The impacts of conventional El Niño-Southern Oscillation (ENSO) and ENSO Modoki on wintertime Southeast Asian rainfall and related mechanisms are studied using the method of partial regression and correlation and numerical simulations of a simple baroclinic model. Results show that the Southeast Asian rainfall associated with these two kinds of ENSO exhibits different spatial distributions. In the case of El Niño, wet conditions are observed over south China, and dry conditions are seen over the Philippines, Borneo, Celebes, and Sulawesi. In contrast, for El Niño Modoki, the negative rainfall anomalies around the Philippines are weaker and are located more northward compared to the El Niño counterpart. The different Southeast Asian rainfalls that are related to ENSO and ENSO Modoki are attributed to the different anomalous Walker circulation and low-level anticyclone around the Philippines. Both the Philippine anticyclone and the descending branch center of the Walker circulation over the western North Pacific occupy a smaller domain and are located more northward during El Niño Modoki than during El Niño. All of these factors favor the difference in the Southeast Asian rainfall anomalies between the two events. Numerical experiments also suggest that the different low-level atmospheric responses are mainly induced by different diabatic cooling over the western North Pacific related to El Niño and El Niño Modoki.
 During boreal winter many Southeast Asian areas experience a wet monsoon season, while the Indo-China peninsula and the Philippines are in the dry season [Cheang, 1987; Chang et al., 2005]. Because the inhabitants of this region depend heavily on the rains for agriculture and many other activities, the variability of wintertime rainfall can exert both social and economical influences on Southeast Asian countries [Cheang, 1987]. Wintertime Southeast Asian rainfall can also extend its effects on remote regions, in addition to its local influence. For example, the rainfall in this region is usually associated with strong convective activities, especially over the Maritime Continent, which represents a dominant heat source for the atmospheric circulation [Neale and Slingo, 2003]. The upper-tropospheric divergent outflow from this region has been shown, at least in an idealized model, to be a major source of wave activity due to the generation of global rotational flow [Sardeshmukh and Hoskins, 1988]. Therefore, it is important to investigate the variation of wintertime Southeast Asian rainfall and its related mechanisms.
 In recent years a new type of tropical Pacific sea surface temperature (SST) warming pattern, named “El Niño Modoki,” has been proposed [Ashok et al., 2007]. When it occurs, a warm SST anomaly is observed in the central equatorial Pacific, and a cold SST anomaly is observed in the western and eastern Pacific. It has been proven that the El Niño Modoki is totally different from the conventional El Niño in both spatial pattern [Ashok et al., 2007; Kao and Yu, 2009; Trenberth and Smith, 2009] and formation mechanism [Kao and Yu, 2009]. Moreover, it has also been noticed that the influence of the El Niño Modoki on global climate is different from that of the conventional El Niño [Ashok et al., 2009; Kim et al., 2009; Taschetto et al., 2009; Weng et al., 2009]. Weng et al.  gave a general comparison of the global climatic influences between the two kinds of tropical Pacific Ocean warming events in boreal winter. Chen and Tam  also suggested that the tropical cyclone frequency during El Niño Modoki and during conventional El Niño events varies over different regions of the western North Pacific. Given the importance of the wintertime Southeast Asian rainfall, it is therefore necessary to investigate whether it exhibits different responses to these two types of tropical Pacific Ocean warming events and to make clear the related mechanisms. These issues are investigated here.
 The data set and methods as well as the numerical model used in this study are described in section 2. Section 3 then reports the different wintertime rainfall anomalies in Southeast Asia associated with the conventional El Niño and El Niño Modoki, respectively. In section 4, we elucidate the related mechanisms through observational analysis and a simple baroclinic model. Finally, a summary is given in section 5.
