Relationship between South China Sea Precipitation Variability and Tropical Indo-Pacific SST Anomalies in IPCC CMIP5 Models during Spring-to-Summer Transition

Received 14 November 2014; revised 26 February 2015; accepted 5 March 2015

Utilizing climate models is a key approach to the understanding of historical climate changes and projections of climate in the future. The evaluation of model performances in various aspects is necessary for unraveling model biases, which then benefits the further development and application of models. Given the availability of the state-of-the-art models involved in the Coupled Model Intercomparison Project Phase 5 (CMIP5) for the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5), we evaluate the precipitation variability over the South China Sea (SCS) and its relationship to tropical Indo-Pacific sea surface temperature anomalies during spring-to-summer transition (April-May-June, AMJ) simulated by 23 coupled models in this study.

The SCS displays remarkable atmosphere-ocean interaction that is related to the climate variability in the surrounding regions (Wang et al., 1997; Wu and Wang, 2001; Wu, 2002, 2010; Lestari et al., 2011; Roxy and Tanimoto, 2012; He and Wu, 2013a, 2013b), as it is a part of the tropical eastern Indian Ocean-western Pacific warm pool and the Asian monsoon region. The SCS is a region of frequent tropical cyclone/typhoon activity (Wang et al., 2007; Chen et al., 2012; Feng et al., 2013) and a pathway of moisture transport from the North Indian Ocean and the Southern Hemisphere to East Asia (Tao and Chen, 1987; Ding, 1994; Wu et al., 2006a), both of which have a great impact on the weather and climate of China, especially South China. The onset of the SCS summer monsoon mostly occurs in May (Wu and Wang, 2000, 2001). So, the AMJ period signifies the transition from spring to summer in the SCS and the surrounding regions. Therefore, it is important to evaluate model capacities in terms of the SCS climate variability during AMJ.

Previous studies have revealed that the atmosphere-ocean relationship in the SCS shows obvious seasonality and regionality. (He and Wu, 2013b) pointed out that atmospheric forcing is dominant in the northern and central SCS during the local warm season, and that oceanic forcing occurs in the northern SCS during the local cold season. For AMJ in the SCS, which is the season of transition to the warm and rainy season, the atmospheric forcing in the central SCS is characterized by remarkable cloud-radiation and wind-evaporation effects across the central basin, and wind-driven oceanic upwelling effects along the west coasts. Furthermore, the SCS rainfall variability is subject to remote forcing from tropical Indo-Pacific SST anomalies (SSTAs). El Niño-Southern Oscillation (ENSO), one of the strongest signals in short-term climate variability, influences the climate variability in remote regions both directly and indirectly (Wu et al., 2009; Wu and Kirtman, 2011; Wu et al., 2012; Yun et al., 2013). After El Niño dissipates during spring, the tropical Indian Ocean (TIO) SST warming persisting into summer (June-August) forces a Kelvin wave response into the northwest Pacific, which triggers suppressed convection and an anomalous anticyclone there (Xie et al., 2009). In addition, (He and Wu, 2014) further pointed out that North Indian Ocean SSTAs serve as a medium for an indirect impact of preceding equatorial central Pacific SSTAs on the SCS summer rainfall variability.

The SCS rainfall variability during AMJ has been found to be influenced by SSTAs in the equatorial Pacific (EP), TIO, and western North Pacific (WNP) (Hu et al., 2014). Above-normal precipitation in the SCS corresponds to lower EP and TIO SST and higher WNP SST. Anomalous precipitation in the SCS occurs via anomalous circulations induced by different combinations of SSTAs in the above regions. Does a link between the SCS precipitation variability to SSTAs in the tropical Indo-Pacific regions during AMJ exist in the CMIP5 models? The main objective of the present study is to understand how climate models perform in this respect.

The organization of the text is as follows: The data and methods used in this study are described in section 2. In section 3, the leading bias modes in the SCS and the tropical Indo-Pacific are examined. Then, we evaluate the model performance in simulating the AMJ precipitation variability in the SCS and the corresponding precipitation and SSTA pattern in the tropical Indian and Pacific oceans in section 4. Following that, we classify the models into different types and analyze the relationship between SCS precipitation variability and tropical Indo-Pacific SSTAs during spring-to-summer transition in different types of models separately in section 5. A summary of the results is provided in section 6.

