1. Introduction
The Indochina Peninsula (IDP) is a geographically unique entity situated between the Indian subcontinent and East Asia, where two contrasting Asian monsoon systems exhibit distinct characteristics [
1,
2,
3]. During the boreal winter, the climate condition over the IDP is dominated by the meridional circulation associated with the intensified Siberian high (SH) [
4]. It is generally characterized by ascending warm air over the Indonesian maritime continent and descending cold continent air over China, resulting in two major branches of low-level wind fields blowing toward the subtropical western Pacific and South China Sea [
5]. A succession of northeasterly surges with outbursts of SH is one of the most conspicuous features of the East Asian winter monsoon (EAWM) [
6]. The active EAWM exerts a strong impact on the extratropical and tropical planetary-scale circulations and influences the convection over the tropical western Pacific [
6], and its variability is linked to the forcing from sea surface temperature anomalies in the tropical Pacific Ocean [
3,
6,
7]. Although the Pacific Ocean is absolutely crucial, the Indian Ocean is also important for the Asian monsoon variability [
8]. Available evidence has indicated that the variability of EAWM exerts social-economic impacts on many Asian countries [
4,
5,
6]. Hence, a better understanding of EAWM variability is still a great scientific challenge and concern.
The variability of EAWM and its strength can be characterized by wind behavior at many levels [
5,
6,
9,
10,
11]. These winds have been used to construct the indices, such as wind at 10 m [
5] and wind at 1,000 hPa [
6,
10]. Many indices derived from low-level wind have been introduced to reveal the intensity of the EAWM [
12]. In general, 850 hPa wind circulation is less influenced by surface roughness than the wind at the 10 m and 1,000 hPa levels. Furthermore, upper-level wind, sea level pressure (SLP) and trough behavior can be used to characterize the monsoon strength as an index representing different aspects of the EAWM [
12]. For example, the index derived from zonal wind at the 300 hPa level represents the association between the East Asian jet stream and EAWM variation [
12]. To further extend our current understanding, the relationship between EAWM indices and winter monsoon over the IDP is of great interest, because the IDP is a tropical regional scale that has different climate features from the larger scale. Furthermore, this is also due to the fact that the winter monsoon variability over the IDP has less studies than its summer counterpart [
2,
11,
13,
14,
15,
16]. These are importance for the IDP countries and their neighbors.
It is widely accepted that sea surface temperature (SST) variability in the equatorial Pacific Ocean plays an important role in atmospheric processes and climate variability. Many previous studies have shown that SST variability associated with El Niño-Southern Oscillation (ENSO) events strongly affects the EAWM [
7,
17,
18,
19,
20], and the SST in the Indian Ocean is important [
8]. Thus, there is a possibility to have some roles of the SST variability playing on the winter monsoon over the IDP.
In this study, we analyzed the monthly gridded data using an analysis of the empirical orthogonal function for complex numbers to reveal spatio-temporal structures of wintertime winds over the IDP and to disclose any possible connections to the EAWM. The relationships between the SST and wintertime low-level wind, representing winter monsoon, and its impact on precipitation are further illustrated. These analyses aim at providing empirical evidence in support of a better understanding of winter monsoon over the IDP, demonstrating some connections to the EAWM and its association with SST variation.
