The study area encompasses the coastal regions of the southern Asia-Pacific Ocean, including the South China Sea, Taiwan Strait, and Taiwan. Figure 1 illustrates two distinct seafloor terrains in the region. The eastern coast of Taiwan features steep terrain with water depths reaching 1000 m within 10 km of the shoreline, while the Taiwan Strait has a wide continental shelf with an average width of 180 km and length of 350 km, where water depths are generally less than 100 m. The water depth along the coast of South China is relatively shallow, at approximately 10-20 m. In the middle of the Taiwan Strait, water depths range from 40 to 70 m, with Proudman resonant long wave speeds estimated to be between 20 and 26 m/s.

Four types of dataset in the southern Asia-Pacific Ocean were used during 16 years (2005-2020). First, 6-min water level data were from the Central Weather Administration (CWA), Taiwan (https://ocean.cwa.gov.tw/V2/data_interface/datasets) and the Water Resources Agency (WRA), Taiwan (https://www.wra.gov.tw). Water level was measured by pressure and acoustic sensors in the harbors. Second, 1-min atmospheric pressure, wind speed, and wind direction from 23 weather stations (circles in Figure 1 ) were obtained from the CWA. Third, 30-min composite radar reflectivity images were from four weather stations (triangles in Figure 1 ), provided by the Data Bank for Atmospheric Research and Hydrologic Research, Taiwan (https://dbar.pccu.edu.tw/). Lastly, Hourly precipitation intensity data with 0.1 x 0.1 degree resolution were retrieved from the Global Satellite Mapping of Precipitation (GSMaP), provided by the Japan Aerospace Exploration Agency (JAXA) (https://sharaku.eorc.jaxa.jp/GSMaP, Kubota et al., 2007). The precipitation was estimated from passive microwave and infrared data (Ushio et al., 2009).

In the northern Asia-Pacific Ocean, meteotsunami occurrences were collected from South Korea and Japan. Wave heights of meteotsunamis were from the west coasts of South Korea during 2010-2019, as reported in Kim et al. (2021a). Wave heights obtained from 70 stations around the coast of Japan during 2009-2020 have been processed by the Japan Meteorological Agency (JMA) (https://www.data.jma.go.jp/gmd/kaiyou/db/tide/gaikyo/index.php). The Japan dataset with a 1-min interval, locally called secondary undulations (Kowalik and Murty, 1993; Rabinovich, 2009; Heidarzadeh and Rabinovich, 2021), has similar features to NSLOTTs. Pressure disturbances associated with the South Korea meteotsunami events were recorded by Kim et al. (2021). 10-min meteorological data around Japan were available from JMA (https://www.data.jma.go.jp/stats/etrn/index.php) since 2009. Radar reflectivity images from South Korea and Japan during the time span of maximum events were obtained from four articles (Kim et al., 2019; Kim et al., 2021a; Kim et al., 2021b; Kim et al., 2022) and the Research Institute for Sustainable Humanosphere of Kyoto University (https://database.rish.kyoto-u.ac.jp/arch/jmadata/synthetic-original.html), respectively.

Radar reflectivity imagery and satellite-derived precipitation data were used to classify weather storm systems into six categories: cluster, complex, bow, linear, frontal, and typhoon ( Figure 2 ). The first four storm structures followed the definition by of Workoff et al. (2012) and Bechle et al. (2015). Main features of frontal and typhoon structures were summarized from our literature review. All storm structures are described as follows.

Cluster: Unorganized convection with isolated peaks of >40-45 dBZ reflectivity in small areas (<40 km²) surrounded by larger areas of<35 dBZ reflectivity.

Complex: Organized convection and nonlinear structures with reflectivity of >45 dBZ in an area of larger than 500 km².

Linear: Organized linear-shaped convection with a line width less than 50 km and a length-width ratio of at least 3:1 (Fowle and Roebber, 2003). The line area has the reflectivity of >45 dBZ.

Bow: Organized convective bow-shaped line with a separated length of less than 100 km. The leading edge of bow has the reflectivity of >45 dBZ.

Frontal: Cold and Mei-Yu fronts with extensive rain belts and heavy rainfalls (Chien and Kuo, 2006). Rainband areas, with the horizontal scale greater than 100 km, have orientated convective lines for the reflectivity of >35 dBZ or the precipitation rate of >6 mm/hr (Meng et al., 2019).

Typhoon: A structure consisting of a primary eyewall and spiral-shaped rainbands in the inner and outer regions (Hence and Houze, 2012; Xiao et al., 2019). In the outer region, the rainband complex has a broken line of convective cells spiraling around the cyclone (Houze, 2010). Areas of convective cells range from 35 to 45 dBZ reflectivity or from 6 to 24 mm/hr covering a horizontal scale of hundreds of kilometers.