This study utilizes monthly mean precipitation, wind fields and skin temperature from historical simulations of 23 CMIP5 coupled models, which are available online at http://pcmdi9.llnl.gov/. Due to the large number of experiments included in the CMIP5 framework, only one member for each of the 23 model outputs from the historical experiments is analyzed. Furthermore, the analysis in the present study is based on a common period from 1979 to 2005 to enable a fair comparison with the observations. The information for the selected models is given in Table 1.

The National Oceanic and Atmospheric Administration (NOAA) extended reconstructed SST, version 3 (ERSST3; Smith et al., 2008) is used in the present study. The ERSST3 is available on a 2^°× 2^° grid from 1979 to 2005. We also use monthly mean rainfall from version 2 of the Global Precipitation Climatology Project (Adler et al., 2003). This dataset is available on a 2.5^°× 2.5^° grid from 1979 to 2005. Monthly mean horizontal winds are adopted from the National Centers for Environmental Prediction-U.S. Department of Energy Reanalysis 2 (Kanamitsu et al., 2002) on a regular 2.5^°× 2.5^° grid for the period 1979-2005. The reanalysis product is provided by the NOAA/Cooperative Institute for Research in Environmental Sciences Climate Diagnostics Center, Boulder, Colorado (available via anonymous ftp at ftp://ftp.cdc.noaa.gov/).

The rainfall and SSTAs in both observations and model outputs have been interpolated to 1^°× 1^° grid for a better coverage of the SCS domain and for convenience of comparison between observations and model simulations. The significance level of correlation is determined based on the Student's t-test.

Li and Xie (2012, 2014) introduced an intermodel Empirical Orthogonal Function (EOF) analysis to isolate the dominant patterns of model biases. The advantage of this method is maximizing the variance of variable differences among models instead of the temporal variance in conventional EOF analysis (Li and Xie, 2014). Here, we carry out a similar analysis to examine the main model biases of the AMJ-mean precipitation. First, the AMJ-mean precipitation is normalized in two dimensions (x-y) for each CMIP5 model. Second, an EOF analysis is applied to each model's precipitation deviations from the multimodel ensemble (MME) mean for a specific region. The obtained series of each intermodel principal component (IPC) represent the degree of deviation with respect to each intermodel EOF bias pattern in the individual models.

We use the Taylor diagram (Taylor, 2001) to provide a concise statistical summary of how well patterns match with each other in terms of their correlation and the ratio of their standard deviations, which are simply indicated by a single point on the diagram. Through comparison with observations in both pattern and magnitude, we can clearly distinguish the degree of accuracy in the simulation of each model.

Figure 1.  Regression patterns of precipitation (contours), SST (shading), and 850-hPa horizontal wind (vectors) onto the (a) first and (b) second IPCs of intermodel variability in AMJ-mean averaged normalized precipitation over the SCS in 23 CMIP5 models. (c) The first two IPCs. The explained variances are given in the top right corner of (a, b).

First, we use the intermodel EOF analysis described above to obtain the dominant bias patterns of AMJ precipitation, SST and low-level winds over the SCS and the tropical oceans in the models. Figure 1 shows the regressed patterns of precipitation, SST and 850-hPa horizontal winds onto the first two IPCs as well as the IPCs with the intermodel EOF analysis applied to the SCS domain.

Figure 2.  As in Fig. 1, but for the tropical Indo-Pacific (30^°S-30^°N, 40^°E-90^°W).

