We present δ18O_c records from two Korean stalagmites obtained from Eden Cave (ED; Jo et al., 2014) and Gwaneum Cave (GE; Jo et al., 2010) (see locations in Figs. 1 and 2a), which grew intermittently. The δ18O_c records for ED1 and GE1 stalagmites were previously measured along with stable carbon isotope (δ13C_c) within the scope of Jo et al. (2010, 2014). However, the ED1 δ18O_c record has remained unpublished to date. Further details regarding the GE1 and ED1records and their covered time spans are provided in Table S1 in the Supplement.

Previous studies of the ED1 and GE1 stalagmites documented their growth fabrics, depositional histories, stable cave environment, and stable-isotope characteristics, thereby constraining the conditions under which their isotopic signals were preserved. The principal growth phase of GE1 was interpreted to have formed under conditions approaching isotopic equilibrium, based on its calcite textures and stable-isotope characteristics, whereas intervals associated with coralloid growth and limited dripwater supply were identified as distinct depositional phases (Jo et al., 2010). ED1, similarly, contains well-characterized growth intervals separated by hiatuses and changes in depositional fabric, supported by an extensive U-series chronology (Jo et al., 2006, 2014). Although certain intervals exhibit high δ18O_c variability, partially due to the low resolution of the records, the absence of a pronounced precessional cycle is notably observed in the δ18O_c profiles (Fig. S1 in the Supplement). From another Korean cave, Jo et al. (2017) identified clear millennial-scale resemblances among late Holocene δ18O_c records from both Korean and Chinese stalagmites, providing an in-depth discussion on the issue in question. These observations suggest that, despite potential site-specific modifications of the calcite isotopic signal, the well-characterized portions of these records can provide observational constraints for evaluating the modeled δ18O_p response over the KP.

For this study, we performed a series of fully coupled paleoclimate simulations with the iCESM1.2 (Brady et al., 2019), using a model resolution of approximately 1.9° × 2.5° in the atmosphere and land and 1° × 1° in the ocean and sea ice, respectively, providing simulated precipitation oxygen and hydrogen isotopes (δ18O and δ2H) (Brady et al., 2019; Hu et al., 2019; Tabor et al., 2018). We also combined simulated precipitation stable-isotope data to examine precessional variations in precipitation d-excess (δ2H - 8 × δ18O) (Dansgaard, 1964). To validate our iCESM1.2 present-day (iCESM-PD) simulations, we have used the available IAEA Global Network of Isotopes in Precipitation (GNIP) datasets for the Asian domain (accessible at https://nucleus.iaea.org/wiser, last access: 25 April 2023), published precipitation isotope records (Sinha et al., 2019), and data from the isotopes-incorporated Global Spectral Model – isoGSM2 (Yoshimura et al., 2008).

First, we conducted an idealized transient simulation covering the orbital forcing history of the past 130 000 years (iCESM-130ka), using fixed pre-industrial conditions of greenhouse gas (GHG) concentrations, ice sheets, and bathymetry. The transient simulation employs an orbital temporal acceleration (Lorenz and Lohmann, 2004; Timmermann et al., 2007) factor of 100, consistent with previous studies, which compresses the 130 000-year forcing history into 1300 model years. The acceleration choice is justified by the fact that we focus here on atmospheric processes primarily controlled by changes in land and ocean surface temperatures, with adjustment timescales that are orders of magnitude faster than the precessional cycle forcing. In our 130 ka-orbitally forced iCESM1.2 simulation, we analyzed the spatiotemporal patterns of annual-mean anomalies of precipitation amount and δ18O_p. We further compared our simulated precipitation patterns with the transient CESM1.2 simulation over the past three million years (CESM-3Ma) (Timmermann et al., 2022; Yun et al., 2023), which was forced by time-varying orbital, GHG, and ice-sheet forcings. CESM-3Ma simulation uses a lower horizontal resolution (3.75° in the atmosphere and 3° in the ocean), and an orbital acceleration of only 5, but does not include isotopes.

