The Luzon Strait is located between the Taiwan Island and Luzon Island, spanning from 18.5°N to 22°N, covering a distance of over 350 km, with a maximum depth exceeding 3 000 m. It serves as the only channel for material exchange between the northwestern Pacific Ocean and the SCS. The study area is situated in the East Asian monsoon region. During summer, the southwest monsoon dominates with weak winds, which is not favorable for Kuroshio intrusion. In contrast, the northeast monsoon prevails during winter, bringing strong winds that facilitates Kuroshio intrusion (Yuan et al., 2006). In addition, the study area is affected by tropical cyclones from the northwestern Pacific Ocean, which frequently occur from July to October. When tropical cyclones form, they have the potential to move northwest or turn northward near the Luzon Stait, affecting nearby waters (Sun et al., 2017).

Typhoon No. 1918 MITAG originated as a tropical depression in the Northwest Pacific (13°N, 138.9°E) at 08:00 (UTC+8) on September 27, 2019. It passed east of the Luzon Strait and Taiwan Island as a typhoon with a maximum wind speed of 40 m/s and a translation speed range of 20–40 km/h from 05:00 on September 29 to 00:00 on October 1 (Fig. 1). Typhoon data were obtained from the Typhoon Network of the Central Meteorological Observatory (NMCS, http://typhoon.nmc.cn). The study area was primarily influenced by the southeast monsoon, and the degree of Kuroshio intrusion was weak before the observation period.

Gliders are a data collecting vehicle that is capable of diving and ascending, by adjusting its in-water buoyancy, along an underwater saw-tooth trajectory, or maintaining gliding mode using hydrodynamic force (Wang et al., 2017). The Petrel-4000 glider (Wang et al., 2017) used in this study equipped with an Idronaut OS304 Plus CTD that records depths from 0 m to 7 000 m with accuracy of 0.05% F.S. (full scale) and resolution of 0.001 5% F.S.; temperatures from –5℃ to 35℃ with accuracy of 0.002 and resolution of 0.000 1; and conductivities from 0 mS/cm to 90 mS/cm with accuracy of 0.003 and resolution of 0.000 3 (Idronaut, 2019). This type of glider has been used in several field studies with satisfactory reliability (Ma et al., 2018; Li et al., 2019, 2020). One Petrel-4000 glider was deployed on 14 September 2019 in northeastern SCS (Fig. 1). Its primary goal was to measure temperature and salinity profiles along the upper reach of the Manila Trench coordinated by a different research project (Liu et al., 2023). In 41 d until recovery on October 25, 2019, a total of 35 valid temperature and salinity profiles (from September 28 to October 12), sampled at 1 Hz, were recorded (Figs 1a and b). For this study, only the data in depth range of 0–500 m are used.

The following steps were followed to process data. (1) Thermal lag errors in the conductivity data from unpumped CTD sensors on the glider were corrected using the method proposed by Morison et al. (1994) and Garau et al. (2011). The temperature reported by the CTD showed a slight time-lag behind the actual value inside the conductivity cell, resulting in an error of –0.1 to 0.2, mostly less than ± 0.05 (Fig. 2a). (2) The difference between the salinity before and after being burred by moving average filtering (Liu et al., 2015; Yi et al., 2019) was mostly within ±0.05 (Figs 2b and c), which is of the same order of magnitude as the thermal lag error. (3) The corrected data were interpolated to create a profile with a vertical bin size of 0.5 m (Troupin et al., 2015). For each pair of dives, the measurements of diving and climbing profiles were averaged to produce the binned vertical profile. This profile was assigned to the middle position between the end coordinates of the first dive and the start coordinates of the next dive. The mixed layer depth (MLD) is defined as the depth at which the potential density increase relative to the surface (10 m) equals the increase in surface potential density when the surface temperature decreases by 0.5℃ (de Boyer Montégut et al., 2004; Wang et al., 2023).

Monthly climatological temperature and salinity data from World Ocean Atlas 2018 (WOA18), produced by NOAA National Oceanographic Data Center, were obtained from https://www.nodc.noaa.gov/OC5/woa18/ with a horizontal resolution of 0.25°. The data of Argo Float (Serial No. 2902753), released on 29 March 2019, are obtained from ftp://ftp.argo.org.cn/pub/ARGO/global/core/ and included pressure, temperature and salinity measurements. Daily sea level anomaly (SLA) and surface geostrophic current with a horizontal resolution of 0.25° were provided by CMEMS (Copernicus Marine Service, http://marine.copernicus.eu). The wind filed data were obtained from the NCEP Climate Forecast System Version 2 (CFSv2) Selected Hourly Time-Series Products (https://rda.ucar.edu/datasets/ds094.1/), with a spatial resolution of 0.2°.

