Impact of Tropical Cyclone Avoidance on Fishing Vessel Activity over Coastal China Based on Automatic Identification System Data during 2013–2018

Tropical cyclones (TCs) may cause severe impacts on the activities of coastal fishing vessels. Due to the unavailability or unacceptability of detailed Automatic Identification System (AIS) data that are capable of differentiating fishing activity from navigation, as well as the lack of detailed models and observation data of TC winds, few studies have provided quantitative and reliable assessment of the impacts of TCs on fishing activities. In this study, we modeled snapshots for the TC winds of 52 TCs over the Northwest Pacific (NWP) basin from 2013 to 2018, as well as daily fishing hours and daily hours of presence (hereafter “vessel hours”) of fishing vessels. Based on these data, the spatiotemporal pattern of fishing vessel activity over offshore China was first analyzed and mapped. Then, a TC wind hazard index and absolute and relative impact indices were proposed to assess the impact of the 52 TCs on fishing and vessel hours. Their relationship was then fitted with the cumulative distribution function (CDF) of the log-normal distribution. The results show that in the 2013−2018 period, the most active fishing areas were located in the South China Sea. In each instance, an increase was first observed in the initial several years; then a decrease followed in the yearly total fishing hours per vessel in the remaining years. The relative impact index was significantly correlated to the TC wind hazard index proposed in this study. Based on the quantitative relationship between the specified TC hazard index and the impact indices, it is possible to implement a pre-cyclone rapid loss assessment due to TC avoidance in the future.


Introduction
Tropical cyclone (TC) disasters may have severe impacts on the activities of coastal fishing vessels. China's total marine economic output value reached CNY 8.34 trillion (about USD 1.26 trillion) in 2018, and the contribution rate of fisheries exceeded 50% (MNR 2019). At the same time, marine fisheries in China face the threat of TC disasters in the Northwest Pacific (NWP) every year. In 2017, disasters in China caused direct economic losses to fisheries of approximately CNY 17.36 billion (about USD 2.57 billion), of which 64.3% were caused by TCs (MARA 2018a). Approximately 80−100 TCs are generated every year around the world, and approximately 1/3 are generated in the NWP (Fang and Shi 2012). About eight TCs make landfall on the China mainland every year in average. It is necessary to clarify the spatial and temporal patterns of China's marine fishing vessel activities and to quantify the impact of TCs on them.
Detailed Automatic Identification System (AIS) data, capable of distinguishing between fishing activity and sailing to and from fishing grounds, are difficult to access. Most existing studies of the spatiotemporal patterns of fishing Abstract Tropical cyclones (TCs) may cause severe impacts on the activities of coastal fishing vessels. Due to the unavailability or unacceptability of detailed Automatic Identification System (AIS) data that are capable of differentiating fishing activity from navigation, as well as the lack of detailed models and observation data of TC winds, few studies have provided quantitative and reliable assessment of the impacts of TCs on fishing activities. In this study, we modeled snapshots for the TC winds of 52 TCs over the Northwest Pacific (NWP) basin from 2013 to 2018, as well as daily fishing hours and daily hours of presence (hereafter "vessel hours") of fishing vessels. Based on these data, the spatiotemporal pattern of fishing vessel activity over offshore China was first analyzed and mapped. Then, a TC wind hazard index and absolute and relative impact indices were proposed to assess the impact of the 52 TCs on fishing and vessel hours. Their relationship was then fitted with the cumulative distribution function (CDF) of the log-normal distribution. The results show that in the 2013−2018 period, the most active fishing areas were located in the South China Sea. In each instance, an increase was first observed in the initial several years; then a decrease followed in the yearly 1 3 vessel activities have been based on data from public reports or datasets such as the yearbooks of the Food and Agricultural Organization of the United Nations (FAO). These data sources include global annual fishing vessel information on the country scale (Anticamara et al. 2011). However, spatially explicit maps of fishing efforts over different months or seasons remain unavailable (Guiet et al. 2019). Some researchers have tried to use the gravity model, which distributes total fishing effort to units, like grounds or cells, based on a weighted index of attractiveness or importance that varies among different units, to reconstruct the global spatial distribution of fishing activities (Gelchu and Pauly 2007;Watson et al. 2013). This method can only estimate theoretical fishing activities, based on the assumption that fishing activities are determined by the defined weighted index. Thus, it is still difficult to quantify a realistic, detailed picture of fishing activities .
Consequently, studies on losses to the fishing industry caused by TCs have mostly focused on direct losses, such as shipwrecks and damage to onshore processing and storage facilities, while little attention has been paid to the losses due to fishing suspension and production reduction. For example, some studies have evaluated direct losses based on field survey data, such as the number of shipwrecks in specific research areas, while there are few quantitative assessments of the losses caused by the suspension of fishing vessel activities (Xu et al. 2005;Monteclaro et al. 2018). Although some studies have taken into account indicators of fishermen's income loss due to the suspension of fishing vessel activities, most of these efforts have been based on statistical data available in provincial or national units (Ren 2009;Han et al. 2016). Few have used dynamic tracking data of fishing vessels to identify the patterns of fishing vessel activities during a specific TC, but when undertaken these efforts have lacked quantitative assessments (Zheng et al. 2016). The common problem of such studies is the lack of both quantitative and dynamic analyses on fishing activity reduction during TC events. It is difficult to meet the requirements of dynamic and quantitative indirect loss assessment.
In recent years, the application of vessel monitoring systems (VMSs) and AIS has made it possible to quantify real-time fishing vessel activities at high spatial resolutions. Researchers have integrated global AIS data, and have quantitatively estimated the distribution of global daily fishing vessel activity intensity. The spatial resolution of the results has reached approximately 1 km (Kroodsma et al. 2018). This dataset provides support for further clarification of the spatial and temporal characteristics of regional fishing vessel activities and enables a dynamic assessment of the impact of TCs on fishing vessel activities.
Based on this dataset, our study had two goals. The first was to quantitatively analyze the spatiotemporal characteristics of fishing vessel activities throughout China's coastal area. The other was to quantitatively analyze the impact of TCs on China's offshore fishing vessel activities. Specifically, we first obtained the global daily fishing vessel activity distribution product based on AIS data and analyzed the spatiotemporal characteristics of fishing vessel activities in the offshore areas of China from 2013 to 2018. Then, we defined two absolute impact indices and two relative impact indices of a single TC on fishing vessel activities. This enabled us to quantitatively analyze the impact of 52 TCs on fishing and vessel activities near China's offshore during the 6 years between 2013 and 2018. Our study provides support for understanding the large-scale tendencies of fishing vessel activities as well as a basis to dynamically and quantitatively assess the indirect losses to fisheries caused by TCs.

