For each Argo profile, the following variables were calculated: SST, MLD, ILD, BLT, z 26 , OHC and TCHP. SST is considered as the temperature at the shallowest observation depth. T 10 and σ 10 represent the temperature and density at 10 m, respectively, in the Argo profile data. Since most Argo initial observation depths are less than 10 m and the ILD and MLD are generally deeper than 10 m in the NWP, the ILD is defined as the depth where the temperature is 0.5°C less than T 10 . The MLD is considered the depth with a density difference of Δ σ from σ 10 . The difference between the ILD and MLD is the BLT. The equation for Δσ is:

where S 10 represents the salinity at 10 m.

The equations for OHC and TCHP are:

where C p = 3.85 k J ( k g °C ) − 1 is the specific heat of water at a constant pressure and T i and ρ i are the temperature and density at the ith level, respectively. Δ z i is the thickness of the ith level.

To reduce the data complexity, principal component analysis (PCA) is used here to reduce the dimensionality of abundant data, as shown in the Figure 3 . In this research, a sample matrix including the MLD, ILD, BLT, SST, z 26 , and OHC data is created, and the Z score (Zero-mean normalization) is chosen to standardize these data so that each variable has a mean of 0 and a variance of 1 ( X S S T i , X M L D i , X I L D i , X B L D i , X z 26 i , X O H C i ). The covariance matrix is to measure the relationships of these ocean matrices, since there is a certain correlation between each ocean metric. Moreover, calculating the covariance matrix is a key procedure in PCA. Because the dataset involves six variables ( X S S T , X M L D , X I L D , X B L D i , X z 26 , X O H C ), the shape of this covariance matrix is 6 × 6 . The eigenvectors of the covariance matrix form the principal component matrix and are also the new, uncorrelated modes that are linear combinations of the standardized variables. Only the first k ( k < 6 ) modes with large variance ( λ 1 ,…, λ k ) values are retained, while the other modes with nearly 0 variance are ignored. The first k modes ( A 1 ,…, A k ) can provide a reasonable explanation for the patterns and physical background. Since the cumulative variance contribution of the first k modes reaches a high level, it is possible to use fewer than six new modes to illustrate significant patterns.

The one-dimensional Price-Well-Pinkel (1DPWP; Price et al., 1986) model is adopted here to investigate the thermodynamical response of the upper ocean, focusing on the different structures. The sea surface heat flux and precipitation are set as 0 in this model, and only the wind stress is considered in this model. The gradient Richardson numbers and critical bulk values are set as 0.25 and 0.65, respectively.

The wind stress is calculated by the following equation:

where ρ = 1.29 k g   m − 3 is the air density, C D is the drag coefficient, and U is the maximum wind speed obtained from the JTWC dataset. According to Powell et al. (2003) and Shay et al. (2015), the value of C D is calculated through the following equation:

The eigenvalues for the six modes are 118725, 100390, 24566, 6423, 8181, and 0. Since the eigenvalues of the first three modes contribute to more than 94% of the sum of all the eigenvalues in the data (mode 1: 46%, mode 2: 39%, mode 3: 10%), only the first three modes are considered in this paper ( Figure 4 ). Linear combinations of the standardized variables for the first three modes are shown in the following equations:

Applying these standardized variable data to the equations, mode scores for each profile are calculated and the loading for each variable determines its contribution. Moreover, a significance test is performed on the modes. The results show that modes 1, 2 and 3 significantly differ (p ≪ 0.05) every month.

Since the loadings of SST, z 26 and OHC are all negative, mode 1 OHC has a reciprocal relationship with SST and z 26 , and all of them decrease the mode 1 score. In contrast, the ILD and BLT both markedly increase the mode 1 score. Therefore, a low mode 1 score represents a structure with high SST, high OHC and deep z 26 ( Figure 5A ). The high mode 1 score shows a structure with low OHC, deep ILD and thick BLT ( Figure 5B ). Unlike mode 1, mode 2 dominantly describes that OHC has a strong relationship with other variables except SST which plays little or no role in increasing the mode 2 score. Profiles with a high mode 2 score have a relatively warm SST, large OHC, and thick MLD, ILD and BLT values ( Figure 5C ) due to their positive loadings. For mode 3, an MLD with a large positive coefficient greatly decreases the mode score. Therefore, an observation with a high mode 3 score generally has an extremely shallow MLD, high OHC, warm SST, and relatively thick ILD, BLT and z 26 values ( Figure 5D ).

