How did the urban land in floodplains distribute and expand in China from 1992–2015?

Urban land in floodplains (ULF) is a vital component of flood exposure and its variations can cause changes in flood risk. In the context of rapid urbanization, ULF is expanding rapidly in China and imperiling societal sustainability. However, a national-scale analysis of ULF patterns and dynamics has yet to be conducted. Therefore, this study aims to investigate the spatiotemporal changes in China’s ULF at different spatial scales (the country, region, basin, and sub-basin scales) from 1992–2015. We found that ULF accounted for 44.41% of the total urban land in China in 2015, which was 3.68 times greater than the proportion of floodplains relative to the total land area in China (12.06%). From 1992–2015, the ULF area increased by 26.43 × 103 km2, or 542.21%. Moreover, the ULF area is expected to grow by 16.89 × 103 km2 (53.38%) between 2015 and 2050. ULF growth was strongly associated with the flood occurrence in China, and continued growth will pose a considerable challenge to urban sustainability, particularly in basins with poor flood defenses. Greater attention should thus be paid to ULF dynamics in China.


Introduction
Floodplains are prone to riverine floods due to their low-lying topography and location adjoining a river, stream, lake, or other inland water bodies (White 1945). Improper floodplain development increases flood exposure and aggravates flood risk (White 1945, Mileti 1999. Urban land in floodplains (ULF) is a vital component of flood exposure, and changes in this component can lead to variations in flood risk , Jongman et al 2012, Güneralp et al 2015, Di Baldassarre et al 2013.
China is amongst the most flood-prone and rapidly urbanizing countries worldwide. During 1992-2015, China experienced 157 riverine floods, which inflicted 21,660 deaths, affected 1.48 billion victims, and caused an economic damage of 182.18 billion dollars (Guha-Sapir et al 2016). In the same period, urban land expanded by nearly five times (Xu et al 2016). Previous studies argued that the exacerbated floods and rapid ULF expansion are likely interrelated (Yu 2012, Cheng and.
Many studies have examined China's ULF at regional scales. For example, using Landsat TM/ETM images, Yuan and Jiang (2005) found that the ULF in the Jingzhou flood retention area of the Yangtze River Basin increased by 111% during 1986-2002. Also on the basis of Landsat TM/ETM images, Xu (2012) revealed that ULF encroached upon 4.76% of water bodies and 34.87% of paddy fields in the Yangtze River Delta during 1991-2006. Based on multiyear topography data, Zhou et al (2011) found that ULF occupied 17% (or a total length of 355.4 km) of the rivers in Shenzhen during 1980-2005. Because of a lack of sufficient data, a national-scale analysis of the ULF patterns and dynamics remains to be elucidated. First, extracting large-scale and long-term datasets of urban land is a costly and timeconsuming endeavor (Liu et al 2012). Previous studies mostly examined China's ULF at regional scales using Landsat TM/ETM images (Yuan andJiang 2005, Xu 2012). Second, a national floodplain map is still lacking because flood inundation modeling has not been conducted at a national scale (Yin et al 2015). In previous studies, floodplains were approximated by flood retention areas (Yuan and Jiang 2005) or water bodies (Xu 2012, Zhou et al 2011. Recent progress in remote sensing techniques and flood inundation modeling has provided the necessary data for a national-scale ULF analysis. First, based on the nighttime light data of the DMSP-OLS and the NPP-VIIRS, researchers have derived urban lands in China over the past two decades (Liu et al 2012, Xu et al 2016. Second, flood hazard maps are now a practical reality for large scales (Trigg et al 2016, Rudari et al 2015 due to improvements in numerical algorithms, global datasets, computing capacity, and modeling frameworks. The resulting flood maps have played essential roles in consistently quantifying floodplains and reliably assessing flood risk (UNISDR 2015).
Against this background, our objective was to systematically assess the patterns and dynamics of ULF in China during 1992-2015. We first investigated the spatiotemporal changes in ULF at four scales ranging from country, region, basin, to sub-basin. We then discussed the reliability, consequences, causes, and implications of the ULF patterns and dynamics. Such an investigation is essential for a comprehensive understanding of the flood risk dynamics in China.