Meteotsunamis were identified using the following steps described in Bechle et al. (2015). First, a high-pass Butterworth filter (Williams et al., 2021) with a cut-off of 6 h was applied to remove tides. In the case of storm surges, we applied a 2 h high-passed filter to retain the desired signals. Second, the zero-crossing method was used to calculate maximum and mean heights and mean periods of waves. Third, georeferenced radar reflectivity imagery and satellite derived precipitation product were used to detect storm systems, as described in the previous section. At last, a meteotsunami event was identified if it met the following three criteria: (i) mean wave periods less than to 2 h (Monserrat et al., 2006), (ii) maximum wave heights exceeding 0.2 m, a site-specific threshold in this study, and (iii) associated with a detected storm structure within 3 h (Bechle et al., 2015). A wave height 0.2 m is among the thresholds used in other studies: 0.1 m in the Laurentian Great Lakes (Bechle et al., 2016), 0.2 m along the U.S. East Coast (Dusek et al., 2019), and 0.25 m across Northwest Europe (Williams et al., 2021).

To assess annual meteotsunami exceedance, we used cumulative frequency analysis with the Peaks-Over-Threshold method (Bechle et al., 2015) and then we fit the identified meteotsunami occurrences to two distributions. First, the Pareto Type I distribution is used in the Laurentian Great Lakes (Bechle et al., 2016) and the survival function is given by:

where x is a meteotsunami wave height, x_m is the threshold parameter, and β is the shape parameter. Second, the survivor function for the Generalized Pareto Distribution (GPD), successfully characterizing extreme significant wind wave height statistics (Anderson et al., 2015), and extreme meteotsunamis in Lake Michigan (Bechle et al., 2015) and U.S. Northeast Coast (Geist et al., 2014), is expressed as

where x_m is the threshold parameter, σ is the scale parameter, and ξ is the shape parameter. To establish x_m , a failure-to-reject method was employed by sorting the wave heights and deleting the lowest values until the distribution is no longer rejected by the one-sided Anderson-Darling goodness of fit test (Choulakian and Stephens, 2001; Bechle et al., 2015). To assess the fitness of the distributions, we used root-mean-square error (RMSE) and the probability plot correlation coefficient (PPCC) value calculated as the Pearson correlation coefficient of the quantile-quantile plots (Filliben, 1975). Given a specified wave height, x, the annual meteotsunami exceedance (AME) (Equation 3) is expressed as

where Φ(x) is the survival function which can be obtained from the cumulative distribution function, i.e., Equations 1, 2, and N_event and N_year are the number of identified events and years of the whole data, respectively.

Co-occurrence analysis is used to assess the spatial scale of meteotsunamis by estimating the number of storm-induced meteotsunami events that occur at affected stations within a 12-h time span (Bechle et al., 2015). For example, one co-occurrence indicates that a meteotsunami occurs at a single station, while co-occurrences greater than 2 indicates that multiple stations, at least two, are caused by the same storm-induced meteotsunami event. Previous studies of storm-induced meteotsunamis have found that the spatial scale of storm systems can reach hundreds of kilometers, allowing a single storm to generate meteotsunami waves that propagate to multiple stations (Šepić et al., 2009; Bechle et al., 2015; Bechle et al., 2016).

Figure 3 illustrates an extreme meteotsunami event on 2 April 2007. Filtered water level oscillations ( Figure 3A ), and nearby air pressure ( Figure 3B ) and wind speed observations ( Figure 3C ) were observed from four stations. The results show that water level oscillations closely follow air pressure and wind disturbances, consistent with the traveling bow storm. The bow storm originated off the coast of South China and rapidly propagated toward the west coast of Taiwan. The corresponding radar reflectivity imagery can be found in Wang et al. (2022), and is omitted here for brevity. During the storm, a 6 hPa air pressure drop was first observed at St. 2, followed by a pressure spike that occurred at St. 3 to St. 8 and 9 within 3 h. High winds also occurred in the time span of 18:00-21:00. At the Jiangjun Harbor (St. 8), the onset of meteotsunamis was at 18:06 and lasted for 1.5 h, with a maximum wave height of approximately 1.5 m. A sudden increase in air pressure of 3.2 hPa occurred between 19:00 and 19:28, accompanied by an average wind speed of 8.0 m/s at 19:25. Additionally, a gust of 20.3 m/s was observed at St. 10 as reported by Wang et al. (2022). These air pressure and wind disturbances suggest that they are likely a potential forcing for generating meteotsunamis. Therefore, such water level oscillations at four stations meet the identification criteria and are identified as a meteotsunami. This demonstration primarily describes the approach to meteotsunami identification and establishes the relationship between the identified storm and meteotsunmai event.

