Risk assessment of shanghai extreme flooding under the land use change scenario

Environmental changes have led to non-stationary flood risks in coastal cities. How to quantitatively characterize the future change trend and effectively adapt is a critical problem that needs urgent attention. To this end, this study uses the 2010 Shanghai land use data as the base and utilizes the future land use simulation (FLUS) model to simulate future land use change scenarios (2030, 2050, and 2100). Based on the results of storm and flood numerical simulations, probabilistic risk, and other multidisciplinary methods, extreme storm and flood risks of various land uses (residential, commercial and public service, industrial, transportation, agricultural, and other land) in Shanghai are analyzed. Our findings demonstrate that the future land use simulated results show that the simulation accuracy is very high, meeting the needs of our research. We evaluated future land use exposure assets and losses and found that their spatial distribution patterns are consistent, ranging from a sporadic distribution for 1/10-year to a banded distribution for 1/1000-year under the two emission scenarios. In terms of economic loss, the losses of total land use in Shanghai for 1/1000-year in 2100 are 1.8–2.7 times that of 2010 under the RCP8.5 scenario. The expected annual damage (EAD) of Shanghai’s land use in 2030, 2050, and 2100 is 189.9 million CNY, 409.8 million CNY, and 743.5 million CNY under the RCP8.5 scenario, respectively, which is 1.7–3.0 times the EAD under the RCP2.6 scenario. Among them, residential, commercial and public service land as well as industrial land has the highest EAD. Risks are mainly distributed in the city center, the lower reaches of the Huangpu River, the northern shore of Hangzhou Bay, the Qingpu (QP)–Songjiang (SJ) depression in the southwest, and Chongming (CM) Island (southwest and northeast). Our work can provide meaningful information for risk-sensitive urban planning and resilience building in Shanghai. These multidisciplinary methods can also be applied to assess flood risk in other coastal cities.