Next, we performed model sensitivity experiments with iCESM1.2 to evaluate the equilibrium climate responses to maximum (Pmax⁡) and minimum (Pmin⁡) precession conditions (iCESM-Pmax⁡ and iCESM-Pmin⁡). These 100-year-long fully coupled time-slice sensitivity experiments branched off from year 900 of a 1000-year-long iCESM-PD control experiment. The two idealized simulations use a high eccentricity (Berger and Loutre, 1991) of 0.0493 to amplify the precessional effects. The simulations reach a quasi-equilibrium state for important atmospheric variables within 100 years. Much of our analysis from these two simulations is based on the June–July–August (JJA) seasonal averages of the last 40 model years. Furthermore, we compared iCESM Pmin⁡ and Pmax⁡ precipitation and circulation patterns with ultra-high-resolution simulations (1/4° in the atmosphere and 1/10° in the ocean) conducted for Marine Isotope Stages (MIS)-5d (115 ka) and MIS-5e (125 ka) using the CESM1.2 (Chu et al., 2020; Huang et al., 2021; Raavi et al., 2023; Wengel et al., 2021) (hiresCESM-5d and hiresCESM-5e). This analysis provides further insights into the robustness of our results, enabling a more nuanced characterization of regional aspects.

To quantify the KP precipitation and δ18O_p responses to the Pmin⁡ and Pmax⁡ conditions, we performed water-tagging experiments using the iCESM to track changes in source regions and their relative contributions to the KP. The experiments were conducted using orbital parameters corresponding to 124 ka (Pmin⁡) and 119 ka (Pmax⁡), along with current atmospheric and oceanic conditions (including the initial isotopic composition) derived from the iCESM-PD simulation. We tagged a total of 14 boxes (7 each over land and ocean), based on the major contributors to the Asian region (Hu et al., 2019). The box latitude and longitude were chosen following Wen et al. (2024), although we have excluded regions from the calculations with minimal contribution (< 10 %) to the total rainfall and no effect on the anomalous δ18O_p. The regional pattern of the simulated δ18O_p is determined by the δ18O of the vapor from the tagged regions, followed by hydrological processes and rainout history (Tabor et al., 2018). Each tagging experiment was conducted over 40 model years, with the mean values from the last 20 years (both summer and annual) used for analysis. Further information on the tagging experiment is provided in the Supplement (Sect. S1).

To extract the dominant modes of variability in Eastern Asia, we used precipitation amount and δ18O_p annual anomalies from the transient iCESM-130ka simulation. We calculated the first Empirical Orthogonal Function (EOF1) and the corresponding principal component (PC1) time series for the selected Eastern Asian region (22–45° N, 100–150° E), including eastern China, the KP, and Japan (Fig. 1). The leading EOF of annual mean precipitation variability exhibits a north-south dipole in the selected EASM domain (Fig. 1a), characterized by strong precessional variability and correlating with the NHSI at 30° N (Fig. 1b). The precipitation anomalies associated with the change in boreal summer insolation exhibit the strongest signal in an east-west elongated structure between 25 and 35° N. For Northern Hemisphere summer perihelion conditions (low precessional index), this area is characterized by anomalously dry conditions, whereas for northern-hemisphere summer aphelion conditions (high precessional index), it is characterized by anomalously wet conditions (Fig. S2). Most notably, Korea lies on the nodal line of the precipitation dipole pattern, suggesting a weak annual-mean precession signal over the KP (Fig. 1a). A similar precipitation pattern also emerges for the CESM-3Ma transient simulation (Yun et al., 2023) over the study area (Fig. S3), indicating that the effects of GHGs and ice sheet forcing are relatively minor in driving the EASM spatial precipitation patterns on precession scales. However, in this case, the nodal line lies slightly south of the KP. The ultra-high-resolution simulations confirm the overall large-scale response pattern in summer precipitation to changes in the precessional index, albeit with somewhat drier conditions over the KP, relative to the coarser-resolution CESM-3Ma and iCESM-130ka simulations (see comparisons in Fig. S2).

Contrasting the meridional dipole in precipitation, the leading mode of δ18O_p variability exhibits an east-west (zonal) dipole pattern over our selected study region (Fig. 1c). The δ18O_p dipole, which is characterized by relative ¹⁸O-depletion (enrichment) for western (eastern) Asia during Northern Hemisphere summer perihelion, has previously been presented as a notable feature over the broader Asian summer monsoon region on precession scales (Wen et al., 2024). In the western part of this dipole (China), we observe δ18O_p signals at pronounced precessional timescales; in the eastern part (Japan, Korea, and nearby seas), we generally find weaker variability that is antiphase with the west (Fig. 1c and d). For the KP region, the simulated precessional signal in δ18O_p is strongly suppressed (similar to the rainfall pattern), and no distinct precession cyclicity is evident in the simulated time series (Fig. S4).