2. Data and Analytical Methods
The monthly gridded data of the Japanese 25-year reanalysis dataset (JRA-25) with 1.25° × 1.25° horizontal resolution [
21] during the boreal winter months (December-January-February) covering the period of 1979–2010 were used for this study. The wind data at 850 hPa were extracted for the IDP region (5°–30°N, 90°–110°E) to analyze the dominant spatio-temporal modes by the empirical orthogonal function (EOF) method. Because winds have magnitude and direction, the vector analysis is more suitable than the traditional scalar analysis to keep the meaning of the wind. Some previous studies used EOF to analyze wind components by forming the complex numbers [
22,
23,
24]. The wind vector is in complex exponential form and is defined by the direction measuring from East to North and the magnitude of the vector as the wind speed [
22]. Furthermore, the wind vector in complex rectangular form consists of the zonal and meridional wind components, represented by the real and imaginary parts, respectively, and can be used for the analysis, as well [
23]. In this study, the wind components at 850 hPa were used to form a complex number in a rectangular form with respect to the zonal and meridional axes. A matrix (
S) of dimension
N ×
M consists of complex number elements (
ekm) formed as follows:
where
u',
v',
k = 1,...,
N, and
m = 1,...,
M denote for zonal wind anomaly, meridional wind anomaly, the location and time of the data, respectively. To analyze a matrix of complex number elements by EOF, a symmetric composition of complex numbers, except for diagonal elements, named the Hermitian matrix (
H), is required. There are important properties of
H, which are as follows: (1) eigenvalues are real; (2) eigenvectors corresponding to the distinct eigenvalues meet orthogonally; and (3) it is unitarily diagonalizable. The matrix used to form a Hermitian matrix for the analysis presents as:
where
H is the Hermitian matrix and
S† is the complex conjugate transpose of the sample matrix (
S) that has
M time series records. Therefore, eigenvectors (
Êj) are determined by:
and satisfy the unitary condition:
where
Êi† is the complex conjugate transpose of
Êi and
δij is the Kronecker delta function. Thus, the eigenvalues ,
λj, are real, and the complete set of
N eigenvectors meets the orthogonality condition. The eigenvectors,
Êj, are called modes of empirical orthogonal analysis. They are used to expand data as:
where
The
ckm is unique and in complex number form (hereafter referred to as the principal component (PC)). Nevertheless, there is an arbitrary phase factor (
θk) associated with the
k-th eigenvector; the set of vectors is exp(
iθk)
Êk. We take this into account by choosing
θk when the mean value of the argument is nearly zero to orient the corresponding eigenvector and PC [
22]. Thus, the real component variation represents variation in magnitudes of corresponding eigenvector elements, whereas the imaginary part that is perpendicular to the vector indicates the strength of the vector rotation in a counterclockwise direction [
23]. The EOF analysis for complex numbers dealing with two components, which is suitable to present co-varying spatial patterns [
25], was performed. It differs from the EOF analysis for complex numbers dealing with one variable to capture propagating patterns and differs from the complex Hilbert EOF analysis (sometimes, it is referred to as the complex EOF analysis) related to the frequency domain [
25].
Since the EAWM is the prominent climate feature during the winter season, it is important to reveal the relationship between the winter monsoon over the IDP and the EAWM. The leading PC of the wintertime low-level winds over the IDP given by EOF analysis of complex numbers was then used to present the variability of the winter monsoon over the IDP. Any relationships of them were explored based on correlation analysis between the leading PC and the EAWM indices derived from wind fields and SLP. Eight EAWM indices were calculated based on the JRA-25 data for the boreal winter period (
Table 1). The regression technique [
26] was utilized to separate the signal into two parts, which are the low-level wind variability over the IDP related to the feature of interest and the unrelated part.
Finally, the SST data obtained from the Hadley Centre sea ice and sea surface temperature dataset (HadISST) with 1° × 1° resolution [
27] and the monthly gridded data of precipitation from the University of Delaware (UDel) with 0.5° × 0.5° horizontal resolution [
28] from the National Oceanic and Atmospheric Administration/Office of Oceanic and Atmospheric Research/Earth System Research Laboratory (NOAA/OAR/ESRL) were used for analyses of its relation to SST variability and its impacts on precipitation, respectively.
Table 1.
The selected East Asia winter monsoon (EAWM) indices for the correlation analysis. SLP, sea level pressure.
Table 1.
The selected East Asia winter monsoon (EAWM) indices for the correlation analysis. SLP, sea level pressure.