Introduction
Climate change, sea-level rise, land subsidence, and rapid urbanization are likely to increase flood risk in low-lying coastal cities in the future, which already face huge risks of extreme storm floods (Najafi et al. 2021;Wolff et al. 2020;Oppenheimer et al. 2019;Shen et al. 2019;Chan et al. 2018;Nicholls et al. 2014). The Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) estimates a sea-level rise that ranges from 0.45 to 0.82 m in 2100, following the representative concentration pathway (RCP) 8.5 scenario (Hinkel et al. 2015). In addition, if the effects of land subsidence are considered, the impact of sea-level rise will be more prominent in the future, posing severe challenges to the sustainable development of coastal cities (Catalao et al. 2020;Abadie et al. 2020;vousdoukas et al. 2018). There is an urgent need to understand the spatiotemporal dynamics and future trends of flood risk under environmental change scenarios to ensure the sustainable development of coastal cities (Jongman 2018;Hinkel et al. 2018).
Some scholars have already carried out research on the dynamic changes of flood risk in coastal cities under the background of sea-level rise, land subsidence, and climate change. For example, several studies have focused on the impact of coastal flooding on people and assets under the scenario of sea-level rise for New York (Garner et al. 2017;Lin and Shullman 2017;Lin et al. 2016;Reed et al. 2015;Horton et al. 2015;Aerts et al. 2013), London (Hall et al. 2019;Hinkel et al. 2018), Ho Chi Minh City (Ngo et al. 2020;Bangalore et al. 2019;Scussolini et al. 2017), and Jakarta (Ward et al. 2011). Due to rising sea levels, the 1/500-year storm flood event in New York City before the Industrial Revolution has currently become a 1/25-year storm event (Garner et al. 2017;Reed et al. 2015). Hinkel et al. (2018) discussed the risks to coastal cities under the sea-level rise scenario in the twenty-first century. Among them, Ho Chi Minh City and New York City will face considerable risks. Scussolini et al. (2017) adopted scenarios of sea-level rise and socioeconomic changes using two risk indicators, asset damage, and potential casualties and assessed the current and future (2050 and 2100) flood risks in Ho Chi Minh City, Vietnam. Other studies usually regard urban land or agricultural land as relatively homogeneous elements at risk and evenly allocate GDP and assets to pixels of urban land or agricultural land (Winsemius et al. 2016;Jongman et al. 2015).
In recent years, significant progress has been made in the analysis of coastal flood loss and risk in Shanghai (Shan et al. 2021;Shi et al. 2020;Du et al. 2020;Shan et al. 2019;Ke 2014). Low-probability and high-impact compound storm flood risk has always been the focus of researchers (Wu et al. 2019;Shan et al. 2019;Li et al. 2019;Ke 2014;Hallegatte et al. 2013;Aerts et al. 2014;Wang et al. 2012), because such risks may cause damage and huge losses to infrastructure such as buildings, assets, industries, and transportation, seriously threatening urban safety (Shan et al. 2021;Du et al. 2020;Li et al. 2019;Ke 2014;Aerts et al. 2014;Hallegatte et al. 2013). These studies reflect the emergence of urban flood disaster risk dynamic changes and analyze changes in hazard factors under scenarios such as climate change, sea-level rise, and land subsidence, characterizing the state of the elements at risk with past and current land use to carry out scenarios of possible loss analysis (Du et al. 2020;Aerts et al. 2014). However, Shanghai's elements at risk are highly dense and complex, and the future storm and flood risk evolution of elements at risk is not systematically understood. The urban spatial model is an effective tool for understanding the process of urbanization and future scenarios, supporting urban planning and management policies, and evaluating the impact of the urbanization process on the environment and ecosystem (Liu et al. 2017;Filatova et al. 2016). In the field of flood disaster research, urban spatial models such as FLUS model have been applied to future scenario analysis of elements at risk Wu 2020;Kim and Newma 2020;Liang et al. 2018;Song et al. 2017) to better reveal the dynamics of flood risk. For example, Lin et al (2020) developed on an approach that employs a future land use simulation (FLUS) model for scenario-based 100-year coastal flood risk assessment. Guangzhou Metropolitan Area (GMA) is used as a case study, the results indicate that urbanization is the main driving factor and will increase flooding of built-up areas in Guangzhou in 2030 and 2050. Wu (2020) took Chongming's land use as the research object, considered its different future development directions, designed 3 kinds of land use change scenarios, and selected a series of driving factors, then, FLUS model was used to predict Chongming distribution of land use in 2035. Ultimately, the risk of future flood exposure in Chongming was studied, and the results showed that under the superimposed effect of sea-level rise and the high tide level of storm surge, the eastern part of Chongming will be the most exposed area affected by topography in the future. The above studies showed that scenario-based flood exposure and flood risk assessment using the FLUS model not only facilitate a more comprehensive understanding of future land use changes due to the ongoing urbanization process in relation to projected environmental changes (sea-level rise, storm surge, and land subsidence), but also enable us to have a more comprehensive insight into the spatial distribution of high-risk flood areas in the future. This allows us to promote the priority of flood control measures at the most critical locations in the region. Therefore, it is necessary to predict future land use changes based on urban spatial model to construct future elements at risk spatiotemporal change scenarios. This not only effectively assesses the future risks of extreme storm floods, but also has important significance for improving the understanding of the dynamic evolution of risks.
This study is based on the land use distribution data of the base year (2010) combined with the FLUS model to predict the future (2030, 2050, 2100) land use distribution pattern, which is further combined with the future extreme storm and flood scenarios in Shanghai to carry out different types of land use loss and risk analysis. Our work can provide meaningful information for risk-sensitive urban planning, flood risk adaptation, and resilience building in Shanghai. The methodology can also be used for risk analysis in other coastal cities facing the threat of storm flooding.

