1 Introduction

Freshwater is vital to humans. People have inhabited places close to rivers and lakes to ensure water supply and navigation since ancient time (Bertuzzo et al. 2007; McCool et al. 2008; Best 2019; Viero et al. 2019). However, close water proximity is not always a positive factor and can cause devastating flooding, which forces human settlements to keep a certain distance from waterbodies (Di Baldassarre, Kooy et al. 2013; Alfieri et al. 2017). Socioeconomic development promotes the implementation of engineering measures, such as dams and levees, to control floods. However, the reduced flood frequency and a sense of safety because of the engineering measures will likely boost floodplain development and increase flood exposure surrounding waters—the so-called “levee effect” (White 1945; Di Baldassarre, Viglione et al. 2013; Jongman 2018; Du et al. 2019). The distance between human settlements and waterbodies is vital to understanding the choices of society in facing the socio-hydrological dilemma between using waters and floodplains and avoiding flood risk (Loucks 2015; Sivapalan 2015).

The proximity of human activities to water has drawn increasing attention (Becker and Grunewald 2003; Fang, Ceola et al. 2018; Ceola et al. 2019). Kummu et al. (2011) examined the distance from human settlements to freshwater and found that over 50% of the world’s population lived within 3 km of freshwater and only 10% lived in areas farther than 10 km away from freshwater. Based on nightlight data as a proxy for human activities, Ceola et al. (2014) analyzed the increase in nightlight digital numbers (DN) along river networks from 1992 to 2012 and revealed a strong correlation between increasing nightlights and intensifying flood damages. Ceola et al. (2015) observed a high human concentration in the vicinity of rivers during 1992–2013. Mård et al. (2018) analyzed the dynamic proximity of total nightlight DN values to main rivers during 1992–2013 in 16 countries with different flood protection levels. The results showed that settlements tend to move farther from rivers in countries with higher flood fatalities and lower flood protection levels. Fang and Jawitz (2019) analyzed the dynamic relationships between settlements and water in the United States and found that people moved closer to major rivers during 1790–1870 but farther from major rivers thereafter.

However, none of these studies considered the differences between human settlements in floodplains and those outside floodplains. In contrast to human settlements in floodplains, the water proximity of human settlements outside floodplains is not directly related to flood exposure and risk. The indicators of human activities in these studies—population density or nighttime light intensity—are also characterized by uncertainties, particularly when they are compared over time. Population distribution is a result of dasymetric map** methods that disaggregate the population numbers from census units to pixels using proxies like built-up land, whose uncertainties originate from both the dasymetric map** methods and the proxy data (Hay et al. 2005; Smith et al. 2019). The nighttime light DNs have difficulties for interannual comparisons (Liu et al. 2012), reducing the usefulness of the data in analyzing water proximity of human settlements. Therefore, we used built-up lands as a direct representation of human activities and employed the built-up land in floodplains (BLF) proxy to investigate the spatiotemporal patterns of human proximity to rivers and their implications for flood risk management.

We chose China as the study area as the proximity of BLFs to rivers may be more complicated and significant than in other countries. China is one of the countries that suffer from devastating floods (Han and Kasperson 2011; Wallemacq and House 2018; Du et al. 2019). From 1990 to 2014, a total of 157 riverine floods occurred in China, annually resulting in 952 fatalities, affecting 69 million people and causing direct economic losses of USD 7.78 billion per year (Guha-Sapair et al. 2016). But China has also developed rapidly during recent decades (Chen et al. 2019). The gross domestic product (GDP) increased 34-fold, from USD 270 billion in 1990 to USD 9,212 billion in 2014 (in 2014 prices); the urban population increased from 3.0×108 in 1990 to 7.5×108 in 2014, more than doubling the urban proportion of the total population from 26.4% to 54.8% (National Bureau of Statistics 2015). Therefore, in the context of China’s severe flood damages and rapid socioeconomic development, the spatial relationship between people and waterbodies may be changing dramatically, which is a key to understanding the flood risk in the world’s second largest economy (Du et al. 2018).

However, the relationship between BLFs and waterbodies in China needs to be clarified. Previous studies mainly focused on the regional-scale impact of urban expansion on waterbodies, such as reclaiming lands from rivers and lakes for croplands, aquaculture ponds and built-up areas (Chen et al. 2013; Fu et al. 2014; **e et al. 2017). Du et al. (2018) assessed the spatiotemporal changes in Chinese BLFs from 1992 to 2015 and Han et al. (2020) analyzed the BLF growth modes in the Yangtze River Economic Belt during 1990–2014. Neither of these studies mentioned the spatial relationship of BLFs with waterbodies. Little is known about the dynamic proximity of BLFs to waterbodies in China. To fill this research gap, this study examined the dynamic relationships between BLFs and waterbodies in China and discussed their implications for flood risk management. Such an assessment is of great significance for understanding the changes in the socio-hydrological system and flood risk in China.