No. | Index | Index Representative | Reference |
---|
1 | IChen | An average of the meridional wind component at 10 m over the region of the East China Sea (25°–40°N, 120°–140°E) and the South China Sea (10°–25°N, 110°–130°E). | [5] |
2 | ILiu-M | An average of the meridional wind component at the 1,000 hPa level over the mid-high latitude area (30°–50°N, 110°–125°E). | [10] |
3 | ILiu-L | An average of the meridional wind component at the 1,000 hPa level over the low-latitude area (10°–25°N, 105°–135°E). | [10] |
4 | IJi | An average of the meridional wind component at the 1,000 hPa level over the area of 10°–30°N, 115°–130°E. | [6] |
5 | ILu | An average of the meridional wind component at the 1,000 hPa level over the South China Sea (7.5°–20°N, 107.5°–120°E). | [9] |
6 | IXu | A sum of the zonal SLP differences (110°E minus 160°E) from 20°N to 50°N. | [29] |
7 | IWu | A sum of the zonal SLP differences (110°E minus 160°E) from 20°N to 70°N. | [30] |
8 | IJhun | A difference in the area-averaged zonal wind speed at the 300 hPa level between a region of 27.58°–37.58°N, 110°–170°E and a region of 50°–60°N, 80°–140°E. | [31] |
4. The PC1s-SST Relationship
A possible cause of a higher correlation between PC1s and low latitude EAWM wind indices rather than the others of mid-high latitudes, is the modulation with other active climate features, such as ENSO. This plays a role in the maritime continent during the winter monsoon [
36]. The ENSO is an important phenomena revealed by SST variability in the equatorial Pacific Ocean. Many studies indicate that there is an association between EAWM and ENSO [
3,
17,
18,
19]. In addition, there is a significant correlation of the northeasterly wind over the IDP represented by PC1s (hereafter, referred to as the northeast (NE) monsoon) with the low latitude EAWM characteristic. It is a possibility that there is another possible climate forcing influencing the variability of wintertime low-level wind over the IDP. This would be the forcing from SST variability in the Pacific Ocean and is interesting for revealing the relationship between NE monsoon and SST in the Pacific Ocean, which has not been completely investigated. The PC1s were employed to construct the correlation maps of these to sea surface temperature anomalies
(SSTAs) for investigation, as shown in
Figure 6.
Figure 6.
Correlation maps between (a) the primary part of the leading principal component (PC1pri) to sea surface temperature anomalies; (b) is the same as (a), but for the secondary component (PC1sec). Shading presents correlation coefficients, and the thick and thin contour lines denote the significance levels at 0.01 and 0.05, respectively. SSTA, sea surface temperature anomaly.
Figure 6.
Correlation maps between (a) the primary part of the leading principal component (PC1pri) to sea surface temperature anomalies; (b) is the same as (a), but for the secondary component (PC1sec). Shading presents correlation coefficients, and the thick and thin contour lines denote the significance levels at 0.01 and 0.05, respectively. SSTA, sea surface temperature anomaly.
The figure shows a negative correlation between PC1s and SSTA in the Indian Ocean, with smaller magnitudes than those in the Pacific Ocean, while the significantly correlated areas in the Pacific Ocean are wider than those in the Indian Ocean. Based on this evidence, it is possible that the SST variability in the Pacific Ocean plays a greater role in the wintertime low-level wind variability over the IDP than that of the Indian Ocean.
Over the tropical central-east Pacific Ocean, there are negative correlations between PC1s and SSTAs. Increasing PC1s values correlated with decreasing SSTAs. This indicates that the cooling (warming) of SST over the tropical central-east Pacific Ocean is relatively in phase with the strengthening (weakening) of northeasterly winds over the IDP. For the west Pacific Ocean, this shows a positive correlation, which differs from the correlation presenting for the tropical central-east Pacific Ocean. The positive correlation shows a V-shape in the horizontal plane pattern, which extends eastward and poleward, indicating that an increase in PC1 values is correlated with high SSTA. This implies that the warming (cooling) of SST over the tropical central-east Pacific Ocean is related to the strengthening (weakening) of northeasterly wind over the IDP. These results are quite consistent with studies [
5,
20] and agrees with a previous study [
7] that suggests the warm SSTA in the tropical central-eastern Pacific Ocean leading to the weakening of the western Pacific Hadley cell, causing the EAWM to be weakened. On the other hand, the cold SSTA in the central-eastern Pacific Ocean leads to presenting a strong EAWM, with air ascending in the equatorial western Pacific Ocean, resulting from the ENSO [
7].
