Indo-Pacific Warm Pool Area Expansion, Modoki Activity, and Tropical Cold-Point Tropopause Temperature Variations

The tropical cold-point tropopause temperature (CPTT), a potentially important indicator of global climate change, is of particular importance for understanding changes in stratospheric water vapor levels. Since the 1980s, the tropical CPTT has shown not only interannual variations, but also a decreasing trend. However, the factors controlling the variations in the tropical CPTT since the 1980s remain elusive. The present study reveals that the continuous expansion of the area of the Indo-Pacific warm pool (IPWP) since the 1980s represents an increase in the total heat energy of the IPWP available to heat the tropospheric air, which is likely to expand as a result. This process lifts the tropical cold-point tropopause height (CPTH) and leads to the observed long-term cooling trend of the tropical CPTT. In addition, our analysis shows that Modoki activity is an important factor in modulating the interannual variations of the tropical CPTT through significant effects on overshooting convection.

1980s are related to El Niño Modoki events. These uncertainties, along with observed climate change, motivated the present analysis of the trend and interannual variations in the tropical CPTT since the 1980s.

Results
An increase (decrease) in the tropopause height corresponds to a cooling (warming) of the tropopause 9,21 . The tropical-averaged CPTH shows a strong increasing trend and significant anti-correlation (r 5 20.79) with the decreasing tropical-averaged CPTT since the 1980s (Fig. 2a). This suggests that the variations in the tropicalaveraged CPTT since the 1980s have largely been determined by the tropical-averaged CPTH. It is interesting that the regions with the largest decrease in tropical CPTT since the 1980s are concentrated over the Indo-Pacific warm pool (IPWP) (Figs. 2b and c). The IPWP area has been increased significantly since 1980 22,23 . It implies that the tropical-averaged CPTH trend, which modulates the tropical CPTT trend, is probably associated with IPWP area changes.
To evaluate the potential impact of changes in the IPWP area, we first define a simple energy index (EI), which approximates the enthalpy changes over a region of sea surface, as follows: where S is the area; C is the heat capacity of seawater (4096 J/(kg?K)); l and Q are the longitude and latitude coordinates, respectively; and dS 5 a 2 ? cosQ ? dldQ, where a is the radius of the earth. Here, we calculated the EI over the IPWP area, known as EI (IPWP) . The IPWP is defined as the area where the SST is higher than 28uC between 30uS-30uN and 50uE-180uE-140uW, based on HadISST data. The tropical-averaged CPTH and EI (IPWP) are strongly correlated, and both show strong increasing trends since the 1980s (Fig. 3a). However, the linear trends in tropical-averaged CPTH and EI (IPWP) are less distinct between 1950 and 1980, and the correlation between them is weak (Fig. 3b). The 30-year sliding correlations of tropicalaveraged CPTH and EI (IPWP) between 1950 and 2010 clearly show that the correlation has steadily strengthened since the 1980s (Fig. 3c). This implies a drift in tropopause changes since the 1980s, possibly related to the areal expansion of the IPWP. Furthermore, IPWP expansion, which leads to variations in the tropical-averaged CPTH, may be an important factor influencing the variations and trends in the tropical-averaged CPTT since the 1980s.
The thermal convection and air expansion caused by a heating source can contribute significantly to an increase in tropopause height 24,25 . The IPWP is a heating source, but the question is whether its areal enlargement enhances thermal convection or air expansion to a degree sufficient to increase the tropical CPTH. This question is addressed here through statistical analysis and model simulations. Figure 4a shows time series of tropical-averaged CPTH and the tropical-averaged vertical velocity at 100 hPa (v 100 ), which represents the changes in convection at the tropopause. The CPTH has an evident increasing trend but the v 100 does not, and the correlation between them is weak, suggesting that thermal convection may not be the primary mechanism responsible for CPTH lifting. To further investigate the relationship between tropical CPTH lifting and the areal expansion of the IPWP, we perform two time-slice experiments integrated by the CAM3 model. Experiment 1 (E1) is the control experiment and Experiment 2 (E2) is the sensitivity experiment. The SST used in both experiments is the monthly mean climatology for 1986-2010, except that in E2, the IPWP area is expanded by ,25% in the region 30uS-30uN and 50uE-180uE-140uW to simulate the ,25% expansion of the IPWP area that occurred between 1980 and 2010. Figure 4b shows the expanded IPWP area in E2 compared with E1 (for further details regarding the simulations, see the Methods section). Table 1 lists the differences in annual tropical averaged CPTH, CPTT, and v 100 between E2 and E1 (i.e., E2 -E1). The results indicate that warm pool area expansion can indeed significantly lift the tropical-averaged CPTH, resulting in the decrease in CPTT. However, the difference in v 100 between the simulations is not significant, supporting the notion that enhanced thermal convection may not be the primary cause of tropical CPTH lifting; instead, air expansion caused by enlargement of the IPWP area is likely to be the key factor in lifting the tropical CPTH.