Materials
The study area covers the main fishing areas near the coastline of China. The data used in this study mainly include the historical TC track and intensity dataset and daily fishing hours (total hours that all vessels were fishing in the specific grid cell during a day) and vessel hours (total hours that all vessels were present in the specific grid cell during a day) dataset. Based on the two variables, fishing ratio, which is the ratio of fishing hours to vessel hours, has been calculated to reflect the overall utilization efficiency of fishing vessels.

Fishing Moratorium in the Study Area
The study area is defined as the sea area within the extent of 105°E-128°E and 17°N−42°N. Due to China's seasonal fishing moratorium policy, fishing activities in the study area are suspended from approximately May to September every year. According to the regulations of the Ministry of Agriculture of China, the starting and ending times of the fishing moratorium are shown in Table 1. The boundaries of the different fishing moratorium zones are shown with white dotted lines in Fig. 1, where Line #1 and Line #2 are 35°N and 26.5°N, respectively. Line #3 is the Fujian-Guangdong

Best Tracks in the Northwest Pacific (NWP)
Two data sources were used. One is the best track dataset of historical TCs in the NWP from the China Meteorological Administration (CMA), as shown in Fig. 1 (Ying et al. 2014). The other source is the International Best Track Archive for Climate Stewardship (IBTrACS) from the World Meteorological Organization (WMO) (Knapp et al. 2010). The information used in the CMA dataset includes the central positions of TCs every six hours (part of observation intervals are three hours after 2017) and the central minimum pressure. The information used in the WMO data includes the wind radii of the TC track points observed by the Joint Typhoon Warning Center (JTWC).