The first intermodel EOF mode (Fig. 1a), explaining 31.2% of total intermodel variability, features more precipitation and warmer SST in most regions of the SCS, accompanied by an anomalous cyclonic circulation. Models with higher or lower IPC1 values would have larger biases of precipitation in the central SCS and SST in the northern SCS than other normal ones (Fig. 1c). The three models with the highest IPC1 values are CanCM4, CanESM2 and HadCM3, while the three models with the lowest IPC1 values are GISS-E2-H, MIROC5 and NorESM1-M. The second intermodel EOF mode explains 17.0% of total intermodal variance (Fig. 1b). This mode has negative precipitation anomaly centers located in the northeastern and southwestern SCS and a positive center in the southeastern SCS, accompanied by an anomalous cyclonic circulation and warmer SST in the central SCS. The three models with the largest IPC2 values are BCC-CSM1.1, CSIRO-MK-3-6-0 and GFDL-ESM2G, and the three models featuring the opposite situation are ACCESS1-0, MIROC5 and MPI-ESM-LR. Both bias patterns show that the positive precipitation anomaly corresponds to a positive SSTA, indicative of the feedback of local SST change on the atmospheric change. However, (He and Wu, 2013b) pointed out that from April to June, an atmospheric forcing of the ocean with a significantly negative precipitation-SST tendency correlation is dominant in the northern SCS and the central SCS. The bias patterns suggest an inaccurate description of the precipitation-SST relationship over the SCS in some models.

(Hu et al., 2014) suggested a remote forcing of the tropical Indo-Pacific SST to the SCS AMJ rainfall variability. Thus, we further evaluate the main model biases in the tropical Indo-Pacific region during AMJ. Figure 2 shows the leading bias modes of precipitation, SST and low-level winds and the corresponding IPCs in the tropical Indo-Pacific domain (30^°S-30^°N, 40^°E-90^°W).

Similar to Fig. 2a in the study of (Li and Xie, 2014), the first intermodel EOF mode (Fig. 2a), explaining 23.2% of total variability, reflects the equatorial cold tongue SST bias. It features deficient precipitation and colder SST in the EP, with excessive precipitation and warmer SST on the flanks of the cold tongue. A large area of negative SSTA covers the SCS, the northern Indian Ocean and the northwest of Australia. In contrast, a positive SSTA center is located in the southern Indian Ocean. The three models with the highest IPC1 values are GFDL-ESM2G, IPSL-CM5A-LR and IPSL-CM5A-MR, and the three models with the lowest IPC1 values are ACCESS1-0, FGOALS-g2, and MIROC-ESM.

The second intermodel EOF mode (Fig. 2b), explaining 14.7% of total variance, exhibits positive precipitation anomaly areas in the SCS and the Maritime Continent, in the tropics of the Southern Hemisphere, and in the eastern Pacific. Furthermore, the most pronounced negative rainfall anomaly center is located in the central and western Pacific. Positive SSTAs tend to be located in excessive rainfall areas, suggestive of an oceanic forcing of the atmosphere. The three models with the highest (lowest) IPC2 values are CanCM4, CSIRO-MK3-6-0 and HadCM3 (FGOALS-s2, GISS-E2-H, and NorESM1-M).

Figure 3.  Spatial distributions of the first EOF mode of the precipitation anomalies (shading, units: mm d^-1) over the SCS in AMJ simulated by (b-x) 23 models compared with the (a) observation.

Based on the intermodel EOF analysis, excessive precipitation in the central (Fig. 1a) and southeastern (Fig. 1b) SCS stands out as the most prominent error in some of the CMIP5 models. In this section, the SCS rainfall variability is investigated and evaluated through conventional EOF analysis and Taylor diagrams. The least squares linear trends are removed from all of the variables before the following analysis is conducted. Figure 3 shows the first EOF modes of the AMJ precipitation anomalies over the SCS simulated by 23 IPCC CMIP5 models compared with the observation.

The first mode accounts for about 40.8% of the total variance in the observation (Fig. 3a), which is separated from the other modes according to the rule of (North et al., 1982). A large positive loading is located over the central SCS, while a negative loading is confined to the southern coast of China. The percentage variance explained by the leading EOF mode in the models varies from less than 20% (CNRM-CM5) to over 60% (GFDL-ESM2M). About two thirds of the models have the capacity to capture the main features of the observed leading mode (Figs. 3b-x). However, some models show remarkable deviations from the observed pattern. The CanESM2 and IPSL-CM5A-LR models display loading that extends to the northern SCS (Figs. 3e and q). The distribution of loading in BCC-CSM1.1, IPSL-CM5A-MR, MIROC5, and MIROC-ESM is dissimilar to the observations (Figs. 3c, r, t and u). In the GFDL-CM3 model, large loading shifts to both the northern and southern SCS are apparent (Fig. 3k). The differences in the pattern of SCS rainfall variability among the models should be related to the AMJ-averaged model errors shown in the previous section.

