Analysis of the Distribution and Seasonal Variability of the South China Sea Water Masses Based on the K-means Cluster Method

Abstract

Influenced by local mixing and coastal runoff, water masses in the South China Sea degenerate significantly. The K-means algorithm is used to classify the water masses based on WOD13 temperature and salinity observations from 1966 to 2013 because its principle is consistent with the definition of a shallow water mass. The numbers and initial centers of the water masses are determined using functions of in-cluster distance and density values. The result shows that there are ten water masses in the South China Sea. In combination with the T-S scatter diagram, the properties of the South China Sea water masses were analyzed, including their distribution, the seasonal variability, and the degeneration processes. The temperatures of water masses were higher in summer and lower in winter, with the amplitudes of variation gradually reduced from the surface to the bottom. The seasonal variation in salinity of the surface water masses was high in winter and low in summer, which mainly depends on the amount of river discharge and precipitation. The subsurface water masses were strongly affected by water from the Pacific Ocean; thus, the seasonal variability of these water masses is weak, especially for the intermediate water mass that characterized by prominent low salinity. The water mass formed by the Kuroshio water invading the South China Sea has insignificant seasonal variations in temperature and salinity. The properties and seasonal variabilities of the water masses derived from the K-means algorithm are in agreement with the existing conclusions, suggesting that the improved K-means algorithm is efficient and accurate in the shallow water mass division.

1. Introduction

The largest marginal sea in China’s offshore area is the South China Sea (SCS), with abundant resources and complex sea-power issues. A full understanding of the hydrological characteristics of the SCS and its variability is of great significance for resource development, economic construction, and defense deployment [1]. An indirect reflection of the changes in circulation paths at multiple scales and mesoscale phenomena, such as leapfrog, ocean fronts, and upwelling, can be found in the distribution and hydrographic features of water mass [2]. The tongue-like distribution of water mass boundaries has a close relationship with the direction of currents and thus can be used as circumstantial evidence for circulation analysis [3]. For marine fish reproduction, water mass distribution and variation are crucial [4,5,6,7,8]. In the area where two water masses mix, fish are more likely to congregate and form good central fishing grounds [9].

The SCS exchanges water with the western Pacific Ocean (WPO), the East China Sea, the Indonesian Seas, and the Indian Ocean through multiple straits. In addition, internal waves in the SCS are very strong due to their intense generation in the Luzon Strait [10]. Influenced by local mixing caused by internal waves and coastal runoff, the water masses in the SCS degenerate significantly; thus, the definition and division method of water mass for the open ocean are not fully applicable to the SCS. Su (1980) [11] first proposed the concept of “denatured water mass” based on the properties of shallow marine water masses. According to previous discussions on the definition of water masses, Li (1986) [12] summarized the definition of water masses as macroscopic water bodies with both “internal homogeneity” and “external anisotropy.” In this study, the shallow water mass is defined as a great body of water with similar physical and chemical properties, regular seasonal variability, and both homogeneity and anisotropy [11,12], which is consistent with the principle of the K-means clustering algorithm.

Although shallow water masses cannot be classified based on homogeneity and conservatism, their non-conservatism can be used to study seasonal variations in water masses and the development of water mass degeneration [13]. In addition to homogeneity in the physical and chemical properties of water masses, homogeneity in other characteristics such as evolutionary and dynamic properties can be taken into consideration. By using the denaturation characteristics of water masses, we can analyze the mixing process from the external sea into the SCS, distinguish the external water from the SCS water, and evaluate the invasion of the external water. The relationship between chemical organisms and water masses can be studied based on the seasonal variation of water masses and their biological distribution, which can further indicate the distribution of fishing grounds.

Many previous studies have investigated water mass division in the SCS and Taiwan Strait (TS), as well as water exchange between the SCS and external waters [14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40]. In the northern SCS, a variety of fuzzy mathematics techniques have been applied to the analysis of water mass [14]. Li et al. used fuzzy clustering methodology to separate the surface water masses in the northern and central SCS [15,16]. Li et al. used the multidimensional fuzzy clustering algorithm to analyze seven oceanographic parameters of seawater around the Taiwan Shoal before discussing the water masses division and the seawater intersection [17]. Huang et al. used Q-type multidimensional clustering methodology to divide the western TS water masses based on the concentration mixing theory founded by Helland-Hansen, with salinity as a conservative element [18]. Li et al. distinguished the Kuroshio water and SCS water in the Luzon Strait (LS) and the northern SCS using a temperature–salinity diagram and then analyzed the Kuroshio water’s intrusion into the SCS [19].