Study area
Shanghai, located on the eastern edge of the Taihu dish-shaped depression in the Yangtze River Delta, is the world's largest port city with an area of 6340 km 2 and a permanent population of 24.28 million in 2019. In terms of topography, the east of Shanghai is slightly higher than the west, with an average altitude of about 4 m. It is surrounded by water on three sides, and the Huangpu River and Suzhou River pass through the city (Fig. 1). Shanghai is affected by land subsidence due to loose Quaternary strata and human activities. Storm flooding caused by typhoons is the main natural disaster in Shanghai. An extreme storm flood event in 1905 caused more than 29,000 deaths; Typhoon Winnie killed seven people and flooded more than 5000 households in 1997 (Wen 2006). Although the construction of flood control measures in the past 50 years has effectively reduced the risk of storm floods, Typhoon Matsa in 2005, the 2013 Typhoon Fitow, and the 2019 Typhoon Lekima substantially threatened Shanghai. Particularly, during Typhoon Fitow, the sea winds were fierce and the Hangzhou-Jiaxing-Huzhou area experienced torrential rains and extreme rainstorms. In the context of climate change, relative sea-level rise, and urban expansion, Shanghai will face greater risks and challenges from typhoon storm floods in the future (Wang et al. 2018).

Data
The data used in this work mainly include Shanghai extreme storm flood scenarios in 2010, 2030, 2050, and 2100, land use data in 2010, and land price data in 2010.

Extreme storm floods
We use a simplified 2D flood inundation model (FloodMap Inertial) to simulate dike failure-induced flooding and to derive inundation maps. FloodMap Inertial, as well as its earlier diffusion-based version (FloodMap) (Yu and Lane 2006), has been tested and implemented in fluvial, pluvial, and coastal environments in Shanghai (Yin et al. 2020) and New York city (Yin et al. 2016), showing acceptable accuracy of flood prediction at a city scale. The model assumes that the coastal/river floodplain is protected by a continuous, broadcrested embankment through which water flow exchange occurs between the sea/channel and hinterland. A simplified solution has been adopted in the model for the treatment of overland flood routing in a raster-based environment.
Considering the risk decision of sea-level rise, the typical concentration paths RCP2.6 and RCP8.5 were selected, and multiple scenarios and probability estimates of future sealevel rise were carried out. The relative SLR in the present study were estimated by incorporating the climatically driven absolute SLR decomposed from the projections at Lvsi station and the linear approximation of future subsidence rate (6 ± 1 mm/year) in Shanghai. The latter is based on citywide leveling measurements since 2000 and the long-term target of subsidence control. All relative sea-level projections by 2100 are given with respect to the baseline year of 2010 (Yin et al. 2020). Then, according to the height and structure of the river embankments along the Shanghai coast, the overwhelming and breaching discharges (more than 20 l/m/s) of storm surges in 1/10, 100, and 1000 years are calculated. Next, estimation results of absolute sea-level rise in the typical RCP scenario of the Yangtze River Estuary and the InSAR inversion are used to predict the settlement of coastal seawalls; it is estimated that the dikes will be flooded and broken in the 10-, 100-, and 1000-year storm surge scenarios in 2010, 2030, 2050, and 2100. The results show that, in terms of overtopping/overflowing, this floodwall can withstand a 100-year return period (RP) event under the current condition (2010), with only two potential failure locations on the left bank of the Huangpu River's middle reach. For the 1/1000-year storm, wave overtopping and overflowing spread to the upstream reach and the right bank of the Huangpu River, with additional failures scattered throughout the middle reach. In terms of breaching due to structural failure, the floodwall along the Huangpu River is expected to remain intact under the 100-year RP event until 2050, while dike breaching could occur at two locations in the upstream reach of the Huangpu River during the 1000-year RP event with the baseline sea level (2010). Breaching further spreads to other sections of the upstream reach in the year 2030 and sparsely affects the downstream reach ( Fig. 2 and Fig. S1). Chongming Island (southwest and northeast), the Qingsong low-lying area in the west, and the north bank of Hangzhou Bay are the current and future dikes with the lowest safety and the highest inundation risk area (Yin et al. 2020). Due to the effects of sea-level rise and land subsidence, still water levels would increase proportionately at all return periods. Under the RCP 8.5 scenario, the current 10-and 100-year flood levels in Shanghai could be exceeded approximately two times more frequently in 2030, three to five times more frequently by 2050, and over fifty times more frequently with a 136-cm rise in relative sea level (50th percentile) by the end of the twenty-first century. The flood inundation scenarios (degree and depth) caused by flooding increase significantly as the return period increases. The spatial resolution of flood inundation map is 50 m by 50 m. Finally, the simulation results are verified, which are based on a set of typhoon random events affecting Shanghai under current climate conditions (Yin et al. 2020). The ADCIRC storm surge model is used to simulate the 1/10-, 100-, and 1000-year storm surge-level distributions, and the three most severely impacted Shanghai areas of the 5,180 typhoon storm surge events were extracted, including one which landed on Chongming Island head-on, and one on Jinshan (due to the asymmetric structure of the typhoon field, the strong wind field on the right caused the Yangtze River estuary to have a higher water level than Hangzhou Bay). All results show that Chongming Island faces the greatest risk of storm surge and flooding.

