As a dominant greenhouse gas in the atmosphere, water vapor has a profound impact on global energy budgets through not only latent heat release upon phase transitions (Held and Soden, 2000) but also cloud formation that reflects longwave radiation from below and shortwave radiation from above (Stevens et al., 2017); thus, water vapor plays a substantial role in the formation and evolution of the climate system. The tropical Pacific is a major convection center and a region with abundant water vapor. Sea surface temperature (SST) anomalies in the tropical Pacific have an important influence on water vapor transport, cloud cover, and precipitation distribution due to the tropical circulation changes caused by El Niño–Southern Oscillation (ENSO). ENSO is characterized by anomalous SST in the tropical Pacific. During ENSO, there is significant precipitation variability in the Euro-Mediterranean (López-Parages and Rodríguez-Fonseca, 2012), Middle East (Sandeep and Ajayamohan, 2018), southwest central Asia (Mariotti, 2007), western Africa (Okazaki et al., 2015), Pacific Ocean (Quartly et al., 2000), and continental USA (Lee et al., 2014). ENSO has an effect on seasonal rainfall in East Asia by inducing a weaker and later onset of the Indian monsoon circulation (Dai and Wigley, 2000; Zhao et al., 2010; Yan et al., 2018). Vertical cloud anomalies in the tropical Atlantic from Aqua Moderate-Resolution Imaging Spectroradiometer are linked to ENSO-induced shift and weakening of the Walker circulation and Hadley cell near the Equator (Madenach et al., 2019). The strong 1997/98 El Niño resulted in cloud structure anomalies and their radiative property changes over the tropical Pacific (Sun et al., 2012) and increased upper tropospheric cirrus over the mid-Pacific but decreased cirrus over Indonesia (Massie et al., 2000). Numerical investigation also indicated that warm-water volume transport and precipitation change are associated with ENSO (Ishida et al., 2008; Hill et al., 2009).

El Niño is generally classified into central Pacific (CP) El Niño, also known as El Niño Modoki, and eastern Pacific (EP) El Niño based on distinct spatial distributions of warming SST anomaly averaged over the Niño4 and Niño3 regions (Ashok et al., 2007; Yu and Kao, 2009; Yeh et al., 2009), respectively. The 2006/07 and 2015/16 events are the EP El Niño because of the stronger SST anomaly during the boreal winter (December to February, as DJF) in the Niño3 region than in the Niño4 region, while correspondingly the 2009/10 and 2018/19 events are categorized as the CP El Niño (Yeh et al., 2009). The two types of El Niño have different effects on precipitation, surface temperature, moisture transport, and carbon cycle over many parts of the world (Weng et al., 2008; Kug et al., 2009; Wang et al., 2013; Yeh et al., 2014; Gu and Adler, 2016; Wang et al., 2018). Su and Jiang (2013) and Takahashi et al. (2013) suggested that the water vapor anomaly over the tropical ocean was mainly controlled by thermodynamic process during the 2006/07 EP El Niño but by both dynamic and thermodynamic processes during the 2009/10 CP El Niño.

The EP El Niño in 2015/16 winter is one of the strongest ENSO events on record. Compared to the strong 1982/83 and 1997/98 El Niños, the 2015/16 El Niño shows distinct aspects that indicate that the largest SST anomalies are extended toward the central Pacific (Paek et al., 2017; L'Heureux et al., 2017). Due to their unusual characteristics, the global effects of the 2015/16 event have attracted much attention. Palmeiro et al. (2017) proposed that an early stratospheric final warming over the polar region and anomalous precipitation over southern Europe in 2016 were related to the 2015/16 super El Niño. Li et al. (2018) revealed that the combined effect of the 2015 ENSO warm phase and Madden-Julian Oscillation (MJO)-4 index negative phase caused a significant deficit of precipitation on the Canadian Prairies in May and June 2015. A striking freshwater anomaly was observed in the equatorial Pacific during the onset of the 2015/16 event (Gasparin and Roemmich, 2016), and rainfall δ18O in southern Papua was generally enriched by 1.6 ‰–2 ‰ during the 2015 El Niño more than during the 2013/14 ENSO-normal period (Permana et al., 2016). Owing to convection anomaly during the 2015/16 El Niño, water vapor in the tropical lower stratosphere was increased by hydration of the lower stratosphere through convectively detrained cloud ice (Avery et al., 2017), and quasi-biennial oscillation in the tropical stratospheric wind was disrupted because of dramatic relocation of deep convection (Dunkerton, 2016; Newman et al., 2016). Hence, the 2015/16 El Niño had important influences on the circulation and composition transport and the mass exchange between the troposphere and stratosphere. In this paper, we investigate the water vapor anomaly over the tropical western Pacific in the CP and EP El Niño events from radiosonde and satellite observations, particularly the extreme anomaly in the 2015/16 super El Niño winter, and explore the contributions of the tropical Hadley, Walker, and monsoon circulation changes to the observed water vapor anomalies in the different El Niño events.