2. Data, Methods, and Model Introduction
 The monthly mean atmospheric data used in this study are the reanalysis data set from the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR), spanning the period 1948 to 2008 [Kalnay et al., 1996]. This data set has a 2.5° × 2.5° horizontal resolution and extends from 1000 to 10 hPa, with 17 vertical pressure levels. The SST used here is the monthly mean Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST) data set. It is a unique combination of monthly globally complete fields of SST and sea ice concentrations on a 1° latitude-longitude grid from 1870 to the present [Rayner et al., 2003]. The rainfall data set includes the monthly global land precipitation data set called Precipitation Reconstruction Over Land (PREC-L), which was produced by the National Oceanic and Atmospheric Administration (NOAA) Climate Prediction Center (CPC) [Chen et al., 2002]. This data set was constructed on a 2.5° × 2.5° grid using the optimal interpolation technique applied to gauge observations at more than 17,000 stations collected in the Global Historical Climatology Network and the Climate Anomaly Monitoring System data sets [Chen et al., 2002]. The analysis has been updated for an extended period, longer than 50 years, from 1948 to the present at the NOAA CPC. To illustrate rainfall over oceans we also use a second rainfall data set, from the Global Precipitation Climatology Project (GPCP), covering the period from 1979 to the present on a 2.5° latitude-longitude grid [Adler et al., 2003].
 Because El Niño Modoki was rarely observed before the 1980s, the time period considered in this study is from 1980 to 2008, so that the numbers of El Niño and El Niño Modoki cases are comparable. Moreover, the availability and quality of data sets are better after 1980. Seasonal means are considered throughout this paper, and they are constructed from the monthly means by averaging the data for December–January–February (DJF), which results in 29 winters, from 1980 to 2008. Here the winter of 1980 refers to the 1979–1980 winter.
 In this study the wintertime Niño3 index and the El Niño Modoki index (EMI) are used to describe the conventional El Niño and El Niño Modoki, respectively; the latter follows the method proposed by Ashok et al. . The definitions of the two indexes are as follows:
where [SSTA]EP is the SST anomalies averaged over the tropical eastern Pacific (150°W to 90°W, 5°S to 5°N), and
where [SST]A, [SST]B, and [SST]C stand for SST anomalies averaged over regions A (165°E to 140°W, 10°S to 10°N), B (110°W to 70°W, 15°S to 5°N), and C (125°E to 145°E, 10°S to 20°N), respectively. Figures 1a and 1b present the normalized Niño3 index and EMI for 29 winters, respectively. The two indexes can successfully capture the distinction of the zonal SST anomaly gradient features (Figures 1c and 1d). Partial correlation and regression analyses are employed throughout this study to exclude the possible influence of one event from that of another. When we test the significance of the results with the two-tailed Student's t test, the degrees of freedom are set to 26. This is the total sample size minus 3, as proposed by Ashok et al. , because rainfall anomalies have little persistence from one year to the next [Ashok et al., 2007].
 The dry baroclinic model used in this study is based on the dynamic core of the Geophysical Fluid Dynamics Laboratory, AGCM [Held and Suarez, 1994; Wang et al., 2003]. It consists of primitive equations linearized by a realistic three-dimensional basic state but retains full nonlinearity in the second-order perturbation terms of the prediction equations. The model has a horizontal resolution of T42 (corresponding to roughly 2.8° × 2.8° in latitude and longitude), and in the vertical there are five uneven σ levels. It uses a form of Laplacian raised to the fourth power in the horizontal diffusion, Newtonian damping, and Rayleigh friction. A basic circulation state and an external forcing should be prescribed to run the model. In this study the basic state is set as the wintertime (DJF) long-term mean for the period from 1980 to 2008 calculated from the NCEP-NCAR reanalysis data set. In the vertical direction the data are linearly interpolated from the original standard pressure levels to the model σ levels. The observed diabatic heating patterns associated with El Niño and El Niño Modoki are prescribed as the external forcing for the model.