Fishing Vessel Activities in the Northwest Pacific (NWP)
The dataset of global gridded daily fishing hours and vessel hours by Kroodsma et al. was used in this study. This dataset is derived from tens of billions of AIS records around the world since 2012, using convolutional neural network (CNN) methods to distinguish between the fishing and non-fishing status of fishing vessels with an overall accuracy greater than 90% (Kroodsma et al. 2018). The main information of the dataset includes gear types, flags, daily fishing hours (t f ) (Fig. 2c, d), daily vessel hours (t v ) (Fig. 2a, b), and the Maritime Mobile Service Identity (MMSI) of fishing vessels present in every grid. The spatial resolution of the data is 0.01°, and the temporal resolution is one day.

Methods
In order to analyze the fishing activity patterns and quantify the relationship between TC hazard and its impact on fishing hours and vessel hours, this study defined the fishing activity intensity indicator, the TC impact index (including absolute and relative indices) on fishing activities, TC hazard index, and the fitting function.

Spatiotemporal Patterns of Fishing Vessel Activities
To analyze the interannual and intra-annual changes in fishing vessel activities in the study area, the average total fishing hours and vessel hours per vessel (mt f and mt v , respectively) were calculated by year and by month using Eq. 1.
where N is the total number of fishing vessels present in the study area during the calculation year or month; t f , t v are daily fishing hours and vessel hours in each 0.01° grid cell, respectively.

Impact of Tropical Cyclones (TCs) on Fishing Vessel Activities
Before defining the TC absolute/relative impact indices, the impact area was first defined. Then the baseline fishing hours and vessel hours were calculated to reflect the reference value under nondisaster conditions.

Boundary of the Impacted Area
According to the standard from CMA on issuing weather warning signals, the Blue warning for TCs will be issued if vessels might or have been affected by a TC within 24 hours and the maximum gust wind speed has been greater than the minimum wind speed of Beaufort scale eight (17.2 m/s). Under this circumstance, fishing vessels are required to take a detour or return to harbor (CMA 2007).
In actual situations, the impacted area of TCs on fishing vessel activities should be greater than the scale eight wind radius range for two main reasons: (1) errors might exist (1) mt f ∕v = Sum t f ∕v of all grid cells in the study area N , in the prediction of TC tracks; and (2) fishermen outside the forecasted scale eight wind radius will also take proactive measures for the purpose of risk prevention. In previous studies, an empirical method to delineate the boundary of the area impacted by TCs was to use the 500 km buffer zone (Elsberry 1987;Ren et al. 2007). Considering the two problems mentioned above, the radius of the TC-impacted area was determined to be 500 km.

Impact Indices of Tropical Cyclones (TCs) on Fishing Vessel Activities
The first step is to specify TC's impact start and end dates. The TC's impact start date is the first day that the TC center reaches the 48 hour forecast area regulated by CMA. The boundary of the 48 hour forecast area is the line linked by Weather 2010). The TC's impact end date is the day that a TC goes beyond 500 km from the China land boundary after its landfall. The second step is to define the baseline vessel hours b v and fishing hours b f during a TC. With reference to the definition used by the World Bank (Jovel and Mudahar 2010), b f and b v represent the assumed t f and t v under nondisaster conditions. In this study, the average t f and t v within 10 days before a TC's impact starts and 10 days after TC's impact ends are calculated as b f and b v , respectively. The formulas are as follows: ( where t f(d) (Fig. 3b) and t v(d) represent the daily fishing hours and daily vessel hours on the date d; d 0 represents the TC's impact start date; and d 1 represents the TC's impact end date.
The second step is to define the absolute impact indices Δt v and Δt f , which refer to the total reduction in vessel hours and fishing hours, respectively, during a TC in each grid of the impacted area (Fig. 3c), as compared to the baseline hours. For each TC, Δt v and Δt f are calculated as follows: where b v and b f (Fig. 3a) are the baseline vessel hours and baseline fishing hours during the TC, respectively; and D is the total number of TC impact days.
We then calculate the maximum reduction rate of the total fishing hours and total vessel hours, respectively, across the impacted area during the TC. For each TC, the variations in T v and T f with time ( Fig. 4) are first constructed, where T v and T f are the sum of t v and t f values of all grids in the impacted area for each day, respectively; then, the baseline values B f and B v (Fig. 4) are calculated using a similar method as mentioned before; next, the minimum values of T f and T v during the TC are identified, and their relative reduction rate to B f and B v is calculated. The results are relative impact index on fishing hours (IR f ) and relative impact index on vessel hours (IR v ), as shown in Eqs. 8 and 9, respectively.  where T f (d) and T v(d) are T f and T v on the date d, respectively; d 0 and d 1 are the dates of TC impact start and impact end, respectively (grey dashed lines in Fig. 4); D is the total number of TC impact days.
When calculating the baseline, if part of the subject days is within the fishing moratorium period, the B f and B v are separately calculated for the moratorium status and nonmoratorium status. For example, the B f for the moratorium status is the average T f in all moratorium days within 10 days before TC's impact start and 10 days after TC's impact end. Then the IR f and IR v should be separately calculated depending on whether the dates on which T v and T f reach the minimum are in moratorium status or not. That means if the minimum T f is in fishing moratorium status, the IR f should be the relative reduction rate to B f to account for the moratorium status.