Figure 4.  Taylor diagram describing the relationship between the regressed AMJ precipitation to each model's CPC1 (solid dot)/CPC2 (hollow hexagon) and that to the observational CPC1 for the SCS domain (0^°-25^°N, 98^°-122^°E). The radial distances from the origin to the points represent the ratio of the standard deviations; the azimuthal positions are determined by the PCC. Only the value over the ocean is calculated.

Figure 5.  As in Fig. 4, but for the regressed precipitation in the (a) TIO and (b) TPO, and for the regressed SST in the (c) TIO and (d) TPO.

To quantify the similarity of the precipitation anomaly distribution, we use the Taylor diagram to compare models with observations based on both the pattern correlation coefficient (PCC) and the ratio of standard deviations (RSD) of the modeled and observed fields in the SCS domain [(0^°-25^°N, 98^°-122^°E), with land grid points excluded)]. The PCC and RSD values are calculated based on the precipitation anomalies obtained by regression with respect to the first conventional principal component (CPC1) of the observation and model, respectively. Since the pattern of observed AMJ precipitation variability may not be represented as EOF1, but may be captured by EOF2 in the models, we also calculate the PCC and RSD between the observation CPC1 and model CPC2. The related Taylor diagram is given in Fig. 4. The radial distances from the origin to the points represent the RSD, while the azimuthal positions are determined by the PCC.

Both the PCC and RSD for the model CPC1 (solid dot) show an obvious diversity. For example, the PCC is only 0.136 for BCC-CSM1.1, but reaches 0.855 for NorESM1-M. The value of RSD varies from 0.551 (MIROC-ESM) to 1.546 (CanCM4). This indicates that the performance in the simulation of the SCS precipitation anomaly pattern is model dependent. The regressed precipitation of seven models (CanESM2, FGOALS-s2, GFDL-CM3, inmcm4, IPSL-CM5A-LR, MIROC5, and MIROC-ESM) shows a higher PCC corresponding to the model CPC2 (hollow hexagon) than to the model CPC1. For these seven models, the distribution of loading in EOF1 shows obvious differences from observations (Figs. 3e, j, k, p, q, r, t and u). As such, the corresponding PCC is relatively low for the model CPC1. There are five models (CCSM4, CSIRO-Mk3-6-0, FGOALS-g2, GFDL-ESM2G, and HadCM3) that have both high a PCC for CPC1 of more than 0.6 and an RSD of around 1, suggestive of a better capacity to simulate the SCS rainfall variability. The PCC and RSD values of the MME for CPC1/CPC2 are 0.465/0.799 and 0.048/1.603, respectively.

The SCS rainfall anomalies in AMJ are influenced by SST and convection anomalies in both the tropical Indian and Pacific oceans (Hu et al., 2014). Therefore, we evaluate the regressed SST and precipitation anomaly patterns in the tropical Pacific and Indian Ocean regions to the CPC1/2 in the models to understand which regional SSTAs contribute to the SCS AMJ rainfall variability. The evaluation is based on the PCC and RSD values calculated for precipitation and SST in both the tropical Pacific Ocean (TPO; 30^°S-30^°N, 120^°-270^°E) and the TIO (30^°S-30^°N, 40^°-100^°E). The results for SST and precipitation are given in Fig. 5.

The PCCs of simulated SST and precipitation in the TPO and TIO vary widely among the models, indicative of diversity in model performance. Some models have a high positive PCC corresponding to the model CPC1 in both the TPO and TIO for both precipitation and SST, e.g., CNRM-CM5, FGOALS-g2, GFDL-ESM2G, GFDL-ESM2M, GISS-E2-H, and IPSL-CM5B-LR. Some models show a high PCC for SST in the TPO (Fig. 5d), but a low PCC in the TIO (Fig. 5c), e.g., ACCESS1-0, GFDL-CM3, HadCM3, and inmcm4. Two models (CSIRO-Mk3-6-0 and IPSL-CM5A-LR) show a high PCC for precipitation in the TIO (Fig. 5a), but a low PCC in the TPO (Fig. 5b). There are two models (CanCM4 and CanESM2) that have large negative PCCs for precipitation in both the TPO and TIO, with the PCC for SST being either negative or small in the TIO. Two models (FGOLAS-s2 and IPSL-CM5A-MR) have negative PCCs for SST in both the TPO and TIO. Two models (MIROC5 and MIROC-ESM) in which EOF2 is more similar than EOF1 to observation EOF1 display large positive PCCs corresponding to CPC2. As for the distribution of model RSD, almost all the models simulate a smaller (larger) magnitude for precipitation and SST in the TPO (TIO) than that in the observation.