To confirm the association between PC1s and SSTA in the equatorial Pacific Ocean, the correlation analysis for −11- to +11-month lags was performed. An important activity in the Pacific Ocean, El Niño, has been measured and monitored through SSTA variation [
37]. The Niño3.4 index, which is averaged SSTA over 5°S–5°N, 170°–120°W, was used for the analysis. For the importance of the Indian Ocean, the influence of SST anomalies in the tropical Indian Ocean (TIO) is measured by the index (hereafter, referred to as the TIO index) introduced by [
8], which is determined by the average of SSTAs over the TIO area (40°–90°E, 10°S–10°N). It was used to confirm the association between PC1s and SSTAs in the Indian Ocean. Three-month running means were performed on SSTA to reduce the influence of seasonal means before the analysis, as suggested by a previous study [
38]. The result is shown in
Figure 7.
The result shows significant negative correlations of PC1s to Niño3.4 index on previous months and following months, which starts from three months for PC1pri (
Figure 7a) and five months for PC1sec (
Figure 7b) before DJF (0), and both persist for three months later. Thus, the air-sea interaction over the equatorial Pacific Ocean affects the variability of NE monsoon over the IDP represented by PC1s. This result indicates that there is an association between NE monsoon variability and ENSO. The ENSO can trigger the monsoon over the IDP, and its relation persists a few months after the winter. The negative correlations between PC1s and the Niño3.4 index imply that the strengthened NE monsoon is associated with the negative phase of Niño3.4, which is considered as La Niña when the Niño3.4 values exceed −0.5 °C, consecutively, and
vice versa for the weakened NE monsoon associated with El Niño.
For the TIO, there is a significant correlation between PC1pri and the TIO index before winter from January to June, but the PC1sec significant correlation is absent, as shown in
Figure 7c,d. The result implies that the variability of SSTA in the TIO correlated to the variation of the NE monsoon over the IDP, due to the wind in the parallel direction to the eigenvector field of the leading mode, rather than that in the perpendicular direction. The relation of PC1s to the TIO index also shows early and less significant influence on the wintertime low-level winds over the IDP, as compared to the SSTA variation in the tropical Pacific Ocean. This agrees with a previous study, that suggested the SST anomalies in TIO are likely to be the factor driving the EAWM [
8], but our result shows little difference in the time period, showing early significant correlation. Although this is different, it is interesting to study further, including the direct and indirect influences or the modulation of SST anomalies in TIO with ENSO. For the next analysis, we only focus on ENSO during the boreal winter.
Figure 7.
Correlation between (a) the primary part of the leading principal component and Niño3.4 represented by the three-month running means of SSTA (0 represents the corresponding DJF mean of SSTA variation, and the negative and positive running means represent the previous and following months considered at the middle point, respectively); and (b) is the same as (a) but for the secondary component. (c) and (d) are similar to (a) and (b), but for the tropical Indian Ocean (TIO) index. The shading presents statistical significance at the 95% confidence level.
Figure 7.
Correlation between (a) the primary part of the leading principal component and Niño3.4 represented by the three-month running means of SSTA (0 represents the corresponding DJF mean of SSTA variation, and the negative and positive running means represent the previous and following months considered at the middle point, respectively); and (b) is the same as (a) but for the secondary component. (c) and (d) are similar to (a) and (b), but for the tropical Indian Ocean (TIO) index. The shading presents statistical significance at the 95% confidence level.