Liu et al. [20,21] and Li et al. [22,23,24] analyzed the structure of SCS water masses using various systematic clustering methods. Zhu et al. used Q-type multidimensional clustering methodology to analyze the vertical distribution of the water masses in the SCS [25]. Tian et al. used the tetragonum quantitative analysis method to obtain the distribution of water masses in the northern SCS and the region near the Bashi Channel [26]. Liu et al. analyzed the seasonal variation and annual average of water in the SCS subsurface and intermediate [27]. Liu et al. [28] and He et al. [29] analyzed the water masses’ structure and seasonal variation near the LS based on Argo buoy data. Cheng et al. [30] used fuzzy clustering methodologies to analyze the distribution of water masses in the northern SCS. Tan et al. divided the water masses and explored the relationship between community structure and water masses with different physicochemical properties in the LS [31]. Zhang et al. analyzed the impact of various water masses on the distribution characteristics of suspension in the southern SCS [32]. Li et al. analyzed the relationship between water mass and nitrogen, phosphorus nutrients, and the distribution of plankton [33].

Up until now, researchers have held different ideas on the division of water masses in the SCS. Most of these studies analyzed the water masses in parts of the northern SCS and the LS, and the data are generally for a certain year or even a certain month due to the limitation of the survey area and the lack of information. Whether the results can represent the division of water masses in the SCS needs to be confirmed by further exploration.

In this paper, the K-means clustering algorithm is used to divide the water masses in the SCS in the horizontal and vertical directions. Climatological monthly data of temperature and salinity obtained from WOD13 in the whole SCS region are selected and used in this study. There are two disadvantages to the traditional K-means clustering algorithm. One is that the number of clusters cannot be determined, and the other is that the initial clustering center is chosen at random. To remedy these drawbacks, an in-cluster distance function and a density value function are proposed to determine the number and the initial clustering centers of water masses, respectively. The distribution and characteristics of each water mass’s temperature and salt content are analyzed in conjunction with a temperature–salinity (T-S) diagram throughout the four seasons of spring (April), summer (July), autumn (October), and winter (January).

2. Data and Methods

2.1. Data

The data from World Ocean Database 2013 (WOD13), developed by the National Oceanographic Data Center (NODC), is used in this study. The temperature and salt information are mainly obtained from high-resolution CTD, water temperature probes (MBT and XBT), drifting buoys (DRB), profiling buoys (PFL), and moored buoy (MRB) observation data. Since the WOD13 data are measured from a wide range of sources over a long period of time, the data quality is uneven. With a view to ensuring the credibility of the conclusions, the data were quality controlled before analysis. After quality control processing, the data were essentially consistent with the SCS water characteristics, but there were still a few unqualified points that will not have an impact on the clustering analysis of the water masses. However, the unqualified points are removed in the clustering process for more reasonable results.

The WOD13 data in this paper is selected for the regions 105° E–122° E and 3° N–26° N, and the time range is from 1 January 1966 to 1 January 2013. The spatial distribution of the stations is shown in Figure 1a, which spreads over the whole SCS. After quality control processes, 29,317 profiles remained. The number of profiles for each season is shown in Figure 1b. There were 7105 profiles in the winter, 7378 profiles in the spring, 7445 profiles in the summer, and 7389 profiles in the autumn, with relatively uniform distribution across the seasons.

2.2. Methods

The K-means clustering algorithm is a more efficient unsupervised clustering algorithm that can effectively handle large-scale and high-dimensional data sets. According to the given measurement criteria, i.e., objects in the same cluster are more similar and objects in different clusters are less similar, n data samples are divided into k clusters so that the obtained clusters satisfy the characteristics of “internal homogeneity and external dissimilarity” of water clusters. Some research has shown that the K-means clustering algorithm can be used for the classification of water masses [41,42]. Classes obtained from the K-means algorithm have similar properties in the same class and different properties between different classes, which is consistent with the definition of SCS water masses. Considering the large size of marine data and the complexity of water masses, the K-means algorithm is simple and easy to understand, with fast convergence and low time complexity, and can handle large-scale data sets effectively.

When researchers used cluster analysis, correspondence analysis, and discriminant analysis in multivariate statistical analysis for shallow marine water mass analysis in the past [1,2,3,4,5,6,7,8,9,12,13,25,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65], the artificial determination of the number of water masses and water mass centers led to the subsequent findings with strong subjective factors. The classical K-means algorithm has similar drawbacks, and the determination of the number of clusters and cluster centers affects the final clustering results. In this paper, according to the characteristics of water clusters, “the same inside and different outside”, and the principle of the K-means clustering algorithm, the rate of change of in-cluster distance is used to analyze the number of water clusters. The in-cluster distance is defined as:

$$f_C={\displaystyle \sum _{k=1}^C{\displaystyle \sum _{r=1}^{m_k}{(x_r-{\overline{x}}_k)}^2}}$$

(1)

where C is the number of categories, m_k is the total number of samples, r is the new number of samples in the k category, and ${\overline{x}}_k$ is the mean value of the k category. As the number of categories increases, the in-cluster distance will gradually decrease and will tend to level off after a certain number of clusters. The number of classifications corresponding to the shift of the f_c curve versus the number of categories from a sharp drop to a flat one is the number of water masses.