The above method is based on a comparison of the pattern of anomalies. It does not consider the value of SST or precipitation anomalies in each SST key region. An evaluation of SSTAs is also needed. Here, we briefly evaluate the SSTAs in some key regions. (Hu et al., 2014) defined three key SST regions based on the distribution of the SST correlation with the SCS precipitation variability during AMJ (shown by the rectangular boxes in Fig. 6b) in the observations; namely, the equatorial Indian Ocean (EIO), WNP, and EP. The SSTAs averaged over the EIO, WNP and EP in the observations are -0.130, 0.168 and -0.356 K, respectively. The above analyses indicate the complexity of model simulations of the SCS precipitation anomaly pattern and associated SSTAs in the tropical Indo-Pacific regions. For illustrative purposes, we classify the models into different types based on both the PCCs of SST and precipitation in the TPO and TIO and the comparison of SSTAs averaged over the EIO, WNP and EP in models with the observations (Table 2). The classification is determined based on relative values of PCCs and SSTAs. The precise classification of each model may need a strict criterion, which is hard to reach due to the complexity involved. The five types of models and the corresponding values of PCCs and SSTAs are listed in Table 2. Note that the BCC-CSM1.1 model is not classified due to low PCCs for both SST and precipitation.

Figure 6.  Simultaneous regression of (a) precipitation (shading; units: mm d^-1) and 850-hPa wind (vectors; units: m s^-1) and (b) SST (shading, units: K) with respect to normalized time series of the leading mode of AMJ precipitation anomalies for the observations. Red dots (a) and black dashed contours (b) denote regions where anomalies are significant at the 95% confidence level according to the Student's t-test. Only significant wind vectors are plotted. The three rectangular boxes in (b) denote the key SST regions that are used to calculate area-mean SSTAs in Table 2.

Figure 7.  As in Fig. 6a, but for the TPO-TIO type models.

Figure 8.  As in Fig. 6b, but for the TPO-TIO type models.

In the TPO-TIO type, the PCC for SST or precipitation in the TPO and TIO is large and positive, while the SSTAs in the EIO, WNP and EP are relatively close to the observations. It includes the following ten models: CCSM4, CNRM-CM5, GFDL-ESM2G, GFDL-ESM2M, IPSL-CM5A-LR, IPSL-CM5B-LR, MIROC5(CPC2), MIROC-ESM(CPC2), MPI-ESM-LR, and NorESM1-M. Note that the CCSM4 and NorESM1-M models have relatively low PCCs for SST in the TIO, but they are classified as the TPO-TIO type because they have large PCCs for precipitation in the TIO, comparable to that in the TPO. We classify CNRM-CM5 as the TPO-TIO type because it has a large EIO SSTA and EP SSTA, comparable to those in the TPO type, though the SSTA in the WNP is small and negative. The IPSL-CM5A-LR model has a higher PCC corresponding to CPC2 (0.319) than CPC1 (0.068) for precipitation in the SCS (Fig. 4). Nevertheless, a visual inspection of the distribution of EOF1 and EOF2 loading tells us that the EOF1 resembles more than the EOF2 to the observed EOF1. Furthermore, the PCC corresponding to CPC2 is small for both precipitation and SST in both the TPO and TIO (Fig. 5). Also note that the percentage variance explained by CPC2 (15.8%) is less than half of that by CPC1 (35.0%). Thus, the IPSL-CM5A-LR model is classified as the TPO-TIO type based on large positive PCCs for both precipitation and SST in both the TPO and TIO. The EOF2 loading of the MIROC5 and MIROC-ESM models is more similar than the EOF1 loading to the observed EOF1 loading (Fig. 4). The PCC for SST in the TPO or TIO is larger corresponding to CPC2 than CPC1. Thus, MIROC5 and MIROC-ESM are classified to the TPO-TIO type based on CPC2.

