Seasonal and interannual ( ENSO ) climate variabilities and trends in the South China Sea over the last three decades

1 We present a short overview of the long-term mean and variability of five Essential Climate 2 Variables observed in the South China Sea over the last 3 decades, including sea surface 3 temperature (SST), sea level anomaly (SLA), precipitation (P), surface wind and water 4 discharge (WD) from the Mekong and Red Rivers. At the seasonal time scale, SST and SLAs 5 increase in the summer (up to 4.2°C and 14 cm, respectively), and P increases in the north. 6 The summer zonal and meridional winds reverse and intensify (mostly over the ocean), and 7 the WD shows positive anomalies. At the interannual time scale, each variable appears to be 8 correlated with El Niño Southern Oscillation (ENSO) indices. Eastern Pacific El Niño events 9 produce basin-wide SST warming (up to 1.4°C) with a 6-month lag. The SLAs fall basin-wide 10 (by up to 9 cm) during an El Niño event, with a 3-month lag. The zonal and meridional winds 11 strengthen (up to 4 m/s) in the north (weaken in the south) during all types of El Niño 12 events, with a 3-5-month lag. A rainfall deficit of approximately 30% of the mean occurs 13 during all types of El Niño phases. The Mekong River WD is reduced by 1/3 of the mean 7-8 14 months after all types of El Niño events. We also show increasing trends of SST as high as 15 0.24°C/decade and SLAs by 41 mm/decade. Increasing trends are observed for zonal wind, 16 which is possibly linked to the phase of the Pacific Decadal Oscillation, and decreasing trends 17 are observed for P in the north and both WD stations that were analyzed. The likely driving 18 mechanisms and some of the relationships between all observed anomalies are discussed. 19 20 21 22 23 24 25 26 27 28


Introduction
The South China Seas (hereafter, SCS) is the largest marginal sea in Southeast Asia and covers an area of 3.8 million km² from approximately 0-23°N and 99°E-121°E (Fig. 1).This semi-closed basin is surrounded by South China, the Indochina Peninsula, Borneo and the Philippines.It is open to the East China Sea through the Taiwan Strait, to the Pacific through the Luzon Strait and to the Indian Ocean through the Malacca Strait and other narrow straits.The SCS is the second busiest maritime route in the world; its topography is rather complex with an average depth of 2000 m and maximum depths reaching 5000 m in the northeast.There are wide and shallow continental shelves in the northwest and southwest of the SCS and numerous islands such as Hainan, Paracel and Spratly.Located in the tropical Northern Hemisphere, the SCS region is influenced by both a tropical and a subtropical climate, by the four adjacent monsoon subsystems, as well as by the inflow from and the outflow to the surrounding oceans (see the reviews by Wang et al., 2009, andQu et al., 2009).At the annual time scale, the climate is mainly determined by two alternative monsoon cycles, a southwest monsoon that brings wet and warm weather over the area in the boreal summer (hereafter, all seasons will refer to the northern hemisphere seasons) and a northeast monsoon that brings cold and dry conditions in the winter (Wyrtki, 1961;Wang and LinHo, 2002;Nguyen et al., 2014).Moreover, at the interannual time scale, the SCS region is further influenced by the El Niño-Southern Oscillation (ENSO) phenomena.
As a complement to the large and growing body of regional climate-related publications stemming from different authors, journal articles, datasets, and analyses of specific time periods and time scales, the goal of the present study is to make an original contribution to the literature by attempting to provide a concise and integrated analysis of five key oceanic, atmospheric and terrestrial variables.To our knowledge, this study is actually the first co-analysis of regional (SCS) oceanographic variables conducted over a Page 4 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.multi-decadal time frame made possible by the presence of long-term data products.For this co-analysis, our approach is based on a coherent methodology, using the same technique for each key variable, comparing each variable with the same ENSO indices, computing trends over the same period, and using either new or longer datasets than the ones currently found in the literature.Furthermore, this study provides an analysis of altimeter-derived Mekong and Red Rivers discharge, which has not been previously documented, as well as in situ P changes in near-coastal stations of Vietnam.The five key variables studied here are sea surface temperature, sea level anomalies, surface winds, precipitation and river discharge, all of which are considered an Essential Climate Variable (ECV) in the frame of the Global Climate Observing Systems (GCOS, 2016).(In total, 50 variables covering three domains, atmospheric, oceanic and terrestrial, are referenced as ECVs).
The five ECV datasets and the common data analysis methods based on temporal filtering and EOF analysis are detailed in section 2. Their long-term means and standard deviations (denoting the overall variability) are described in section 3 to set the context.Then, their seasonal and interannual (ENSO) variabilities are analyzed in sections 4 and 5, respectively.The long-term trends are finally documented in section 6. Aiming to be a short overview paper, the knowledge gained from our study is compared to a variety of previous results in all sections.A conclusion and discussion section is given in section 7.

