1. Introduction
Urbanization is an integral element of rapid income growth and industrialization throughout the world, and it changes population distribution, production modes, life-styles, and the ecological environment [
1]. In the past three decades, China’s urbanization has been a notable event that has attracted widespread global attention [
2,
3]. As the economy grows, China is experiencing a vigorous rural–urban migration and expansion of cities [
4,
5]. The number of cities reached 657 by the end of 2012—an increase of 431 compared to 1981 [
6]. This urbanization process brought immense urban land expansion. The total urban built-up land area expanded by 513%, from 7438 km
2 to 45,566 km
2, and the annual rate of increase reached 6% between 1981 and 2012 [
7]. The country’s urbanization has brought about remarkable achievements; however, economy-centered urbanization also triggered a series of ‘urban diseases’ [
8,
9,
10,
11,
12], which prevents such urbanization from being sustainable.
Currently, flood and water-logging are occurring with an increasing frequency during the process of rapid urbanization, which is a serious global problem [
13]. When water-logging and floods happen, they cause serious urban flood losses; e.g., serious casualties and property losses, the destruction transport and communication, epidemic disease, the pollution of drinking water and damage to the living environment [
14,
15,
16]. In China, the annual economic loss caused by flooding has been over 100 billion RMB since the 1990s, which is roughly equivalent to 1.5% of the annual average gross national product during the same period [
17,
18]. It was reported that 62% of Chinese cities experienced water-logging and flood disasters during 2008–2010, and 137 cities experienced these more than three times during that period [
19].
Research and practice in the last decades has shown that urbanization has an adverse impact on the urban hydrological processes, such as the expansion of impervious areas accelerating runoff flow velocity and enlarging peak flow and runoff volume, resulting in increased risk of urban flood and water-logging [
20]. In order to mitigate urban water-logging and flood disasters and reduce the flood losses, a large number of researchers have explored the quantitative impacts of urbanization on urban hydrological processes under extreme precipitation using the combination of field observations and hydrodynamic and hydrological models [
21,
22,
23,
24,
25,
26,
27]. In addition, urban flooding risk assessment methods were developed based on various risk assessment theories [
28]; e.g., the risk matrix method [
29].
Currently, the studies of urban flooding have mainly focused on the floods occurring in metropolitan regions. In these regions, it was demonstrated that the dominant cause of urban flood disasters is water-logging induced by impeded drainage systems and extreme rainfall [
21,
22,
23,
24,
25,
26,
27,
28]. However, flood disasters occurring in small mountainous cities are significantly different from those of large cities [
30]. Due to the terrain restrictions of mountainous areas, numerous small cities are constructed in the plain areas near to the river systems; consequently, the river terrace and flood plain are usually occupied by building construction to meet the demand of urban development. This occupied river channel increases the risk of urban flood disasters induced by floodwater out of the river channel when heavy rainstorms occur. In addition, the severe soil erosion in the mountainous region results in the transport of vast amounts of sediment into the river channel [
31], and the cities near to the river channels are therefore seriously threatened not only by the inundation but also the sedimentation from floodwater. The destruction caused by sedimentation is much more severe than water inundation for the urban operating mechanisms and ecological environment; thus, small cities in mountainous areas are more vulnerable than metropolitan cities in the face of rainstorm floods.
Investigating the impact of river channel occupation on urban flood disasters during the urbanization process is critical for making effective plans and implementing measures to mitigate flood disaster and achieve sustainable urban development in rural mountainous areas. In recent years, the intimate connection between the variation of river channels and the watershed floods has been paid increasing attention [
32,
33,
34]. Deng et al. [
33] discovered that urbanization has significantly altered the flood regulation function of river systems in the Taihu Basin, eastern China, which has led to an increased flood risk for urban cities. However, few studies have quantitatively explored the role of river channel occupation on urban flood disasters for small cities in mountainous areas. Long-term satellite observations and efficient flood modeling are beneficial tools for river channel dynamics and the impact assessment of river channel variation on urban flood disasters. In this study, a torrential storm flood event that occurred on 26th July 2017 in Suide County, which is a typical small city on the Loess Plateau, China, was taken as an example. The main objectives of this study are to (1) develop a detection method for river channel dynamics caused by urbanization; (2) assess the quantitative impacts of river channel occupancy on rainstorm flood disasters using two-dimensional flow and sediment hydrodynamic modeling; (3) reveal the underlying reasons for the increased risk of urban flood disasters caused by river channel occupation.