To the east of the Luzon Strait, an Argo float (Serial No.: 2902753) was deployed on March 29, 2019, and passed by the east of Luzon Island in September and October 2019. The trajectory of float from September 23 to October 17 is shown in Fig. 1 with a 3-d interval. These trajectories located at the path of the Kuroshio (Fig. 1), and the observed temperature and salinity were similar to the KW in the upper layer (Fig. 5a). Also, a recent observation from Ma et al. (2022) found that the core of the Kuroshio east of Luzon Island was located at 122.7°E, which near the trajectories of the Argo float. Therefore, the Argo dataset could represent the variation in KW during the study period. It revealed that the HSW was mainly distributed at depths ranging from 80–250 m (Fig. 6b), which was deeper than the observations by glider. The subsurface maximum salinity exceeds 38.95. After the typhoon, the surface temperature and salinity decreased by about 0.8℃ and 0.4, respectively (Figs 6a and b). The MLD decreased from around 60 m to 40 m, and the HSW in subsurface was uplifted (Fig. 6b). Until October 9, the surface temperature, salinity and MLD recovered as before.

Previous studies have demonstrated that the 34.7 salinity isohaline effectively distinguishes between KW and SCSW (Qu, 2000; Nan et al., 2011; Wang et al., 2021). Kuroshio intrusion inevitably pushes high salinity KW through Luzon Strait and sometimes wells into the SCS (Xue et al., 2004; Caruso et al., 2006). The monthly climatology salinity distributions from WOA18 indicate that KW is widespread west of the Luzon Strait on September and October, but no water with salinity greater than 34.8 has crossed the Luzon Strait (Figs 5b and c). The high salinity (>34.9) in the subsurface from the Argo dataset agreed with the climatology data (Fig. 6b). However, the glider monitoring results indicated an unusually large range of HSW in this season, extending to 20.3°N (Fig. 3b), which is broader when compared to previous observations in same season (Zhou et al., 2009; Huang and Hu, 2010; Wang et al., 2021, 2023).

To explore the processes of HSW intrusion, we plotted the time evolution of sea level anomaly (SLA) and geostrophic currents from September 28 to October 14, 2019 with a 2-d interval (Fig. 7). According to the types of Kuroshio intrusion identified by Nan et al. (2011), the Kuroshio leaped the Luzon Strait before typhoon MITAG. The eastern side of the trajectories of glider corresponded to the Kuroshio main axis, while the western side was the center of a cyclonic eddy. The convex-upward isopycnal in the cyclonic eddy resulted in a decrease in surface temperature, consistent with observations by glider (Figs 3a, b, and c). Meanwhile, the surface water with high temperature (>28.5℃) and low salinity (<34) appeared to originate from the anti-cyclonic eddy south of the observation area (Figs 3a, b, and 7a). Following the typhoon, significant changes were observed in the Kuroshio main axis and the eddies east of the Luzon Strait. In Fig 6a, we defined the Kuroshio main axis location as the longitude with the maximum v-component velocities within the green rectangular area (20.125°–21.875°N, 120.125°–121.875°E) (Kuo et al., 2018) and denoted the cyclonic and anti-cyclonic eddies as “CE” and “AE” (Fig. 7a). Figure 8 depicts the variations in the Kuroshio main axis and related eddies (CE, AE) before and after the typhoon. Prior to the typhoon, the Kuroshio main axis near 21°N was positioned around 121.125°E (Fig. 8a), with a northward velocity of about 0.5 m/s (Fig. 8b). During the typhoon, the Kuroshio current weakened by about 0.1 m/s but remained in the same position. After the typhoon, the main axis shifted westward by about 0.5° and its velocity decreased to about 0.25 m/s. It wasn’t until Octorber 6 that the main axis returned to its original position, although its velocity did not recover to the initial level (Figs 8a and b). Additionlly, we extracted the velocity magnitude of the Kuroshio upstream (near 18.5°N) during the study period. It exhibited a decrease after the typhoon but rebounded to it initial level after the typhoon on October 6 (Fig. 8c). Therefore, we concluded that the influence of typhoon on the Kuroshio persisted for approximately 7 d, from October 2 to October 9, which aligned with the observations by glider, indicating a deeper MLD (Fig. 3d) and the appearance of HSW (Fig. 3b) during this period eddies in the upper ocean can be influenced by typhoons and may not fully recover in the short term, as noted in previous studies (Sun et al., 2009; Hu et al., 2012; Hsu et al., 2018; Zhang et al., 2023). The variations in SLA and geostrophic currents (Fig. 7), coupled with the increase of total relative vorticity and the decrease of SLA east of the Luzon Strait (blue box in Fig. 7a; Figs 8d and e), suggested that CE was strengthened while AE was weakened after the typhoon. The rise of the pycnocline following the typhoon, as observed from the Argo dataset, might be partially attributed to the weakening of AE (Fig. 6). These changes in the eddies also resulted in a reduction in the velocity of the Kuroshio current. As the Kuroshio flows northward through the Luzon Strait, the β effect and potential vorticity conservation cause it to bend lead clockwise (anti-cyclonic), promoting Kuroshio intrusion. The decrease in Kuroshio velocity renders its main axis more susceptible to this clockwise bending, further facilitating Kuroshio intrusion (Yaremchuk and Qu, 2004; Nan et al., 2015; Gao et al., 2022).