Analysis of the correlation between SST and CPTT in each grid cell across the entire tropics for the time series between 1981 and 2010 showed no significant correlation over the western Pacific (Fig. 1). This is because the method considers only the effect of SST on the tropical-averaged CPTT in a single grid cell. However, we considered the total effect of SST in all grid cells throughout the IPWP on the tropical-averaged CPTT. This is why EI (IPWP) has a significant impact on tropical CPTT variations.
An empirical orthogonal function (EOF) analysis of the variability in tropical CPTT shows that the leading principal component (PC1), accounting for 70% of the variance, is associated with tropical CPTH variations, which are themselves related to warm pool area variations  Figure 5b shows the EOF1 pattern, which is in agreement with the CPTT trend distribution during the past 30 years (Fig. 2b). PC2, which accounts for 8% of the variance, is significantly correlated with the NINO3 index (Fig. 5c). The EOF2 pattern is consistent with tropical CPTT anomalies during canonical ENSO events (Fig. 5d). PC3 accounts for 6% of the variance, and is strongly correlated with the QBO index (Fig. 5e). The corresponding EOF3 pattern is shown in Fig. 5f. Several previous studies performed EOF analyses on the tropopause temperature. Based on tropical tropopause temperature from ECMWF reanalyses, the leading modes are associated with the canonical ENSO and QBO 26 . However, as we discuss in the Data subsection of the Methods section, the variations in the ECMWF tropopause temperature since the 1980s are not in good agreement with the observations, suggesting that the tropical tropopause temperature in ECMWF reanalyses have biases. A previous study based on NCEP1 data found that the first leading mode of tropical CPTT variations from 1979 to 1999 is associated with canonical ENSO 13 . However, the zonal mean signal was removed from CPTT in that study. Figure 6a shows that the interannual variations in the tropicalaveraged CPTT are anti-correlated with the interannual variations in the tropical-averaged CPTH. However, the correlation coefficient decreases to 20.53 (the correlation coefficient between CPTT and CPTH variations is 20.79, Fig. 2a). This implies that there are factors other than the tropical-averaged CPTH interannual variations that are also important in controlling the interannual variations in the tropical-averaged CPTT. To uncover these processes, we separated the interannual variations of the tropical-averaged CPTT into two parts: one caused by tropical-averaged CPTH interannual variations [CPTT (H) ], and the other unrelated to tropical-averaged CPTH interannual variations [CPTT (O) ] (see the Methods section for details). Divergence of the vertical eddy heat flux near the tropopause caused by convection can warm or cool the tropopause, which controls the interannual variations in tropopause temperature 27 . v 100 interannual variations are strongly correlated with the interannual variations in tropical-averaged CPTT (O) (Fig. 6b), suggesting that the interannual variations in the tropical-averaged CPTT are caused mainly by an integrated effect of both tropical-averaged CPTH and convection interannual variations (Fig. 6c).
The interannual variations in EI (IPWP) are strongly and significantly correlated with the interannual variations in the tropical-averaged CPTH (Fig. 6d). This implies that the interannual variations in EI (IPWP) modulates the interannual variations in the tropical-averaged CPTH, which then partly affects the interannual variations in the tropical-averaged CPTT (Fig. 6a). The second leading mode of SST tropical variations from 1980 to 2010 relate to the new type of ENSO; namely, El Niño Modoki 16,28 . We find that v 100 interannual variations are significantly correlated with the Modoki index from 1980 to 2010 (Fig. 6e). El Niño (La Niña) Modoki events correspond to strong (weak) convection, positive (negative) divergence of the vertical eddy heat flux, and a warm (cool) tropopause. That is, El Niño Modoki events result in a positive temperature anomaly in the tropical upper troposphere and tropopause (Fig. 6f), and modulate convection at the tropopause level, which affects interannual variations in the tropical-averaged CPTT. This analysis illustrates that the interannual variations in the IPWP area and Modoki activity may be the primary factors in modulating the interannual variations in the tropical-averaged CPTT.