Data and Methods
Five ECVs are investigated: sea surface temperature (SST), sea level anomaly (SLA), surface wind (SW), precipitation (P) and water discharge (WD).Among the multiple available databases, all of the products that are described next have been selected based on a compromise between several criteria: the dataset is as long as possible, grid resolution must be as fine as possible, confidence, robustness and/or 'clear' documentation of the product (partly based on a literature review).
Page 5 of 48 SST.The primary datasets we selected are the 1° daily optimum interpolation SST (daily OISST) version 2.0 developed by the National Oceanic and Atmospheric Administration (NOAA; Reynolds et al., 2007;Reynolds 2009).(Note that all websites we used to get the data are listed in the Acknowledgements section below).We extracted the merged and gridded SST data covering the SCS from 0-24°N to 100-120°E and spanning from 01/1982 to 12/2015 (34 years).
SLA.We used the multi-mission gridded sea level anomaly (MSLA) product produced by AVISO+ (Archiving, Validation and Interpretation of Satellite Oceanographic Data) that was based on TOPEX/Poseidon, Jason 1, ERS-1 and ERS-2 data.This product provides SLAs relative to a 20-year mean from 1993 to 2012.This weekly data set is averaged to produce the monthly mean data set in the present work and covers the same area as the SST data from 01/1993 to 12/2015 (23 years) with a resolution of 0.25°x0.25°.The mean dynamic topography (called the CNES-CLS13 MDT) was also obtained from AVISO+ and gives the surface height above geoid over the period 1993-2012 (Rio et al., 2014).This product is based on two years of GOCE (Gravity Field and Steady-State Ocean Circulation Explorer) data, seven years of GRACE (Gravity Recovery and Climate Experiment) data and 20 years of altimetry and in situ data.SW.To investigate the SW, we selected the 10-m ERA-Interim merged and gridded wind data developed by the European Centre for Medium-Range Weather Forecasts (ECMWF; Dee et al., 2011).The data are available monthly on a 0.75°x0.75°spatial grid covering the entire region for the period 01/1979 to 12/2015 (37 years).Kouraev et al., 2004;Frappart et al., 2015).The time-series of altimetry-based water levels was obtained using the Multi-Mission Altimetry Processing Software (MAPS - Frappart et al., 2015) to process the along-track altimetry data from ERS-2 (1996ERS-2 ( -2003)) Index (SOI) (Allan et al., 1996), the oceanic ENSO indices through the SST Niño3.4 (5°S-5°N, 170°W-120°W), SST Niño4 (5°S-5°N, 160°E-150°E) and SST Niño1+2 (10°S-0°, 90°W-80°W) indices (Trenberth andStepaniak, 2001 Rayner et al., 2003), and the El Niño Modoki index (EMI) (Ashok et al., 2007).The SST Niño1+2 index is preferentially used to characterize Eastern Pacific (EP) ENSO events.The SST Niño4 index is preferentially used to characterize all types of El Niño events, but the authors could have chosen to use either the SOI or the SST Niño3.4,as the phases of these three indices are almost equally sensitive to El Niño events (Hanley et al., 2003).
Data processing.The climatological means, standard deviations (STDs), seasonal and interannual variations were quantified for each ECV.The climatological means and STDs were calculated over the entire length of each dataset (ranging from 23 to 37 years).The seasonal variations were estimated by constructing a typical year for each ECV.This method consists of computing the mean of each month (J, F, …, D) of the year over the multiyear series.The interannual variations were estimated by subtracting the typical year from the original series, which removes the mean seasonal cycle and filtering the remaining signal with a 13-month low-pass Hanning filter.analyses were then performed on the typical years and low-pass filtered time series, which were first detrended.EOF analyses allow us to extract the main spatial modes of variability and determine how they can change over time (e.g., Hannachi et al., 2007).Only statistically consistent EOFs, based on North et al. (1982), are shown here, and all of the values presented below are significant at the 90% confidence level (except for the P and WD stations, as we did not perform EOFs on these series).Although statistically significant, some EOF modes equal to or greater than two are not presented below because we did not find any simple physical mechanisms possibly accounting for their related temporal and spatial functions.
The long-term trends were computed by least-square fitting a straight line to the monthly time series over the entire length of each dataset.The long-term trends over a common period of measurement of 27 years  are also calculated and presented when available (for P and SW from Era-Interim, P in situ and WD at ST).

ECV means and standard deviations
The mean and standard deviation are presented for each oceanic and atmospheric variable in Figures 2 and 3, respectively, and the atmospheric variables are available over both the ocean and continent.
SST.The mean SST ranges from 24.5 to 29.5°C with the lowest values in the northern part of the SCS basin and along the coasts of Vietnam (Fig. 2a), in agreement with Tuen (1994), Chu et al. (1997) and Qu (2001) to name a few.Liu et al. (2004) also emphasized a (relatively) cold tongue of water as the result of the advection of cold waters from the north via the western part of the cyclonic gyre that develops in the winter in the area.The standard deviation (Fig. 2b) ranges from 0.5 to 3.5°C with a high variability in the northern part of the basin.On both the mean and standard deviation maps, the northeast-southwest oriented isotherms are observed to have low temperatures and high standard deviations in the north-northwest (see also Chu et al., 1997, andQu, 2001).found similarly low sea level heights at approximately 18°N-117°E northwest of Luzon Island that extended southeastward to the east of Vietnam.Interestingly, the MDT resembles the mean 0/400dbar dynamic topography derived from in situ temperature profiles and mean TS curves (Qu, 2000).This similarity corroborates the existence of the cyclonic West Luzon and East Vietnam eddies, which are two major features of the mean upper circulation that are centered at the minimum values noted above.The SLA standard deviations range from 6 to 18 cm (Fig. 2d), which is consistent with the results from Zhuang et al. (2010).
These results are comparable with previous results from TOPEX/Poseidon altimeter data and Argo-tracked ocean surface drifter measurements presented by Ho et al. (2000).
SW.The mean surface zonal and meridional components are stronger over the sea than over the land area, and the winds mostly blow W-SW (negative values in Fig. 3ac  the Chinese coastline (Fig. 3e).As discussed in Wang et al. (2009), the sharp P contrast between the NW and SE parts of the SCS arises from a similar SST contrast (see Fig. 2a).The standard deviation of P ranges from 1.5 to 5.5 mm/d (Fig. 3f).The rainfalls data from in situ measurements at 17 Vietnam stations (Table 1) range from 8.2 to 14.4 mm/d (corresponding to approximately 3 to 5.3 m/yr).These measurements are consistent with the results from Gobin et al. (2016) who noted that rainfall on land was measured within a range of approximately 0.65 m/yr to more than 7.3 m/yr for the period 1960-2009.Here, the mean rainfall for the in situ stations that were considered is approximately 3.8 m/yr with the highest rate in the central part (Da Nang), as found in Gobin (2016).The standard deviations range from 6.05 to 15 mm/d.
WD.At the ST Red River River station, the mean water discharge is equal to 3540 m 3 /s (for the period 1960 to 2010) with a standard deviation of approximately 2960 m 3 /s, which is similar to the results from Vinh et al. (2014).At the CCV Mekong station, the mean water discharge from the gauge stations and satellites appear to be consistent (11600 and 12400 m 3 /s, respectively).