The data used are briefly described in Sect. 2. In Sect. 3, water vapor anomalies in four El Niño winters are presented, and the relationship between the ENSO intensity and the water vapor anomaly at the observational stations is explored. In Sect. 4, we decompose the tropical circulation into the Hadley, Walker, and monsoon circulation components and estimate the roles of these circulations in the water vapor variation. Tropical cloud and outgoing longwave radiation (OLR) are investigated in Sect. 5. A discussion of the water vapor data quality is provided in Sect. 6. Finally, we summarize the results in Sect. 7.

In present study, we investigate the atmospheric water vapor by using radiosonde observations at five tropical stations for 15 years from January 2005 to December 2019, which are provided by the National Oceanic And Atmosphere Administration (NOAA) at the following website: https://www.ncei.noaa.gov/pub/data/igra/derived/ (last access: 6 September 2021). The five radiosonde stations are at Koror (7.33∘ N, 134.48∘ E), Yap (9.48∘ N, 138.08∘ E), Guam (13.55∘ N, 144.83∘ E), Truk (7.47∘ N, 151.85∘ E), and Ponape (6.97∘ N, 158.22∘ E), located in the western Pacific warm pool. A balloon was launched twice daily at 00:00 and 12:00 UTC, and during balloon ascent the sensing payload on the balloon can obtain many meteorological parameters, such as atmospheric pressure, temperature, relative humidity, and wind speed and direction. We plot daily temperature, relative humidity, and wind speed time series observed by radiosonde to identify potential outliers, and then the high resistant asymmetric biweight technique is applied to weed out the outliers (Lanzante, 1996). The outlier data are very few, and the outliers of temperature, wind, and relative humidity account for only 0.09 %, 0.08 %, and 0.02 % of all observational data at the five stations over 15 years, respectively. The radiosonde data are linearly interpolated to a vertical grid of 50 m, and the interpolated data below 10 km are utilized to analyze the atmospheric water vapor variation. The burst height of the balloon is usually more than 30 km; thus, the data availability below 10 km is high. In the period that we focus on, the data are missing for about 4, 2, 1, and 4 months over Yap, Guam, Truk, and Ponape, respectively, and they are almost entirely missing from the several continuous observational data rather than the balloon burst data below 10 km.

Specific humidity can be derived from the profile of meteorological parameters observed by radiosonde. The saturated vapor pressure es is calculated according to a modified version of the Magnus formula as follows (Murray, 1967): 1es=6.1078×exp⁡17.269T-273.16T-35.86, where T is the temperature in units of K. Following this, the specific humidity q (g kg-1) is determined from the following equations: 2e=RH×es,3q=0.622ep-0.378e, where e is the vapor pressure, RH is the relative humidity, and p is the pressure in units of hPa.

In addition, we use the monthly specific humidity and horizontal winds from the surface to 300 hPa during the period of 2005–2019, obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis data, to investigate the water vapor anomaly and tropical atmospheric circulation in the region of the radiosonde stations. The reanalysis data are produced by a sequential 4D variational data assimilation scheme, with a latitudinal and longitudinal resolution of 0.25∘ × 0.25∘ at 37 pressure levels from 1000 to 1 hPa (Hersbach et al., 2020). The data are available at the website of https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-pressure-levels-monthly-means?tab=form (last access: 6 September 2021).

