Compound flood potential from storm surge and heavy precipitation in coastal China

The interaction between storm surge and concurrent precipitation can cause greater flooding impacts than either in isolation. This paper investigates the potential compound effects from these two flooding drivers along the coast of China. Statistically significant dependence between them exists at the majority of locations that are analysed, but the strength of the correlation varies spatially and depending on how extreme events are defined. In general, we find higher dependence at the south-eastern tide gauges 15 (TGs) (latitude < 30°N) compared to the northern TGs. Seasonal variations in the dependence are also evident. Overall there are more sites with significant dependence in the typhoon season, especially in the summer. Accounting for past sea level rise further increases the dependence between flooding drivers and future sea level rise will hence likely lead to an increase in the frequency of compound events. We also find notable differences in the meteorological patterns driving compound and non-compound events. 20 Compound events at south-eastern TG sites are caused by low-pressure systems with similar characteristics across locations, including high precipitable water content (PWC) and strong winds that generate high storm surge. Based on historical disaster damages records of Hong Kong, compound flood events account for the vast majority of damages and casualties, compared to univariate flooding events, where only one flooding driver occurred. Given the large coastal population and low capacity of drainage 25 systems in many Chinese urban areas, these findings highlight the necessity to incorporate compound flooding and its potential changes in a warming climate into risk assessments, urban planning, and the design of coastal infrastructure and flood defences.

and precipitation to investigate dependencies and incidences of compound flooding associated with storm 85 surge and heavy precipitation along the coast of China, as well as the large scale weather systems causing compound events.
In this context our three main objectives are to: 1) identify and collate compound events from storm surge and precipitation, and analyse their dependence, including the role of sea level rise and seasonality; 2) understand the driving weather patterns of compound/non-compound events; and 3) compare damages 90 caused by compound and non-compound events.

Data
Most tide gauge (TG) data are kept confidential in China; thus, we obtained hourly sea-level data of 11 TGs with at least 20-year lengths along the Chinese coast from the University of Hawaii Sea Level Center (Caldwell et al., 2015). Locations of TGs and the time series' lengths are shown in Fig. 1. The stations 95 are located south of the Shandong peninsula in China, where tropical cyclone impacts are most severe.
Nine of the 11 TG stations have about 20 years of data , Xiamen and Hong Kong have 46 years  and 52 years , respectively.

[Fig. 1]
Storm surge is extracted using the MATLAB t_tide package (Pawlowicz et al., 2002) by applying a year-100 by-year harmonic tidal analysis with 67 constituents. It also effectively removes the MSL influence. The data has been checked for common errors and 75% completeness of each year is required. An offset in the Hong Kong data is adjusted by shifting the earlier data by 1.02 cm, following Ding et al. (2002).
Cumulative daily precipitation records from 1951-2015 are collected from China Meteorological Administration. The closest meteorological station is chosen to match each TG station, and 9 out of 11 105 TGs are less than 25 km distance. The time series of precipitation observations are usually longer and more complete than TG observations; thus TG data availability determines the lengths of overlapping periods available for the analysis presented here. https://doi.org/10.5194/hess-2020-377 Preprint. Discussion started: 30 July 2020 c Author(s) 2020. CC BY 4.0 License.
To identify weather patterns typically associated with compound and non-compound events, sea surface pressure (SLP), precipitable water content (PWC), and wind fields are used from the Twentieth Century 110 Reanalysis Project Version 2c (Compo et al.,2011).
To compare impacts caused by compound and non-compound events, we employ a typhoon database developed by Yap et al. (2015), which includes historical typhoon records from 1951 to 2012, with the main focus on Hong Kong, Taiwan, and the south-eastern Chinese coastal provinces of Zhejiang, Fujian, Guangdong and Hainan. The database contains information of 853 typhoons in total, with records of direct 115 normalized economic loss (in US$), death toll, and number of people affected.

Selecting compound events
The combination of storm surge and precipitation can exacerbate the flood impacts in different ways (Wahl et al., 2015). First, both heavy precipitation and extreme sea levels (storm surge with high tides) 120 can coincide, leading to more severe floods. This often happens during typhoon events. Second, impacts of a storm surge already causing flooding will increase when significant precipitation occurs at the same time, although the precipitation itself may not be considered extreme. Third, a moderate storm surge can block freshwater water drainage and high precipitation occurring at the same time can lead to more serious flooding (as compared to the same rain event coinciding with low sea level). To capture all of those 125 mechanisms, we investigate the relationship of storm surge and precipitation for two distinct cases: in Case 1 we select extreme storm surge events and the corresponding precipitation within ± 1 day of the surge; in Case 2 we select extreme precipitation events and the corresponding storm surges within ± 1 day of the precipitation (Fig. 2).
[ Fig. 2] 130 We use the peaks over thresholds (POT) method to select extreme events. The POT method refers to selecting events over a high threshold within a certain time span. The annual maximum approach is widely used for sampling extreme events. However, it would be not appropriate here as time series of 9 out of 11 https://doi.org/10.5194/hess-2020-377 Preprint. Discussion started: 30 July 2020 c Author(s) 2020. CC BY 4.0 License.
TGs in China only have around 23 years of data, which would lead to small sample sizes. Furthermore, the second or third largest values in a given year may be larger than the annual maximum in another year 135 (Coles et al., 2001;Arns et al., 2013). To test the sensitivity of the results to the threshold selection, we employ thresholds related to eight percentiles ranging from 95% to 99.5%, i.e., 95%, 96%, 97%, 98%, 98.5%, 99%, 99.25% and 99.5%. Independence of the POT sample was achieved using a declustering time of 3 days. We also test how the inclusion of sea level rise affect the compound events (in Section 4.2).