Seasonal variability
SST.The first seasonal EOF mode on SST (94% of the total variance) represents an annual cycle (Fig. 4b).The spatial function is positive over the entire basin and high in the northwestern part of the basin and along the southeastern Vietnam coast (Fig. 4a).The The second seasonal EOF mode (5% of the total variance) represents a semi-annual cycle with peaks in April and November and troughs in January and August (Fig. 4d).The spatial function is positive over the majority of the basin and negative along the Chinese and northern Vietnamese coasts (Fig. 4c).In April, the SST can increase by as much as 1.5°C in the Gulf of Thailand and south of Vietnam, and can decrease by -1.5°C in the Gulf of Tonkin.
As noted by Huynh et al. (2016), this semi-annual variability is mostly driven by oceanic To further analyze the SST seasonal changes, typical years are represented in four boxes, following the criteria of Qu (2001) based on mixed layer depths (boxes A to D in Fig. 5).A fifth box (E) was added to represent the region of high variability in the Gulf of Tonkin.
The results correspond to those of Qu and highlight the main influence of monsoon cycles on SST in boxes A and E, while box C (and B and D to a lesser extent) also seems to be influenced by the thermal advection described above.Vietnam eddy, although to a lesser extent).Logically, the timing of those seasonal SLA drops corresponds to the maximum development of the two eddies inferred from hydrographic observations (Qu, 2001).In addition, the seasonal sea level drops that are observed in the summer along the coasts are consistent with the upwelling-favorable surface winds that blow to the northwest at that time of the year (see below).
SW.The first seasonal EOF mode on surface wind anomalies contributes to up to 93% of the variance for both zonal (U) and meridional (V) components.The two temporal functions show an annual cycle, and the spatial functions show positive values over the entire region and more intense values over the sea (Fig. 6a-d).In the summer, the eastward (U) and northward (V) anomalies are then more intense (up to 17 m/s).The situation reverses in the winter with negative anomalies up to 10 m/s.The January-February-March (JFM) and June-July-August (JJA) mean wind vectors are shown in Figure 7, which stress the seasonal monsoon reversal to ease the interpretation of the EOF analysis performed separately on each wind component.The wind anomalies strongly depend upon the orographic features of the region: the seasonal wind anomalies are stronger above the sea.
Page 18 of 48  P. The first seasonal EOF mode on P contributes to 69% of the total variance (Fig. 6ef).A clear annual cycle appears on the temporal function.North of 8°N during the summer, the entire region undergoes positive rainfall anomalies as intense as 6 mm/d, except along the central coastline of Vietnam where the anomalies are slightly lower, which possibly corresponds to the upwelling-favorable wind effects and relatively cold SST patterns near the coast (Chu et al., 1997).During the winter, the signs of the anomalies reverse and the region north of 8°N undergoes seasonal rainfall deficits compared to the mean.
The seasonal P anomalies relative to the means (see values in Table 1) of the three   3e and 6e (see also Table 1).The units are in mm/d.WD. Figure 9 presents the seasonal variability relative to the means (see values in Table 1) of the water discharge from the Red River (at ST) and Mekong River (at CCV).The discharges observed at the ST station (red line) are much lower than those at the CCV station (black and blue lines), with anomalies ranging from -1.10 4 to +2.10 4 m 3 /s.At the CCV station, the highest discharge flux is observed in September (from both gauge station and satellite measurements) with anomalies as high as 1.8x10 4 and 1x10 4 m 3 /s for the in situ and satellite data, respectively.The entire water discharge series shows a peak in the summer and a trough in the winter.This variability is probably driven by the monsoon dynamics with a peak in water flow during the rainy season (summer monsoon) and negative anomalies during the dry season (winter monsoon), as illustrated in Figures 6ef and    discussed above.
To explain the different amplitudes between the satellite and gauge station data at the CCV Red River station (black and blue lines in Fig. 9), measurements over the common   1.