In the TPO type, the PCC for SST or precipitation is large in the TPO, but small in the TIO. It contains four models: ACCESS1-0, GFDL-CM3, HadCM3, and inmcm4. In the TIO type, the PCC is large in the TIO but small in the TPO. It includes CSIRO-Mk3-6-0 and GISS-E2-H. In the western EP (WEP) type, the PCC (SST) or PCC (precipitation) in the TPO and TIO are significantly negative. It consists of two models: CanCM4 and CanESM2. As shown later, in the WEP type models, there is an overly strong impact of WEP SST that overcomes the eastern TPO and TIO SST influences on the AMJ SCS precipitation variability. In the local type (FGOALS-g2, FGOALS-s2, IPSL-CM5A-MR, and MRI-CGCM3), the PCCs for SST in both the TPO and TIO are relatively small or even negative, or the SSTAs in the EIO, WNP and EP are relatively small.

In this section, we analyze in detail the precipitation, 850-hPa wind, and SSTA distribution in different types of models. For comparison, we first analyze the observed anomaly distribution. The regressed precipitation, 850-hPa wind and SST with respect to the normalized CPC1 of AMJ precipitation anomalies for the observation are shown in Fig. 6, which is similar to Fig. 3 of (Hu et al., 2014). Significant positive precipitation anomalies cover most parts of the SCS, the Indo-China Peninsula, and the WNP, with a cyclonic circulation consisting of northeasterly winds from South China and westerly winds from the Indian Ocean (Fig. 6a). Three negative precipitation anomaly centers are located over the western TIO, northwest of Australia, and the equatorial central Pacific. Strong easterlies prevail over the equatorial central Pacific. The SSTA field shows two negative regions over the TIO and the EP, and one positive region over the WNP (Fig. 6b). The relevant physical processes of the relationship between AMJ SCS precipitation variability and tropical Indo-Pacific SSTAs are shown by (Hu et al., 2014). Overall, both anomalous cross-equatorial flows from the southwestern TIO induced by negative SSTAs there, and an anomalous Walker circulation forced by negative EP SSTAs, contribute to enhanced convection over the SCS and the surrounding regions. Additional contribution comes from positive WNP SSTAs via a Rossby wave-type response (Hu et al., 2014).

For the TPO-TIO type, the simulated patterns of rainfall, 850-hPa wind, and SSTAs obtained by regression with respect to the normalized CPC of AMJ precipitation are given in Figs. 7 and 8. Seven models (CCSM4, GFDL-ESM2G, GFDL-ESM2M, IPSL-CM5A-LR, IPSL-CM5B-LR, MIROC5 and NorESM1-M) simulate the observed anomalies well (Figs. 7a, c, d, e, f, g, j and 8a, c, d, e, f, g, j). For example, negative SSTAs in the TPO are accompanied by below-normal precipitation and anomalous easterlies over the EP. Negative SSTAs in the southern TIO are followed by negative rainfall anomalies and cross-equatorial southerlies there. Furthermore, positive SSTAs in the WNP are accompanied by cyclonic wind anomalies over the SCS and the Philippine Sea due to the destabilization of the lower troposphere (Wu and Wang, 2000, 2001; Hu et al., 2014). These features are similar to observations, indicating the contribution of tropical Indian and Pacific Ocean SSTAs to the SCS precipitation variability (Hu et al., 2014). The other three models capture the precipitation and wind anomalies over and surrounding the SCS, but precipitation and wind anomalies over the broad TIO-TPO are relatively weak (Figs. 7b, h and i), which is related to relatively small SSTAs (Figs. 8b, h and i). The SCS precipitation variability in these three models appears to be mainly influenced by SSTAs in the TIO and WNP.