The near absence of precessional variability in hydroclimate predicted by the models can now be tested using our speleothem isotope data from the KP (Fig. 2a). The Korean speleothem stable carbon isotope (δ13C_c) records aligns well with Chinese δ18O_c during overlapping periods (Fig. 2b), and the growth frequency of Korean speleothems exhibits clear signals that correspond to orbital-timescale hydrologic changes (Jo et al., 2014). However, Korean stalagmitesrecords show no apparent precessional signal (Fig. 2b). The weak covariation between δ13C_c and δ18O_c (ED1, R2 = 0.16, p < 0.01; GE1, R2 = 0.04, p = 0.02) suggests that kinetic effects are unlikely to be the primary control on their variability. Moreover, the precessional-scale variability in the Korean δ18O_c records is strongly suppressed relative to the relatively pronounced (∼ 4 ‰) variability observed in the Chinese composite δ18O_c record (Fig. 2b). This observational evidence is consistent with the muted response of δ18O_p over the KP to precessional forcing in the simulations (Fig. 1c). This raises the question of which physical processes govern the Korean δ13C_c on precessional timescales. Previous studies from the KP (Jo et al., 2010, 2011, 2014) suggested that glacial-interglacial changes in δ13C_c could be related to summer-temperature and rainfall-driven anomalies in terrestrial productivity. However, a more detailed investigation of the underlying processes driving precessional-scale variations in δ13C_c is beyond the scope of our comprehensive water-isotope analysis of the KP.

According to the transient iCESM-130ka and CESM-3Ma simulations, precipitation over Eastern Asia (Fig. 3b) exhibits precessional signals, which are out of phase with insolation. It is important to note that simulated precessional scale precipitation is suppressed during glacial maxima in eastern Asia, documenting the influence of the 80–120 kyr eccentricity forcing on the amplitude modulation of precession (Fig. 3a and b). There are several East Asian hydroclimate proxies, such as ¹⁰Be in loess from central China (Beck et al., 2018), pollen records from Japan (Hayashi et al., 2017, 2021; Tarasov et al., 2011) (Fig. 3c), and total nitrogen content in Lake Biwa (Iwamoto and Inouchi, 2007), which support and demonstrate the above-mentioned features in regional-simulated precipitation. Speleothem records from Japan (Kato et al., 2021; Mori et al., 2018) lack sufficient data points within the 0–130 ka simulated period, and the intermittent time span does not encompass high-insolation periods, which would be necessary to validate the simulated isotopic composition. Some mismatches between the pollen records and the model occur during MIS11 and low-CO₂ conditions, suggesting that vegetation dynamics in the region may, in part, be driven directly by CO₂, either through temperature or the CO₂ fertilization effect. However, uncertainties in the chronology of the palynological reconstruction preclude a definitive conclusion at this time. Moreover, the simulations indicate a stronger 41 kyr obliquity signal, particularly during periods of low eccentricity, which seems absent in the paleo proxy records (Fig. 3b).

As discussed in the previous section, under low-precession conditions, a region of low precipitation in the eastern EASM domain zonally extends from KP across southern Japan into the central North Pacific. Our simulation shows that the region of low precipitation is co-located with an anomalously high sea level pressure (SLP; contour lines, Fig. 4a) and suppression of atmospheric convection (Fig. S6c, contour), in concordance with the intensification of the anomalous WNPSH (geopotential height, Fig. S6c). Supporting evidence comes from an integrated moist static energy budget analysis by Wen et al. (2024), which showed that the reduction in rainfall is linked to descending air motion. Our simulated precipitation pattern, meridional and zonal wind components, and strength of anomalous WNPSH are well supported by the difference between the hiresCESM-5e and hiresCESM-5d experiments (Figs. 4b and S6), even though the anomalies over the KP are slightly stronger compared to the ∼ 2° resolution iCESM.