Interannual variability
As noted in section 2, EOF analyses are performed to extract the main interannual variability and compare with ENSO indices.The maximum correlation coefficients at given lags between the interannual temporal function of the EOF and all (low-pass filtered) climate indices are listed in between the ENSO indices and reported interannual EOF time functions (as shown in Figs. 11 and 13).A positive lag indicates that the indices lead the variable.All R values are significant at the 90% confidence level.
SST.The spatial pattern of the first interannual EOF mode on SST (75.4% of the total variance) is positive over the entire SCS with highest values along a northeast-southwest diagonal (Fig. 10a).The associated temporal function (Fig. 10b) shows a maximum correlation coefficient value of +0.59 with the Niño1+2 index (Table 2) with an index lead of 6 months.Thus, it appears that 6 months after the mature phase of an Eastern Pacific El Niño event (represented by the Niño1+2 index, see section 2), a basin-wide SCS peak warming of SST occurs.This peak is especially clear during the strong events of 1986-87 and 1997-98 when the SST increased by up to 0.7 and 1.0°C, respectively, in the central northern parts of the SCS.
The spatial pattern, the anomalous amplitude and the lag are quite similar to those of Chu et al. (1997) and Fang et al. (2006) that were obtained from different data sets and time periods.They further resemble the averaged February and August double-peak structures of the SCS SST composite anomalies derived from seven El Niño events covering 1950-2002 by Wang et al. (2006).The lagging response of SST to ENSO events (observed here for Niño1+2, and in Table 2 for Niño3.4,Niño4 and EMI) is described as a common feature in tropical oceans.Klein et al. (1999) and Wang et al. (2004)  SLA. Figure 10ef displays the first interannual EOF mode on the SLA, which accounts for 48.1% of the total variance.The spatial patterns of this EOF mode are characterized by the same negative sign throughout the basin (but with large horizontal gradients in amplitude), which indicates that SLAs occur consistently over the entire SCS.The associated temporal function is highly correlated (R=+0.78;lag=3 months) with the Niño4 index (Table 2).This EOF mode reveals that the sea level drop is largest three months after the mature phase of an El Niño event.The negative SLA can be as strong as 5-10 cm in the eastern part of the basin and along approximately 15°N.A reverse situation occurs during a La Niña year.Rong et al. (2007) found similar spatial and temporal patterns for the period 1993-2005 with high correlation with the SOI (the only index tested).The observed anomalies are consistent with the results presented by Fang et al. (2006Fang et al. ( ) based on 1993Fang et al. ( -2003 satellite altimetry data and with the results from a data assimilation model ran for the period 1993-2002 (Wu and Chang, 2005).
SW.The first interannual EOF mode on zonal wind contributes to 64% of the total variance of the signal (Fig. 11ab).The spatial function is positive in the southern part of the filtered again with the H25 filter here but are described in the text with the TY-H13 filtration method.The black dots in Figure 11e are the same as those in Figure 3e.
Figure 12 represents the composites of the anomalous wind vectors for the winter during four strong El Niño events, noted JFM, and for the following summer, noted JJA.The four El Niño events considered here are the 1982-83, 1986-87, 1991-92, and 1997-98 events, which were selected according to the studies of Chiodi andHarrison (2010, 2013).
The authors identified the abovementioned El Niño events as strong events that were characterized by strong peaks in both Niño3.4SST and outgoing long-wave radiation (OLR) anomalies over the eastern-central equatorial Pacific. Figure 12a shows that the anomalous winds were blowing N-NE, and when compared to the mean conditions in JFM when the winds blow S-SW (Fig. 7a), this potentially leads to a weakened winter monsoon.The anomalous JJA composites (Fig. 12b) show anomalous winds blowing S-SW, compared to the mean conditions in JJA (when the winds blow N-NE (Fig. 7b)).Furthermore, each JFM and JJA of each strong El Niño year were tested separately (not shown here), and the results showed the same wind patterns as the JFM and JJA composites.Figure 12 compares the anomalies over three months of a composite of strong El Niño events and is thus not directly comparable with Figure 11abcd where the anomalies are presented relative to the long-term mean, comprising all types of El Niño and La Niña events.Lastly, the same analyses were performed on four La Niña years (1983-84, 1988-89, 1995-96, 1998-99) 1983, 1987, 1992, and 1998, representing the 1982-83, 1986-87, 1991-92, and 1997-98 El Niño events, respectively.The arrows denote the vectors, with the longest arrows equal to 1 m/s.P. The spatial function of the first interannual EOF mode on P (48.4% of the total variance) is negative over most of the region, except north of Vietnam, Laos and along the Chinese coast, and it shows maximum values in the southeastern part of the region (Fig. 11e).The temporal function that is correlated with the Niño4 index (R=0.64at 0 lag) reveals that precipitation decreases during El Niño events and increases during La Niña phases.For the case of the 1997-98 event, which was one of the strongest ENSO events ever recorded, the amount of rainfall associated with EOF 1 (again representing only 48.4% of the variance) reduces by as much as 1.6 mm/d in the southeast region, which is a reduction of 28% relative to the mean value (Fig. 3f).The important P decreasing rate that was observed can be explained by the eastward shift in convective activity in the equatorial western Pacific during El Niño phases.Such shift leads to altered Walker circulation, generating a subsidence area over the SCS (Wang et al., 2006).However, when examining the period prior to the Hoa Binh Dam impoundment (from 1960 to 1978), the signal shows a statistically significant correlation to Niño4 (R=-0.47 at 12month lag).This suggests that the dam, by regulating the water flow according to the dry/wet seasons and demand, may interfere with the interannual signal.At the CCV Mekong station, both the 1960-2002 in situ and 1996-2002 satellite anomalies are correlated with the SOI (+0.51 and +0.53, respectively, with an SOI lead of 6 months) and with the Niño4 index (-0.42and -0.50 with 7-8-month lags, respectively).As an example, when considering the 1997-98 event, the water discharge is reduced by approximately 4000 m 3 /s (i.e., 34% of the mean) approximately 7 to 8 months after the event.Xue et al. (2011) showed a similar connection between ENSO and Mekong runoff (at Pakse, Laos, approximately 400 km upstream of the CCV station) for the period from 1993-2005.It is important to consider the WD data with caution as the precipitation anomalies over the entire Red River and Mekong River watersheds (143,000 and 795,000 km 2 , respectively) are integrated and may relate processes that occur outside of the area considered here and potentially far away from any ENSO influence.