During the K-means clustering process, the average value of each new cluster is calculated as the new cluster center. If the two adjacent cluster centers have not changed, this indicates that the cluster center adjustment is over and the cluster is complete. Otherwise, readjust the cluster center and continue the cluster. Therefore, the selection of the initial clustering center will affect the efficiency of the K-means clustering. For the determination of the initial clustering center, a density value function is used [54]. For the n data points to be clustered in the data set $x_1,x_2,...,x_n$, the density of the data points x_i is defined as:

$$den\left(x_i\right)={\displaystyle \sum _{j=1}^n\sqrt{{\left(x_{is}-x_{js}\right)}^2+{\left(x_{it}-x_{jt}\right)}^2}}$$

(2)

where n is the number of data points in the data set to be clustered and ${x_{*}}_s$ and ${x_{*}}_t$ are the salinity and temperature of $x_{*}$, respectively.

For a certain point, the point contains two elements: temperature and salinity. According to the density value formula, it represents the difference in temperature and salinity between points. The degree of evacuation or densification of this point was determined based on its density value. Combined with the property of the water mass, the points in the same water mass are more concentrated. Therefore, the smaller the density value, the denser the point is with the surrounding data, which can be used as the initial clustering center.

Assuming that the number of water masses is m and the points to be clustered are n, the initial cluster center is determined as follows: firstly, calculate the density value of all points, the point with the smallest density value as the first cluster center x₁, then remove the n/m points with close temperature and salinity values of x₂. Secondly, calculate the density of the remaining points, and also select the point with the smallest density value as the second cluster center x₂, then remove the n/m points with close temperature and salinity values of x₂; repeat this step until the full initial clustering center is obtained.

The traditional K-means clustering algorithm automatically generates the initial cluster centers, and the isolated points will affect the clustering results. The number of clusters is also artificially given, which is more subjective. In this paper, the numbers and initial centers of the water masses are determined using functions of in-cluster distance and density values. These are added to the traditional K-means clustering algorithm, which is called the “improved K-means cluster method”.

3. Results

3.1. Water Masses Division

Based on the observed temperature and salt data from WOD13, the SCS water masses are divided into five layers in the vertical direction by using the improved K-means clustering algorithm. The vertical water masses include surface water, subsurface water, subsurface-intermediate water, intermediate water, and deep water. The results of clustering are shown in Figure 1. In the T-S diagram, the scattered points of different colors represent different water masses obtained from the clustering. The division of different water masses is not strictly divided by a certain line, but there is a certain range of a mixing area between the water masses. The black area is the “body” of each water mass, i.e., the 50% of water masses closest to the central temperature and salt value; the other area is the mixing area between the water masses (Figure 1). The properties of water masses in different depth layers are determined by the range of values for the properties (including temperature, salinity, and depth), which is based on the maximum and minimum values of the “body” of each water mass in the vertical division. Among them, the phenomenon of branches in the deep water in autumn is due to the difference in the data years (Figure 2d). The scattering of the surface layer in the T-S diagram indicates a large temperature and salt interval, which is due to its complex composition, i.e., flushing water from continental runoff and the intrusion of Kuroshio water with high salinity. Due to the invasion of Kuroshio water, the subsurface water is characterized by high salinity and clear seasonal changes. The surface and subsurface water can be divided into different water masses in the horizontal direction, which will be analyzed later.

The SCS subsurface–intermediate mixed water (UI) is located between subsurface water with high salinity and intermediate water with low salinity (Figure 2). This water mass is formed in the northwest Pacific Ocean near 122.5° E, between 300 and 500 m in the winter and between 200 and 350 m in other seasons [35]. The salinity values of the source water range from 34.40 to 34.70. The salinity range decreases after the water enters the SCS through the LS, ranging from 34.35 to 34.65 in winter and from 34.30 to 34.70 in summer (Figure 2a,c). This water mass is located in the SCS between 200 m and 350 m. The salinity difference in the vertical direction in the southern SCS is smaller than that in the northern SCS. Compared with winter, the salinity values of Kuroshio subsurface waters increased by 0.1 and the WPO intermediate water (122.5° E) decreased by 0.1 in spring, leading to an increase in the salinity value difference in the vertical direction of the UI [35]. The temperature ranges from 9.8 °C to 15.0 °C in the winter and ranges from 11.0 °C to 16 °C in other seasons.