To investigate the mechanism related to the influence of ENSO, regression analysis was applied [
26]. The PC1s-res, which represent the variability of northeasterly wind over the IDP without the influence of the thermal contrast in the temperate region, were analyzed to separate them into the ENSO-related part (PC1s-res-EN) and the ENSO-unrelated part (PC1s-res-uEN) by the regression technique, as described in a previous study [
26]. The contributions of ENSO to PC1pri-res and PC1sec-res are 17% and 25% of their total variances, respectively, whereas the contributions of the ENSO unrelated part are 83% and 75% for PC1pri-res and PC1sec-res, respectively. The ENSO-related patterns show the positive anomalous SLP over the temperate mainland and the north Eastern Pacific Ocean (
Figure 8a,c). They also present the anomalous low pressure in the west Pacific Ocean with the cyclonic circulation around the maritime continent. These indicate that the variability of northeasterly wind over the IDP is influenced by wind blowing from the mid-latitude to the IDP, which is induced by the cyclonic circulation. These reveal the role of ENSO. On the other hand, the ENSO-unrelated part of PC1pri-res (
Figure 8b) presents wider positive anomalous SLP over the eastern Pacific Ocean, which implies the high pressure over this area influencing the wind variability over the IDP in a parallel direction to the directions of wind vectors given by the eigenvector. For the PC1sec-res, which is related to the variation of winds in the perpendicular direction, the pattern (
Figure 8d) shows two poles of the positive anomalous SLP over the areas around Siberia and the Gulf of Alaska without the obvious SLP forcing in the tropical area. These indicate that the two poles in the mid-latitude influence the variability of northeasterly wind over the IDP without the ENSO forcing by driving winds in the high latitude to the tropical region. Therefore, the variability of northeasterly wind over the IDP is influenced by the modulation of the tropical and temperate forcing, which are the ENSO and the SLP forcing in the mid-latitude.
Figure 8.
Spatial patterns of composite differences between positive and negative phases of (a) the El Niño-Southern Oscillation (ENSO) related part and (b) the unrelated part of PC1pri-res; (c) and (d) are the same as (a) and (b), but for PC1sec-res. The vector and contour line are the differences of wind (meters per second) and sea level pressure (hectopascals), respectively, and the shading contour is the 0.05 significance level.
Figure 8.
Spatial patterns of composite differences between positive and negative phases of (a) the El Niño-Southern Oscillation (ENSO) related part and (b) the unrelated part of PC1pri-res; (c) and (d) are the same as (a) and (b), but for PC1sec-res. The vector and contour line are the differences of wind (meters per second) and sea level pressure (hectopascals), respectively, and the shading contour is the 0.05 significance level.
Nevertheless, a previous study [
39] reports that the relationship between the ENSO and the EAWM is not stationary, and the ENSO exerts influence on the EAWM through the Pacific-East Asian teleconnection (an anomalous anticyclonic circulation around the western North Pacific to the east of The Philippines). The influence of ENSO on EAWM has been weak since 1976 [
39], but we observed the association with ENSO. This can be explained by the meridional winds, which are always significant, and the winds associated with the anticyclone (see
Figure 3 in [
39]). Hence, the non-stationary ENSO-EAWM relationship does not influence the IDP region.
For the next analysis, it is important to present evidence of the NE monsoon impact on precipitation, because there is the association between PC1s and ENSO, the linkages between rainfall and ENSO [
36,
40,
41,
42] and the linkage between ENSO with EAWM [
7]. The monthly gridded dataset of precipitation from the University of Delaware (UDel) [
28] from the National Oceanic and Atmospheric Administration/Office of Oceanic and Atmospheric Research/Earth System Research Laboratory (NOAA/OAR/ESRL) was used for the analysis.
The spatial distribution of climatic precipitation covering a period 1979–2010 during the boreal winter shows more precipitation on the southern part of IDP, along the coastline area of Vietnam, and some parts of Guangxi and Guangdong of China, than the mainland of IDP, which includes Myanmar, the upper part of Thailand, Laos and Cambodia, except some parts of Arunachal Pradesh of Northeast India and the northern part of Myanmar (
Figure 9a). The possible impact of the NE monsoon on precipitation over the IDP is revealed by regression analysis of precipitation on PC1s.
Figure 9.
Spatial distributions of (a) climatic mean during the boreal winter of monthly total precipitation from 1979 to 2010, (b) the regressed precipitation anomaly on PC1pri, and (c) is the same as (b), but for the PC1sec. The shading color shows the regression results in units of centimeters per month, whereas the dashed line presents the significance level at 0.05.
Figure 9.