Trends
Page 31 of 48 Trends are calculated over the total length of each dataset.Since the series differ in lengths, they are also calculated over a common 27-year period (1979-2006) for most of the datasets (P in situ, P and SW products from ERA-interim, WD at ST).No significant differences appeared between the trends that were calculated over the entire length of the dataset and over the common period (both values are discussed when possible).
SST.The 1982-2015 trend shows an increasing SST trend with the highest values in the northeastern part of the basin and maximum values west of the Luzon Strait and along the Borneo coastlines of approximately 0.24°C/decade (or a total rise of 0.82°C of for the period 1982-2015).A small SST decrease is observed along a very narrow coastal band off central Vietnam and in the northern part of the Gulf of Tonkin of about -0.06°C per decade (Fig. 13a).The computation of the wind stress curl trend (not shown here), an alias for Ekman pumping trend, proves to be inconsistent with the spatial variations and even with the sign of the SST trends.Fang et al. (2006)  SLA.The map of the sea level trends for 1993-2015 shows an increasing trend throughout the SCS basin with high spatial variability from 3 to 70 mm/decade and a mean rising rate of approximately 41 mm/decade (Fig. 13b).A tongue of rapidly increasing rates is observed from 14 to 20°N and along the longitudes 110 to 116°E, corresponding to the location of the minimum of MDT (observed in Fig. 2c).This finding suggests that the sea level must be rising in the low center of the cyclonic loop current that is generated by the intrusion of a branch of the Kuroshio Current through the Luzon Strait (Farris and Wimbush, 1996;Ho et al., 2000), which results in a slowdown of the related Luzon Strait and East Vietnam cyclonic eddies.Peng et al. (2013) found similar sea level rise rates of approximately 39 mm/decade for the period 1993-2009 using the same data product.These results are, as well as the results found here, faster than the global mean rate of sea level rise of 28±0.4 mm/decade (obtained for the period 1993-2003 with the TOPEX/Poseidon altimeter) (Cazenave and Nerem, 2004) and the global mean sea level for 1993-2015 of approximately 3.3 mm/decade that was recently obtained by Dieng et al. (2017).In line with the analysis of Meyssignac et al. (2012), Cheng et al. (2016)  These results are consistent with the results from Fang (2006), who found an area-mean trend of the zonal component of 0.56±35 m/s/decade for the period 1993-2003.However, we did not find an obvious trend for the meridional component.Part (not estimated here) of the observed trend in the zonal wind is directly linked to the phase of the PDO, as suggested by England et al. (2014) and others, since trade winds have considerably strengthened over the past two decades.Figure 13e summarizes Figures 13cd in a vector form and shows that the maximum rate of the trend is 1x10 -2 m/s/decade, which occurs toward the south along the Vietnamese coastlines and in the Gulf of Thailand.
P. The map of the linear trend of P clearly shows a rainfall deficit north of 8°N of approximately 6 mm/d/decade over the period considered , with maximum values inland (Fig. 13e).Over the common period of measurements from 1979 to 2006, the decreasing trend is approximately -6.8 mm/d/decade.The Lục Yên station also shows a decreasing rainfall trend (-2.64x10 -1 mm/d/decade).However, the Đà Nẵng and Mỹ Tho stations both show increasing precipitation trends with rates of 1.48x10 -1 and 6.73x10 -1 mm/d/decade, respectively.So far, we are not able to determine the origin of the observed discrepancies between the satellite and in situ measurements (as well as the discrepancies about the mean, as mentioned in section 2).Comparison with other satellite products is recommended here.For the period 1961 to 1998, Manton et al. (2001) also showed that the number of rain days over land (at least 2 mm of rain) has significantly decreased throughout Southeast Asia.They associated this decrease with the predominance of El Niño events since the mid-1970s (Trenberth and Hoar, 1997).
Page 36 of 48 This decreasing trend has also been emphasized by Vinh et al. (2014); it is partially due to the impoundment of the Hoa Binh Dam in the late 1980s, which has decreased the water flow since then.The trends at the CCV Mekong station also showed decreasing rates of -1220 m 3 /s/decade (0.98% of the mean WD) and -1340 m 3 /s/decade (0.87% of the mean WD) for satellite and in situ data, respectively.Lu and Siew (2006) investigated the disruption in water discharge at stations on the Lower Mekong River that was induced by the cascade dams in the upper part of the main stream of the Mekong River and found a declining trend during the dry season.Furthermore, Xue et al. (2011) showed that the runoff of the lower Mekong River was more closely connected to precipitation and ENSO in the post-dam period (1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005) than in the pre-dam period .Distinguishing and isolating an effect from another one (precipitation, water use and water regulation of a dam, ENSO) on water discharge would provide crucial information on the behavior and possible forecasting of a trend.
ENSO and long-term trend.The study periods and time series lengths are responsible for not only the slight trend discrepancies observed between the literature and our results found for each ECV but also the way the timing and the amplitude of ENSO may affect the computation of long-term trends.To illustrate this issue, the 25-month Hanning filtered SST and the corresponding linear trends were computed over the box 14.5-19.5°N,112.5-117.5°Eroughly centered in the middle of the northern SCS (see the black square in Fig. 13a).Looking at Figure 14