The SCS intermediate water (I) is mainly distributed in the 300–900 m depth, with the temperature ranging from 4.5 °C to 10.0 °C in winter and from 5.5 °C to 11.0 °C in other seasons, and a salinity ranging from 34.25 to 34.50 (Figure 2). This water mass originates in the WPO intermediate water near 122.5° E, with a low salinity core primarily distributed in the 19° N–24° N latitudes [35]. After entering the SCS near 119° E, the salinity value of the low salinity core increases to 34.40 and is mainly distributed in the 350–550 m water layer between 17° N and 21.5° N. After that, the salinity value of the low-salinity core basically does not change anymore in the west, but the horizontal distribution range becomes wider and wider. When water enters the SCS from the WPO, the salinity values of this water mass increase primarily near the LS. When water enters the SCS, it retains its low salinity characteristics in comparison to the upper and lower water masses.

The SCS deep water (D) is mostly found in seas more than 900 m deep. It is the body of water with the greatest thickness, and its salinity and temperature are essentially identical to those of the deep waters in the WPO. It has been confirmed that a large amount of deep Pacific Ocean water enters the SCS via the LS (1.5 Sv) [66]. The low-temperature, high-salinity Pacific Ocean deep water enters the SCS and sinks, which may form a cyclonic deep circulation. In winter, the salinity ranges from 34.35 to 34.62 and the temperature is below 5.0 °C (Figure 2a). In other regions, the salinity ranges from 34.30 to 34.70 and the temperature is below 5.5 °C (Figure 2b–d). This water mass is the most stable in the SCS, and its seasonal and regional variations in temperature and salinity are not obvious.

The T-S diagrams show that the temperature and salt distribution of seawater have obvious divergence and convergence characteristics in the upper and deeper layers, respectively (Figure 2). The divergence at the upper layer of the T-S scatter reflects the seasonal variability of shallow sea water masses, while the deep layer shows the relatively stable characteristics of water masses [14]. Considering the “internal homogeneity” of water masses, i.e., the scattered points of the same water mass in the T-S diagram will cluster into clusters, indicating that the surface and subsurface layers contain more than one water mass. The surface water is then further separated into the Kuroshio surface water (KS), the southern SCS shelf water (SS), the coastal diluted water (F), the nearshore mixed water (M), the SCS surface water (S), and the southern SCS shelf water (SS), which only occurs in the summer (Figure 3). The SCS subsurface water (U) and the Kuroshio subsurface water (KU) are further separated from the subsurface water (Figure 4).

The coastal diluted water (F) is primarily distributed in the coastal shelf sea with water depths less than 20 m, which is a low salinity water zone formed along the continental coast after river runoff into the sea with a salinity value less than 31 (Figure 3). It is mainly distributed along the coast of Guangdong Province, the mainland coast of Beibu Gulf, and the southeasterly China–Indochina Peninsula coast. The Pearl River, Red River, and Mekong River are the sources of the water that fills the three sea areas. This water mass is significantly influenced by the seasonal changes of the continental climate, with low temperatures in winter (about 19.5 °C) rising to about 22.0 °C in spring. The temperature ranges from 27.0 °C to 32.1 °C in the summer and decreases to about 23.0 °C in the autumn. Under the influence of the northeast monsoon in winter, the Pearl River water enters the sea and flows westward along the coast of Guangdong Province into the Beibu Gulf, becoming one of the sources of fresh water flushed along the coast of the Beibu Gulf. In summer, the increase in river inflow leads to the expansion of the water mass.

The nearshore mixed water (M) is mainly distributed in the seas of continental shelves shallower than 75 m (Figure 3). The water mass is formed by the mixing of the coastal flushing water and SCS water. Compared to winter, the water mass’s distribution widens in the spring; the reason is that the increased temperature increases the mixing of different properties of water masses. The temperatures range from 21.4 °C to 22.9 °C in the winter and from 22.53 °C to 30.31 °C in the summer. The salinity ranges from 31.90 to 32.30 in the winter and from 31.70 to 32.10 in the summer. In the summer and winter of 1998, Liu et al. analyzed the water masses in the SCS and found that the salinity values were higher in the summer than in the winter, contrary to our conclusion [20]. In Liu’s article, the Kuroshio water was not obtained in relation to ENSO. In this paper, the clustering analysis was performed after monthly averaging of the data for many years, the interannual signal was eliminated, and the reasons for the different results obtained from Liu need to be discussed and analyzed in the context of the interannual variation of water masses.