Study area and data
Four scales (the sub-basin, basin, region, and country scales) were used to analyze the ULF in China (figure 1(a)). The study area contained 529 sub-basins and 21 basins. The delineations of these sub-basins and basins were first derived from the Food and Agriculture Organization (FAO) and then rectified based on datasets of rivers and elevation, which were obtained from the National Geomatics Center of China. The 21 basins were aggregated into four regions according to the climate zonation of China (Zheng et al 2010), i.e. southwest, northwest, northeast, and southeast China.
Two major datasets were employed to derive the floodplain and ULF areas. First, a flood depth dataset from the CIMA Foundation (Rudari et al 2015) was used to define the extent of floodplains. This dataset was produced based on hydrological and hydraulic models at a resolution of 1 km, which were validated against historical floods. The dataset contains floods with different return periods, e.g. 100 year and 1000 year return periods ( figure 1(b)).
Second, we employed the urban expansion dataset of China for 1992-2015 (Liu et al 2012, Xu et al 2016 to derive the ULF and its changes over time. This dataset was produced based on the DMSP-OLS and NPP-VIIRS nighttime light, vegetation index, and land surface temperature data ( figure 1(a)). This dataset has a resolution of 1 km. Based on a comparison to the urban land estimates extracted from Landsat TM/ETM+ images, this dataset reliably represents urban expansion in China from 1992-2015, with an average overall accuracy of 92.62%.

Defining floodplain and ULF extents
We defined the floodplain as the maximum extent of the 100 year return period flood (UNISDR 2015, Rudari et al 2015. This definition is consistent with the flood risk assessment by Shi et al (2015) and the flood exposure analysis by Jongman et al (2012). Then, we extracted the ULF. To validate the reliability of the floodplain definition, we further compared the defined ULF and the ULF of the 1000 year return period floodplain.

Analyzing the ULF pattern and dynamics
We used equation (1) to calculate the ULF ratio (ULFR), which is defined as the ULF divided by the floodplain area (FP) for zone i (supplementary figure S1 available at stacks.iop.org/ERL/13/034018/mmedia).
In addition, the change rate (CR) and average annual change rate (AACR) were calculated for ULF during 1992-2015 following equations (2) and (3). (2) where 1 and 2 refer to the area of ULF in years t1 and t2, respectively.

Features of floodplains in China
The total area of floodplains in China is 1140.35 × 10 3 km 2 , accounting for 12.06% of China's total land area. A large proportion of the floodplains (36.25%, or 413.42 × 10 3 km 2 ) is located in southeast China. Northeast, northwest, and southwest China contain 29.83%, 21.22%, and 12.69% of the floodplains, respectively (figure 1(b)).

The ULF pattern in 2015
In 2015, ULF encompassed a total area of 31.32 × 10 3 km 2 . It accounted for 44.41% of the total urban land in China, which was 3.68 times greater than the ratio of floodplains to total land in China. The ULFR was 2.74% while urban land ratio outside floodplains was only 0.48% (table 1). Urban land in China was thus disproportionally skewed toward ULF. An overwhelming majority of China's ULF (96.76%) was located in eastern China-southeast and northeast China contained 73.43% (23.00 × 10 3 km 2 ) and 23.33% (7.31 × 10 3 km 2 ) of the national total ULF, respectively. Southeast China featured the largest ULFR (5.56%), which was twice as large as the national average (2.74%). Additionally, northeast China featured an ULFR of 2.15%. In contrast, only a small proportion of China's ULF (3.24%) was located in western China. The ULF in northwest and southwest China had an area of 0.44 × 10 3 km 2 and 0.58 × 10 3 km 2 , respectively, and the ULFRs in these two regions were quite low (0.44% and 0.58%, respectively).
At the basin and sub-basin scales, the ULF was clustered in the eastern coastal areas (figure 2). A large portion of China's ULF (68.93%) was concentrated in the eight coastal basins (figure 2), which together account for only 13.27% of China's total land area. In these coastal basins, 17 sub-basins of the Beijing-Tianjin-Hebei, Yangtze River Delta, and Pearl River Delta-China's three primary urban agglomerations, had a combined ULF of 14.24 × 10 3 km 2 . These subbasins accounted for 45.48% of China's total ULF area but only 1.79% of China's total land area. In addition to the coastal basins, the ULF was also clustered in other basins in eastern China, such as the middle Yangtze River (MYZ) basin.