An EOF analysis of the interannual variability in the tropical CPTT shows that PC1 accounts for 50% of the variance and is associated with the integrated effect of the interannual variations in IPWP area and Modoki activity (Fig. 7a). The EOF1 pattern (Fig. 7b) shows that the maximum anomalies are located in the IPWP region. PC2, which accounts for 10% of the variance, is significantly correlated with the NINO3 index (Fig. 7c). The EOF2 pattern is consistent with CPTT anomalies during canonical ENSO events 2,13,15 (Fig. 7d). PC3 accounts for 6% of the variance, and is strongly correlated with the QBO index (Fig. 7e). The corresponding EOF3 pattern is shown in Fig. 7f. The explained variances imply that the integrated effect of IPWP area and Modoki activity on tropical CPTT interannual variations may be more significant than that of the canonical ENSO and the QBO since the 1980s.
Although both canonical El Niño and El Niño Modoki are related to SST anomalies over the tropical Pacific, the present results suggest that El Niño Modoki has a more significant influence on the CPTT than canonical El Niño. Figure 8a shows the monthly SST anomalies, based on HadISST data, over both the equatorial eastern Pacific (5uN-5uS, 150u-90uW), where they are used to define canonical El Niño events, and the equatorial central Pacific (5uN-5uS, 160uE-150uW), where they are used to define El Niño Modoki events. It is apparent that during the past 30 years the amplitudes of SST anomalies are similar overall during the two types of El Niño events, except for 1997. However, the sea surface background temperatures over the two regions are significantly different; specifically, over the equatorial central Pacific they are 28-30uC, but over the equatorial eastern Pacific they are less than 26uC (Fig. 8b). The SST anomalies can cause convective activity to increase sharply over regions with sea surface background temperatures above 26uC [29][30][31][32] , which implies that convective activity is much stronger during El Niño Modoki events than during canonical El Niño events. Figures 8c and 8d show the composite anomalies of the overshooting number, which can reach 15 km (close to the tropopause level in the tropics) during both El Niño Modoki and canonical El Niño, according to Tropical Rainfall Measuring Mission (TRMM) observational data (the calculation of the overshooting number and composite anomalies is explained in the Methods section). The composite results are in agreement with the above analysis. The different intensities of convection give rise to different divergences of the vertical eddy heat flux, which lead to differences in adiabatic heating at the tropopause. This may be the reason that El Niño Modoki has a more significant influence on the tropical CPTT than canonical El Niño. It is important to note that the patterns of deep convection anomalies are similar during the two types of El Niño event 20 . This suggests that both El Niño types generate similar levels of lower-middle

Discussion
From the present analysis, we indicate that the variations in the tropical-averaged CPTT since the 1980s have been caused mainly by an integrated effect of the areal expansion of the IPWP and El Niño Modoki activity. The impact of this integrated effect on the tropical-averaged CPTT may be more significant than that of the canonical ENSO and the QBO from 1980 to 2010. The continuous areal expansion of the IPWP during the past three decades has caused the decreasing trend of the tropical-averaged CPTT. This expansion represents an increase in the total heat energy of the IPWP that is available to heat the troposphere, likely causing expansion of the air, lifting of the tropical CPTH, and a cooling trend in the tropical CPTT. The data analyzed here suggest that the interannual variations in the tropical-averaged CPTT are caused mainly by interannual variations in the IPWP area and Modoki activity. Modoki events can evidently change the distribution and intensity of overshooting convection, which influences the divergence of the vertical eddy heat flux 27 . This in turn changes the temperature in the upper troposphere and tropopause.