In addition, the simulated western branch of the anomalously strong WNPSH (Fig. S6c, f) might help transport more moisture from the Pacific Ocean toward northern China and the northern part of KP, resulting in regional positive precipitation anomalies in the model. The rainfall pattern corresponds to an overall weakening and a slight northward shift of the Korean and Japanese summer monsoon systems, specifically the Changma and Meiyu-Baiu fronts, respectively. On the other hand, the KP hydroclimate reconstructed from speleothem proxies (viz., growth frequency) suggests signals of a wetter climate during interglacial periods (Jo et al., 2014), which are characterized by Northern Hemisphere summer perihelion. These observations highlight the importance of considering speleothem-specific characteristics, including site location, local orography, cave processes, and chronological precision, when comparing individual speleothem records with spatially resolved simulations, particularly for distinguishing hydroclimatic changes between closely spaced intervals such as MIS 5e and MIS 5d. It is worth noting that the Korean caves are primarily located in the mountainous Gangwon-do region of South Korea and range in altitude (Jo et al., 2014). It is possible that orbital forcing may also influence the precipitation intensity and the extent of the nodal line, relative to local topographic features (Fig. 4b). Thus, the apparent contrast between the speleothem growth-frequency records and the simulated muted response over the broader KP region may partly reflect spatial heterogeneity in the regional hydroclimate response.

We further explored the mechanisms underlying the precessional-scale muting of precipitation and δ18O for the KP and other parts along the nodal line, both for summer (JJA) and annual-mean conditions. Using the water tagging method, we estimated the contributions from different moisture sources to precipitation and δ18O_p at the KP in JJA (Fig. 5). The Pacific Ocean transports more moisture to the KP region during the Northern Hemisphere summer perihelion (Fig. 5a). This anomalous moisture predominantly originates from the subtropical North Pacific (SNP, 0.40 mm d⁻¹) due to the strengthening and northwest shift of the NWPSH. Land moisture (Fig. 5b) from the East Asia region also contributes to the anomalous Pmin⁡ precipitation over the KP in JJA (EAS, 0.11 mm d⁻¹). However, during Pmin⁡ less moisture from the northern North Pacific (NNP, -0.15 mm d⁻¹) and the equatorial Indian Ocean (EQI, -0.09 mm d⁻¹) arrives in KP, which compensates for the increase with a small cumulative change in summer precipitation of 0.28 mm d⁻¹ as a result. The annual-mean KP precipitation anomaly between Pmin⁡ and Pmax⁡, shows only a small change, ranging from 0.023 mm d⁻¹ (SNP) to -0.15 mm d⁻¹ (EAS) in terms of moisture contributions (Fig. S7, Table S2).

JJA δ18O_pover the KP, will be determined by both regional rainfall and water vapor transport contributions from the different source regions. Using the isotope tagging method, we calculated summer precipitation-weighted δ18O_p values (Sect. S1 and Table S2) at the KP (inset, Fig. 5). We find that the moisture transported from the SNP oceanic region to the KP is characterized by anomalously high δ18O values (i.e., enriched in ¹⁸O) during Pmin⁡ relative to Pmax⁡ by +0.22 ‰, with EAS land moisture also contributing higher δ18O values (Pmin⁡–Pmax⁡ = +0.31 ‰). However, the moisture transported from the distant Indian Ocean (EQI) and the land (SAS) regions shows a compensating effect, with lower δ18O values (EQI + SAS = -0.38 ‰). Combining the positive δ18O contributions from the SNP and EAS regions with the negative contributions from the EQI and SAS regions results in a small net increase in δ18O, corresponding to higher JJA δ18O_p values (+0.22 ‰) over the KP during Pmin⁡. On annual timescales (inset, Fig. S7), this compensation is even more pronounced, leading to a slight net depletion in ¹⁸O (lower in δ18O_p values) over the KP by -0.09 ‰ (Pmin⁡–Pmax⁡). Furthermore, the tagging experiment reveals that, for annual means, the Indian region plays a negligible role in precessional-scale changes in precipitation and δ18O_p over the KP (Fig. S7, Table S2).

Overall, there is a tug-of-war between contributions from different oceanic and continental regions to KP δ18O_p, leading to the muting of the precessional signal. For instance, the anomalous WNPSH leads to a westward (eastward) shift of the ISM branch towards northern China under low (high) precession conditions, as documented by an analysis of the vertically integrated moisture transport (VIMT, Fig. S8). The schematic in Fig. 6 summarizes these findings by highlighting changes in circulation patterns and simulated δ18O_p values for Pmin⁡ and Pmax⁡ conditions.