Meteotsunami events are identified at 17 tidal stations in the coastal areas of Southern Asia-Pacific Ocean over a 16-year period, with a total of 1994 occurrences. For this study, we divide the study area into four zones: Zone A (coast of South China, St. 1-2), Zone B (Taiwan Strait, St. 3), Zone C (western coast of Taiwan, St.4-11), and Zone D (eastern coast of Taiwan, St. 12-17), as shown in Figure 4A . In Figure 4A , the spatial distribution shows that three stations, St. 3, 4, and 8, account for approximately half of all meteotsunami occurrences. Zone A, along the coast of South China, has only 62 meteotsunami events, while St. 10 and 11 in Zone C, and St. 13, 14, and 16 in Zone D, have less than 70 events each. In Zone B and C, it is observed that meteotsunami occurrences may be rare, yet the impact can be catastrophic. Specifically, meteotsunamis with a maximum wave height exceeding 1.0 m are documented at St.7, 9, and 16. Figure 4B shows the 872 co-occurrences of meteotsunamis. Nearly 50% meteotsunami events occur at a single station, and the other half occur at 2 to 7 co-occurring stations. In addition, four events are associated with 10 and 11 co-occurring stations. The spatial distribution of co-occurring events at each station is shown in Figure 4C . Co-occurring stations are grouped into two regions: Zone B-C and Zone D (northeast coast of Taiwan). The first region has frequent events across 2 to 7 co-occurring stations, especially between St. 3 and St. 8. The second region, between St. 15 and 17, has fewer events. Six stations, St. 1, 2, 10, 11, 13, and 14, in Zones A, C, and D have the lowest meteotsunami co-occurrences. Over 50 single-station events are found at St. 3, 4, and 12 each. In other words, stations in the Taiwan Strait and on the west coast of Taiwan are frequently affected by meteotsunami co-occurrences than those on the east coast of Taiwan, while events in the South China area are weakly correlated. Findings reveal that the spatial scale of meteotsunamis can reach tens to hundreds of kilometers, along the west coast to the northeast coast of Taiwan.

Figure 5 shows the map of six storm types associated with meteotsunami occurrences. Three major findings are drawn upon here. First, meteotsunami occurrences are more frequent on the west coast than the east coast, with a ratio as high as 5.5. Cluster storms primarily dominate the generation of meteotsunamis on the west coast, specifically at St. 3 in Zone B and St.4, 5, 6 and 8 in Zone C. While complex, linear, bow, and frontal storms exhibit similar spatial distributions, their combined occurrences account for less than 50% of those induced by cluster storms. Second, frontal and complex storm-induced meteotsunamis are about four times more common than those caused by linear and bow storms. Third, typhoons can generate meteotsunamis in all four zones. Of notice is that St. 12 on the southeast coast has close to 80 meteotsunami occurrences, equivalent to an average of 5 events per year. Figure 6 displays the distribution of storm-induced meteotsunami events aggregated in the four zones. In Zones B and C, over 60% of meteostunamis are generated by cluster storms, with an additional 20% being induced by complex and frontal storms. In addition, approximately 50% of meteotsunamis are caused by typhoons in Zones A and D. Overall, cluster, typhoons, complex, and frontal storms account for 57.2%, 18.1%, 9.8%, and 10.2% of meteotsunami occurrences, respectively. Linear and bow storms are responsible for less than 5% of meteotsunamis.

Figure 7 illustrates three temporal distributions of meteotsunami occurrences varied with six storm types in the four zones. In Figure 7A , the diurnal distributions of meteotsunamis in Zones A, B, and D are varied randomly. Interestingly, Zone C shows a diurnal pattern with a peak at midnight (00:00) and a drop in the afternoon (16:00) by a factor of three. It may be likely linked to the diurnal conditions that give rise to weather storms, e.g., the local land–sea breeze and the large-scale diurnal circulation changes (Kerns et al., 2010; Huang and Chang, 2018). Figure 7B depicts the monthly distribution results. Across all four zones, the highest fractions of meteotsunamis occur from winter to early spring (December to April). Cluster storms account for more than 60% of identified meteotsunami occurrences during these months. In Zone D, the fraction of cluster storm-induced meteotsunamis reaches up to 50%. During late summer to autumn (July to October), the fractions are relatively low and typhoon-induced meteotsunamis are more prevalent in Zones A and D. Figure 7C shows that the annual meteotsunami occurrences display a slight declining trend from 2005 to 2011 in all four zones. In Zones A and C, there is an increasing trend from 2011 to 2019. Before 2011, meteotsunami occurrences are mainly associated with typhoon impacts in Zone A. In Zone B, there is a declining trend for cluster storm-induced meteostuanmis from 2016 to 2020. In Zone D, annual meteotsunami occurrences are strongly correlated with typhoon events throughout the study period, except for the year 2020. Notably, during 2020-2022, Taiwan has experienced a 56-year period without any typhoon making landfall for the first time (Narvaez et al., 2022).