3. Rainfall Anomalies Associated With ENSO and ENSO Modoki
 To elucidate the importance of considering the different impacts of the two types of Pacific Ocean warming, we first give a simple comparison of wintertime Southeast Asian rainfall anomalies between the two cases. Three strong conventional El Niño events (1982–1983, 1991–1992 and 1997–1998) and four strong El Niño Modoki events (1979–1980, 1990–1991, 1994–1995 and 2004–2005) are selected according to the criteria that the normalized winter mean (DJF) Niño3 index is >1.0, and the EMI is >0.9. Figure 2 shows the composite winter rainfall anomalies in all seven El Niño winters, three conventional El Niño and four El Niño Modoki winters, and the difference between the latter two. Figure 2 reveals that moderate negative rainfall anomalies are observed over Southeast Asia if we mix all El Niño cases together (Figure 2a). However, the negative anomalies are about twice as strong in conventional El Niño winters (Figure 2b) and much weaker in El Niño Modoki winters (Figure 2c). In addition, a band of positive rainfall anomalies is observed over the subtropics between 20°N and 30°N in conventional El Niño winters (Figure 2b) but not in El Niño Modoki winters (Figure 2c), suggesting different rainfall patterns between the two cases (Figure 2d).
Figure 3 presents the partial regression-correlation map between the Niño3 index-EMI and the land precipitation in boreal winter based on the PREC-L data set. The results also display remarkable differences in the land precipitation anomalies between the two types of tropical Pacific Ocean warming events. In the case of the conventional El Niño significant negative rainfall anomalies are observed in the Philippines, Borneo, Celebes, and Sulawesi (Figure 3a). Positive rainfall anomalies appear in the vicinity of Sumatra and the western part of Borneo, which is out of phase with its surrounding regions owing to the effect of high topography [Chang et al., 2004]. In addition, positive rainfall anomalies are observed in south China, which is consistent with Figure 2b and previous results [e.g., Zhang et al., 1996; Chen et al., 2005]. In contrast, for the El Niño Modoki cases, positive rainfall anomalies are rarely observed over the Maritime Continent. Significant negative rainfall anomalies are located in south China, the southern Indo-China Peninsula, the Malay Peninsula, and the Philippines (Figure 3b). In addition to the differences in the spatial pattern, the negative signals in Figure 3b are much weaker than those in Figure 3a, suggesting a relatively weaker influence of El Niño Modoki on Southeast Asian rainfall compared with the conventional El Niño.
 Because the PREC-L data set contains only land precipitation, we further confirmed the previous results using the GPCP data set, which consists of the precipitation on both the land and the ocean. The results shown in Figure 4 are almost identical to those in Figure 3 but provide more pattern information for better spatial continuity. During conventional El Niño winters, a belt of positive rainfall anomaly stretches from south China to the east of Japan, while negative rainfall anomalies cover the southern South China Sea and most of the tropical western Pacific (Figure 4a). In El Niño Modoki cases, however, significant negative rainfall anomalies are mainly located in the region around the Philippines (Figure 4b). Weak negative rainfall anomalies are also observed in south China and the south of Japan. Therefore, the impacts of the conventional ENSO and ENSO Modoki on Southeast Asian rainfall are significantly different in boreal winter.
4. Possible Mechanism
4.1. Results From Observations
 The responses of Southeast Asian rainfall to ENSO are usually attributed to the anomalous Walker circulation and the associated anomalous anticyclone-cyclone around the Philippines [Wang et al., 2000; Chang et al., 2004], so the circulation anomalies were examined to understand the different impacts of the two types of ENSO events. Figure 5 shows the partial regressions of the 850 hPa divergent winds and velocity potential with the Niño3 index and the EMI, which reflects the anomalous Walker circulation and related large-scale divergent motions. In conventional El Niño episodes the anomalous Walker circulation features a dipolar pattern, with descent (ascent) over the tropical western (eastern) Pacific (Figure 5a). The significant descending center is located around the eastern Maritime Continent. In El Niño Modoki episodes, in contrast, the anomalous Walker circulation exhibits a tripolar pattern [Ashok et al., 2007], with ascent over the tropical central Pacific and descent over the tropical western and eastern Pacific (Figure 5b), respectively. The descending center over the western Pacific is located around the South China Sea, and the strength of this descent is only half that of its counterpart in the conventional El Niño case. This result suggests that both the location and the strength of the Walker Circulation are different during the two types of ENSO.