Land use type
The land use type data of Shanghai in 2010 and 2013 come from the Shanghai Institute of Surveying and Mapping. The land use type data describe the land use type hierarchically, are divided into 3 departments (one-digit code identifier), and then subdivided into 15 sub-sectors (2 digit code identification), and 73 land use sub-categories (3 digit code identification) that are further divided into 6 categories, which are residential land (including urban residential land and rural homestead), commercial and public service land (including commercial service land, public management, and service land), industrial land, transportation land (including railway, highway, port, airport, square, and other transportation facilities), agricultural land (including farmland, garden land, woodland), and other land (including water area, park green space, grassland, land under construction, land for other facilities, unused land, etc.).

Land use price
Land classification and benchmark land price for Shanghai in 2010 is determined based on different uses (Cao, 2013). The benchmark land price reflects the regional average price within the same level at the benchmark time. The land price data take into account the influence of time factors, age, floor area ratio, regional factors, individual factors, floors, and other conditions. Among them, the benchmark land price factor correction coefficient considers regional factors (prosperity, traffic conditions, infrastructure conditions, environmental conditions, population conditions, and urban planning) and individual factors (frontage conditions, parcel shape, parcel area). Ultimately, residential land was 10,706 CNY/m 2 , commercial and public service land was 13,365 CNY/m 2 , industrial and mining storage land was 3604 CNY/m 2 , transportation land was 1927.2 CNY/m 2 , agricultural land was 5 CNY/m 2 , and other land was 840 CNY/m 2 . Since land prices and housing prices are closely related (Du et al. 2011), the estimation of future (2030 and 2050 year) land prices is based on the future average annual growth rate of housing prices in first-tier cities (Chen 2014). Due to the uncertainty of future land prices, this study assumes that the land price in 2100 is the same as that in 2050.

Methods
This research integrates elements at risk modeling, hydrology and hydrodynamic models, and risk analysis to develop a comprehensive research framework and apply it to the future flood risk analysis of Shanghai under the land use change scenarios (Fig. 3). Firstly, the future land use simulation (FLUS) model is used to simulate the spatial distribution of future land use. Secondly, the two-dimensional (2D) hydrodynamic model FloodMap is adopted for simulating the range and depth of storm flood inundation, and the extreme storm flood inundation maps with different return periods in 2010, 2030, 2050, and 2100 are drawn. Thirdly, the value of Shanghai's land use assets in 2010, 2030, 2050, and 2100 is analyzed with the extreme storm and flood scenarios to obtain exposed assets. Then, the land use vulnerability curve and exposed assets are combined to determine economic losses. Finally, the loss of multiple return periods is integrated and the expected annual damage (EAD) is quantitatively analyzed.

