive SST anomaly appears in the east-central basin (Fig. 4a), accompanied by a large positive FWF anomaly in the west-central basin of the Pacific, SPCZ, and ITCZ (Fig. 4b). The direct effect the positive FWF anomaly is to strengthen a negative SSS anomaly (Fig. 4c) in the west-central Pacific. Correspondingly, the surface ocean density becomes smaller, which tends to stabilize the upper ocean and depress the mixing at the base of the mixed layer. In addition, the MLD becomes shallower in the west-central equatorial Pacific (Fig. 4d), which also tends to suppress the entrainment of subsurface water into the mixed layer to enhance upper-ocean warming in the eastern Pacific. At the same time, the Q _S anomaly is positive in response to the positive FWF in the west-central basin (Fig. 5a). Thus, as part of Q _B, the positive Q _S anomaly acts to compensate for the negative Q _T anomaly (Fig. 5b), leading to a smaller negative Q _B anomaly (Fig. 5c). The reduced negative Q _B anomaly (i.e., due to the contribution of the positive Q _S anomaly) tends to decrease the MLD in the west-central equatorial region. These oceanic processes, which have been demonstrated by (Zhang and Busalacchi, 2009), are favorable for more warming in the surface layer.

Similar to the above observation, the SST EOF1 simulated by GFDL-ESM2M represents ENSO (Fig. 4e). Compared with the observation, there is a higher positive FWF anomaly into the ocean in the west-central equatorial Pacific (Fig. 4f), leading to a relatively stronger negative SSS anomaly (Fig. 4g) than that of the observed anomaly. The high volume of freshwater entering the ocean causes a low surface ocean density anomaly simulated by GFDL-ESM2M, renders the upper ocean more stable and weakens the mixing of the cold water into the mixed layer (Fig. 4h), which leads to stronger warming of SST during El Niño. As such, a warmer than observed SST is likely to be simulated by GFDL-ESM2M, as shown in Fig. 4e.

In addition, a stronger positive Q _S anomaly (Fig. 5d) is seen in the west-central basin and a negative Q _T anomaly (Fig. 5e) is seen in the east-central equatorial Pacific, compared with the observation. The positive FWF anomaly results in a positive Q _S as part of the anomaly increases in the west-central equatorial Pacific, where Q _S makes the interannual Q _B a weaker negative anomaly (Fig. 5f), which makes the upper ocean more stable and thus depresses the mixing of cold water into the mixed layer. As such, the SST anomaly simulated by GFDL-ESM2M can be more positive than observed, as shown in Fig. 4e.

The above results can be considered as positive feedback between FWF and SST during ENSO cycles through the response of SSS, MLD and Q _B, because a stronger FWF anomaly simulated by GFDL-ESM2M leads to a stronger SSS anomaly in the west-central Pacific, and a stronger positive feedback to SST during El Niño is observed in the model results when compared with the observation. However, as exemplified by GFDL-ESM2M, the differences in FWFs simulated by the multi-model can exert striking differences in the modulation of SST in the equatorial Pacific; it is difficult to analyze a quantitative relationship based only on the spatial distribution of FWF, SSS and SST.

In this section, based on the differences in the correlations among the FWF, SSS and SST interannual variabilities simulated by the multi-model and those observed, specific areas are selected to analyze the linear relationship. According to the distribution of interannual variability, FWF is selected from the box (2^°S-2^°N, 160^°-180^°E), SSS is selected from (2^°S-2^°N, 160^°-180^°E), and SST is selected from (2^°S-2^°N, 160^°-160^°W). Then, according to the regional averages of the three interannual variabilities indicated above, the quantifications of those linear relationships are analyzed. Table 2 shows that FWF interannual variability ranges from 1.0 to 5.2 mm d^-1, SSS is in the range of 0.08-0.8 psu, and the SST ranges from 0.8^°C-1.96^°C. As seen from the scatter plots, there are obvious linear relationships among the variables. For example, the FWF, SSS and SST variabilities averaged regionally in the results of BNU-ESM, FGOALS-s2 and HadGEM2-ES consistently correspond to larger variabilities, but those of CSIRO-MK3-6-0, MPI-ESM-LR and MRI-CGCM3 correspond to smaller values.

The corresponding relationships in the SST-FWF linear regression are analyzed according to the two pairs, SSS-FWF and SST-SSS. First, as shown in Fig. 6, the scattered distribution of FWF and SSS illustrates an obvious linear relationship simulated by the multi-model and based on observations. With the increase of FWF variability, the SSS increases in linearity, and the slope is 0.12 (Table 3). The physical mechanism responsible is that each 1 mm d^-1 FWF anomaly can cause a 0.12 psu SSS anomaly in the tropical Pacific.

Similarly, Fig. 7 shows scatter plots for SSS and SST interannual variability, which can induce the linear relationship. For a slope of 0.93 (Table 3), it means that every 1.0 psu of SSS anomaly can bring about nearly 0.93^°C of SST interannual change; therefore, SSS can be considered as a factor of positive influence on SST change.