2 Data and Methods

The following subsections present the employed datasets and their validations, followed by the methods to calculate the BLF and its water proximity dynamics.

2.1 Data

We used three datasets to examine the proximity of BLFs to waterbodies: built-up lands, freshwaters and floodplains. Built-up lands were derived from the Global Human Settlement Layer (GHSL) data packages—produced by the European Commission Joint Research Centre as free access data—for the years 1990, 2000 and 2014. The built-up land in the GHSL dataset is defined as building surfaces of human settlements—which include buildings, associated structures and civil works (Pesaresi et al. 2015)—with a resolution of 38 m. In order to evaluate the quality of the GHSL built-up land data in China, a sample of 1,000 random built-up points and 1,000 random non-built-up points was selected for each of the three years 1990, 2000 and 2014. These random points were then interpreted using high-resolution remote sensing images of Google Earth to evaluate the accuracy. The evaluation showed that the overall accuracy of the built-up land data is 91.25% in 1990, 89.45% in 2000 and 91.05% in 2014 (Table 1).

Table 1 Accuracy assessment of built-up land data in China (n=1,000)
Full size table

The freshwater bodies refer to the major lakes and rivers, which exclude seawaters following Kummu et al. (2011). The data were obtained from the China National Basic Geographic Information Center.Footnote 1 The 100-year riverine flood depth map was provided by the Centro Internazionale in Monitoraggio Ambientale (CIMA) foundation (Rudari et al. 2015). The data are produced based on regional runoff frequency analysis and hydrodynamic model simulation and are verified based on historical records. This map has been used for analyzing global flood risk (UNISDR 2015) and studying floodplain urbanization in China (Du et al. 2018).

Floodplains are defined as the maximum extent (flood depth>0 cm) of the 100-year riverine flood depth map, following the flood risk assessment by Shi et al. (2015) and the flood exposure analyses by Jongman et al. (2012), Du et al. (2018) and Fang, Du et al. (2018). The total area of floodplains in China is 1,131.65×103 km2, accounting for 12.11% of China’s total land area. A large proportion of the floodplains (36.10%, or 408.57×103 km2) is located in southeast China. Northeast, northwest and southwest China contain 29.91%, 21.12% and 12.87% of the floodplains, respectively (Du et al. 2018; Fang, Du et al. 2018).

2.2 Methods

The BLF was calculated by overlaying the floodplains and the built-up land datasets using ArcGIS 10.4 (Han et al. 2020). Two major indices were then examined to represent the dynamic spatial relationships between BLFs and waterbodies: (1) water proximity, in terms of distance between built-up lands and waterbodies; and (2) BLF growth within different distances to waterbodies. The analyses were conducted at multiple scales.

2.2.1 Distance between Built-up Lands and Waterbodies

We calculated the water proximity based on BLF patches, which is a basic component of the BLF landscape and refers to a spatial entity of BLF that is qualitatively different from its surrounding environment (that is, non-built-up land) (Turner et al. 2001; Han et al. 2020). The water proximity of a BLF patch is defined as the Euclidean distance from its gravity center to waterbodies (Di). Regarding a spatial unit z (that is, the country, a region, a basin, or a subbasin) that has n BLF patches, its water proximity (Dz) is defined as the weighted average of the water proximities of its BLF patches, which can be expressed as Eq. 1:

$$D_{z} = {{\sum\limits_{n}^{i = 1} {D_{i} A_{i} } } \mathord{\left/ {\vphantom {{\sum\limits_{n}^{i = 1} {D_{i} A_{i} } } {\sum\limits_{n}^{i = 1} {A_{i} } }}} \right. \kern-\nulldelimiterspace} {\sum\limits_{n}^{i = 1} {A_{i} } }}$$
(1)

where Ai refers to the area of the patch i.

2.2.2 Measuring Built-up Land in Floodplains Growth by Different Distances to Waterbodies

We also examined how BLFs were distributed across the water proximity classes. Following Kummu et al. (2011), we divided the water proximity into three classes: low distance (<3 km), moderate distance (3–6 km) and high distance (>6 km). For each proximity zone, we calculated the BLF area (BLFd) and its proportion (rd) to the total BLF, using Eq. 2:

$$r_{d} = \frac{{BLF_{d} }}{{BLF_{TOT} }} \times 100\%$$
(2)

where BLFd is the BLF area in the water proximity class d and BLFTOT is the total BLF area in the spatial unit z.