Profiles with low mode 1, high mode 1, high mode 2, and high mode 3 scores are hereafter called M1-L, M1-H, M2-H, M3-H water, respectively. There are obvious difference among the averaged metrics of these water ( Table 1 ). Notably, the MLD calculation of the M1-H water is not sensitive to the density threshold value. Hence, the ILD of the M1-H water is regarded as the MLD. M1-L water has the highest SST and the shallowest BLT. M1-H water has the lowest SST. M2-H water has the thickest ILD, BLT and z 26 . M3-H water has the shallowest ILD and MLD.

Waters with a TCHP greater than 60 kJ cm -2 in the TC season are further analyzed (Mainelli et al., 2008), as shown in Table 2 . Because the TCHP of the M1-H water is 0, Table 2 does not show the results for M1-H water. The number of M1-L water bodies with TCHP ≥ 60 kJ cm -2 reaches a minimum in January and a maximum in September. The average TCHP of the M1-L water reaches a maximum in May ( Table 2 ). The M2-H water has the highest average TCHP during the whole year ( Table 2 ). The number of M2-H events with TCHP ≥ 60 kJ cm -2 reaches a maximum in winter and decreases in spring and summer. M3-H water with TCHP ≥ 60 kJ cm -2 accounts for a small proportion of the waters with TCHP ≥ 60 kJ cm -2 . The number of M3-H waters reaches at a maximum in September. Generally, the number of M3-H waters with TCHP ≥ 60 kJ cm -2 is much smaller than that of M1-L and M2-H waters.

The temporospatial distribution of these waters is shown in Figure 6 . The data are categorized into north and south categories. The north and south are separated at 20°N. M1-H water mainly occurs in the northern NWP, and the number reaches a maximum in winter and a minimum in the TC season. Compared to the north of the NWP, the south of the NWP is dominated by M2-H water in winter. M1-L water generally appears in the south of the NWP, and M3-H water occurs in the north, both of which prevail in the TC season. Notably, a considerable amount of M2-H water exists in the TC season.

To further analyze the upper ocean structure, which is deeply influenced by upper ocean temperature and salinity, monthly mean temperature and salinity are averaged between 134.5°E and 135.5°E. In winter, the ILD is generally thick in the whole NWP. However, because of the latitude difference, the upper ocean at low latitudes is warmer than that at high latitudes ( Figures 7A, K–L ). Hence, the thermal conditions of the southern NWP are beneficial for TC development ( Figure 1B ). Moreover, due to the difference in local rainfall or meridional advection, sea surface salinity at high latitudes has a higher SSS than that at low latitudes. Convective precipitation is mainly distributed over the lower latitudes (Liu et al., 2012). Subsurface salinity at low latitudes is far higher than surface salinity ( Figures 8A, K–L ), and the difference between surface and subsurface salinity generates a thick barrier layer since the upper ocean at low latitudes is largely influenced by the North Equatorial Current (NEC), which usually carries high-salinity water (Schönau and Rudnick, 2015). The high solar radiation, plentiful precipitation, and NEC give the upper ocean at low latitudes characteristics of M2-H water. The small amount of precipitation and low solar radiation give the upper ocean at high latitudes characteristics of M1-H water. From spring to autumn, with increasing solar radiation, the SST in the whole NWP is maintained at a high level ( Figures 7B–K ), and a seasonal thermocline is generated in the whole NWP, greatly decreasing the mode 1 score and increasing the mode 3 score. Moreover, because the region with high precipitation moves to high latitudes (Liu et al., 2012), sea surface salinity decreases at high latitudes and thus generates a thick barrier layer ( Figures 8B–K ), which greatly increases the mode 3 score. Due to changes in solar radiation and precipitation, the upper ocean is characterized by M1-L water at low latitudes and M3-H water at high latitudes.