Given the oceanic-continental moisture contributions to the orbital-scale characteristics of precessional-scale changes in KP hydroclimate, we anticipate that the dynamics will also be well captured in d-excess, which reflects the source region (Araguás-Araguás et al., 2000; Jouzel et al., 2013; Pfahl and Sodemann, 2014) and evaporative conditions (Masson-Delmotte et al., 2005; Uemura et al., 2008) of the moisture source. Seasonal variations in d-excess, to some extent, can serve as an analog for precessional changes, as precessional forcing often modulates the amplitude of the seasonal signals in precipitation and isotopic composition, due to the strength and position of atmospheric circulation (Bosmans et al., 2018). As such, we examined d-excess variations on precessional timescales in our iCESM simulations.

d-excess, a secondary parameter derived from stable water isotopic ratios, provides additional insights into hydrological processes. Non-equilibrium conditions can lead to enrichment of ²H in the less stable phase (e.g., the gas phase) due to the higher diffusivity of hydrogen isotopes relative to oxygen. For instance, during evaporation from the ocean surface, the strong gradient in relative humidity accompanied by strong winds leads to higher d-excess in the vapor (Craig and Gordon, 1965; Uemura et al., 2008). It has been suggested that climatic influences on relative humidity, mediated by changes in atmospheric circulation, are captured in the d-excess proxy in precipitation-based paleoarchives (Pfahl and Sodemann, 2014) (e.g., ice core, speleothem). Studies have shown that the d-excess in paleowater can, in fact, be recovered from fluid-inclusion data in certain speleothems (Held et al., 2025; Matthews et al., 2021).

The precipitation d-excess is sensitive to various hydrological processes between the “source” and the “sink” (Xia et al., 2023), such as evaporation at the moisture source location (Craig and Gordon, 1965; Merlivat and Jouzel, 1979), moisture recycling from the land (Gat, 1996; Gat et al., 1994), raindrop evaporation (Risi et al., 2008; Stewart, 1975), and Rayleigh distillation (Xia et al., 2023). On precessional timescales, one crucial parameter is the near-surface relative humidity (Rh_s) at the source location. Over the tropical monsoon region, Rh_s changes may occur mainly in conjunction with wind anomalies, as the annual mean surface temperature changes are quite small on precession scales (Clement et al., 2004). We have analyzed the characteristics of simulated Rh_s at plausible source locations and d-excess variations at the sinks. The spatial pattern and gradient of surface relative humidity over the Asian region are reasonably well captured in the present-day simulation (Fig. S9). Brady et al. (2019) reported that the d-excess values simulated by iCESM1 exhibit a positive median bias of 3.3 ‰ and a poor correlation with GNIP observations. Our reinvestigation of simulated d-excess in the Asian region for present-day conditions shows that GNIP and model data align reasonably well (Fig. S10). A closer look at Fig. 4b in Brady et al. (2019) supports this conclusion. For the Asian region, we found good agreement (r = +0.52, p < 0.01) between the annual mean d-excess in GNIP observations and our iCESM-PD simulation (Fig. S11), with a smaller median bias of +1.75 ‰, compared to the 3.3 ‰ global mean bias. Moreover, the strong seasonality in the observed (GNIP) d-excess for East Asia, specifically for northern and eastern Asian sites, is well captured in iCESM-PD (Fig. S12), which is crucial in the current context because precessional shifts are likely to influence the amplitude of the seasonal cycle, rather than the annual mean.

To refine our interpretation of d-excess as a proxy for precessional signals over Asia, we further investigate Rh_s variability in the Asian monsoon oceanic source domains, including the Indian and Pacific Oceans. In the calculated EOFs for the Rh_s from the iCESM-130ka simulation, EOF2 represents precessional-scale changes (Fig. 7). Over the southeast Indian Ocean and northwest Pacific Ocean (rectangular boxes in Fig. 7), we observe strong precessional-scale variability in Rh_s (Fig. 7b), with an amplitude change of approximately 3 %–4 % (Fig. 7c, d). The chosen blue and red boxes, which exhibit strong out-of-phase variability (Fig. 7a), respectively, closely correspond with the dominant moisture source regions to the South Asian (Tabor et al., 2018) and Eastern Asian (Wen et al., 2024) monsoon regions under Pmin⁡ climate (black dashed boxes, Fig. 7a). Subsequently, our results indicate that the South Asian and East Asia (mainly Northeast China) continental domains also exhibit a substantial precessional-scale variability in Rh_s (Fig. S13a, b), with an amplitude of about 10 % change and in phase with the NHSI (Fig. S13c, d).