To assess the atmospheric water vapor as compared to the reanalysis data and the radiosonde observations, a further evaluation is carried out using Aqua atmospheric infrared sounder (AIRS) water vapor mass mixing ratio data from 2005–2019. AIRS is a hyperspectral infrared spectrometer orbiting on the National Aeronautics and Space Administration (NASA) Aqua spacecraft launched in May 2002, which can provide accurate measurements of temperature, moisture, and other atmospheric variables (Aumann et al., 2003). The data used here are water vapor vertical profiles from Level 3 monthly standard gridded retrieval product version 6, AIRS3STM (Susskind et al., 2014), which are available at https://disc.gsfc.nasa.gov/datasets/AIRS3STM_006/summary?keywords=AIRS3STM (last access: 6 September 2021). The water vapor data contain eight levels from 1000 and 300 hPa with a latitudinal and longitudinal grid of 1∘ × 1∘ derived from the average of two observations in two orbital overpasses per day. The ascending and descending orbits have an equatorial crossing time at 13:30 local time (LT) and 01:30 LT, respectively.

The oceanic Niño index (ONI) is applied to discuss the correlation between the ENSO and the observed water vapor anomaly. ONI is the measurement of ENSO strength, which is provided by the NOAA at https://catalog.data.gov/dataset/climate-prediction-center-cpcoceanic-nino-index (last access: 6 September 2021). The ONI is defined as a 3-month moving average of extended reconstructed sea surface temperature (ERSST) V5 sea surface temperature anomalies in the Niño3.4 region at 5∘ N–5∘ S and 120–170∘ W (Huang et al., 2017).

Cloud occurrence probability and OLR flux are also examined since they are sensitive to water vapor variation (Stevens et al., 2017; Soden et al. 2008). The OLR data are measured by the NOAA-18 satellite, which travel in sun-synchronous orbit with a 13:55 LT equatorial crossing time (Kramer, 2002). We use the monthly OLR data between 2005 and 2019 from the NOAA archives with a latitudinal and longitudinal grid of 2.5∘ × 2.5∘ (Liebmann and Smith, 1996), which can be accessed through the website of https://psl.noaa.gov/data/gridded/data.interp_OLR.html (last access: 6 September 2021). The Cloud Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) satellite is able to clearly identify cloud vertical structure (Winker et al., 2007). The satellite has a sun-synchronous orbit with an equatorial crossing time around 01:30 and 13:30 LT (Stephens et al., 2002). Here, we use the CALIPSO version 1.00 lidar level 3 cloud occurrence monthly data in a latitudinal and longitudinal grid of 2∘ × 2.5∘ with an altitude resolution of 60 m above the mean sea level, and the available data are from June 2006 to December 2016, downloaded from the website of the NASA at https://asdc.larc.nasa.gov/project/CALIPSO/CAL_LID_L3_Cloud_Occurrence-Standard-V1-00_V1-00 (last access: 6 September 2021)

Using the cloud occurrence from the CALIPSO from June 2006 to December 2016, we calculate tropical cloud fraction between 0 and 15∘ N in the 2006/07, 2009/10, and 2015/16 winters and its climatic mean in winter, which is shown in Fig. 11. We also compute the OLR anomalies over 30∘ S–30∘ N in the four El Niño winters based on the monthly OLR data between 2005 and 2019. Figure 12 shows the OLR anomalies in the four El Niño events. In the western Pacific, the strong rising flow carries abundant water vapor to high level due to the convergence of horizontal wind field in winter, as shown in Fig. 4, and then the water vapor condenses to form clouds as it cools, and thus there are clouds over the tropical western Pacific. In the El Niño events, the cloud amount decreases from about 80 to 160∘ E but tends to increase between about 160∘ E to 120∘ W because of the tropical circulation changes. Owing to the reflection effect of cloud on OLR, the OLR change is opposite to the variation of cloud amount. In the 2009/10 and 2015/16 strong El Niño winters, the OLR is obviously enhanced in the tropical northwest Pacific and significantly reduced in the equatorial central eastern Pacific as the cloud occurrence changes. Hence, the cloud and OLR have a clear response to the water vapor anomaly in the El Niño events.

As described above, the reanalysis CWV anomalies at the radiosonde stations in the 2009/10 winter have almost the same intensity as that in the 2015/16 winter, but the radiosonde observation indicates that the water vapor reduction is evidently lower in the 2009/10 winter than in the 2015/16 winter. As shown in Figs. 11 and 12, the satellite observation shows that there is less cloud occurrence and more OLR at the radiosonde stations in the 2015/16 winter compared with in the 2009/10 winter. Therefore, this supports the radiosonde observation that the water vapor over the radiosonde stations in the 2009/10 winter may be moister than in the reanalysis.