The changes in ULF during 1992-2015
The ULF increased from 4.88 × 10 3 km 2 in 1992 to 31.32 × 10 3 km 2 in 2015, corresponding to a change rate of 542.21%. The average annual change rate (AACR) was 8.42%, which was higher than the AACR of urban land outside floodplains (8.22%) and 5.33 times greater than the AACR of global urban land (1.58%) during 1990-2014 (Martino et al 2016).
Most of the growth in China's ULF (96.37%) occurred in eastern China. Southeast and northeast China experienced an increase in ULF of 19.47 × 10 3 km 2 and 6.01 × 10 3 km 2 , respectively, with corresponding AACRs of 8.50% and 7.80% (table 2). In contrast, western China had a combined ULF increase of 0.95 × 10 3 km 2 , accounting for only 3.63% of the national total increase. However, the combined AACR for ULF in western China was 13.42%, which was much higher than the national average (8.42%). Therefore, this change likely aggravated the flood risk in western China because the flood defense systems are relatively poor in the region (Yu 2012, Cheng and. At the basin and sub-basin scales, a large portion of China's ULF growth was clustered in the eight coastal basins (figure 3). The ULF growth in these eight coastal basins reached 18.04 × 10 3 km 2 , or 68.22% of China's total increase in ULF. The AACR of ULF was 8.17% in these basins but averaged 9.05% in other basins. The coastal basins thus exhibited the greatest increases in  Note: the numbers in the brackets are the average annual change rates (%).  (table 2 and supplementary  figure S2). This phenomenon was more significant for basins in western China, which experienced 32.82%

The reliability of the floodplain data in China
Correlation analyses indicate that the 100 year return period flood used in this study is reliable for defining floodplain and ULF. First, the defined floodplain and the 1000 year return period flood floodplain were significantly correlated across sub-basins (p < 0.001, R = 0.99). Second, the defined ULF was significantly correlated with the ULF associated with the 1000 year return period floodplain at the sub-basin scale (p < 0.001, R = 0.99). Additionally, the temporal trends of ULF were similar among different flood depths (figure 4). All of the Pearson correlation coefficients among the time series of ULF at different flood depths were greater than 0.99 and significant at the level of 0.001 (supplementary table S1). The defined ULF (ULF with a flood depth greater than 0.00 m) was also significantly correlated with the ULF values at different flood depths (p < 0.001, R > 0.83).
However, we did not consider the dynamics of the floodplain itself in the context of climate change for two reasons. First, our study period is relatively short (23 years, from 1992-2015) and the effects of climate change on floodplain dynamics may not be significant in this period. Second, a national-scale dataset of floodplain dynamics in the context of climate change is not available. Nevertheless, the ULF results in this study have much wider implications-they present not only the ULF in the 100 year return period floodplain but also reflect the ULF characteristics in different flood depths of the 100 year return period as well as in the 1000 year return period floodplain.