3. Methodology
In this study, an evaluation strategy was proposed to assess the influences of river channel occupation on urban inundation and sediment deposition induced by flood water, as illustrated in
Figure 3. A river channel dynamic detection algorithm was developed using the combination of high-resolution imagery and Landsat normalized difference vegetation index (NDVI) time-series. A two-dimensional water and sediment hydrodynamic model for the river channels of tributaries of the Yellow River was then employed and calibrated based on field observations of flood events. Consequently, scenario simulations were conducted to calculate the contributions of different river channel occupations on urban flood disasters.
3.1. River Channel Dynamic Detection Using NDVI Time-Series And High-Resolution Imagery
In this study, all Landsat TM/ETM+/OLI images acquired from 1990 to 2017 (
Table 1) were used to construct the normalization difference vegetation index (NDVI) time-series dataset. The NDVI is defined as follows [
36]:
where NIR is the surface reflectance of the near-infrared band, and R is the surface reflectance of the red band.
GF-2 satellite images acquired in 2017 and GIS on-screen manual digitization tools were firstly employed to visually delineate the built-up area boundary and river channel boundary in Suide County according to the brightness and texture of different land cover types [
37,
38].
In this region, local farmers have been growing crops in the riparian area as a result of the local terrain restrictions and convenient irrigation; therefore, the land cover type of the original riparian area later occupied for urban construction used to be cropland. Based on this prior acknowledge, a total number of 214 pixels with a conversion from cropland into built-up area in the riparian area were selected as samples based on the SPOT imagery acquired in 2004 and GF-2 imagery acquired in 2017 in order to assess the accuracy of the detection result.
By comparing the histograms of NDVI for the cropland and built-up area samples, it was found that the threshold value of 0.21 was optimal for separating these two classes [
37,
39]. The pixels with an NDVI value above 0.21 were assigned as non-built-up. For each pixel, the NDVI time-series of the pixels within the built-up area interpreted by GF-2 high-resolution images were assessed by referencing the detection rules of the previous study of Wang et al. [
40], as follows.
From the beginning of 2017, if there was vegetation in the ith year, and there were at least two years with vegetation cover from (i − 4)th year to the ith year, the ith year was defined as the final year of the vegetation coverage period, and the (i + 1)th year was defied as the beginning year of river channel occupation for urban construction.
The NDVI time-series dynamics for a sample pixel are shown in
Figure 4. The year of destruction of crops detected by NDVI time-series is 2006, which is consistent with the remarkable visual distinction from vegetation (dark red) to built-up land (bright white) interpreted from multi-temporal Landsat images with pseudo-color (RGB:432). Therefore, the period (2006–2017) with the occupation of the river channel was accurately detected by the proposed algorithm.
3.2. Calculation of Urban Inundated Area and Sedimentation Based on Field Survey Data
In this study, the inundated area caused by this flood event was determined according to field survey data of 71 sites and high-resolution images. Due to the fact that there was a great discrepancy between the sediment depths for different sections within the inundated area, we divided the entire inundated area into five sections: Wuliwan Bridge–Zhangjiabiancun Bridge, Zhangjiabiancun Bridge–Diaoyin Bridge, Diaoyin Bridge–Dali River Bridge, Dali River Bridge–Qianshi Bridge, and Qianshi Bridge–Longfeng Bridge. The co-kriging method [
41] was then used to spatially interpolate the sediment thickness in each section, and the high-accuracy DEM (1:5000) in the urban region was employed as a covariate. It should be noted that the calculated values of the flood-inundated area and sediment deposition were taken as the ‘true value’ for calibrating and validating the Yellow River Conservancy Commission of the Ministry of Water Resources in China (YRCC2D) model.