Quantification of the Tropical Cyclone (TC) Wind Hazard Intensity
Existing indices for measuring the TC wind hazard usually include the Accumulated Cyclone Energy (ACE), the Power Dissipation Index (PDI) (Emanuel 2005), the Chicago Mercantile Exchange Hurricane Index (CHI) (CME Group 2009), and the Willis Hurricane Index (WHI) (Owens and Holland 2010). Among these indices, the ACE and PDI consider the maximum wind speed (MWS) and duration of TCs. The CHI adds the maximum wind speed radius based on the ACE and PDI. The WHI further considers the moving speed of TC centers. Based on these indices, Li and Fang (2012) developed the Rapid Loss Index (RLI) to rapidly assess losses. For all the indices mentioned above, the MWS is used to reflect the instantaneous maximum wind speed in a TC. However, the MWS is not able to reflect the spatial distribution of the instantaneous wind speed field, which should be considered to assess TC wind hazards more accurately. Parametric wind field models can improve the hazard indices by quickly simulating the instantaneous spatial distribution of the TC wind speed field in any specified resolution (Tan and Fang 2018).
Following the parametric wind field model combination for the NWP in Tan and Fang (2018), an advanced accumulated spatiotemporal cyclone wind hazard index (CWI) is proposed. First, a 1 km instantaneous wind field is simulated for each point of a TC track with a time interval of 6 hours. The Georgiou gradient model (Georgiou et al. 1983) was used to simulate the asymmetry of the TC gradient wind field and the Yan Meng boundary model (Meng et al. 1997) was used to convert gradient winds to surface winds at a height of 10 m by considering the effects of surface roughness. The input parameters include TC locations (longitudes and latitudes), the minimum central pressure (Pc), TC heading speed (C), TC heading direction (Vd), the maximum wind radius (RMW), the Holland shape parameter (B), and surface roughness (Z0), which assumes that the surface roughness is uniform and equal to a sea surface roughness of 0.0003 m). All the parameters can be found in the CMA best track dataset except for the RMW and B. We used the statistic model by Lin and Fang (2013) to calculate RMW and the Vickery model (Vickery and Wadhera 2008) to calculate B. Then, the simulated surface wind speed field for each point is spatially integrated to reflect the TC's instantaneous intensity. Finally, the CWI is obtained by integrating the instantaneous intensities of all 6 hour TC points. The CWI takes into account the duration, moving speed, spatial distribution of the wind speed, and area impacted by TCs. The formula is as follows: where T represents the duration from TC's impact start date to the impact end date (in 6 h intervals); A represents the distribution area of the instantaneous wind speed field for each TC point; V represents the simulated wind speed at each grid in the instantaneous wind field (wind speed greater than scale six or 10.8 m/s is selected); C represents the moving speed of TC centers; V 0 represents the reference value of the wind speed, which is set as 33.44 m/s (Owens and Holland 2010); and C 0 represents the reference value of the TC moving speed, which is set as 18 km/h (Lin and Fang 2013).

Relationship between Impact Indices and Hazard Index
First, the correlation coefficients between the TC relative impact indices (IR f and IR v ) and the TC hazard index CWI were analyzed. The Pearson correlation coefficient and Spearman rank correlation coefficient were used. Second, there are two main steps to quantify the relationship between the TC impact indices and the hazard index. One is to select a proper function for fitting curves. Another is to quantify the uncertainty by using the upper and lower bounds of the standard deviation intervals (Mo and Fang 2016).
There are two main principles for selecting the fitting function: one is that the value domain should be [0, 1] and the function should be monotonically increasing. Another is an acceptable goodness of fit. According to the distribution characteristics of the data of the research results, the cumulative distribution function (CDF) of the log-normal distribution is selected as the fitting function (Porter 2021). The formula is as follows: where ϕ is the CDF of the standard normal distribution, and μ and σ are the parameters.