Figure 6.  Scatter plot of SSS STD and FWF STD in multi-model simulations and observations. Data points are the regional mean in the box (2.0^°S-2.0^°N, 160^°-180^°E). The line represents linear regression fits to the 23 CMIP5 models, and the 24th point is for observation (correlation coefficients r are shown in Table 3). The units are mm d^-1 for FWF, and psu for SSS.

Figure 7.  Scatter plot of SST STD and SSS STD in multi-model simulations and observations. Data points are the regional mean of SST in the box (2.0^°S-2.0^°N, 160^°-120^°W) and SSS in the box (2.0^°S-2.0^°N, 160^°-180^°E). The line represents linear regression fits to the 23 CMIP5 models, and the 24th is for observation (correlation coefficients r are shown in Table 3). The units are ^°C for SST and psu for SSS.

The above analyses verify the interannual FWF modulations of SST in the equatorial Pacific through modulating SSS interannual anomalies. Furthermore, we can also comprehen sively analyze the linear relationship between FWF and SST interannual anomalies (Fig. 8). A linear regression equation is derived for SST-FWF based on the scatter plots for FWF and SST simulated by the multi-model and based on observations (Table 3). The slope of the equation is 0.11, which means that, every 1.0 mm d^-1, FWF can indirectly result in a 0.11^°C SST change in the tropical Pacific.

The contributions of ocean fields to SST anomalies can be defined by one value based on a single-variable linear regression equation. As an example of observed SST interannual variability in a regression equation, an FWF change of 3.22 mm d^-1 can result in a 0.35^°C SST anomaly (3.22 mm d^-1 multiplied by 0.11). For 0.99 ^°C of SST average anomalies, the contribution rate of FWF to SST interannual anomalies is nearly 36% (0.35/0.99). Similarly, the FWF contribution rate to SST interannual anomalies can be derived from the CMIP5 multi-model, and we can generalize a conclusion from the facts that FWF interannual anomalies in the western equatorial Pacific can lead to approximately 12%-40% SST anomalies in the eastern equatorial Pacific, with a mean contribution rate of 24%.

Figure 8.  Scatter plot of SST STD and FWF STD in multi-model simulations and observations. Data points are the regional mean of SST in the box (2.0^°S-2.0^°N, 160^°-120^°W) and FWF in the box (2.0^°S-2.0^°N, 160^°-180^°E). The line represents linear regression fits to 23 CMIP5 models, and the 24th point is for observation (correlation coefficients r are shown in Table 3). The units are ^°C for SST and mm d^-1 for FWF.

Although many studies have been conducted to investigate the effect and modulation of FWF and relevant salinity on SST in the tropical Pacific through diagnosis of observations and simulations, considerable uncertainty remains in determining clear linear relationships of the FWF effect on SSS and SST. Quantifying the direct influence of FWF on SSS, and the FWF modulation of SST in the equatorial Pacific, not only promotes the effectiveness of the model simulation of ocean fields but also helps to better predict SST interannual anomalies. This paper quantifies the linear relationship of the positive feedback of FWF to SST using the simulations of the CMIP5 multi-model and observations. The key conclusions can be summarized as follows:

(2) The interannual FWF variability in the western equatorial Pacific influences interannual SSS variability directly, and modulates SST in the eastern equatorial Pacific. When comparing the spatial distribution characteristics of FWF and related SSS and SST variability simulated by the CMIP5 multi-model and based on observations, it is found that a large FWF variability is located in the western equatorial Pacific, and the displacement of larger variability simulated by most models moves eastward across the date line and increases SSS in the western equatorial Pacific to warm SST in the eastern equatorial Pacific. CMIP5 multi-models can reproduce the ocean processes observed, but the simulation capabilities of the multi-models are different with regard to the magnitude and position of FWF and related SSS and SST.

(2) The differences in model intrinsic properties among multi-models may lead to an inconsistent feedback effect on oceanic fields. Based on the intensities and positions of the positive feedback to SST, GFDL-ESM2M is selected from among the CMIP5 multi-model simulations to analyze the differences in ocean physical processes with respect to FWF feedback to SST in the equatorial Pacific. The fact that the interannual FWF variability simulated by GFDL-ESM2M is stronger than observed can lead to a larger negative SSS anomaly and a more negative ML anomaly in the western equatorial Pacific. In other words, the stronger FWF can cause Q _S to compensate for the Q _B more effectively. The less negative Q _B anomaly and shallower MLD create more positive feedback to an SST anomaly to warm the upper ocean. Differences in FWF in the western Pacific lead to the differences in ocean physical processes among the models and observations, such that the feedback to SST is highly sensitive to FWF and related ocean fields in the equatorial Pacific.

(3) Simple linear relationships exist in the interannual variability of FWF, SSS and SST. There are positive linear relationships among the interannual variabilities of FWF with SSS, SSS with SST, and FWF with SST. Based on the slopes of linear regression equations, an FWF anomaly of 1.0 mm d-1 causes an SSS anomaly of nearly 0.12 psu in the tropical Pacific, and a 1.0 psu SSS anomaly leads to an SST anomaly of 0.93^°C. In general, a 1.0 mm d^-1 FWF anomaly indirectly corresponds to a 0.11^°C SST anomaly.