3.3. Simulation of Urban Inundation and Sedimentation Using the YRCC2D Model
3.3.1. Two-Dimensional Flow and Sediment Hydrodynamic Model
In this study, the water and sediment evolution process of the flood was simulated using the two-dimensional flow and sediment hydrodynamic model developed by the YRCC2D [
42]. In this model, two-dimensional shallow water equations for depth-averaged continuity and momentum conservation are as follows [
42]:
where
h is water depth;
zb is bed elevation; g is the gravitational acceleration;
u,
v are the depth-averaged velocities in
x and
y directions, respectively;
Sox and
Soy are the bottom-slope source terms in x and y directions, respectively;
Sfx and
Sfy are the bottom-slope friction source terms in
x and
y directions, respectively; and the dimensionless coefficient of n is used to represent bed roughness, which is calibrated based on field observations [
42].
Turbulent viscosity
νt is modeled with the zero-equation turbulence model,
where
h is water depth,
Axy is a calibrated parameter with a range of 1–10, frictional velocity
= 6.47 cm/s, and empirical constant
Cs = 1/6 [
43].
The suspended sediment transport is computed using the formula of flow sediment carrying capacity proposed by Yu et al. [
44] in the following manner:
where
μr represents the dynamic viscosity coefficient, which is calculated by the formula proposed by Fei et al. [
45];
Pk and
Puk are the gradation of the suspended and bed sediment, respectively;
and
are settling velocities of uniform and mixed sediment, respectively, which are determined according to the lookup table describing the relationship between the settling velocity and gradation of suspended sediment [
44]; U is flow velocity;
and
are the volume-weights of the sediment and water, respectively,
= 2650 kg/m
3,
= 1000 kg/m
3; R represents hydraulic radius;
α,
β are the empirical constants,
α = 6.1962,
β = 0.0433 [
44]; and
k0 and
m are calibrated parameters.
3.3.2. Numerical Simulation Using YRCC2D Model
According to the spatial domain of the available river channel and urban topographic map, the scope of calculation for this study is defined as follows: Wuliwan bridge is the inlet of the Dali River, the location 3.5 km downstream of the Dingjiagou gauging station is the inlet of the Wuding River, and the location 3.98 km downstream from the junction of the Dali River and Wuding River is the outlet of the Wuding River, as illustrated in
Figure 5. The water and sediment process data measured from 12:00 a.m. on 26 July 2017 to 12:00 a.m. on 27 July 2017 at the Suide and Dingjiagou gauging stations were used as the upper boundary conditions for the YRCC2D model. Due to the fact that there were no measured water and sediment data at the outlet of the scope, the water level–discharge relationship developed based on the terrain condition of the outlet section and the Manning formula [
46] were employed as the lower boundary conditions for the YRCC2D model, as shown in
Figure 5.
The initial gradation distribution of suspended and bed sediment used in the YRCC2D simulation is shown in
Table 3 [
42], and the sediment gradation changes with the variation of erosion and deposition of the riverbed during simulation processing. An initial value of the roughness n in the Equation (5) was set to 0.025 based on the historical studies in this area [
42]. A triangular mesh with a maximum grid side length of 30 m was used for mesh generation in the non-river channel area, whereas a quadrilateral mesh was adopted for mesh generation in the river channels. The maximum grid side lengths in the river width direction for the Dali River and Wuding River are 6 m and 10 m, respectively, and the grid side length in the direction of water flow of both rivers is 30 m. Meanwhile, the density of the grid will be increased in the area near bridges and house construction; consequently, a total of 78,000 grid elements were constructed within the calculation range, as illustrated in
Figure 6.
The key parameters of the YRCC2D model were then calibrated based on the calculated urban inundation area and sediment deposition derived from field survey data. In the simulation process, the water blocking section of the buildings located in the river channel was treated as an impervious section characterized by the measured spatial geometric parameters of the house or bridge. Consequently, two disaster indicators including the inundated area and sedimentation in the urban area were calculated based on the simulation results derived from YRCC2D.