Land use simulation and verification
The Markov chain (MC) is a random mathematical process in a finite time sequence, which can predict dynamic variation characteristics and has high operational precision and high prediction accuracy (Lu et al. 2018). In this study, MC model was used to calculate the probability of the conversion of various land types and then predict the number of future land change.
The future land use simulation (FLUS) model is a land use change simulation model that combines a top-down system dynamic (SD) model and bottom-up cellular automata (CA). Liu et al. (2017) studied the working mechanism of the FLUS model, which incorporated natural factors including future global warming and precipitation variations and socioeconomic developments into both the SD and CA models and considered the influence of neighborhood influence, weight factors, self-adaptive land inertia, and conversion costs. It was found that FLUS model has higher simulation accuracy than the CA and CLUE-S models, which has been applied to research at different scales and for different purposes Wang et al. 2019;Huang et al. 2018;Liang et al. 2018;Dong et al. 2018). Therefore, the FLUS model was adopted for spatial simulation in the present study.
To quantitatively assess the simulated result, samples in the test set were used to build the grid-by-grid confusion matrix of the simulated result versus the actual land use pattern, from which the overall accuracy and the Cohen's Kappa coefficient for all land use types were calculated (Liu et al. 2017). Additionally, the agreement of the changes was validated using the figure of merit (Fom), which is superior to the Kappa coefficient in assessing the accuracy of simulated changes (Pontius et al. 2008). The Cohen's Kappa and Fom index can be expressed as the following equation: where P m represents the correct proportion of the simulation, P i represents the correct proportion of the model in the random case, and P n represents the proportion of the correct simulation in the case of ideal classification. The value range of Kappa coefficient is 0-1. The larger the Kappa coefficient, the higher the accuracy.
(1) Kappa = (P m − P i )∕(P n − P i ) Fig. 3 The methodology framework where A is an area of error due to observed change predicted as persistence, B is an area of accuracy due to observed change predicted as change, C is an area of error due to observed change predicted as changing to an incorrect category, and D is an area of error due to observed persistence predicted as change.

Asset value assessment
Using 2010 as the base year, the asset value can be obtained by evaluating information such as the area and land use value in 2030, 2050, and 2100. The calculation formula for the valuation of land use assets is: where B asset (2030 2050, 2100) is the value of land use assets; S is the area of land use; and P is the land price.

Exposure analysis
An overlay analysis of extreme storm flooding scenarios and various land use value distribution maps in multiple return periods is used to assess land use exposure assets. The calculation formula is: where E (2030, 2050, 2100) is the exposed asset; B asset (2030 2050, 2100) is the asset value (Eq.(3)); and H (2030,2050,2100) is hazard. (

Loss analysis
The stage-damage function developed by Yin et al. (2012) was used to evaluate the losses of different land use in this work (Fig. 4). The stage-damage function, derived by summarizing a variety of previous studies and empirical events in Shanghai (Yin et al. 2012), indicates the relationship of land use loss rate with water depth at an interval of 0.5 m. Six types of land use vulnerability curve types were obtained, including residential land, commercial and public service land, industrial land, transportation land, agricultural land, and other land in Shanghai (Fig. 4). The calculation formula for direct economic loss of land use is: where T loss (2030,2050,2100) is the loss; E (2030, 2050, 2100) is the exposed assets (as shown in Eq. (4)); and R is the loss rate of different land use under different inundation depth.

Risk expression
The estimation of risk adopts the most commonly used method in the world, that is, to express the risk in terms of expected annual damage (EAD) (Du et al. 2020;Lin and Shullman 2017). The calculation formula is: where EAD is the expected annual loss; x is the flood loss (or risk value); and f(x) is the probability of flood loss. EAD in the study area can be used as the basic basis for flood disaster cost-benefit analysis.