Wave heights corresponding to annual meteotsunami exceedances in the four zones are examined here. Table 1 lists the fitted parameters with RMSE and PPCC for the Pareto Type I distribution and GPD. All PPCC values are close to 1, indicating good linear relationships between observations and two distributions. Based upon the smallest RMSE, we select the Pareto Type I distribution to fit meteotsunami wave heights at the study site. In comparison, the Pareto Type I distribution has been successfully applied in the Laurentian Great Lakes (Bechle et al., 2016), while the GPD has effectively represented meteotsunamis in the US East Coast (Dusek et al., 2019) and the eastern coast of Japan (Lin and Wu, 2021). Figure 8 shows the annual exceedances of meteotsunami wave heights in the four zones, each fitted with the Pareto Type I distribution (color lines). Three key points are highlighted here. First, in Zone A, the meteotsunami occurrences are not significant, as the 1-yr predicted wave height, equivalent to one event per year, is less than 0.3 m. Second, in the Zone B, C and D, 1-yr predicted wave height ranges from 0.44 to 0.72 m, approximately 1.5-2.6 times that in Zone A. Third, in the same three zones, the predicted wave height of > 1.0 m may be expected to occur more than once in a 3-yr interval. In summary, our findings indicate that the dangerous meteotsunami, wave height exceeding 0.3 m (Bechle et al., 2016), can occur up to 44 events per year. This highlights a potential risk in the Taiwan Strait and along the coasts of Taiwan. Therefore, heightened attention to meteotsunami risk in the coasts of southern Asia-Pacific Ocean is essential.

Figure 9 presents identical findings to those shown in Figure 8 , but with a focus on storm-induced occurrences. In all storm types, the maximum wave heights exceed 0.5 m and reach the dangerous magnitude. For cluster, complex, linear and frontal storm types, the predicted wave heights in a 100-yr interval remain below 1.2 m. However, for bows and typhoons, the wave heights could exceed 2.0 m. Based on these results, it’s expected that typhoons have the potential to trigger meteotsunami waves to the greatest extent, especially along the east coast of Taiwan, owing to their high intensity in terms of pressure and wind disturbances. Studies conducted by Lin and Wu (2021) and Shi et al. (2020) suggest that triggered meteotsunamis are primarily affected by the outer and inner spiral rainbands of cyclones. As reported by Houze (2004), the vertical structure of convective and stratiform features (Marks, 1985; Marks and Houze, 1987) within the rainbands plays a crucial role in the evolution of cyclone systems, indicating their potential to disturb the sea surface. Furthermore, typhoon-induced infra-gravity waves typically exhibit periods less than 300 s (e.g., Herbers et al., 1994; Herbers et al., 1995), which partly overlap with the oscillation periods of meteotsunamis. However, the meteotsunamis identified in this study may face limitations in separating two components due to aliasing effects caused by a 6-min sampling rate in water level. As a result, typhoon-induced sea level oscillations may arise from the combined effect of infra-gravity waves and meteotsunamis (Medvedev et al., 2022). As for the bow storms, although the number of meteotsunami events is the lowest, the extreme waves, such as the 2007 event, are likely to be triggered by a strong rear inflow jet (Wang et al., 2022). This phenomenon leads to damaging downflow wind on the sea surface, suggesting that Proudman resonance is likely to occur in the rear of the bow structure around the Taiwan Strait.

Figure 10 provides insights into the impact of meteotsunami on the coastal areas. Hazardous levels are determined by the annual average of 125 meteotsunami events occurred at the wave height >0.2 m, and the observed fractions of storm types at each station. The results prominently highlight that meteotsunamis can have a substantial impact with the highest occurrence of 29 events per year observed at St. 4 near Taipei City, the capital of Taiwan, which has approximately 6.8 million residents. In regions with lower population density but a high concentration of aquaculture industries, such as St. 3, 8, and 9 in the southwest regions, the risk level ranges from 13 to 16 events per year. In contrast, the east coastal areas of Taiwan, characterized by lower population density, exhibit a significantly lower potential risk. However, the sparsely populated northeast regions, with a low risk of approximately 5 events per year, may still be vulnerable to accidents, such as meteotsunami-induced rip currents (Linares et al., 2019; Liu and Wu, 2022). Typhoon-induced meteotsunamis show a similar risk level but are confined to St. 12.