Changes in ULF are significantly related to the flood occurrence in China
To assess the possible influence of ULF changes on the flood occurrence, we employed two flood datasets: historical flood extent data from the Dartmouth Flood Observatory (DFO) (Brakenridge 2017) and annual flood frequency data from the EM-DAT (Guha-Sapir et al 2016), both of which span from 1992 to 2015. From the DFO flood extents, we derived the cumulative flood area (CFA) during 1992-2015 for sub-basins, which revealed the spatial variations in flood areas in the study period.
Sub-basins with larger ULF increases were found to have higher CFAs. At the national scale, the changes in ULF were significantly correlated with the CFAs across sub-basins (R = 0.22, p < 0.01). In the four climate regions, the correlations were all positive between the changes in ULF and the CFAs across sub-basins. At the basin scale, the correlation was found to be positive in 19 of the 21 basins and significantly positive in 11 of the basins (p < 0.1) (figure 5). The changes in ULF are thus highly associated with CFAs.
The national ULF time series and the EM-DAT flood frequency dataset during 1992-2015 were also significantly correlated (P < 0.01, R = 0.62), implying that the increase in flood frequency was attributable to the increase in ULF. Following the relative contribution analysis of Johnson (2000), the results further revealed that changes in ULF may have driven 92.62% of the variation in the flood frequency; thus, the ULF contribution was dramatically higher than the contribution of annual storm rainfall (6.38%). Consequently, the ULF expansion was one of the major causes of the increases in flood frequencies.
Our findings are also consistent with those of previous studies. For instance, previous studies found that even with strengthened flood protection, improper ULF expansion was responsible for the increasing flood frequencies and losses in the United States (White 1945, Mileti 1999. Several studies have also confirmed the roles of ULF growth in increasing flood losses (Ceola et

Policy implications
China is urbanizing at an unprecedented rate, which is typically seen as an opportunity for realizing a national dream of socioeconomic rejuvenation (Bai et al 2014). However, such a dream is also impeded by many challenges, among which an aggravated flood risk stands out (Cheng andLi 2015, Yu 2012). At the national scale, we revealed that ULF accounted for a disproportional ratio (44.41%) of China's total urban lands in 2015 and grew by more than five times during 1992-2015. The disproportional distribution and the rapid increase of ULF may imperil the capacity to cope with flood damage . At the regional scale, most of the ULF (96.76% of China's total ULF) was clustered in eastern China in 2015. Meanwhile, the ULF in eastern China is home to a population of more than 300 million people (about a quarter of China's total population) (Martino et al 2016). The concentration of ULF and population in this flood-prone region may augment the risk, especially in light of land subsidence and sea level rise in coastal cities . Moreover, the ULF growth in western China has accelerated, although the amount of ULF is small in this region. The accelerating ULF increase should be given significant attention because the flood defense systems in western China are relatively poor (Cheng and Li 2015).
China's ULF is likely to continue increasing rapidly through 2050. The ULF was linearly related to the urban population during 1992-2015 (p < 0.001, R = 0.99, figure 6(a)), when the urban population increased from 338 million to 779 million (or proportionally from 27.46% to 55.60% of the total population). According to the National New-type Urbanization Plan (2014-2020) published by the central government, China's urban population fraction is expected to rise by 1% a year and reach 60% by 2020 (Bai et al 2014). By 2050, China's urban population is expected to reach more than 1 billion people (United Nations 2014). If the linear relationship between ULF and the urban population holds, the ULF could increase to 48.20 × 10 3 km 2 by 2050, corresponding to a growth rate of 53.58% during 2015-2050 ( figure 6(b)).
ULF expansion is typically a result of multiple causes. First, ULF can facilitate the water supply, transportation, and military defense because of the proximity to water bodies (Ceola et al 2014, Kummu et al 2011, Gu 1997. Second, even after disastrous  MHURD 1998China's National Disaster Reduction Plan (1998−2010 Hydraulic projects should be constructed for comprehensive disaster reduction; core cities should construct flood protection works and implement comprehensive urban disaster reduction plans. Hydraulic projects should be constructed and reinforced for flood prevention, particularly for middle-and small-sized rivers; disaster prevention and reduction should be integrated with regional development plans; integrated risk maps of different scales should be produced.