Spatial distributions of (a) climatic mean during the boreal winter of monthly total precipitation from 1979 to 2010, (b) the regressed precipitation anomaly on PC1pri, and (c) is the same as (b), but for the PC1sec. The shading color shows the regression results in units of centimeters per month, whereas the dashed line presents the significance level at 0.05.
The regressed precipitation patterns (
Figure 9b,c) are positive over the area from the south central coast of Vietnam to the Mekong Delta, the southern part of Cambodia and the southern tip of IDP, which indicate the enhancement of northeasterly wind strength (the increasing of PC1s values), resulting in more precipitation over these areas, and
vice versa for the weakening of the NE monsoon (the decreasing of PC1s values), showing the reduction of precipitation there in terms of a linear relationship. These results agree with the blowing pathway of the wind connected with the EAWM. It passes the South China Sea, penetrating into the IDP, which carries moisture from the SCS and the Gulf of Thailand to the eastern coast area and the southern area of the IDP, respectively. On the other hand, the negative presence over areas around Guangxi and Guangdong of China and the northern part of Vietnam, significantly, indicates the positive phase of the northeasterly wind, resulting in wind blowing more southward, carrying a dry air mass, causing less precipitation over these areas (
Figure 9b,c). According to a previous study [
20], these results (
Figure 6 and
Figure 9) are like those reported in [
20], which indicate that the impact on precipitation over the IDP was influenced by the conventional ENSO. Therefore, there is a possibility to indicate possible impacts on precipitation in the IDP by the PC1 indices, utilizing the relationship among the PC1, EAWM and ENSO indices.
5. Conclusions
Over the IDP, we revealed spatio-temporal wintertime wind variability and have identified the principal mode of the wind at 850 hPa by using the EOF analysis for complex numbers. The spatial patterns given by the regression of wind on the leading PC and the patterns given by the composite analysis show prominent northeasterly wind over the IDP, known as the NE monsoon, which agree with the characteristic of the EAWM and its influence on the tropical region. Namely, these results are consistent with the circulation characteristic of EAWM, exhibiting wind blowing along the coast of East Asia and the influence of the Borneo vortex. Thus, the first mode of wintertime wind variability over the IDP is affected by the EAWM via wind blowing that passes through the SCS to the IDP.
We further presented the possible connection of wintertime low-level wind variability over the IDP to the EAWM on the basis of the correlations between the PC1 and EAWM indices. The PC1 shows good correlations with the EAWM indices, characterized by lower latitude wind than others. The comparison of correlations between the PC1 and EAWM indices indicates that the ILu index is a suitable EAWM index to reveal the variability of the winter monsoon over the IDP.
Hence, the relation of PC1s to EAWM indices is not fully correlated. The correlation analyses focusing on SSTA show that there is a more significant correlation between PC1s and SSTA in the Pacific Ocean than in the Indian Ocean, quite similar to the ENSO pattern. Thus, another possible forcing influencing the NE monsoon variability over the IDP is mainly related to SSTA in the Pacific Ocean. The ENSO represented by the Niño3.4 index was used to reveal its association with the NE monsoon variability over the IDP. The results show that the variability is influenced by wind blowing from the mid-latitude to the IDP, which is induced by the cyclonic circulation related to the ENSO-related part of PC1s. Although the relationship between the ENSO and the EAWM is not stationary, the meridional wind is important [
39]. Therefore, the non-stationary ENSO-EAWM relationship does not influence the IDP region, and the wintertime low-level wind variability over the IDP is influenced by the modulation of the tropical and temperate forcing.
Furthermore, the result of the analysis showed the strengthening (weakening) of the NE monsoon, causing more (less) precipitation over the central coast of Vietnam to the Mekong Delta, the southern part of Cambodia and the southern tip of the IDP. It is an interesting challenge to understand more about the winter monsoon over the IDP associated with ENSO by a modeling study and utilizing the Niño3.4 index to develop a forecast system of monsoons and their impacts. Furthermore, it is required to obtain a better understanding, particularly in terms of the non-linear relationship between another forcing, such as the Indian Ocean Dipole, and the winter monsoon over the IDP.