In the ERA5 reanalysis data, water vapor is calculated by a humidity analysis scheme introduced by Hólm (2003), which involves nonlinear transformation of the humidity control variable to render the humidity background errors nearly Gaussian. The transformation normalizes relative humidity increments by a factor that varies as a function of background errors of relative humidity and vertical level (Dee et al., 2011). For the ERA5 humidity analysis, measurements from radiosondes, surface synoptic observations, aircraft, and satellite observations are assimilated (Andersson et al., 2007). To date, the reliability and accuracy of ERA5 water vapor products have been extensively estimated. Overall, ERA5-retrieved precipitable water vapor (PWV) performs well over the Indian Ocean (Lees et al., 2020), central Asia (Jiang et al., 2019), the Antarctic (Ye et al., 2007), the East African tropical region (Ssenyunzi et al., 2020), and Varanasi (Kumar et al., 2021), which can be seen via comparisons with ground-based observations, satellite retrievals, and other reanalysis datasets. Nevertheless, some discrepancies can be noticed over small tropical islands characterized by steep orography (Lees et al., 2020), and it is reported that although PWV from the ERA5 reanalysis is in good agreement with the retrievals from the Global Navigation Satellite System over 268 stations, there is a bias of 4 mm PWV in the southwest of South America and western China due to the terrain limitations and fewer observations (Wang et al., 2020).

Since the CWV anomalies look more or less different between the radiosonde and reanalysis data, we compare the CWV in the ERA5 reanalysis with that in the radiosonde and satellite observations at the five stations and attempt to explain the different CWV anomalies between the reanalysis data and radiosonde observations in the 2006/07 and 2009/10 events. By using the reanalysis data and measurements of radiosondes and AIRS on the Aqua satellite for the 15-year period from 2005 to 2019, we calculate the monthly mean CWV at the five radiosonde sites, and Fig. 13 depicts the monthly mean CWV in winter as scatterplots of the reanalysis vs. radiosonde data and the reanalysis vs. AIRS data. Following this, the climatic mean difference is derived from these monthly mean CWV series in 2005–2019, which is also presented in Fig. 13. At the five stations, the monthly mean CWV in winter is distributed between 30 and 60 kg m-2 in all three datasets, and the CWV is obviously shifted to the low values in the El Niño winter, indicating the negative anomaly in the El Niño event. The correlation of the mean CWV series between the reanalysis and observations is quite high with the minimum coefficient of 0.88, and the root mean square (rms) of the mean CWV differences between the reanalysis and observations is less than 2.32 kg m-2. Meanwhile, the difference in the climatic mean CWV is mainly concentrated in the range of 0–2 kg m-2, except for several months at the Guam station, and thus the relative difference in the monthly mean CWV between the reanalysis and observations is generally smaller than 5 %. These comparisons and analyses confirm a fine confidence level of the ERA5 reanalysis and observational datasets. Nevertheless, there are still very small discrepancies among these data, and the discrepancy is relatively large between the radiosonde and reanalysis data compared to between the satellite and reanalysis data, which may be attributed to the different sampling times between the radiosonde and AIRS. It can be noted from Fig. 13 that the red dots representing the reanalysis vs. radiosonde data in the 2009/10 winter show a relatively large scatter around the symmetric axis, indicating a relatively large discrepancy in the CWV anomalies between the reanalysis data and radiosonde observation in this event, as in previous reports of some discrepancies over small tropical islands or in the regions with fewer observations (Lees et al., 2020; Wang et al., 2020). In comparison to the reanalysis data, the CWV derived from AIRS also shows the largest difference of 1.31 kg m-2 in the 2009/10 event, while the differences are less than 1 kg m-2 in the other three events.