The SCS surface water (S) originates from the WPO surface water. It is mostly found in the central and southern SCS, and it is found at 0–120 m in the summer and 0–100 m in the other seasons (Figure 3). The principal reason for the thick water layer in summer is that thermal factors such as solar radiation promote vertical mixing. When the WPO’s surface water enters the SCS during the winter through the LS, the salinity decreases continuously. Compared to the central and southern SCS, the northern SCS has a higher salinity and lower temperature. The main cause of the temperature differential between the northern and southern SCSs is the latitudinal variation in solar radiation. In spring, the surface water temperature gradually increases from the LS southward, but the increase is less than that in winter. The temperature of the water mass increases in spring due to land and meteorological factors, and this temperature difference offsets the temperature difference caused by latitudinal differences. The temperature ranges from 21.5 °C to 29.0 °C, and the salinity ranges from 32.70 to 34.60 in winter. The temperature ranges from 19.0 °C to 30.5 °C, and the salinity ranges from 33.00 to 35.00 in spring. In summer, the temperature ranges from 22.0 °C to 30.5 °C, and the salinity ranges from 32.50 to 34.65. In autumn, the temperature ranges from 21.0 °C to 29.0 °C, and the salinity ranges from 32.60 to 34.55.

In the surface water of the southern SCS, as seen by the T-S figure (Thailand–Sunda Strait), there are two water masses with the same temperature and salinity properties. During the summer, when the southwest monsoon is active, the southern SCS shelf water mass (SS) is formed by the mixing of low salinity water from the Java Sea with water from the Gulf of Thailand [28]. The SS is classified as an independent water mass due to its different sources. This water mass is influenced by the monsoon and exists only in the summer. The southern SCS shelf water is primarily distributed in the 0–120 m of the China–Indochina Peninsula’s southeast sea. This water mass is influenced by the Java Sea and the Gulf of Thailand, which have high temperatures and low salinities, with a temperature of around 28.0 °C and a salinity range of 31.90 to 33.20.

The SCS is invaded by Kuroshio water, which is primarily spread along LS and TS, forming the Kuroshio surface water (KS) and the Kuroshio subsurface water (KU). The KS is mainly distributed in the 0–100 m water layer, ranging from 25.0 °C to 30.9 °C in temperature and 34.45 to 34.80 in salinity. The KU is mainly distributed from 100 m to 250 m, with a temperature range of 16.0 °C to 24.6 °C and a salinity range of 34.65 to 34.90. The next section will provide a detailed analysis of the Kuroshio water intrusion into the SCS.

The SCS subsurface water (U) is mainly spread in the 100–200 m water layer, with a temperature range of 15.0 °C to 21.5 °C in winter and 16.0 °C to 22.0 °C in summer, and a salinity range of 34.00 to 34.70 in winter and 34.00 to 34.65 in summer. Due to the northeast monsoon’s influence, there is extensive and strong mixing of surface and subsurface water masses in winter. The vertical eddy mixing of water bodies is enhanced, and the mixing between water masses is conducive to the transport of organic matter and nutrients, etc.

3.2. Seasonal Variation of Kuroshio Intrusion into the SCS

To analyze the intrusion of Kuroshio water into the SCS, the SCS and the LS (5° N–25° N and 105° E–125° E, respectively) were divided into small regions of 5° × 5° following Qu et al., 2000 [35]. Every single region was clustered separately and combined with the T-S diagram to analyze the spatial variation of each layer of water, as well as the seasonal change of entry of the outer waters into the SCS.

During the winter, the KS is primarily distributed in a 0–100 m layer in the southwest of Taiwan Island and the western LS (Figure 5). This water mass has a temperature range of 25.0 °C to 30.9 °C and a salinity range of 34.45 to 34.70. In the WPO (Section 122.5° E), the KS mainly distributes in the waters shallower than 50 m, and the core salinity value is greater than 34.65 [35]. The water mass flows westward to the LS (121° E), where the core salinity value decreases to 34.60 and the water layer is shallower than 75 m. After entering the SCS in the 119° E section, the KS mainly distributes in the 0–100 m water layer between 21.5° N and 22° N, and the salinity value decreases to 34.50 (Figure 3). The strong mixing of SCS surface water with the KS is the main reason for the rapid decrease in salinity values.