Given these meteotsunami outcomes, we have identified an instance of boat damage in the northern Taiwan (Lin and Liang, 2017). Interestingly, even during the most significant meteotsunami occurrence in 2007, no damage linked to it was documented in newspaper articles or social media reports. This underscores the need for further investigations to enhance our understanding of the hazard risks posed by meteotsunamis.

In the context of the Asia-Pacific Ocean (APO), our study is divided into two main parts. First, we provide a summary of meteotsunami occurrences, associated hazards, and their connection to weather conditions and atmospheric factors. For this, we’ve collected data from published literature, including individual events reported by Rabinovich (2020), as well as papers sourced from the Web of Science database. We present these findings in Table 2 .

Over the past decades, there have been instances of strong meteotsunamis resulting in destructive damage and human casualties, notably in Longkou Harbor, China (Wang et al., 1987), and Nagasaki Bay, Japan (Hibiya and Kajiura, 1982). In recent years, our data includes 17 reports from South Korea, Japan, and China, with three reports collected from Taiwan, India, and Australia, respectively. We analyze the weather conditions and atmospheric forcings that contribute to meteotsunamis, and retrieve information regarding the damages and wave heights from event reports. In Table 2 , weather conditions are categorized as “good weather (GW)” and “bad weather (BW),” referring to calm ocean conditions and stormy weather as defined by Rabinovich (2020). The atmospheric forcings in Table 2 are examined through the analysis of synoptic-scale patterns observed from CWA weather charts, the extraction of rain rates from satellite data (GSMaP), and the determination of storm types based on reported radar reflectivity images.

The findings reveal that 8 GW meteotsunami events were related to specific weather systems and light-to-moderate rains, challenging the notion that GW conditions are always “calm” on the ground. Additionally, it’s observed that the atmospheric disturbances are likely to be associated with the resulting meteotsunamis. However, the classification of storm types based on radar reflectivity imagery remains limited in these reports, indicating that the relationship between meteotsunamis and specific storm types has not been firmly established.

Furthermore, two events with wave heights ranging from 1.4 to 1.7 m resulted in high human casualties, notably along the coast of Jiangsu Province, China, and the southwest coast of South Korea. This suggests that unexpected and overlooked meteotsunamis could be responsible for drowning accidents, even in GW conditions, underscoring the potential danger they pose. It’s important to note that wave heights exceeding 0.5 m can also lead to other forms of damage.

In the second part of our study, we illustrate the maximum meteotsunamis associated with different storm types in Figure 11 . We apply the proposed methodology to identify meteotsunamis and classify storm types in South Korea and Japan. Along the east coast of Taiwan, Japan, and certain Pacific islands, high waves induced by typhoons can reach heights of up to 2.1 m. On the west coast of South Korea, Kyushu Island in Japan, and Taiwan, meteotsunamis are associated with cluster, frontal, and bow types, respectively, with wave heights ranging from 1.0 to 1.6 m. Along the west coast of Japan, few meteotsunamis exceed a wave height of 1.0 m.

Based upon the spatial patterns, extreme meteotsunamis are likely to originate in the Taiwan Strait, East China Sea, and Yellow Sea, particularly in areas with water depths within 100 m. In these regions, the resonance of meteotsunamis may be attributed to the specific storm types, benefiting from favorable water-depth conditions (Hibiya and Kajiura, 1982; Bechle et al., 2015; Šepić et al., 2015a).

Overall, the extreme meteotsunamis in the APO are posed by different storms. In the Japan area, typhoons are significant drivers of meteotsunamis and infra-gravity waves, potentially causing destructive impacts on shorelines and structures (Heidarzadeh and Rabinovich, 2021). Along the west coast of APO, high waves triggered by various storm types are comparable to those generated by typhoons. However, studies focusing on the mechanisms behind meteotsunamis resulting from diverse atmospheric forcings have received limited attention (Tanaka, 2010; Vilibić et al., 2014; Bechle et al., 2015; Williams et al., 2021), especially around the APO. Given that these storm patterns have been linked to mesoscale convective systems (MCSs), the frequent and intense MCSs occurring between subtropical and mid-latitude regions (Schumacher and Rasmussen, 2020) may contribute to the atmospheric disturbances for triggering meteotsunamis. This is exemplified by the spatial correlation between meteotsunamis and winds at the 500-hPa level in the same regions (Vilibić and Šepić, 2017).