Moreover, our results show that the Pmin⁡ and Pmax⁡ conditions exhibit opposite signals in precipitation d-excess values between South Asia and East Asia (Fig. 8a). The d-excess is high over India and low over central China for Pmin⁡. Specifically, in the KP, d-excess is low (high) for high (low) NHSI. It should be noted that the KP d-excess differences for Pmin⁡–Pmax⁡ are predominantly from the changes during the summer season; however, the Indian region's d-excess values are higher throughout the year for Pmin⁡ relative to Pmax⁡ (Fig. S14). Our tagging results also show that transported moisture from the Pacific Ocean and the East Asia continental regions contributes to lower d-excess at KP under Pmin⁡ conditions (Fig. S15), likely due to relatively higher humidity over the source regions (Figs. 7d and S12d). To evaluate the Rh_s and d-excess relationship quantitatively, the translation of the source region's Rh_s change to the sink's precipitation d-excess has been estimated from the iCESM-130ka simulation. A regional Rh_s regression on d-excess shows about a 0.3 ‰–0.4 ‰ increase in summer precipitation d-excess values over South Asia with a 1 % decrease in source region (southeast Indian ocean) Rh_s (i.e., Δ(d-excess)/ΔRhs = ∼ -0.4) (Fig. S16a). And a 1 % increase in Rhs in the northwest Pacific Ocean can cause a 0.7 ‰–0.8 ‰ decrease in summer precipitation d-excess across Eastern Asia (i.e., Δ(d-excess)/ΔRhs = ∼ -0.8) (Fig. S16b).

In this section, we refine our understanding of speleothem records from the KP in the broader context of hydroclimate changes across South to Eastern Asia in our transient simulation (Fig. S4). The iCESM-130ka simulation shows dominant anti-correlated precessional variability in precipitation between India (representative of Southern Asia) and central China (Fig. 9a and e). Furthermore, the amplitude of this variability is stronger (weaker) for high (low) eccentricity values, such as during MIS 5a–e and MIS3. On the other hand, precessional-scale variability in simulated KP precipitation, as well as in parts of southern China, is strongly muted (Fig. 9c and g), as discussed above. Moreover, over the KP, we found that precipitation exhibits an 11 kyr higher-order precessional harmonic, for instance, during MIS 5d and MIS 4 (Fig. 9g). This signal likely arises from the nonlinearity of the moisture transport equation, which multiplies precessional-scale winds with precessional changes in moisture, leading to semi-precessional variability.

In the Indian region, apart from the strong precessional signals in precipitation, we see precessional variability in d-excess and δ18O_p, with high (low) d-excess and depleted (enriched) δ18O_p values for high (low) insolation (Fig. 9a and b). Here, the change in d-excess values is predominantly controlled by the surface conditions at the ISM moisture source region, that is, high and low Rh_s values for low and high precession conditions, respectively, as discussed above. In central China, δ18O_p varies in phase with the anomalies over India (Fig. 9b and f). However, their d-excess variations are essentially anti-correlated (Fig. 9a and e), in particular for MIS5e and 5c, reflecting the complex interplay between the Indian and Pacific Ocean moisture sources and continental moisture effects. Our results support the previously suggested role of the ISM branch in controlling precipitation isotopic ratios across China (see Fig. S4a and b). Furthermore, we observe that KP precipitation and δ18O_p exhibit little precessional variability due to the compensating effects described above. In contrast, d-excess over the KP shows a more pronounced precessional signal (Fig. 9g). This d-excess variability responds to the balance between oceanic and continental moisture transport and changes in the surface conditions of source regions (i.e., Rh_s) on precessional scales. Moreover, the KP d-excess precessional-scale variability is a unique feature as compared to other Eastern Asian regions, with high values occurring during MIS 5d, 5b, 4, and 2. The positive feedback (kinetic effect) to the vapor d-excess could be even stronger for Pmax⁡ relative to Pmin⁡ when dry/cold winds pass over the ocean (Yoshimura and Ichiyanagi, 2009) (e.g., the Yellow Sea).