MHURD
MHURD, MWR, MCA floods, ULF is usually restored in its original location (Ma et al 2016, Kocornik-Mina et al 2015. Third, ULF expansion can be accelerated by the 'levee effect'-a false sense of development safety associated with engineering protection (Cheng andLi 2015, Kates et al 2006). The unprecedentedly rapid ULF expansion in China also reflects policy shortcomings in urban planning and risk management (table 3). First, the urban (and rural) planning laws and disaster reduction plans have only limited power to control the growth of ULF because they cannot forbid ULF expansion in specific regions (Yu 2012). Second, the enforcement of relevant policies can be hampered by the fragmented and overlapping structure of flood risk governance in China. Currently, ULF expansion is relevant to several agencies: the Ministry of Housing and Urban-Rural Development (MHURD), the Ministry of Water Resources (MWR) and the Ministry of Civil Affairs (MCA). These three agencies are responsible for urban planning, flood protection investment, and postdisaster relief, respectively. In this multi-jurisdiction setting, a new administrative structure is needed to coordinate urban planning, risk management, ULF expansion, and floodplain sustainability (Cheng and Li 2015). Finally, a reliable national-scale flood map is currently unavailable (Yin et al 2015), severely hampering the control of ULF expansion.
To control the rapid increase in ULF, China should optimize the governance system and effectively strengthen relevant policies. We recommend that policies related to urban planning and flood risk management impose stringent restrictions on ULF development and specify the responsibilities of local agencies and governments in the implementation of these policies. Furthermore, flood maps at different scales should be developed to ensure location-based risk reductions and ULF control. ULF control should be integrated it into the United Nation Sustainable Development Goals (Abel et al 2016). We also believe that financial instruments, such as flood insurance, could play important roles in controlling rapid ULF expansion (Brody and Highfield 2013).

Future perspectives
This study has several limitations, which need further analyses. First, the flood map can be further improved by incorporating flood defense data into analyses and reinforcing flood models (Rudari et al 2015, Ward et al 2017). Second, the influences of both climate change and urban expansion on ULF can be considered in future studies (Ward et al 2017). Third, the impacts of ULF expansion on other land use/cover types (e.g. wetlands, farmlands, and waterbodies) can be included for a comprehensive analysis of ULF dynamics and landscape sustainability (Wu 2013). Fourth, the ULF expansion and related land use/cover changes may also drive regional climate change (Cao et al 2016), which needs further investigation. Finally, based on the ULF analysis, a flood risk assessment can be implemented by integrating asset density and vulnerability metrics into ULF estimates (Jongman et al 2012, Jurgilevich et al 2017.

Conclusion
ULF represents a large proportion of China's urban land. In 2015, ULF occupied an area of 31.32 × 10 3 km 2 , accounting for 44.41% of China's total urban land, while floodplains accounted for only a relatively small proportion (12.06%) of China's total land area. In other words, the ratio of ULF to urban land was 3.68 times greater than the ratio of floodplains to total land in China in 2015. Moreover, the spatial distribution of ULF in China is highly uneven. The majority of ULF (68.93% of total ULF) is concentrated in eight coastal basins, which account for only 13.27% of China's total land area.
ULF grew rapidly during 1992-2015. During these 23 years, ULF increased by 26.43 × 10 3 km 2 , or 542.21%, with an average annual growth rate of 8.42%. This growth rate is 5.33 times the average annual growth rate of global urban land (1.58%). The majority of the ULF increase (96.37%) was concentrated in eastern China, but the ULF expanded much more rapidly in the west. Moreover, an accelerated increase in ULF occurred across China during the last few years of the study period, particularly in the western basins.
The increases in ULF were strongly associated with changes in the flood occurrence in China. Sub-basins with larger increases in ULF were consistently found to have larger CFAs at the national, regional, and basin scales. In addition, the variations in flood frequency were overwhelmingly controlled by the changes in ULF (92.62%) and relatively little by the changes in annual rainfall (6.38%). Following the current trend, the ULF area will grow by 16.89 × 10 3 km 2 between 2015 and 2050 at a growth rate of 53.38%. This growth will pose a challenge to flood risk management in China, particularly in basins with poor flood defenses. China should therefore optimize the governance system and strengthen relevant policies to effectively control the rapid growth of ULF.