Results
Based on the materials and methods, the temporal and spatial changes or patterns are first analyzed. Then the absolute and relative impact indices of 52 historical TCs are calculated and analyzed. Meanwhile, the TC hazard index for each TC is also modeled, after which the relationship between the TC impact indices and the TC hazard index is fitted using the log-normal CDF. Key details of all the results above are presented in this section.

Spatiotemporal Patterns of Fishing Vessel Activities
In this section, the yearly and monthly total fishing hours and vessel hours per vessel across the study area have been analyzed first. Then the gridded hours summed by year and by month are plotted.

Temporal Patterns
The yearly total fishing hours and vessel hours per vessel in the study area (that is, yearly mt f and mt v , respectively) during 2013−2018 are shown in Fig. 5a. The annual mt f values increased initially, and then decreased after 2015. One possible reason is that China has adjusted its policies on fuel subsidies for fisheries since 2015 for the sake of reducing marine fishing intensity and conserving fishery resources (MOF 2015). The fishing ratio (fishing hours versus total vessel use hours per vessel per day or month or season) fluctuates slightly around approximately 25%.
The monthly total fishing hours and vessel hours per vessel in the study area (that is, monthly mt f and mt v , respectively) during 2013−2018 are shown in Fig. 5b. Both indicators reach the maximum in September, and the minimum mt f is in June, while the minimum mt v is in February. The results suggest strong seasonality in both fishing hours and vessel hours. For fishing hours, the results follow the order of autumn > spring > winter > summer. For vessel hours, the results follow the order of autumn > spring > summer > winter. The main reason for the low fishing hours in summer is China's fishing moratorium policies (Table 1). The fishing ratio shows an upward trend from January to May and reaches a maximum of approximately 40% in May. Then, it drops from May to July, with a minimum of approximately 23%. From August to December, it recovers to a relatively stable level of approximately 34%. The yearly and monthly total fishing hours per vessel in the four seas of China are shown in Fig. 6. The highest yearly fishing hours per vessel are in the South China Sea. Compared to 2013, the total fishing hours per vessel in 2018 dropped by 19.4%, 4.0%, 7.5%, and 18.6% in the Bohai Sea, Yellow Sea, East China Sea, and South China Sea, respectively. The monthly total fishing hours per vessel across the four seas are substantially different in May, August, and December. In other months, monthly total fishing hours per vessel across the four seas are close.

Spatial Patterns
The spatial patterns of the yearly total fishing hours are shown in Fig. 7a. In the Bohai Sea, the fishing hotspot areas are scattered and have partially expanded since 2016, mainly near the southern coastline of Hebei Province. In the Yellow Sea, the fishing hotspot areas are mainly distributed near the Shandong Peninsula and have expanded significantly since 2016. In the East China Sea, the most widely distributed fishing hotspot area of the entire study area has suffered a sharp contraction since 2017. In the South China Sea, fishing hotspots are mainly distributed quite close to the coastline and have slightly expanded since 2016.
The spatial patterns of the yearly total vessel hours are shown in Fig. 7b, and the results resemble the patterns of fishing hours. During 2013−2015, the East China Sea area near the coastlines of Jiangsu Province, Zhejiang Province, and Shanghai was basically the most widely distributed hotspot area. After 2016, the vessel hour hotspot areas in the Bohai Sea, Yellow Sea, and South China Sea expanded slightly, while those in the East China Sea contracted significantly.
The spatial patterns of the monthly total fishing hours (aggregated by month during 2013−2018) are shown in Fig. 8a. In January and February, the fishing hotspot areas were mainly in the East China Sea, near the coastline from southern Jiangsu to northern Fujian Province. In March and April, the hotspot areas in the East China Sea and the South China Sea expanded. In May, the fishing hotspot areas in the East China Sea and the South China Sea decreased due to the fishing moratorium policy (Table 1). In June and July, fishing activities in the study area almost stopped, while a few illegal fishing activities continued. In August, fishing  The spatial patterns of the monthly total vessel hours are shown in Fig. 8b, and the results are similar to the fishing hours.