Conclusion and discussion
In this study, we successively analyze the long-term means, standard deviations, seasonal and interannual (mostly ENSO-related) variability and long-term trends of five key ECVs (sea surface temperature, sea level anomaly, surface wind, precipitation, and water discharge) to further document, validate, corroborate and/or complete the current scientific knowledge on the climate variability in the South China Sea.
The analysis of the first (or first two) EOF mode(s) on each ECV clearly exhibits the seasonal variability of the SCS under the influence of monsoon.In the summer, the SST is      by approximately 0.7 mm/d/yr over almost the entire region.In situ nearcoastal P stations in Vietnam showed increasing trends at lower latitudes, while the northern stations showed slightly increasing to decreasing trends.All of the water discharges that were observed showed small relative decreasing trends with higher rates (compare to the mean) at the Red River ST station (1.5% of the mean), which is located a few kilometers upstream of the city of Hanoi.
As a cautionary note, as discussed in England et al. (2014)  Our results are based on a unique and unprecedented long-term data set we thought to be relevant to each ECV.To confirm these results, it would however be fair to analyze complementary data sets such as NCEP (Kalnay et al., 1996) and JRA-55 (Kobayashi et al., 2015) for P, and TropFlux (Kumar et al., 2013) for P, SST and surface wind.Furthermore, the growing literature that is available on the subject, and particularly the literature mentioned throughout the text, has enabled us to identify mechanisms that likely account for the observed variability at different time scales.Strictly quantifying the (common or not) mechanisms responsible for the observed anomalies (including the ones of the 2015-16 El Niño not discussed here) and trends highlighted in this paper by using outputs from ocean/atmosphere dynamic models validated with our observations seems crucial to better understand the seasonal, interannual and long-term variability of the region.This will be the subject of another study.

Figure 1 .
Figure 1.Bathymetry below sea level and topography above sea level (in m) of the South China Sea region.Our research domain is enclosed with a black rectangle.The acronyms denote, GTh Gulf of Thailand, GTo Gulf of Tonkin, H Hainan island, IO Indian Ocean, JS Java Sea, MS Malacca Strait, LS Luzon Strait, TS Taiwan Strait, P Paracel Islands, PO Pacific Ocean, Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.

P
. The ERA-Interim reanalysis is also used as a source of precipitation data with the same resolution and time coverage as previously noted.To further the study of precipitation, we selected to investigate the land rainfall measurements provided by Vietnam's Hydro-Meteorological Service.Among the 172 rainfall stations available, only 17 are considered in this study.The selection was made regarding the following criteria: a) the data set must Page 6 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.cover at least 28 consecutive years, b) the dataset should not contain any missing data, c) the 125 spatial distribution must be homogeneous across Vietnam, and d) at least one station must 126 be located in each of the eight climatic zones of Vietnam described in Gobin (2016).The 127 locations and primary characteristics of the 17 selected rainfall stations are shown in The lowest SSTs are mainly Page 10 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.caused by the northeast winds that blow in the winter, which first cool the seawater by bringing cold and dry air and second generate a southwest-ward coastal current, which, in return, brings cold coastal waters from the East China Sea into the SCS(Fang et al., 2006).

Figure 2 .
Figure 2. Mean and standard deviation of SST (a and b, respectively) in °C, and mean dynamical topography (MDT) and standard deviation of SLA (c and d, respectively) in m.The values were computed over 1982-2015 for SST, and 1993-2012 for MDT.The color codes differ between the figures.
), particularly in the northern part of the basin where the values are as high as 8.5 m/s (approximately 16.5 knots).A mean southward flux is observed along the south of the Vietnamese coastline.It appears that the mean surface winds and the standard deviations associated with both zonal and meridional winds strongly depend upon the orographic features (lower values above lands).The interpretation of the annual mean surface wind in a region that is highly influenced by strong seasonal wind reversals due to the monsoon cycles (see section 4) does not mean very much.The standard deviation values are the lowest above land (from 0.5 to 4 m/s) and the highest above water (from 5 to 11 m/s) with cores of maximum values near the southern tip of Vietnam.These values mostly reflect the seasonal reversal (chiefly over the ocean) of the wind driven by the summer-winter monsoon cycles.Page 12 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 3 .
Figure 3. Means (a, c, e) and standard deviations (b, d, f) of the zonal and meridional surface wind components in m/s (noted U and V, respectively) and precipitation in m/d (noted P).Positive U and V wind components are directed to the East and the North, respectively.The Page 13 of 48 temporal mode reaches troughs in January-December and peaks in June-July, and they Page 14 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.correspond to the cold northeast winter monsoon and the warm southwest summer monsoon, respectively.Thus, compared to the mean, this mode exhibits a cooling (up to -4.8°C) over the entire basin in the winter and a warming (up to 4.2°C) mainly in the northwest in the summer.