It has been pointed out [19] that the intrusion of Kuroshio water does not exceed 119° E (or 119.5° E), but Figure 4 shows that the black scatter representing KS is scattered in the region west of 119° E to 117° E. Furthermore, an examination of the salinity sectional distribution at 118°E reveals the presence of the KS in the 0–50 m water layer between 20° with a salinity value of 34.45. At 117.5° E, the 0–50 m water layer between 18.5° N and 19.5°N also has the characteristics of the KS. To 117° E, water with salinity higher than 34.45 is located between 17° N and 17.5° N, and from 20 m to 50 m, the KS disappears. This indicates that the KS invades the SCS more strongly in winter, up to nearly 117.5° E.

In winter, the KU is mainly distributed in the 100–200 m water layer in the southwestern TS and the western LS, with temperatures ranging from 16.0 °C to 24.6 °C and salinities ranging from 34.70 to 34.90. The salinity value decreases from 34.80 to 34.70 in the passage from the WPO (122.5° E) through the LS to the SCS at 119° E and is mainly distributed in the 120–200 m water layer between 19° N and 22° N [35]. The salinity value at 118° E is still 34.70, mainly in the 180–250 m water layer between 18.5° N and 20.5° N. The salinity value at 117° E in the core area still remains at 34.70, but its distribution area shifts southward to 17.5° N–20° N and sinks to a 200–250 m water layer. When the KU enters the SCS, its salinity value remains essentially constant at 34.70. Additionally, when compared to the KS’s decreasing salinity value after entering the SCS, this water mass retains more of the high salinity properties of the Kuroshio water.

In spring, the changes in the Pacific surface water after entering the SCS are comparable to those in the winter; both decrease gradually or even disappear (Figure 6). However, unlike in winter, a small amount of Pacific water with a high temperature and salinity still exists southwest of Luzon Island (region d). Liu et al. (2001) point out that this part of the water does not directly come from the Kuroshio water, but most likely from the Sulu Sea (with a threshold depth of roughly 450 m, the Mindoro Strait serves as a conduit for water exchange between the Sulu Sea and the SCS), which moves anticyclonically with the Kuroshio in the 10–250 m water layer in the vicinity of 20° N and 119.5° E, causing the water to sink and mix and thus result in water with high temperature and salinity values in the southwest of Luzon Island [20,21]. The water in the North Pacific Ocean’s middle layer with low salinity values also shows an increase in salinity after entering the SCS through the LS, but the changing process is slower than that in winter. In a region that is far from the southwest of Luzon Island (near 118.5° E), water with lower salinity values than the mid-water of the WPO can still be observed.

In spring, the KS is mainly distributed in the 0–100 m layer in the northwestern LS and the southwestern part of Taiwan Island, with a salinity value of about 34.55 to 34.80. The KS is primarily distributed shallower than 100 m in the WPO (122.5° E), with a salinity greater than 34.65. After entering the SCS from the LS, near 119° E, the KS is distributed between 22° N and 22.5° N in shallow waters of 50 m, with salinity values higher than 34.55. The KS extends westward as far as 118.5° E after entering the SCS in winter (Figure 3a). In spring, no KS is observed at 118.5° E; the KS reaches as far as 119° E (Figure 3b), and the invasion intensity is weaker than that in winter.

In spring, the KU is mainly distributed in the 100–250 m water layer in the northwestern LS and southwest of Taiwan Island, ranging from 16.5 °C to 25.0 °C in temperature and 34.75 to 34.85 in salinity. This salinity range is broader than that in the winter. In the WPO (122.5° E), the KU is mainly distributed in the 100–250 m water layer with a core salinity value higher than 34.85. The Kuroshio water enters the SCS through the LS and travels to a region near 119° E, where the salinity value decreases to 34.80 and is mainly distributed in the 100–200 m water layer between 21° N and 22.5° N. At 118.5° E, the KU is found to be distributed in the 75–175 m water layer between 20.5° N and 22° N, and the core salinity value is higher than 34.75. After the water mass enters the SCS from the LS, the salinity value decreases continuously, the western boundary reaches the region near 118.5° E, and the invasion intensity is weaker than that in winter. The intense mixing of the subsurface water in the SCS with the KU in spring results in an increase in the salinity of the subsurface water in the SCS, and its salinity value is higher than that in winter.

In summer, the KS is mainly distributed in the 0–100 m water layer in the northwestern LS and the southwestern TS, with a temperature range of about 25.0 °C to 30.9 °C and a salinity range of about 34.45 to 34.65 (Figure 7). In the WPO (122.5° E), the KS is mainly distributed in the shallow waters of 75 m, with a core salinity value of 34.65, and enters the SCS through the LS. After the KS enters the SCS through the LS, the core salinity value drops to 34.55 in the shallow sea area of 50 m near 120° E, and only part of the residual water of the KS remains when it reaches near 119.5° E. This indicates that the southwest monsoon in summer hinders the invasion of KS into the SCS, and the invasion range does not exceed 119.5° E. The intensity of the invasion is also weaker than in winter.