During the course of the typhoon, the turbulent local wind field typically induce vigorous mixing, Ekman pumping and transport, leading to upwelling or downwelling in the upper ocean. Wind curl (S), Ekman pumping velocity (EPV, $ {W}_{{\mathrm{E}}} $), and Ekman transport were calculated using the equations $ S= \dfrac{\partial v}{\partial x}-\dfrac{\partial u}{\partial y} $, $ {W}_{{\mathrm{E}}}=\dfrac{\nabla \times \tau }{{\rho }_{{\mathrm{s}}}f} $ and $ {M}_{x}=\dfrac{{\tau }_{y}}{f} $, where $ \tau ={\rho }_{{\mathrm{a}}}{C}_{{\mathrm{D}}}{u}^{2} $ is the wind stress, $ f=2\omega\ \mathrm{sin}\ \theta $ is the Coriolis parameter for latitude $ \theta $, $ {\rho }_{{\mathrm{a}}} $, and $ {\rho }_{{\mathrm{s}}} $ are the densities of air and sea water, respectively, u and v represent wind velocities at 10 m above sea level , $ \omega $ is the rotation rate of the Earth, and $ C_{\mathrm{D}} $ is the speed-dependent drag coefficient. Figure 9 illustrates the distribution of wind stress curl and Ekman transport during the typhoon. Positive wind stress curl prevailed to the east of the Luzon Strait (Figs 9a and b), resulting in robust Ekman upwelling (Fig. 8d, orange line) and westward Ekman transport (Figs 9d and e). However, the Kuroshio main axis remained stable, with only a slight velocity decrease (Figs 8a and b), suggesting the typhoon-induced westward Ekman transport was not the direct cause of Kuroshio intrusion. Typhoon MITAG introduced positive potential vorticity to the east of the Luzon Strait (Fig. 8d, blue and red lines), weakening AE and strengthened CE. This led to a significant reduction and westward shift in the Kuroshio main axis, with upper ocean effects lagging by 1–2 d but persisting for nearly 7 d. Generally, the anti-cyclonic eddies at the edges of typhoon intensified due to strong negative wind stress curl (Zhou et al., 2017; He et al., 2024). The wind stress curl at AE alternated between positive and negative EPV (Fig. 8d, green line), with positive curl ultimately dominating and weakening AE. Furthermore, the center of the typhoon was east of the Kuroshio main axis, and northerly winds influenced this axis. This may have caused a buildup of water levels to the Luzon Strait, resulting in a difference in water level with the south and facilitating a geostrophic flow to the west. This, in turn, contributed the westward shift of the Kuroshio main axis (Gao et al., 2022). The frequent occurrence of tropical cyclones (typhoons) near the Luzon Strait likely played a role in the high-frequency disturbances in the Kuroshio intrusion (Hsin et al., 2010). It is worth pointing out, however, that the mechanisms of short-term strengthening of Kuroshio intrusion is likely more complex than the action of eddies. Other factors such as instability of the Kuroshio, Rossby waves responding to remote forcing, among others, may also have contributed to the processes (Lu and Liu, 2013; Nan et al., 2015). More high-resolution observations and numerical modelling efforts are required to untangle these relationships.

The influence of strengthened Kuroshio intrusion was not only in surface and subsurface, but also in the deeper water. The mixing, upwelling and horizontal advection due to passing of typhoon caused the decreased in temperature in the surface and subsurface, but did not affect the deeper regions (>200 m) (Figs 4a and d). The increase in temperature in the depth of 200 m to 400 m was mainly influenced by the Kuroshio water with high temperature and salinity, also the change of salinity also indicated the influence range of Kuroshio intrusion (Fig. 4b). After October 6, the typhoon process weakened and the Kuroshio intrusion increased the surface and subsurface temperature slightly. Finally, the Kuroshio intrusion speeded up the recovering of temperature after the typhoon.