Dependence analysis
Kendall's rank correlation coefficient τ is employed to measure dependence between storm surge and precipitation. In Case 1, storm surges sometimes could occur without any precipitation, and this leads to ties (i.e., several zero values) affecting the dependence analysis. We use the same method as suggested in Kojadinovic (2010) and Wahl et al. (2015) by assigning ranks randomly, repeating the procedure 100 145 times and calculating the average rank correlation. To better understand the influence of seasonality, dependence is assessed for the full year as well as for summer (June to August), autumn (September to November), and the typhoon season (July to October) separately.
We also assess how compound event frequencies are affected by MSL rise along coastal China. The effects of MSL are initially removed during the harmonic tidal analysis. In order to assess the compound 150 effects under nonstationary conditions, we repeat the analysis but keep the MSL influence and extract surge events by only removing the tidal influence, i.e., total water level minus tide. Then we re-count the numbers of compound events (i.e. falling in Zone 1 in Fig. 2) with MSL included.

Weather patterns
To investigate the meteorological patterns that drive compound and non-compound flood events, sets of 155 the two types of events are selected based on a threshold of 98%. Compound events refer to joint occurrences of high storm surge and heavy precipitation (Zone 1 in Fig. 2). Non-compound events refer to only high storm surge (Zone 2 in Fig. 2) or only heavy precipitation (Zone 3 in Fig. 2). SLP, PWC, and wind fields on the days are selected to match days when compound/non-compound events occurred, then they are averaged into composites to represent reference synoptic-scale weather patterns favouring 160 compound events.

Losses associated with compound and non-compound events
In order to quantify the differences in impacts causes by compound and non-compound events, we employ historical damage records. We take Hong Kong (TG 7) as an example; it also has the most historical damage data available, from 1962 to 2012. The other ten TGs cannot directly be linked to the damage 165 database, as the typhoon database from Yap et al. (2015) only collected damage records at province level. Therefore, to compare damages caused by compound and non-compound events, we match the days when the selected compound/non-compound events (separated in the same way as for the synoptic weather type analysis) occurred with records in the database including information of death toll, people affected, and economic losses.

Dependence between storm surge and precipitation
Figure 3 demonstrates dependence between storm surge and precipitation in Case 1 and Case 2, also indicating the impact of the thresholds (95% to 99.5%) which can influence the correlation. For Case1, south coastal China, which is more affected by TCs (Fig. 1), exhibits higher dependence than the northern 175 part. Overall, Case 1 dependence is also higher than Case 2 dependence and we identify more locations with significant dependence, 11 TGs in Case 1 and 7 TGs in Case 2, respectively. shows the second highest positive dependence for Case 1, and also shows relatively high dependence in 180 Case 2. Lianyungang (TG2) and Beihai (TG9) show insignificant dependence for both cases, indicating that a limited number of compound events occurred at those sites ( Fig. 3c and 3d). Shanwei (TG6) and Zhapo (TG8) show high positive dependence in Case 1, but insignificant dependence in Case 2, meaning https://doi.org/10.5194/hess-2020-377 Preprint. Discussion started: 30 July 2020 c Author(s) 2020. CC BY 4.0 License. that high storm surge is often accompanied by high rainfall but not the other way round. The opposite is true for Lusi (TG3) which has positive dependence in Case 2, but insignificant dependence in Case 1. 185 At most locations the dependence increases when higher thresholds are used to sample extremes ( Fig. 3c and 3d). There are exceptions however, for example, Haikou (TG10) in Case 2 shows higher dependence with a threshold of 99% than 99.5%. At some TGs dependence becomes insignificant due to small sample sizes when thresholds are very high, indicating the trade-off between bias and variance in the threshold selection. [ Fig. 4]

Seasonal variation 205
To better understand the timing of events leading to joint dependence throughout the year (as identified in Section 4.1), the influence of seasons is investigated. TCs are active over the western North Pacific  compound events. However, from the results shown in this study, this pattern is not captured due to 235 limited observation (most observations end in 1997).