Our findings suggest that tropical CPTT changes are closely related to climate change. Within the context of present and future global warming, the IPWP area is expanding 22,23 and the frequency of Modoki events is increasing 16,28 . The implication, therefore, is that IPWP expansion and the occurrence of Modoki events will continue to be the main factors affecting the trend and interannual variations of the tropical CPTT. This study will be helpful for the prediction of future changes in the tropical CPTT, and will provide a good understanding of the variations in future stratospheric water vapor. This study focused on the effect of changes in the areal extent of the IPWP on CPTT. However, further work, looking in detail at the controls on variations in the IPWP and considering why the area of the IPWP changes (see Fig. 3a) is still needed.

Methods
In this study, the variations are calculated by removing the seasonal cycle from the original time series. Interannual variations are calculated by removing the linear trend and seasonal cycle from the original time series. All time series have been normalized and three-point running averages have been obtained.
Data. In this analysis, the meteorological fields (temperature, height, and winds) are based on monthly mean National Center for Atmospheric Research first generation (NCEP1) reanalysis data for the period 1951-2010 (http://www.esrl.noaa.gov/psd/ data/reanalysis/reanalysis.shtml). Figure 9 shows the variations in the tropicalaveraged CPTTs since the 1980s based on National Center for Atmospheric Research first (NCEP1) and second (NCEP2) generation data, compared with those from the Radiosonde Innovation Composite Homogenization (RICH) 33 (Fig. 9a), the Microwave Sounding Unit (MSU) and the Advanced Microwave Sounding Unit (AMSU) (https://climatedataguide.ucar.edu/climate-data/msuamsu-atmospherictemperature-climate-data-record-remote-sensing-systems-rss.) (Fig. 9b), and the Global Ozone Chemistry and Related trace gas Data Records for the Stratosphere (GOZ) (http://disc.gsfc.nasa.gov/datacollection/GozSmlpT_V1.shtml) (Fig. 9c). It should be noted that the CPTT from MSU data is the brightness temperature of the lower stratosphere, which basically represents the variations in CPTT. The CPTT variations from NCEP1 data are in good agreement with NCEP2 and observed CPTTs variations. In our study, we analyzed the spatial distribution of the tropical CPTT from 1951 to 2010. Since there are no NCEP2 data from 1951 to 1978 and the observed data do not include complete information on spatial distributions of the CPTT, we used NCEP1 data. The variations in CPTT from ECMWF reanalysis since the 1980s is not in good agreement with observations (Fig. 9b, d and f). First, the correlations of tropical-averaged CPTTs between observations and NCEP reanalysis are much stronger than that between observations and ECMWF reanalysis. Second, the trends of tropical-averaged CPTTs from observations and NCEP reanalysis are positive, while the trend from ECMWF reanalysis is negative. This means that tropopause temperature in ECMWF reanalysis may have biases, but the NCEP1 tropopause temperature is suitable for this study.
Model and simulations. The CAM3 model has a longitude-latitude resolution of 2.8u 3 2.8u with 26 levels extending from the surface to 4 hPa and a vertical resolution of about 2 km in the tropopause region. Previous studies have shown that the CAM3 can simulate a relatively realistic tropopause and that the common tropopause cold bias problem has been almost eliminated in the model 34 . The SST used in experiments is observed data 35 . All experiments were run for 33 years, where the first 3 years are spin-up and only the remaining 30 years are used for the analysis.
Statistical significance of correlations. Following Li et al. 36,37 , the statistical significance of the correlation between two auto-correlated time series was determined via a two-tailed Student's t-test using the effective number (N eff ) of degrees of freedom (DOF) 38 , which can be estimated by: where N is the sample size, and r XX and r YY are the autocorrelations of two sampled time series, X and Y, at the time lag j, respectively.
Separating the CPTT interannual variations. Following Feng et al. 39 , the interannual variations in the tropical-averaged CPTT was split into two components as follows: Here      NINO3 and Modoki indices. The monthly NINO3 index and the Modoki index were used to identify monthly occurrences of canonical El Niño events and El Niño Modoki events, respectively. The NINO3 index is defined as the area mean SSTA over the region (5uS-5uN, 150u-90uW), while the Modoki index is defined as follows 28 : where the subscripted brackets represent the area mean SSTA over the central Pacific   www.nature.com/scientificreports