Future land use and accuracy verification
In this study, the 2010 land use status data in Shanghai were used as the initial year data, multiple types of land use impact factor data were selected, and the Markov chain model of land quantity prediction and the FLUS model of land use dynamic simulation model were used to calculate the future demand and regional suitability distribution of various  (Liu et al. 2017). The prediction results for the spatial distribution of land use in 2030, 2050, and 2100 with spatial resolution of 50m by 50m show the overall spatial distribution trend of land use in the following three stages. Agricultural land will gradually be transformed into commercial and public service land and residential land. Commercial and public service land in the city center is relatively dense, and public buildings and residential areas in the suburbs have gradually developed (Fig. 5).
In addition, it has been credibly verified that the 2030 land use forecast map is relatively consistent with the Shanghai Municipal Land Use Planning Map of 2035 in the "Shanghai City Master Plan (2017-2035)" (https:// www. shang hai. gov. cn/ newsh anghai/ xxgkfj/ 20350 03. pdf). Among them, the residential land corresponds to the residential living area, and the location is roughly the same; the commercial and public service land matches the commercial office area and the public service facility area and is mainly distributed in the central city of the inner ring; and the agricultural land coincides with the permanent basic farmland and other agricultural spaces and is mainly distributed in the three islands and outer ring of Chongming District. For industrial land, it is in keeping with industrial bases and industrial communities and is mainly distributed outside the outer ring and in areas near the river and the sea.

Exposure assets analysis of land use
The asset value of six types of land use and its spatial distribution in Shanghai can be estimated based on total area and land prices using Eq. (3). The spatial distributions of exposed assets for land use are mapped by overlay analysis to combine the extreme flooding inundation scenarios with the land use asset value maps in Shanghai using Eq. (4) (Fig. 6).
Our analysis shows that the exposure increases rapidly as the return periods of storm flooding increases under the two emission scenarios (Table 1 and Table S1). Under the RCP8.5 scenario, residential, commercial and public service and industrial land has the highest exposed assets, accounting for the largest proportion of 74% in 2100. Moreover, when there is a flood 1/1000-year flood, the sum of the exposure assets of the above three types of land in 2100, 2050, and 2030 will be 1.3-1.7 times that of 2010 (Table 1).

Loss analysis of land use
By combining the exposure assets and vulnerability curves of land use, Eq. (5) can be used to obtain the direct economic losses caused by extreme storms and floods in different return periods of 2010, 2030, 2050, and 2100, and their spatial distribution ( Fig. 7 and Fig. S2). Under the RCP8.5 scenario, the spatial pattern of land use loss during the return period of the three extreme storms and floods is scattered in the 1/10 years, mainly on both sides of the Huangpu River. As the return period increases, the scope and amount of losses continue to increase. When the return period is 1/1000 years, the spatial pattern of loss distribution in 2030 and 2050 is mainly distributed along the banks of the Huangpu River, and the main urban area will form a contiguous distribution area in 2100. At the same time, the loss of the city center concentrated on the Suzhou River mouth is gradually increasing (Fig. 7) and is mainly distributed in the city center, and the lower reaches of the Huangpu River, the northern shore of Hangzhou Bay, the Songjiang area in the southwest, and Chongming Island (southwest and northeast). In the RCP2.6 scenario, the spatial pattern of loss distribution is similar to the RCP8.5 scenario (Fig. S2). Similar to exposed assets, the loss of residential, commercial and public service and industrial land accounted for the largest proportion of total losses, accounting for 82% in 2100 (Fig. 8). Under the RCP8.5 scenario, for 1/1000-year event, the losses of total land use in 2100, 2050, and 2030 will be 2.7 times, 2.0 times, and 1.8 times that of 2010, respectively (Fig. 8).

Risk expression
The loss values of the three types of storm flood return periods are used to establish the extreme storm flood annual exceedance probability (AEP)-loss curve of land use, the 1 3  of Shanghai's land use in 2030, 2050, and 2100 is 189.9 million CNY, 409.8 million CNY, and 743.5 million CNY, respectively (Table 2), which is 1.7 to 3.0 times the EAD under the RCP2.6 scenario (Table S2). Residential, commercial and public service and industrial land has the highest EAD; in 2100, they will be 209.0 million CNY, 220.5 million CNY, and 176.2 million CNY, respectively ( Table 2). The risks are mainly distributed in the city center, the lower reaches of the Huangpu River, the northern shore of Hangzhou Bay, the Songjiang area in the southwest, and Chongming Island (southwest and northeast).