Figure 4 .
Figure 4. Spatial functions (a, c, e) and associated temporal functions (b, d, f) of the first and 305 second (only for SST) EOF seasonal modes of SST and SLA.The products between spatial and 306

Figure 5 .
Figure 5. Locations of boxes (left) and time series of the relative SST monthly means (in °C) over boxes A (black), B (magenta), C (green), D (blue) and E (cyan).SLA.The first seasonal EOF mode on SLA accounts for 80% of the total variance of the signal (Fig.4ef).The spatial function shows positive values in the central and eastern parts of the SCS and negative values in the western part of the basin.The temporal function reveals higher-than-average (up to 0.14 m) SLAs in the summer in most of the basin and lower-than-average SLAs along the coastlines of Vietnam, Cambodia and Malaysia.In the winter, the situation reverses leading to negative anomalies over most of the basin and positive anomalies along the coasts.Interestingly, this EOF mode also emphasizes lowerthan-average SLA in the winter near the center of the West Luzon eddy (as well as the East

Figure 6 .
Figure 6.Spatial functions (a, c, e) and associated temporal functions (b, d, f) of the first seasonal EOF modes on the U and V components of the surface wind, and P. The products between the spatial and temporal functions denote anomalous U and V (in m/s) and P (in Page 19 of 48

Figure 7 .
Figure 7. Mean surface wind in January-February-March (a) and June-July-August (b).The arrows denote the wind vectors, and the longest arrows equal 8 m/s.
selected in situ stations are shown in Figure 8.The anomalous values can be as high as 40 mm/d and as low as -8 mm/d, with a tendency to depict annual, Dirac-like, and semi-annual functions at the Lục Yên, Đà Nẵng and Mỹ Tho stations, respectively.The results for the Lục Yên station and the peak observed in the autumn at the Đà Nẵng station are qualitatively similar to the observations (from different periods) by Nguyen et al. (2014).The authors Page 20 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.emphasize that the months of peak rainfall coincide with the southward migration of the subtropical ridge and the intertropical convergence zone.

Figure 8 .
Figure 8. Seasonal variability of precipitation anomalies relative to the means from in situ land stations, Lục Yên (red), Đà Nẵng (black) and Mỹ Tho (blue).The station locations are denoted by black dots in Figures3e and 6e(see also Table1).The units are in mm/d.
Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.period(1996-2002)  were compared (not shown here).Both in situ and satellite measurements showed similar amplitudes although the satellite measurements showed 1-2month lags in the peak and trough, for currently unknown reasons.

Figure 9 .
Figure 9. Seasonal variability of the water discharge (WD) anomalies relative to the means at the (Red River) ST station from gauge station measurements (period: 1960-2010) in red, and at the (Mekong River) CCV station from gauge station measurements (period: 1960-2002, in black) and satellite measurements (period 1996-2015) in blue.The units are in 10 4 m 3 /s.The mean values are reported in Table1.

Figure 10 .
Figure 10.Spatial functions (a, c, e) and associated temporal functions (b, d, f) of the first and second interannual (only for SST) EOF modes on SST and SLA.The products between the spatial and temporal functions denote anomalous SSTs (in °C) and SLAs (in m),

Figure 11 .
Figure 11.Spatial functions (a, c, e) and associated temporal functions (b, d, f) of the first interannual EOF modes on the U and V wind components and P. The products between the spatial and temporal functions denote anomalous winds (in m/s) and P (in mm/d), . The surface winds showed almost reversed patterns from the ones observed for El Niño years (not shown here).The anomalous JFM composites for La Niña events showed anomalous wind blowing S-SW, and the anomalous JJA composites showed anomalous wind blowing N-NE.Page 29 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.

From the 17
Vietnam rainfall stations tested, it appears that the linkage of precipitation anomalies to ENSO is stronger at lower latitudes, with correlation coefficients Page 30 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.with Niño3.4 ranging from R=-0.19 to -0.45.As an example, at station Đà Nẵng where R=-0.45, the rainfall decreases during El Niño phases by up to 4 mm/d during the 1997-98 El Niño event (which is 70% of the mean).Nguyen (2014) also demonstrated that temperature and rainfall variability are more strongly linked to ENSO events at lower latitudes of Vietnam for the period 1971-2010.A study of the gridded precipitation data over the mainland of Southeast Asia (1980-2013) that was undertaken by Räsänen et al. (2016) confirmed the stronger influence of ENSO in the southern parts of the study area with high correlation to ENSO during the spring of the decaying year.WD.No obvious correlation was found between the interannual water discharge at the ST Red River station and ENSO indices when the period 1960-2010 was considered.

Figure 13 .
Figure 13.Trends of SST, SLA, surface wind components (U and V) and vectors, and P (a, b, 582 c, d, e, f, respectively) in °C/decade, mm/decade, m/s/decade and m/d/decade, respectively.583 Page 34 of 48 indicated that the PDO contributed to 72% of the total sea-level rise in the SCS during the 1993-2012 period.The authors suggested that the intensification of the easterly winds associated with the PDO in the last two decades leads to the increase in the steric sea level by deepening the thermocline in the Western Tropical Pacific.Finally, they indicated that the remaining 28% of the sea level trend corresponded to the global sea-level rising rate (of 18±0.3 mm/decade).SW.The trends of the two wind components (Fig. 13cd) range from -0.48 to +0.18 m/s/decade, respectively, with an increase of the zonal component (U) for the period 1970-Page 35 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.(reduced westward winds).Over the common period of measurements from 1979 to 2006, the SW trends range from -4.3 to 1.7 10 -2 m/s/yr.The meridional component seems to mainly weaken over the region, except in the eastern part of the SCS and above Borneo.
Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.WD.All of the stations show small decreasing trends in water discharge.At the ST Red River station, the water flux declined by -520 m 2 /s/decade (representing 1.5% of the mean water discharge) over the period of 1960-2010.For the common period of measurement (1979 to 2006), the water flux declined by approximately -420 m 2 /s/decade.
, the large SST anomalies that are characteristic of ENSO events (see the ENSO indices in Fig. 10ab) clearly impact the computation of the linear trends.To crudely estimate the expected effects, the 1987-1988 and 1997-1998 El Niño periods were Page 37 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.then removed, and the linear trends were recomputed.Ignoring these four El Niño years increased the SST trends by as much as 1.4°C/decade (from 1.06 10 -1 °C/decade to 1.53 10 -1 °C/decade).