Links to weather patterns
We derived composite plots of synoptic conditions of SLP, PWC, and wind fields that drive compound (both high storm surge and heavy precipitation) and non-compound events (high storm surge or heavy precipitation only) across coastal China. To illustrate the results, we focus on Kanmen (TG4) and Shanwei 240 (TG6) on the east and south coast of China, which both have been frequently affected by typhoons ( Fig.   6 and Fig. 7). Results for the other nine stations are shown in Supplementary Figs. S1-S9. Based on the thresholds we selected to identify compound and non-compound events (see Method 3.3), we identify 15 compound events for Kanmen (TG4) and 21 events for Shanwei (TG6), respectively.
[ Fig. 6] 245 The meteorological patterns in SLP, PWC, and wind fields are distinctly different across the three event types. At Kanmen (TG4), compound events are associated with a well-defined low-pressure system with strong east-west and south-westerly winds transporting moist air toward the south-eastern coast of China ( Fig. 6a-c). Non-compound events with only high storm surge exhibit a distinct pressure gradient along the coast (Fig. 6d). As expected, the wind speed is much stronger along the coast for this case (Fig. 6e) 250 compared to the one where only precipitation is high (Fig. 6h), and the northern high wind drives moist air away from the site of interest. The differences in PWC patterns for compound and non-compound events are more pronounced (Fig. 6c, f, i). There is low PWC for the type of only high storm surge events ( Fig. 6f), while high PWC from the Bay of Bengal and cross-equatorial flow is observed for the other two types of events. 255 [ Fig. 7] At Shanwei (TG 6), similarly to Kanmen, the meteorological patterns in SLP show a cyclone-structure for both compound events and non-compound events with only high storm surge ( Fig. 7a and 7d). For events with high storm surge only, there is a distinct pressure gradient and strong wind speed (Fig. 7e).
The PWC is low for the high storm surge events and high for compound and precipitation only events ( Fig. 7c, f, i). For precipitation only events, flows from the Bay of Bengal and cross-equatorial flow is observed, and south-eastern wind drives moist air to the site of interest for compound events and high precipitation only events (Fig. 7b and 6h).
The results for other stations are similar (Supplementary Figs. S1-S9). For compound events, synoptic weather patterns for south-eastern TG sites (latitude < 30°N) show similar low-pressure systems carrying 265 intense PWC and causing strong wind. For northern TGs, such as TGs 1-3 (Supplementary Figs. S1-S3), the low-pressure systems are less developed compared to other TG sites. As most typhoons make landfall along the south-eastern China coasts, their intensity decreases when they move from south to north (see also Fig. 1).

Impacts caused by compound and non-compound flood events 270
To understand impacts caused by compound and non-compound events, we calculated death toll, people affected, and economic losses for both classes of flooding events. Here we identify compound and noncompound events in the same way we did for the synoptic weather type analysis. The impact data base does not include information on all events that we identified, as not all of them led to significant impacts or the impacts were not recorded. A total number of 42 compound flood events (HH) are identified for 275 Hong Kong, 135 events with high surge and low precipitation (HL), and 160 events for low surge and high precipitation (LH). As shown in Fig. 8 In this study, we demonstrate that compound flood events comprised of high surge and heavy precipitation can occur along major stretches of coastal China. The results show that significant dependence exists between the two flood drivers at many locations, especially at sites in lower latitudes (latitude < 25°N).
The dependence varies when using different thresholds in the event sampling and is also affected by seasonality. The latter shows that compound events occur more often during the typhoon season, especially in summer. In terms of weather patterns, compound events at south-eastern TG sites (latitude < 30°N) are caused by low-pressure systems of similar characteristics carrying intense PWC and causing strong winds that generate storm surges. For Hong Kong, we find that compound flooding events were 290 responsible for the vast majority of the recorded casualties and damages, as opposed to flooding events where only one driver was extreme. We also find SLR plays an important role for increasing occurrence of compound events. As SLR keeps rising, it will keep exacerbating the compound effects of flood drivers.
One of the main limitations of this study is the relatively small number of tide gauge sites and limited length of the time-series available, especially from TGs. For now, publicly accessible datasets considered 295 here constitute the most comprehensive collection of hourly sea level data along Chinese coasts. There is an urgent need for longer data sets to be used in order to better assess compound flood risk, especially for south-east China coasts which are prone to TCs. Here we only consider two drivers of flooding, precipitation and storm surge. The role of other flooding drivers needs to be further explored, as well as compound effects under nonstationary conditions, including bivariate frequency analysis, assessing the 300 relationship to climate indices, and the implications for flood risk management. The latter is particularly important, given the low capacity of drainage systems in many Chinese urban areas.
Ignoring compound effects likely leads to an underestimation of flood risk in coastal China, particularly along the south-eastern coasts. It is therefore crucial that coastal cities and urban planning authorities address compound flood effects (including additional drivers such as river discharge or waves) when 305 designing coastal infrastructure and flood defences or developing adaptation plans to combat the negative impacts of climate change. Fig. 4 Counts of compound events between storm surge and precipitation with/without sea level rise at threshold of 98% (falling in Zone 1 in Fig. 2). + indicates compound events between storm surge considering historical sea level rise trend.
indicates compound events between storm surge by removing annual mean sea level.