Proposed risk-reduction measures
Shanghai is the financial center in China. It has the highest standard flood control system in China (designed using 1/1000-year standards). However, sea-level rise and land subsidence have brought major challenges to Shanghai's flood risk management, which has also caused Shanghai to be considered in foreign research as one of the coastal cities with the highest flood risk and the fastest growth rate in the world. According to estimates by Hallegatte et al. (2013), among the 136 largest coastal cities in the world, Shanghai is one of the cities with the fastest increase in annual expected loss (EAD) (ranking 13th) compared with 2005. Wang et al. (2012) showed that sea-level rise, land subsidence, storm surges, and surface runoff work together to cause more complex, variable, and sudden flood disasters. By 2100, half of Shanghai's city will be affected by coastal floods, and 46% of seawalls and flood walls will be overwhelmed. This type of extreme flood caused by simultaneous typhoons, storm surges, astronomical tides, heavy rains, and river floods is a low-probability and high-impact catastrophe (Aerts et al. 2013). For example, the 2013 Typhoon Fitow was the first time since 1949 that Shanghai suffered the "four encounters" disaster of strong winds, storms, tides, and floods; 121,000 people were affected and the direct economic loss was 370 million CNY. Taking into account the superimposed effects of sea-level rise, land subsidence, and climate change, the risk of compound extreme storm and floods to Shanghai in the future may be further increased. This kind of low-probability and high-impact extreme compound event is the focus of Shanghai's flood risk adaptation and risk decision making. Drawing on the flood control experience of other coastal cities abroad, Shanghai can also implement hard adaptation measures: adaptation measures based on engineering measures such as tidal sluices, flood walls, and sea dikes, which reduce flood risk by changing the probability of flood hazards, as well as implementing soft adaptation measures: adaptation strategies based on non-engineering measures such as building codes and coastal wetlands, which mainly reduce flood risk by reducing exposure and vulnerability. From the results of this study, as the return period of storm floods increases, the city center, the Songjiang district in the upper reaches of the Huangpu River, the north bank of Hangzhou Bay, and the Qingpu-Songjiang depression in the southwest become high-risk areas for disasters and assets exposure. Different measures need to be taken to deal with storm flooding for different regions. (1) Shanghai's city center has dense population and assets. It is necessary to implement rainwater storage and peak reduction facilities such as sponge cities to strengthen the construction of flood control capabilities and enhance the safety of the city center. (2) Shanghai should focus on strengthening the flood control system in the weak sections of the upper and middle reaches of the Huangpu River and newly added urbanized areas, especially in low-lying areas such as upstream Songjiang, the northern shore of Hangzhou Bay, and the Qingpu-Songjiang area in the southwest. Several flood diversion and discharge areas are recommended. Hospitals, schools, residential areas, and other important units and facilities should not be planned within 3 km on both sides of the river. In other areas, the flood control wall should be increased according to the latest analysis of Huangpu River tide level to maintain its designed flood control capacity. On the one hand, the current defensive capacity of the embankments on some of Shanghai's shores is low, and the defense standards need improvement, such as the sea embankments on the nearby shores of Chongming North Coast, Chongming Nanmen Port, Baozhen Port, Xijia Port, Wusongkou Port, and Luchao Port. On the other hand, full attention should be paid to the impact of sea-level rise and ground subsidence on the defensive capacity of embankments. Land subsidence will significantly change the shape of the underlying surface and will lead to, combined with the superimposed effect of seal-level rise on the water level, the reduction of the flood control project fortification standards and the weakening of flood control capacity. Therefore, regular monitoring and settlement analysis of the flood control project should be strengthened. Secondly, the construction of large-scale high-rise buildings and large-scale underground projects within 2 km of the both banks of the Huangpu River should be strictly restricted to control the land subsidence in Shanghai, especially the land subsidence on both sides of the Huangpu River. Finally, it is possible to make full use of the ecological functions of urban green space, plan park green space, use landscape belts in flood-prone areas on both banks of the Pujiang River as flood buffer zones, and plan temporary flood diversion areas in key sections. (3) The Qingpu-Songjiang depression on the upper reaches of the Huangpu River should be full used for water storage and drainage of the river network, increase the water area of the river and lake, interrupt small river channels, dredge the sediment, control the water level of the river, and prevent the low-lying area from being flooded. (4) Due to coastal erosion at north Hangzhou Bay, beach protection measures should be strengthened, especially in frontier parts of seawalls in Luchao Port and Jinshan Petrochemical Factory; increasing dam capacity along the beach may be necessary. In addition, it is necessary to improve the local flood control infrastructure including sluices, pumping stations, embankments, and urban drainage systems to ensure that orderly construction and efficient operation of various water conservancy infrastructures are managed precisely and intelligently, as well as adapting various wet/dry-floodproofing measures to improve the spatial response capacity of risks and enhance urban resilience.