Impact of Tropical Cyclones (TCs) on Fishing Vessel Activities
The total absolute impacts of the 52 TCs in China on the fishing hours (Δt f ) and vessel hours (Δt v ) from 2013 to 2018 are shown in Fig. 9. Due to the impacts of the TCs, the absolute reductions in the fishing hours and vessel hours are mainly distributed in the East China Sea and the South China Sea (blue in Fig. 9). The integrated reduction in fishing hours and vessel hours in the 52 TCs impact period during the 6 years was calculated as well and their percentage relative to the total fishing and vessel hours of the study area during the 6 years are 2.28% and 1.04%, respectively.
The total IR f , total IR v , mean IR f , and mean IR v statistics are shown in Fig. 10. The highest values of the annual total IR f and IR v were in 2013, and the highest values of the annual mean IR f and IR v were in 2015. The monthly total IR f and IR v reached the maximum in July, followed by August and September. The TC landfall points and their corresponding IR f and IR v from 2013 to 2018 are shown in Fig. 11. It can be seen that (1) the landfall points of the TCs were over the coastline between 13°N and 32°N during the 6 years; and (2) the total IR f and IR v in each 1° interval showed a trend of increasing first and then decreasing as the latitude decreased from 32°N to 13°N, and the highest values of the total IR f and IR v appeared in the range of 22°N−23°N. Table 2 provides information on the 10 TCs with the highest IR f values. The table suggests the following characteristics of TCs that caused the most disruption to fishing: (1) Six TCs landed when the fishing moratorium near the landfall site had expired; (2) The minimum and maximum latitudes of the 10 TC landfall points are 18.8°N (near Wenchang, Hainan Province) and 26°N (near Fuzhou, Fujian Province); (3) The maximum CWI and the minimum CWI for the 10 TCs were 48.0 and 11.8, respectively, which demonstrates a medium range of variation; (4) The 10 TCs had a relative impact on the fishing hours (IR f ) of over 94% and a relative impact on the vessel hours (IR v ) of over 59%.

Hazard Index of Tropical Cyclones (TCs)
The interannual changes in the numbers and total CWI of the TCs are shown on the left of Fig. 12. 2013 and 2018 had the largest number of TCs, and the total CWI reached its maximum in 2013. The lowest number of TCs was in 2015, while in this year, the mean event CWI was the highest of all the years. In 2017, the yearly total CWI and the mean event CWI were both the minimum out of the 6 years despite the large number of TCs. Tropical cyclone wind hazard index (CWI) and relative impact on fishing hours (IR f ) and vessel hours (IR v ) aggregated by the latitudes of the landfall points. Event IR f is the IR f of every single historical TC event. Mean IR f is the average IR f of TC events that the longitude of landfall points lies in every 1° range. Total IR f is the sum of IR f of TC events that the longitude of landfall points lies in every 1° range. So are IR v and CWI.
The intra-annual changes in the numbers and total CWI of the TCs are shown on the right of Fig. 12. The top three months with the largest numbers and the highest total CWI are July, August, and September, while no more than one TC event per month appeared from November to May during the 6 years. However, due to the super TC Haiyan in 2013, the monthly total CWI in November was not the lowest.
The spatial distribution of the CWI of the TCs from 2013 to 2018 is shown in Fig. 11. It can be seen that the landfall points are distributed on the coastline between 13°N−32°N. From 32°N to 13°N, the total CWI in each 1° interval increased first and then decreased. In 22°N−23°N, the landfall points (near Guangdong Province) within the interval correspond to the highest total CWI, and the landfall point with the highest event CWI also lies in 22°N−23°N.

Relationship Between Impact Indices and Hazard Index
The Pearson correlation coefficient and Spearman rank correlation coefficient were used to find the correlation between IR f , IR v , and CWI. The results are shown in Table 3. The Pearson correlation test and the Spearman rank correlation test have both passed the 0.01 significance level (P < 0.01). This suggests that both IR f and IR v are significantly related to CWI. Log-normal CDF is used to fit the relationship between IR f , IR v , and CWI. The fitting results are shown with a solid red line in Fig. 13. The goodness-of-fit is measured by R 2 . It can be seen that (1) when CWI < 5, both curves are in a rapid growth stage, showing a near-vertical growth trend; (2) when 5 ≤ CWI < 10, the curves are in a transition stage,    Fig. 13. It shows that (1) when CWI < 5, the dispersion degrees of IR f and IR v are the largest; (2) when CWI ≥ 5, the dispersion degrees of IR f and IR v gradually decrease as CWI increases, which is mainly reflected by the lower bound of the double standard deviation range gradually approaching the mean value; and (3) under the same CWI, the dispersion degree of IR f is less than that of IR v .