The summer KU presents in the northwestern LS’s 100 to 250 m water layer, with temperatures ranging from 16.0 °C to 24.6 °C and salinities ranging from 34.65 to 34.85. In the WPO (122.5° E), the KU is distributed in the 100–250 m water layer with a core salinity value of 34.90. After entering the SCS (119.5° E), this water mass is distributed in the 100–200 m water layer between 21° N and 22° N with a core salinity value of 34.80. The KU can still be observed in the waters near 118.5° E, mainly in the 100–200 m water layer between 20° N and 21.5° N, with a core salinity of 34.65. The above analysis shows that the KU can invade the SCS to the sea near 119° E, and the intensity of the invasion is greater than that of the summer KS but weaker than that of the winter KU. The mixing intensity of the summer subsurface water is greater than that of the surface water, and the mixing of this water mass with the KU results in higher salinity values in the sea near the LS than in the central and southern SCS.

In autumn, the KS is mainly distributed in the 0–100 m water layer in the northwestern LS, with a salinity range of 34.45 to 34.80 (Figure 8). Near the LS (120° E), the KS is mainly distributed in the 0–100 m water layer between 21.5° N and 22.5° N, with a salinity value of about 34.55. The water mass is distributed in depths shallower than 25 m near 119.5° E, with a salinity value of about 34.45. Further west to 119° E, the salinity value in the surface layer between 22° E and 23° E is lower than 34.40 and the KS disappears. The above analysis shows that the KS invades the SCS up to nearly 119.5° E, and the intensity of the invasion is basically the same as that in summer.

The autumn KU is mainly distributed in the 100–200 m layer in the northwestern LS, with a salinity range of 34.70 to 34.85. In the WPO (122.5° E), the KU is mainly distributed in the 100–200 m layer, with a core salinity value of 34.90. Near the LS (120.5° E), the core salinity value decreases to 34.80. After entering the SCS near 119° E, the water mass is mainly distributed in the 100–200 m water layer between 21.5° N and 22.5° N, with a salinity value of 34.75. Further west to 118.5° E, the highest salinity value in the subsurface layer is 34.60, indicating that the KU can invade the SCS near 119° E. The mixing of subsurface water masses and KU in the SCS in autumn mainly occurs near the LS, and the mixing intensity is weaker than that in winter but stronger than that in summer.

4. Conclusions

Based on the WOD13 observations from 1966–2013, the improved K-means clustering method is used to classify the water masses in the whole SCS. The water mass distribution, temperature and salinity properties, and seasonal variation are regularly analyzed by combining T-S point clustering maps. In the clustering process, the density value function and the sum intra-class distance function are used to determine the “initial center” and “number of water masses”, respectively, which enhances the calculation’s efficiency and accuracy. The SCS seawater was divided into ten water masses using the improved K-means clustering method: the coastal diluted water (F), the nearshore mixed water (M), the South China Sea surface water (S), the South China Sea southern shelf water (SS) (only occurs in summer), the Kuroshio surface water (KS), the South China Sea subsurface water (U), the Kuroshio subsurface water (KU), the South China Sea subsurface–intermediate mixed water (UI), the South China Sea intermediate water (I), the South China Sea deep water (D). The properties of each water mass and their seasonal variations are shown in

Table 1. Details of each water mass.

Name	Symbol	Source	Depth/m	Temperature/°C	Salinity
Coastal Diluted Water	F	Continental runoff	<20	About 19.5	<31.0
				27.0–32.1
Nearshore Mixed Water	M	Coastal diluted water mixed with external seawater	<75	22.4–22.9	31.90–32.30
				22.5–30.3	31.70–32.10
South China Sea Surface Water	S	Western Pacific surface water	0–100	21.5–29.0	32.70–34.60
			0–120	22.0–30.5	32.50–34.65
South China Sea Southern Shelf water	SS	The Java Sea water mixed with the Gulf of Thailand water	0–120	About 28.0	31.90–33.20
Kuroshio Surface Water	KS	Kuroshio surface water	0–100	25.0–30.9	34.45–34.70
					34.45–34.65
Kuroshio Subsurface Water	KU	Kuroshio subsurface water	100–250	16.0–24.6	34.70–34.90
					34.65–34.85
South China Sea Subsurface Water	U	Western Pacific subsurface water	100–200	15.0–21.5	34.00–34.70
			120–200	16.0–22.0	34.00–34.65
South China Sea Subsurface–Intermediate Mixed Water	UI	Subtropical thermocline water in the WPO	200–350	9.8–15.0	34.40–34.70
			200–350	11.0–16.0	34.35–34.65
South China Sea Intermediate Water	I	Western Pacific intermediate water	350–900	4.5–10.0	34.40–34.50
			350–800	5.5–11.0	34.39–34.55
South China Sea Deep Water	D	Deep water in the WPO	>900	<5.0	34.35–34.62
			>800	<5.5	34.30–34.70

Temperature and salinity for each water mass: top for winter and bottom for summer.