Limitations and future work
There are some limitations in our study, which need to be improved in the future. Firstly, we simulated the land use in Shanghai in 2030, 2050, and 2100; due to the limitation of data resources, we can only use the Shanghai City Master Plan (2017-2035) to compare land use in 2030 and found that the simulation results are almost the same as the planning results, which further shows that the accuracy of land use simulation in 2030 is very high. For the land use simulation in 2050 and 2100, due to the relatively long-time scale, certain limitations are unavoidable. Secondly, the vulnerability curve established by Yin et al. (2012) is used to evaluate the loss of land use in Shanghai in 2030, 2050, and 2100. With the acceleration of urbanization and the renewal and construction of the city, the flood vulnerability of land use may also change dynamically. Thirdly, the direct economic loss of land use is evaluated, but the indirect economic loss that has a longer impact cycle and a wider scope is not considered. Further work should be combined with sociology and economics to carry out the assessment of indirect economic losses.

Conclusion
This study presents an integrated modeling framework to simulate future land use by using FLUS model and analyzes the exposure, loss, and risk patterns of Shanghai's various land uses (residential, commercial and public service land, industrial, transportation, agricultural, and other land use) in the future (2030, 2050, and 2100) under extreme storm floods. The main conclusions are as follows: (1) The overall accuracy, the Cohen's Kappa coefficient, and the figure of merit (Fom) for all land use types were calculated. The future land use simulated results show that the accuracy of the simulation is very high, meeting research needs. (2) The spatial pattern of exposure and loss distribution is similar, forming the scattered distribution of 1/10-year mainly distributed on both sides of the Huangpu River under the two emission scenarios. As the return period increases, the scope and amount of exposure assets and losses continue to increase. For 1/1000-year, it mainly showed a gradual strip distribution in 2030 and 2050, and formed continuous distribution of the city center and the Qingpu-Songjiang depression in the southwest in 2100. According to the statistics of exposed assets, residential, commercial and public service and industrial land has the highest exposed assets, accounting for the largest proportion of 74% in 2100 under the RCP8.5 scenario. In terms of economic loss, the losses of total land use in Shanghai for 1/1000 years in 2100 are 1.8-2.7 times that of 2010 under the RCP8.5 scenario.
(3) Risks are mainly distributed in the city center, the lower reaches of the Huangpu River, the northern shore of Hangzhou Bay, the Qingpu-Songjiang depression in the southwest, and Chongming Island (southwest and northeast). Under the current scenario, the EAD for land use in Shanghai is 92.4 million CNY. Under the RCP8.5 scenario, the EAD of Shanghai's land use in 2030, 2050, and 2100 is 189.9 million CNY, 409.8 million CNY, and 743.5 million CNY, respectively, which is 1.7 to 3.0 times the EAD under the RCP2.6 scenario. Among them, residential, commercial and public service and industrial land has the highest EAD.