Based on specific humidity in the reanalysis and radiosonde data, the CWV is calculated to be 44.87 (44.10), 43.06 (40.23), 41.16 (39.83), and 44.07 (42.87) kg m-2 in the radiosonde (reanalysis) data in the 2006/07, 2009/10, 2015/16, and 2018/19 events, respectively. In fact, the relative difference in the CWV between the radiosonde and reanalysis data is very small, with only 1.7 % in the 2006/07 winter and 6.6 % in the 2009/10 winter. The CWV average in winter is 45.61 (44.17) kg m-2 in the radiosonde (reanalysis) data from 2005 to 2019, and thus the CWV anomaly in the radiosonde (reanalysis) data is -0.74 (-0.07) kg m-2 in the 2006/07 event and -2.55 (-3.94) kg m-2 in the 2009/10 event. This means that the discrepancy in the CWV anomaly looks quite large in Fig. 9, especially in the 2006/07 event, but the differences in both the CWV and CWV anomaly values are small between the radiosonde and reanalysis. Even so, the relatively large discrepancy between the reanalysis data and the radiosonde and AIRS observations in the 2009/10 event, as shown in Figs. 1 and 13, and the cloud and OLR measurements in Figs. 11 and 12 seem to suggest that the reanalysis data underestimates the tropospheric water vapor over the radiosonde stations in the 2009/10 winter.

In this paper, we report the significantly negative water vapor anomaly in the troposphere during four El Niño winters at five radiosonde stations in the tropical western Pacific based on radiosonde and reanalysis data for 15 years from 2005 to 2019 and study the relationship between the water vapor anomaly and the El Niño index and the contribution of the different tropical circulation anomalies to the observed water vapor anomaly in the El Niño events.

The radiosonde observation shows that the negative water vapor anomaly arises in the El Niño winters, specifically showing an extremely negative anomaly in the 2015/16 super El Niño event. The prominent specific humidity anomaly is concentrated below 8 km in the troposphere with the peak at the height of about 2.5–3.5 km. The local CWV anomaly has a large negative correlation coefficient of -0.63 with the ONI in the Niño3.4 region but has a lag of about 2–3 months. The reanalysis data reveal that the negative water vapor anomaly occurs widely in the tropical northwestern Pacific, while the positive anomaly correspondingly takes place in the equatorial central eastern Pacific. The 2015/16 El Niño event (with ONI = 2.6) is the strongest during the 15 years, leading to the extreme anomaly in the water vapor over the tropical Pacific.

The atmospheric water vapor from tropical sea water evaporation is affected not only by the SST but also by the vertical motion of the atmosphere, which can transport the water vapor from near the sea surface up to high levels. By using the definitions and method introduced by Tanaka et al. (2004), we decompose the tropical circulation into the Hadley, Walker, and monsoon circulations to estimate their contributions to the observed water vapor anomaly in the four El Niño events. In general, the tropical circulation anomaly in the El Niño winter is characterized by divergence (convergence) at 850 hPa in the tropical western (eastern) Pacific, and thus the CWV decreases over the tropical western Pacific as the ascending flow is suppressed. As the large-scale meridional circulation is driven by the differential heating, the variation of the Hadley circulation is pretty small with an anomaly index of less than 2 units. At the radiosonde stations, the anomaly of the Walker circulation makes a considerable contribution to the total anomaly in all the El Niño winters, especially in the 2006/07 and 2015/16 EP El Niño events. The monsoon circulation exhibits an obvious variability from one event to another, and its anomaly is large in the 2009/10 and 2018/19 CP El Niño winters and small in the 2006/07 and 2015/16 EP El Niño winters. Therefore, the observed water vapor anomaly is caused mainly by the Walker circulation anomaly in the 2015/16 super EP event but by the monsoon circulation anomaly in the 2009/10 strong CP event. Based on the reanalysis data back to 1979, we examine the general features of water vapor and circulation anomalies in the two types of composite El Niño events. The roles of the Hadley, Walker, and monsoon circulations in the composite EP and CP El Niños are consistent with those in the EP and CP case events.

Because of the reduction in the upward transport of water vapor over the tropical western Pacific in the El Niño events, the satellite observations show that, relative to the climatic means, the cloud decreases and that the OLR is accordingly strengthened, particularly during the strong El Niño winters of 2009/10 and 2015/16. In addition, a detailed comparison of water vapor in the reanalysis, radiosonde, and satellite data shows a high confidence level of these datasets; nevertheless, the reanalysis seems to slightly underestimate the water vapor over the five radiosonde stations in the 2009/10 winter.