Figure 14 .
Figure 14.Time series of the 25-month Hanning filtered SST averaged over the box 14.5-19.5°N,112.5-117.5°E(in black), the corresponding linear trend (in red) and the linear trend corresponding to the series shortened by four El Niño years (1987-88 and 1997-98) (in green).The location of the box is denoted by the black square in Figure 13a.
found to increase throughout the basin, the SLA increases in the central and eastern parts of the basin, positive rainfall anomalies are observed north of 8°N, the eastward and northward Page 38 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.wind components are stronger (reflecting the seasonal monsoon reversal) and the water discharge increases.A summary of the ENSO-related changes and long-term trends of the five ECVs analyzed is provided in falls throughout the basin during an El Niño event, with a 3-month lag, and more intensely in the eastern part of SCS.Unexpectedly, the SST and SLA thus have an inverse response to ENSO.This was examined byRong et al. (2007).Using the subsurface temperature analysis fromIshii et al. (2006), they show the coexistence of positive temperature anomalies in the upper 75 m (hence >0 SST anomalies) and negative temperature anomalies below 75 m (at least down to 700 m) during El Niño events.These out-of-phase temperature anomalies (0-Page 39 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.75 m versus 75-700 m) result in negative anomalies of the thermosteric sea level component, which account for most of the negative SLA during El Niño events.Previous studies have identified different potential hypotheses to explain the anomalous ENSO-related SST changes.Wang et al. (2000) attributed the SCS warming to an anomalous lower-tropospheric anti-cyclonic circulation located in the western North Pacific through an 'atmospheric bridge'.The anomalies can persist until the following summer.Qu et al. (2004) examined the heat transport through the Luzon Strait from the Pacific into the SCS and suggested that this 'oceanic bridge' between the two basins was a key process in conveying the impact of the Pacific ENSO into the SCS (both for SST and SLA).The respective parts of the atmospheric and oceanic bridges clearly remain to be clarified.The first EOF mode (64% of the total variance) of the zonal wind component revealed that the wind increased eastwards north of 14°N (decreased south of 14°N) during an El Niño event, with a 5-month lag.The first EOF mode (49% of the variance) of the meridional wind component showed a strengthening of the northward winds north of 6°N (weakening south of 6°N) during an El Niño event, with a 3-month lag.When superimposed on the strong mean seasonal cycle, the El Niño events hence produce a slowdown of the winter NE and summer SW monsoon winds.For precipitation, the first EOF mode (48% of the variance) showed a decrease in rainfall over the entire region during El Niño phases.The linkage of near-coastal in situ P stations and ENSO appeared to be stronger at lower latitudes in Vietnam, for instance with P decreasing by up to 70% relative to the mean at the Da Nang station during El Niño phases.Finally, water discharges taken from the Red River (at least at the ST station) before the impoundment of the Hoa Binh Dam showed correlation to ENSO but the post-dam impoundment period did not show a clear correlation to ENSO.The water discharge from the Mekong River (at least at the CCV station) seems to respond to ENSO via reduced water flow (34% of the mean for the 1997-98 El Niño event) with a 7 to 8-month lag.Page 40 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.The patterns of the trends presented in the previous section provide useful elements to determine the long-term variability of the ECVs.SST has risen by 0.24°C/decade basinwide (except along the coastlines of China and central Vietnam).The SLA increased with a mean rising rate of 4.1 mm/yr with the highest rates in the central and the eastern parts of the SCS basin, which is approximately 3 times faster than the global rate of sea level rise.This corroborates the need to include data from the SCS (and semi-enclosed basins in general) to improve the computation of global mean sea level trends(Dieng et al., 2015).The zonal wind decreased over the period 1970-2015 by up to 0.48 m/s/decade in the eastern and western parts of the region.The meridional wind strengthened southwards over the entire region, except in the eastern part of the basin where the northward winds increased by up to 0.18 m/s/decade.The satellite-derived precipitation decreased over the period and others, it is crucial to consider the likely role of the PDO in the calculations of long-term trends.An inversion of the sign of the PDO (e.g., in 2000) can induce a strengthening of the Pacific trade winds, leading to the slowdown of the Pacific Ocean surface warming that has been observable since 2001 and is related to the changes in the regionally analyzed ECVs.Furthermore, as discussed in the previous section and as addressed by Solomon and Newman (2012), it is essential to consider the modulation of long-term trends that is induced by ENSO events (even when estimated over 30 years), especially in the Indo-Pacific region.Page 41 of 48 Ocean Sci.Discuss., https://doi.org/10.5194/os-2017-104Manuscript under review for journal Ocean Sci. Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.

Table 1 .
Means, standard deviations and trends from the 17 selected inland rainfall stations in Vietnam (in mm/d and mm/decade), and the Red River ST, Mekong CCV in situ WD stations and CCV from satellite measurements (in m 3 /s and m 3 /s/decade), along with their coordinates.The values were computed over the 1979-2006 period for the rainfall stations, over the 1960-2010 period for the ST gauge station, over the 1960-2002 period for the in situ measurements at CCV and over the 1996-2015 period for CCV from satellite measurements.WD.Finally, two river systems are considered in this study: the Red River and the

Table 2 .
Maximum correlation coefficients (R) at given lags (given in months in brackets)

Table 3 :
Summary of the main modifications of five ECVs analyzed, as observed in the SCS during El Niño events and on basin-averaged long-term trends during the reported years.