In IPSL-CM5A-LR (Figs. 7e and 8e), the precipitation, wind, and SSTA patterns are quite similar to observations. One notable discrepancy from observations is that the precipitation anomalies over the SCS shift to the northern part compared to observations. This may suggest a systematic shift in the response to tropical Indo-Pacific SSTAs in this model. In MIROC5 (Figs. 7g and 8g), the SSTA pattern is similar to observations, as is the precipitation and wind anomaly distribution. For example, there are anomalous cross-equatorial flows over the TIO and anomalous easterlies over the EP. The precipitation, wind, and SSTA distribution in MIROC-ESM (Figs. 7h and 8h) are also similar to observations, but with weaker magnitude.

Figure 9.  As in Fig. 6, but for the TPO type models.

Figure 10.  As in Fig. 6, but for the TIO type models.

Figure 11.  As in Fig. 6, but for the WEP type models.

Figure 12.  As in Fig. 6, but for the local type models.

Figure 13.  figure* \includegraphics140250fig13.eps

For the TPO type, precipitation, 850-hPa wind, and SSTAs corresponding to the normalized CPC1 of AMJ precipitation are shown in Fig. 9. The SSTA distribution similar to observations is visible in the TPO (right panels of Fig. 9). However, the wind and precipitation anomalies over the EP are weak (left panels of Fig. 9). Above-normal rainfall and the anomalous cyclone around the SCS may be contributed by positive SSTAs in the WNP via a Rossby wave-type response (Wu et al., 2014). Wind and precipitation anomalies over the TIO are similar to observations, but the SSTAs display an east-west pattern of contrast, leading to a low PCC (SST) in the TIO (Table 2).

Two models, CSIRO-Mk3-6-0 and GISS-E2-H, are included in the TIO type. The precipitation and wind anomalies in the TIO and western tropical Pacific are similar to observations, as are the SSTAs in the TIO and WNP (Fig. 10). However, negative SSTAs in the tropical Pacific are confined to north of the equator in CSIRO-Mk3-6-0, and SSTAs in the EP are weak in both models (right panels of Fig. 10). The SCS precipitation and wind anomalies appear to be contributed by both the TIO and WNP SSTAs.

For the WEP type, prominent SSTAs are present in the TIO and TPO, but different from observations (right panels of Fig. 11). The SSTAs in the equatorial central-eastern Pacific and TIO are opposite to observations, whereas the WEP SSTAs are similar to observations. Precipitation and wind anomalies over the SCS appear as a northwestward extension of the Rossby wave-type response to WEP SSTAs (left panels of Fig. 11).

For the local type (FGOALS-g2, FGOALS-s2, IPSL-CM5A-MR, and MRI-CGCM3), the SSTAs in the TPO and TIO are small (right panels of Fig. 12). In FGOALS-s2, prominent precipitation anomalies are present over the SCS and WNP (Fig. 12a). These are accompanied by negative SSTAs over the northern SCS and subtropical WNP (Fig. 12b). Thus, it appears that the SSTA is a response to atmospheric change. A similar feature is seen in FGOALS-g2, IPSL-CM5A-MR and MRI-CGCM3, but with weaker anomalies. These results seem to suggest that the SCS precipitation variability is largely related to internal atmospheric dynamics.

There are three points to note according to the above analysis of anomalies corresponding to SCS precipitation variability during AMJ in different model types. First, the signal of SCS precipitation variability in the TPO and TIO SST and precipitation varies largely among the models. Second, the SSTA pattern in the TPO and TIO is important for capturing the AMJ SCS precipitation variability. And third, a realistic simulation of the WEP and local SST impacts is necessary for reproducing the AMJ SCS precipitation variability.

Previous studies have revealed that SST is influenced by the atmospheric changes through the cloud-radiation effect, wind-evaporation effect, and wind-induced oceanic processes (Klein et al., 1999; Wu and Kirtman, 2007; Wu, 2010). Meanwhile, SST change could modulate regional convection via lower-level moisture convergence, surface evaporation, and lower tropospheric stability (Lindzen and Nigam, 1987; Wu and Wang, 2000, 2001). The performance of precipitation-SST correlation in the climate models could provide information about whether the physical processes of air-sea interaction in the models are realistic. Positive precipitation-SST correlation indicates an oceanic forcing of precipitation, with ocean surface warming inducing more precipitation (Wu et al., 2006b). By contrast, negative precipitation-SST correlation means an atmospheric forcing of SST, with decreased downward shortwave radiation reaching the surface induced by more precipitation and leading to surface cooling (Trenberth and Shea, 2005; Wu et al., 2013). Thus, we compare the SST-precipitation correlation in models against observations during AMJ to understand the biases of the relationship between AMJ SCS precipitation variability and SST key regions. The point-wise and simultaneous precipitation-SST correlations for the observation and the 23 models are given in Fig. 13.