Figure 6 further depicts the associated outgoing longwave radiation (OLR) anomalies with the two types of ENSO events to elucidate the convection activity over the tropical Pacific. It is clear that, for the conventional El Niño episodes, the convections are enhanced over the central and eastern tropical Pacific and suppressed over the regions around the Maritime Continent (Figure 6a). A thin enhanced convection belt is also observed from south China to the east of Japan. For the El Niño Modoki episodes, enhanced convective activities appear over the tropical central Pacific, with suppressed convective activities in the vicinity of the Philippines and the eastern Pacific (Figure 6b). Comparatively, the convection near the Philippines is much weaker, with a small domain in the El Niño Modoki cases. Generally, the distributions of convection anomalies in Figure 6 are consistent with the anomalous rainfall shown in Figure 4. In addition, the convection over the northwestern Indian Ocean (0° to 20°N, 40°E to 60°E) is enhanced for conventional El Niño cases and suppressed for El Niño Modoki cases, respectively. Watanabe and Jin  proposed that during El Niño, the convective cooling over the Maritime Continent is strongly enhanced by the Indian Ocean ascent, by modification of the Walker circulations. Our results confirm their findings but suggest that their mechanism is not valid for the El Niño Modoki case. The results also imply that the convection and related diabatic heating-cooling around the Maritime Continent may contribute to the different rainfall anomalies in Southeast Asia associated with the two types of ENSO.
Figure 7 shows the partial regressions of 850 hPa winds and partial correlations of vertical p velocity with the Niño3 index and the EMI. During conventional El Niño episodes the anticyclone is located around the southern Philippines and central and eastern Indonesia (Figure 7a). The anomalous vertical p velocity is generally positive in this region, which suggests descending motion and favors less rainfall. To the northwest of the anticyclone, sufficient water vapor can be transported by the anomalous southwesterlies to the areas stretching from northern Southeast Asia to the south of Japan. The negative vertical p velocity suggests ascending motion and favors more rainfall in this region. During El Niño Modoki episodes, in contrast, the anticyclone and related descending center shift northward to about 20°N (Figure 7b). The strength of the anticyclone is relatively weak compared with that in the conventional El Niño case. Consequently, the negative rainfall anomaly center shifts northward, inducing less precipitation over south China, the Philippines, the Indo-China peninsula, and the surrounding areas. Because the subtropical western North Pacific anticyclone around the Philippines is the direct factor through which El Niño influences Southeast Asian rainfall [Wang et al., 2000], the differences in both the strength and the location of this anticyclone can explain well the distinct rainfall responses to these two types of ENSO.
4.2. Results From a Simple Baroclinic Model
 To further illustrate the relationship between the atmospheric circulation anomalies and the heating associated with the two types of ENSO, two experiments are conducted to examine the atmospheric responses to different heating sources with a simple dry baroclinic model. The prescribed heating has a vertical profile in the model. It peaks at 500 hPa (σ = 0.5), with a magnitude of 0.6 K d−1, and decreases toward upper and low levels. The column average of this profile is about 0.25 K d−1, equivalent to the latent heating associated with precipitation of 1 mm d−1. In the horizontal direction the heating distribution is estimated by the observed OLR anomalies (Figure 6). To focus on the heating field that is associated only with ENSO, the OLR anomaly is set to 0 beyond the region (30°S to 30°N, 80°E to 90°W). Furthermore, a triangle truncation at 15 wave numbers (T15) is applied to remove small-scale features and highlight large-scale heating patterns. The processed OLR anomaly is then normalized by −4.5 W m−2, whose absolute value is half the maximum OLR (∼9 W m−2) over the western North Pacific during El Niño Modoki; this normalization results in a nondimensional horizontal distribution. Consequently, the obtained spatial distribution and intensity of heating are basically, in agreement with latent heating released by the observed anomalous precipitation (Figures 3 and 4). The prescribed heating patterns at σ = 0.5 are shown as shaded areas for each experiment in Figures 8 and 9. The heating anomaly is set to be steady with time in the two experiments. A steady state is reached after about 10 days of integration. Here we show the results on day 30.