In addition, the change of BLFs during 1990–2014 was calculated following Eq. 3:

$$CR_{d} = \frac{{BLF_{d} \left( {t_{2} } \right) - BLF_{d} \left( {t_{1} } \right)}}{{BLF_{d} \left( {t_{1} } \right)}} \times 100\%$$
(3)

where BLFd(t1) and BLFd(t2) represent the area of BLF in distance class d in years t1 and t2, respectively.

2.2.3 Multiple-Scale Analysis

We analyzed the spatial and temporal changes of the distance between BLFs and waterbodies at four scales—subbasin, basin, region and country. The study area contains 529 subbasins and 21 basins. These subbasins and basins were first derived from the Food and Agriculture Organization (FAO) and then the delineations were rectified based on datasets of rivers and elevation (Du et al. 2018) that 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)—southwest, northwest, northeast and southeast China (Fig. 1).

Fig. 1
figure 1

Notes: Climate regions: NWC (Northwest China), NEC (Northeast China), SWC (Southwest China), SEC (Southeast China). Basin abbreviations: HLJ (Heilongjiang River), LH (Liaohe River), IM (Inner Mongolia Rivers), XJ (** the BLFs at a safe distance from waters through proper land use planning is vital in this region. Elevation, wet-proofing and other flood-proofing building codes should be encouraged particularly for new buildings.

4.4 Strengths and Limitations of the Study

Previous studies have demonstrated that BLF growth can increase flood exposure and exacerbate flood risk (Du et al. 2019; Han et al. 2020). This study provides new insights in terms of the BLF water proximity and the BLF dynamics in different water proximities for understanding the changes in flood exposure in China. It reveals the divergent water proximity between BLFs and BLOFs, suggesting the importance of distinguishing BLFs and BLOFs in the evaluation of human settlement water proximity and its dynamics. It extends our understanding about Chinese BLFs from a high concentration of build-up lands in floodplains (Du et al. 2018) to a high concentration of the BLFs in the vicinity of waterbodies and from a rapid BLF growth to a concentration of the BLF growth in waterbody surroundings.

The study, however, has several limitations that require further analyses. First, the water data used in this study only cover the large and medium-sized rivers and lakes in China, excluding small rivers because detailed data are unavailable. However, small rivers generally lag behind in flood protection and urban planning both in China (Wang 2018) and globally (Jongman et al. 2012). Therefore, BLF expansion near small rivers may mean a higher risk of flooding.

Second, the shape of rivers and lakes may be instable due to diverse factors including human activities and climate change (**e et al. 2017). The spatiotemporal evolution of the relationships between BLFs and waterbodies can be further revealed given the availability of spatially explicit BLF and waterbody data over long time periods. This kind of study might be more interesting in China than in other countries because of China’s long history of civilization, during which the relationship between settlements and waterbodies has evolved over time.

Third, the reasons for the distance changes between the BLFs and waterbodies are complicated and outside of the scope of this article, which only provides a brief explanation. Future research needs to investigate the underlying mechanisms using more case studies and a mixed quantitative and qualitative approach. However, the study of the relationship between BLFs and waterbodies not only helps to reveal the changes in flood exposure, but also provides an enhanced understanding of the socio-hydrological process in terms of floodplain development and flood risk adaptation.

5 Conclusion

This study examined the spatiotemporal dynamics of BLFs in different water proximities at a national scale in China to provide new insight for understanding the changes in flood exposure and resulted in three major findings. First, the water proximity and its dynamics are divergent between the BLFs and BLOFs. From 1990 to 2014 the BLFs moved towards waterbodies by an average of 169 m, less than the BLOFs (223 m). Second, a large proportion of the Chinese BLFs (62.00%, or 25.88×103 km2) are distributed within 3 km from the waterbodies. The proportion is highest in southwest China (87.38%), making this the region with the closest proximity between the BLFs and waterbodies, at only 2.17 km. Third, the BLFs increased rapidly by 81% or 18.68×103 km2 from 1990 to 2014. The BLF growth decreases with the distance from waterbodies, following the law of negative exponents. A large portion (57.16%, or 10.68×103 km2) of the newly developed BLFs are concentrated in waterbody surroundings (≤3 km). The BLF growth concentrates in waterbody surroundings even in areas where the BLFs have an overall increasing distance from waterbodies.

Both the increase of BLFs and their proximity to waterbodies can increase flood exposure and exacerbate flood risk. This relationship should be verified by further studies and evidence to uncover the underlying mechanisms between the water proximity of BLFs and flood risk. Such an analysis can be extended to an integrated assessment of the BLF impacts on the three components of flood risk—flood exposure, vulnerability and hazard. In urban planning and flood risk management, policymakers should pay attention not only to the volume of BLF growth, but also to its spatial relationship with waterbodies.