Discussion
This study provides a quantitative relationship between the specified TC hazard index and the indices of impact on fishing vessel activities. Based on this relationship, the pre-cyclone impact assessment can be conducted to support fishing vessel management. Besides, compared to previous studies on TC impacts on fishing activities or losses in China (Ren 2009;Han et al. 2016), which were based on national and provincial units in annual or multiannual periods, this study provides the assessment with higher spatial resolution based on each individual TC event.
Although the spatiotemporal patterns of fishing activities and the impact of TC avoidance have been explored based on objective data derived from observed Automatic Identification System (AIS) data, it should be noted that the fishing activity data used in this study may have underestimated fishing intensity and vessel numbers, especially for small vessels. The reason is that according to the International Maritime Organization only international sailing ships with a weight of more than 300 gross tons and non-international ships with a weight of more than 500 gross tons are obligated to install AIS (IMO 2004). Based on official Chinese statistics, the number of small fishing and motorized fishing vessels with a length of less than 12 m accounts for more than 50% of the total fishing fleet in China (MARA 2017). At the same time, only approximately 0.4% of small fishing vessels are equipped with AIS (Kroodsma et al. 2018). Therefore, underestimation of fishing activities must be considered or calibrated before further application, and the fusion of statistical data and AIS data should be explored in future studies.

Conclusion
Very few studies on the impacts of tropical cyclone (TC) avoidance to the vessel and fishing hours can be found, because quantitative fishing hour data at the regional and global scales were barely available or very challenging to extract in the past (Kroodsma et al. 2018). In this study, the analysis of the spatiotemporal patterns of fishing activities by vessel and fishing hours during 2013−2018 derived from AIS data was first implemented. Then the cyclone wind hazard index (CWI) was proposed to reflect the comprehensive TC hazard intensity. Two impact indices, IR f and IR v , were defined to reflect the influence of TC avoidance on fishing activities. Based on the data of 52 historical TCs, the relationship between the impacts on vessel activities and the fishing activities due to TC avoidance and TC wind hazard index were quantified. The main findings of this study have been summarized as below: 1. Most fishing activities took place from September to November after the fishing moratorium, and the overall fishing activity levels near China's offshore area have experienced an overall increase first and then a subsequent decrease because of the enhanced fishing regulations in China introduced in 2015. The most active fishing area throughout the year was located in the South China Sea, where the number of fishing hours per vessel has dropped by approximately 18.6%. However, in some other areas, such as the South China Sea and the Yellow Sea, the spatial extent of fishing hotspot areas has expanded. 2. The proposed TC hazard index CWI considers the integrated effects of wind speed on the temporal and spatial dimensions based on the 1 km wind field dataset of every six hours simulated with parametric wind field models. Based on the impact indices IR f and IR v , the highest hazard index and impact indices, excluding the fishing moratorium period, occurred in September. The area with the largest hazard index and relative impact index values is located offshore near Guangdong Province, which was hit by frequent TCs in the study period. 3. Both impact indices have statistically significant correlations with CWI as an independent variable and were fitted with the cumulative probability function (CDF) of the log-normal distribution.
The relationship between CWI and the impact on fishing vessel activity can be used in a variety of disaster management applications. For example, the quantitative findings of this study can help improve emergency response decision making. An optimal avoidance routes can be solved and disseminated to fishing vessels through satellite-based tele-communication system by the authority, according to the assessment on the costs of different vessel avoidance route scenarios. Another possible application is that it can be used in fishing insurance product design. The major risk metrics of fishing activity interruption, including the loss probability distribution, the expected annual loss, and the variation of loss, can be estimated based on the quantitative relationship between CWI and fishing activities, the spatiotemporal pattern of fishing activities, and other datasets such as historical TC tracks, and so on.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.