Figure 8 presents the 850 hPa wind response to the prescribed heating in the conventional El Niño and El Niño Modoki experiments, respectively. The low-level responses indicate the Philippine anticyclone signals in the two experiments. For the conventional El Niño experiment there is a large-scale anticyclone over the Philippines forced by the dipolar heating pattern (Figure 8a). In contrast, the anticyclone in the El Niño Modoki experiment (Figure 8b) is confined to a smaller domain and shifts northward, induced by the tripolar heating pattern, which is in good agreement with the observed results (Figure 7). These results suggest that the atmospheric response to the diabatic heating associated with two types of ENSO can explain well the observed differences in circulation and precipitation. Furthermore, a second experiment was conducted to examine the key diabatic heating source responsible for the formation of the anticyclone. In the experiment we only retain the diabatic cooling over the Maritime Continent, as shown in Figure 9. The simulated results are almost identical to those in the first experiment shown in Figure 8. Thus, the model results suggest that the distinction of the anticyclone results from the difference in the anomalous diabatic heating associated with the two types of Pacific Ocean warming. In addition, the anticyclone around the Philippines is mainly a direct atmospheric response to the diabatic cooling over the Maritime Continent, which is consistent with previous studies [e.g., Wang et al., 2000; Watanabe and Jin, 2002].
 This paper has investigated the different relationships of Southeast Asian rainfall with the conventional El Niño and El Niño Modoki events in boreal winter. The linear partial correlation-regression method is used to isolate the influence from each kind of ENSO. On the basis of the PREC-L and GPCP precipitation data sets, it is revealed that conventional El Niño events generally favor below-normal rainfall across the Philippines, Borneo, Celebes, and Sulawesi and above-normal rainfall over south China. In contrast, for El Niño Modoki events, negative rainfall anomalies are generally observed in south China, the Indo-China peninsula, the Malay peninsula, and the Philippines. Compared with conventional El Niño events, the negative rainfall anomalies over Southeast Asia are relatively weaker and located more northward during El Niño Modoki events.
 The different Southeast Asian rainfall anomalies related to the conventional ENSO and ENSO Modoki can be attributed to the difference in the anomalous Walker circulation and the low-level anticyclone near the Philippines. In the case of the conventional El Niño the anomalous Walker circulation depicts a dipolar pattern, with ascending flow over the eastern Pacific and descending flow over the Maritime Continent. However, the anomalous Walker circulation associated with El Niño Modoki shows a tripolar pattern, with a rising branch in the central Pacific and sinking branches in the eastern Pacific and the area around the Philippines. The OLR anomalies suggest that the suppressed convection over the tropical western Pacific in El Niño Modoki winters is weaker and occupies a smaller domain compared to that in conventional El Niño winters. Hence, the diabatic heating-cooling associated with the two types of Pacific Ocean warming may induce different atmospheric circulation anomalies over the western Pacific. For El Niño episodes an anomalous low-level anticyclone centered near the Philippines controls a large domain including the Philippines, Borneo, Celebes, and Sulawesi where the sinking motion favors less rainfall. The southwesterlies to the northwest of this anticyclone may supply plentiful water vapor to south China, inducing wet conditions there. However, for El Niño Modoki episodes the anomalous anticyclone is weaker and moves northward. Therefore, the negative rainfall anomalies are weaker and shift more northward, to south China, the southern Indo-China peninsula, the Malay peninsula, and the Philippines.
 On the basis of a simple baroclinic model, two numerical experiments were conducted to validate the atmospheric response to the anomalous heating forcing associated with the two types of ENSO. In the first experiment the ENSO-related diabatic forcing in the whole tropical Pacific was used. The simulated atmospheric circulation anomalies are largely consistent with the observations, suggesting that the anomalous low-level anticyclone over the tropical western Pacific is mainly a direct response to the anomalous heating related to ENSO. Furthermore, in the second experiment, with only the convective cooling around the Philippines retained, almost-identical low-level anticyclones can be forced over the tropical western Pacific. Therefore, the differences between the anticyclones are suggested to result from the difference in the diabatic heating associated with the two types of Pacific Ocean warming. The diabatic cooling over the Maritime Continent plays the dominant role in this process.
 We are grateful to two anonymous referees and Steve Ghan for their helpful suggestions. This work was supported by the 973 Program of China (grants 2010CB428603) and the National Natural Science Foundation of China (grants 40775035, 40905026, and 40730952).