.

The coastal diluted water (F) originates from continental runoff, with salinity values lower than 31.0 throughout the year and mainly distributes in a depth shallower than 20 m off the mainland coast. The water mass is significantly influenced by the continental climate, with significant seasonal changes in temperature, high in summer and low in winter, with a variation up to 7.5 °C. The nearshore mixed water (M) is formed by coastal flushing freshwater mixed with outer seawater, which is primarily distributed on the continental shelf shallower than 75 m. Mixed water is rich in nutrients and tends to be an excellent fishing area. This water mass’s temperature is slightly higher in the summer than it is in the winter, but its salinity value is slightly lower.

The SCS surface water (S) mainly comes from the western Pacific surface water, mainly distributed in the 0–100 m water layer, with significant denaturation. The seasonal variation in the temperature of this water mass is about 1.5 °C. Due to the influence of solar radiation, the temperature of the southern SCS is higher than that of the northern SCS. The temperature difference between the northern and southern hemispheres is larger in winter and basically disappears in summer. In addition, due to the weakening of the invasion of KS, the salinity of this water mass in summer is lower than that in winter, with a seasonal variation of about 0.2. The southern shelf water (SS) in the SCS is formed by the mixing of seawater from the Java Sea and the Gulf of Thailand and enters the SCS under the influence of the southwest monsoon, with the characteristics of high temperature and low salinity. This water mass is mainly distributed from the surface to 120 m in the southeast of the China–Indochina Peninsula, which exists only in the summer due to the influence of the monsoon, and it is the only seasonal water mass available in the SCS.

After the invasion of Kuroshio water into the SCS, it is mainly distributed in the northwestern LS and the southwestern TS, with a high temperature and salinity; the seasonal changes in temperature and salinity properties are not obvious. However, there are large seasonal differences in the invasion intensity of this water mass in the SCS. The invasion of Kuroshio surface water (KS) does not exceed 119.5° E in summer. However, after the KS enters the SCS in winter, it can reach further west in spite of a rapid decrease in salinity. In addition, a low-salinity body with a salinity value of 34.60 still exists near 117.5° E. The salinity value of Kuroshio subsurface water (KU) is significantly higher than that of KS. During its invasion into the SCS, the salinity of KU only decreases by about 0.1 near the LS, retaining the high salinity properties of Kuroshio water to a large extent. The invasion of KU into the SCS is still more powerful in winter than it is in summer, and the water mass is able to reach further western areas than the KS in the same season.

The SCS subsurface water (U) is formed by the mixing and denaturation of subtropical subsurface water from the WPO into the SCS, which is mainly distributed in the 100–200 m water layer. The temperature of this water mass has a seasonal variation of about 1 °C. The seasonal variation of salinity values is small, but the regional variation is obvious. Influenced by the invasion of KU, the salinity value in the sea near the LS is higher than that of other waters in the SCS.

The SCS subsurface–intermediate mixed water (UI) mainly comes from the Western Pacific subtropical thermocline water and is distributed in the 200–350 m water layer. The seasonal temperature variation is about 1.2 °C, and the seasonal salinity variation is about 0.05. In the vertical direction, the temperature and salinity of this water mass decrease with the increase in depth; this temperature and salinity vertical difference is the largest near the LS and gradually disappears to the south.

The South China Sea intermediate water (I) is mainly distributed in the 350–800 m water layer, where the temperature is high in summer and low in winter, but the salinity value basically does not have seasonal changes. After the source area water entered the SCS, the salinity value increased. However, the change is only about 0.1, which is not significant. The low salinity characteristics of this water mass were still very obvious compared with the upper layer and the lower layer, which better reserved the low salinity properties of the intermediate water in the WPO. The South China sea deep water (D) comes from the WPO’s deep water and is mainly distributed in the SCS basin area deeper than 900 m, with insignificant seasonal changes in temperature and salt values.

The analysis of the characteristics and seasonal variations of ten water masses in the SCS reveals that each water mass has been degenerated to different degrees compared with the source water, which is the most important characteristic of shallow water masses and the main reason for the difficulty of dividing shallow water masses. The results of the classification and analysis of the SCS water masses are in good agreement with the existing conclusions, indicating that the K-means clustering method has high accuracy in the classification of shallow water masses.