In addition to solar radiation, precipitation and NEC, mesoscale eddies also play a key role in the upper ocean structure (Pegliasco et al., 2022; Shay et al., 2000; Lin et al., 2008; Schiller and Ridgway, 2013). The isopycnal surfaces of anticyclonic eddies (AEs) with positive sea level anomalies (SLAs), known as warm-core eddies, exhibit a concave shape (Zhang et al., 2014). The upper isopycnal surface envelops a generous amount of warm water and outcrops at the sea surface. In contrast, cyclonic eddies (CEs) with negative SLAs, known as cold-core eddies, exhibit opposite structures compared with AEs. Profiles within AEs often also have the characteristics of M1-L water and M2-H water, while profiles within the CEs show characteristics of M3-H water. Since the number of eddies with a lifespan of 30 days (5451) is far more than that of eddies with lifespans of 90 (175) and 180 days (7), only eddies with a lifespan of 30 days are investigated here. Figure 9 depicts the locations of eddies encountered by TCs. The number of AEs (280) is smaller than that of CEs (321) within 300 kilometers of TCs, but the number of AEs (12) within 100 kilometers of TCs is visibly larger than that of CEs (4).

TCs are mainly generated east of the Philippines and move toward the west or north because of the Pacific subtropical anticyclone. The difference in the proportion between the two directions is not obvious. However, the possibility of TC intensification is widely divergent. Figure 10A depicts the location when TCs begin to intensify during the TC season. Most TC intensification occurs in the southern NWP, and the proportion that occur in the north is quite small.

Since this phenomenon is closely related to the upper ocean structure, the amount of each category of water that a TC encounters during the TC season is counted here and the temperature change during TC passage is further analyzed. To accurately calculate the temperature change during TC passage, two paired Argo profiles are identified through the following criteria: (1) the distance between the Argo profile and the TC center must be less than 300 km; (2) the distance between two paired profiles must be within 50 km, which can greatly reduce the differences in spatial variability in the ocean environment; (3) Argo profiles with a minimum observation depth shallower than 10 m and a maximum observation depth deeper than 300 m are extracted; and (4) the pre-TC profile observation time must be fewer than 5 days before TC passage, and the post-TC profile observation time must be within 3 days after TC passage. Following these conditions, 864 pairs of profiles are identified. With a one-dimensional mixed layer model, the temperature response of different waters to TCs is also investigated here. For each water type, all profiles with a maximum observation depth greater than 1000 m are identified. The averaged thermohaline profiles of different water types are created to initiate the 1DPWP model, as shown in Figure 11 . For each simulated experiment, the 1DPWP model is run with 30-min time steps, and the averaged wind speed in TC categories between typhoons and super typhoons is applied.

The numbers of TCs that encounter M1-L water, M2-H water and M3-H water are 542, 306, and 216, respectively. M1-L and M2-H waters are mainly located in the south of the NWP. Notably, there are also considerable numbers of M1-L and M2-H water profiles in the northern Pacific, which is influenced by the Kuroshio. Moreover, the possibility of TC intensification is also large here. M3-H water mainly appears in the northern Pacific ( Figures 6G–J ). These results show that when a TC encounters M1-L water or M2-H water, it tends to intensify. The TCHPs of the M1-L, M2-H and M3-H water encountered by TCs are 68.9, 69.6 and 29.1 kJ cm -2 , respectively. The high TCHP of M1-L and M2-H causes TCs to rapidly intensify ( Figure 10B ), while M3-H water with low TCHP cannot provide sufficient heat energy for rapid intensification.

The mean and maximum SSTC induced by TCs is much lower for M2-H water relative to M1-L and M3-H water ( Table 3 ). Moreover, although the SSTC of M1-L water is obvious, the SST of M1-L water is rather high and the modeled results show the same patterns ( Figure 12 ). This reveals that the upper ocean structure obviously has a significant impact on the temperature response to strong TCs, which is important for TC intensification. Although the SST of M2-H water is the lowest, the BLD of M2-H water is the thickest, which reduces the SSTC by decreasing the entrainment rate caused by strong TCs. Thereby a TC intends to intensify rapidly, when it passes through upper ocean with M2-H characteristic. This result is also supported by Rudzin et al. (2017); Potter et al. (2019) and Rudzin et al. (2020). Moreover, the warm T I L D + 30 of M2-H water ( Table 3 ) is also beneficial for low SSTC. Unlike that of M2-H water, the high SSTC of M1-L water is offset by its highest SST and warmest T I L D + 30 ( Table 3 ). Therefore, the considerably high SST of M1-L water is still maintained, which is beneficial for TCs intensification. As the T I L D + 30 of M3-H water is cool ( Table 3 ), and strong TCs can easily bring cold water to the surface, thus, high SST and thick BLD conditions play a small part in the sea surface cooling induced by strong TCs. Hence, it is hard for a TC to intensify when it meets M3-H water.