In the observations, positive correlation regions cover the eastern TPO, the WNP, and the southern TIO, while negative correlations appear in the eastern Bay of Bengal and most of the SCS (Fig. 13a), which is consistent with (He and Wu, 2013b). This indicates the forcing of SST in the equatorial central-eastern Pacific, southern TIO, and WNP, supported by numerical experiments (Hu et al., 2014). In turn, anomalous convection and winds may affect the SCS precipitation and wind. In the SCS domain, multiple models (CanESM2, CCSM4, CSIRO-Mk3-6-0, GFDL-ESM2M, GISS-E2-H, HadCM3, inmcm4, IPSL-CM5A-LR, IPSL-CM5A-MR, MIROC5, MIROC-ESM, MPI-ESM-LR, MRI-CGCM3, and NorESM1-M) show a sign of correlation opposite to the observed, indicating unrealistic oceanic forcing to the atmosphere in these models. Also of note is that overly strong WEP SST impacts are seen in CanCM4, CanESM2, CSIRO-Mk3-6-0, HadCM3, inmCM4, IPSL-CM5A-LR, IPSL-CM5A-MR, and MIROC5. For example, the positive correlation over the WEP exceeds 0.8 in CanCM4 (Fig. 13d). The warm SST there induces large positive precipitation anomalies that extend northwestward to the SCS (Figs. 11a and b).

Previous studies have indicated a remote forcing of the SCS precipitation variability during spring-to-summer transition in observations. In the present study, using 23 IPCC AR5 coupled models, we first examine the dominant bias patterns over the SCS and tropical oceans, and then evaluate the precipitation variability over the SCS and its relationship to tropical Indo-Pacific SSTAs in AMJ. The main results can be summarized as follows:

Based on the intermodel EOF analysis, the leading model biases in the SCS feature excessive precipitation and warmer SST in the central SCS, with an anomalous cyclonic circulation. Moreover, deficient precipitation and colder SST in the EP is detected as the leading model bias mode, plus negative SSTA over the SCS, the northern Indian Ocean, and the northwest of Australia.

Most of the models have the capacity to capture the AMJ precipitation variability in the SCS. This is related to the TPO-TIO SSTA pattern and its influence on the SCS precipitation and wind. When the models can simulate the SSTAs in the equatorial central-eastern Pacific Ocean, southern TIO, and WNP, either in all three regions or in a combination of two regions, they tend to produce a similar SCS precipitation variability as in observations. This indicates that the SSTA pattern in the TPO and TIO is important for capturing the AMJ SCS precipitation variability.

There are, however, exceptions. One is the case when there is excessive SST forcing in the WEP that disrupts the relationship between the SCS precipitation variability and the tropical Indo-Pacific SST. In this case, the above-normal precipitation over the SCS is mainly due to the positive SSTA in the WEP. Another case is when there is no apparent SST influence. Here, the AMJ SCS precipitation variability is not affected by the SST conditions over the Indian Ocean and the Pacific, but is seemingly dominated by internal atmospheric dynamics. This indicates that a realistic simulation of regional SST impacts and internal atmospheric variability is necessary for reproducing the relationship between the AMJ SCS precipitation variability and tropical Indo-Pacific SST.

The model evaluation presented here provides a general understanding about how these models perform in simulating the SCS precipitation variability and its relationship with the SSTAs in the tropical Indo-Pacific regions. Unraveling the reasons why the models perform well or badly calls for a detailed analysis of the physical connections in various aspects. Further studies should focus on the physical processes that connect the SCS to the tropical oceans, the relative impacts of SSTAs in different regions, and the relative contributions of internal atmospheric dynamics and SST forcing to the SCS precipitation variability.