3.4. Analysis of the Impacts of River Channel Occupations on Urban Flood Disasters
According to the dynamics of the river channel derived from the all available Landsat time-series images, the year of 1990 was selected as the base year for analyzing the impacts of river channel occupation on urban flood disasters.
Several scenario simulations [
47] based on the YRCC2D model were designed in order to evaluate the impacts of the river channel width variation and bridge/house construction within the river channel on the urban inundated area and sediment deposition in the flood event on 26 July 2017:
S1: A control simulation based on the calibrated YRCC2D model with observed runoff, sediment concentration, and the boundary of the river channel and the existing bridges/houses in the river channel in 2017 was conducted;
S2: The same forcing data were used as the control simulation S1, except that bridges/houses in the river channel were not taken into account;
S3: The same forcing data were used as the control simulation S1, except the width of river channel was fixed to the condition in 1990, and bridges/houses in the river channel were not taken into account.
The difference between S1 and S2 (S1–S2) was used to estimate the change magnitude
of the simulated urban inundation area and sediment volume caused by the construction of bridges/houses in the river channel:
where
is the simulated urban inundation area or sediment volume in scenario S1, and
is the simulated urban inundation area or sediment volume for scenario S2.
The difference between S2 and S3 (S2–S3) was used to estimate the change magnitude
of the simulated urban inundated area and sediment volume caused by the river channel variation:
where
is the simulated urban inundation area or sediment volume in the scenario S3.
The total change (
) can be obtained by summing
and
, and the contributions of the river channel width variation (
) and bridge/house construction within the river channel (
) can be quantitatively estimated using the following equations:
where
μC and
μB represent the contributions of the river channel width variation and bridge/house construction within the river channel to the urban-inundated area or urban-sediment deposition in relation to the year 1990.
6. Conclusions
Although the expansion of impervious surface areas will enlarge the urban runoff volume, the area of imperious surface is too small to cause serious flood inundation; thus, floodwater from the river channel is the main threat to urban flood disasters for small cities in mountainous areas. Recently, building construction through the occupation of river channels and land reclamation is the main mode of urban expansion in these cities, and the occupation of river channels increases the risk of rainstorm flood disasters in urban regions. A torrential rainstorm flood event that occurred on 26 July 2017 in Suide County, which is a typical town in the Loess Plateau, China, was adopted as an example to explore the impacts of river channel occupancy on urban inundation and sedimentation caused by floodwater.
In this study, an innovative method of detecting urban area and river channel area dynamics using time-series satellite images was proposed, and the contributions of river channel boundary variation and bridge/house construction in the river channel on urban flood disasters were also analyzed based on different scenario simulations with the YRCC2D model. The results showed that due to the urban expansion along the riverbank, the configuration of the river cross-section has changed from a natural “V” shape to a “U” shape or a rectangular shape, and the width of the river channel has narrowed sharply, causing a significant change in the water level–discharge relationship of the river channel cross-section. This has led to a drastic decline in the flood discharge capacity of the river channels. In addition, the geometric characteristics of bridges and the excessively high distribution density have led to a reduction in flood peak discharge, and an increase of the flood water level, which have increased the risk of flood disaster in urban regions. The framework of the proposed method has significant implications for the planning of effective flood control strategies and the establishment of a sustainable development plan for the numerous small cities near the river channel in the mountainous area, especially for the Loess Plateau. The investigation and research findings together with the other proposed sustainable mitigation measures in this study can also be integrated into the disaster management and mitigation portfolio to help reduce the future effects of flooding in similar cities.
Regarding future studies, more sufficient field survey data about urban sediment deposition could be obtained to better reflect the spatial heterogeneity, instead of utilizing the interpolation results. Furthermore, the proposed assessment approach should be further evaluated by comparing similar approaches in different study areas, and the simulation method using a hydrodynamic model can also be extended to incorporate other hydrological models to figure out the different influences of water-logging and floodwater on urban disasters. A quantitative assessment of the sustainability of urban construction can be conducted by the comprehensive consideration of social, environmental, economic, political and human factors.