Change of stream network connectivity and its impact on flood control

Yu-qin Go *,Yun-ping Liu Xio-hu Lu b,Ho Luo ,Yue Liu

a College of Water Conservancy and Hydropower Engineering,Hohai University,Nanjing 210098,China

b China Water Resources Beifang Investigation,Design & Research Co.,Ltd.,Tianjin 300222,China

c University of Illinois at Urbana-Champaign,Champaign IL 61820,USA

Received 24 March 2020;accepted 26 September 2020 Available online 11 December 2020

Abstract Urbanization can alter the hydrogeomorphology of streams and rivers,change stream network structures,and reduce stream network connectivity,which leads to a decrease in the storage capacity of stream networks and aggravates flood damage.Therefore,investigation of the ways in which stream network connectivity impacts flood storage capacity and flood control in urbanized watersheds can provide significant benefits.This study developed a framework to assess stream network connectivity and its impact on flood control.First,a few connectivity indices were adopted to assess longitudinal stream network connectivity.Afterward,the static and dynamic storage capacities of stream networks were evaluated using storage capacity indices and a one-dimensional hydrodynamic model.Finally,the impact of stream network connectivity change on flood control was assessed by investigating the changes in stream network connectivity and storage capacity.This framework was applied to the Qinhuai River Basin,China,where intensive urbanization has occurred in the last few decades.The results show that stream network storage capacity is affected by stream network connectivity.Increasing stream network connectivity enhances stream network storage capacity.© 2020 Hohai University.Production and hosting by Elsevier B.V.This is an open access article under the CC BY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).

Keywords:Stream network connectivity;Static storage capacity;Dynamic storage capacity;One-dimensional hydrodynamic model

1.Introduction

Stream networks are an important component of a hydrological system and provide a physical foundation for hydrological processes,storage and conveyance of water,aquatic habitats,and human water demands.They play an important role in water quality,the water balance,water use,and aquatic ecosystem integrity(Cabezas et al.,2011;Zhang et al.,2014).However,with land use changes and urbanization,stream networks in urbanized areas are often altered or completely removed,leading to aggravated flood damage in urban areas.Many river channels have been buried,straightened,removed,or blocked during urbanization processes,especially in developing countries.The alteration of natural stream networks often reduces stream network connectivity,which impacts the flood storage capacity of stream networks and likely increases regional flood risks(Shao,2013;Xiu et al.,2019).

Stream connectivity is usually used to describe the spatial connection within a river channel network,and it has gained increasing attention since the early 1990s(Freeman et al.,2010).The commonly used concepts to describe stream connectivity include hydrologic connectivity and stream network connectivity.From the perspective of landscape ecology,Ward(1997)defined riverine connectivity as energy transmission across a river landscape.Based on this concept,Pringle(2010)broadly defined hydrologic connectivity from the perspective of aquatic ecology as the transport capacity of water-based substances,energy,and life forms in water cycle elements.From the perspective of hydrology,Herron and Wilson(2001)stated that hydrologic connectivity mainly refers to the efficiency of runoff movement through a basin.Stream network connectivity is generally defined as the connectivity among river channels,tributaries,lakes,wetlands,and other backwater areas,and it reflects the continuity of water flow and the connectivity of stream networks(YRWRC,2005).The main function of stream network connectivity is to maintain,reshape,or build new water flow connection channels to maintain the stability of water bodies and relevant material circulation(Xia et al.,2012).It plays an important role in flood control,optimal distribution of water resources,and ecological security(Fullerton et al.,2010).Compared with hydrological connectivity,stream network connectivity focuses on the physical and spatial connectivity of stream networks that convey water.

In recent years,stream network connectivity has been investigated from the perspectives of hydrology,geomorphology,landscape ecology,and hydrogeomorphology with a range of methods including graph theory,hydrological modeling,and hydraulic modeling.In terms of the graph theory method,a network model is usually established to evaluate stream network connectivity using the unimpeded water flow and runoff coefficient as the weights of the edges of a graph model(Phillips et al.,2011;Xu et al.,2012;Gao et al.,2018a).Lane et al.(2009)analyzed the sensitivity of structural parameters of a stream network based on a distributed hydrological model.Karim et al.(2012)analyzed hydrological connectivity by constructing a MIKE 21-based hydrodynamic model.Tian et al.(2016)used MIKE 11 to select the optimal connectivity scheme from a series of stream network connectivity schemes.Jencso et al.(2009)quantified the connectivity through time and determined its relationship with runoff dynamics.Martinez-Carreras et al.(2015)explored the ecological connectivity of rivers by studying the migration of diatom along riverbanks.

A change in stream network connectivity directly or indirectly impacts the flood storage capacity of a river system.Stream network connectivity can affect the retention time of flood peaks and flood volumes(Jencso et al.,2009;Phillips et al.,2011).Stream networks with high connectivity not only improve water conveyance,aquatic habitats,and water quality,but also increase the capacity of streams and rivers for flood and drought mitigation(Li et al.,2011;Chen et al.,2019).Decreased flood storage capacity of stream networks is one of the factors that lead to aggravated floods in urban areas.An increase in flood flows and flood frequency is related to the degree of urbanization(Hollis,1975;White and Greer,2006).Urbanization can substantially change the structure and connectivity of stream networks,which usually decreases the diversity and complexity of stream networks,and reduce the area of lakes and other water bodies(Yin et al.,2012).In addition,urbanization can reduce flood plains,alter hydrogeomorphology,and reduce stream network connectivity,resulting in decreased flood storage capacity and increased flood risks(Arnaud-Fassetta,2003;Yates et al.,2003;Wu et al.,2019).

Although stream network connectivity has been investigated from different aspects,most of the assessment methods are not comprehensive,and there have been few studies focusing on systematically quantifying the impact of longitudinal stream network connectivity on flood control(Xu,2012).As for the theory of landscape ecology,the structure of stream networks is usually assessed but the water movement state is ignored.The hydrological modeling method considers water movement in a basin but neglects the connections of the water body and the complexity of connection paths.This study developed a stream network connectivity measurement model.This model is based on water flow resistance and hydrological processes and considers the structural characteristics of stream networks,water transport capacity,and watershed hydrological processes,aiding in the rational investigation of stream network connectivity in different periods.Additionally,the impact of changes in stream network connectivity on storage capacity was assessed,with particular regard to changes in static and dynamic storage capacities of stream networks in different periods and with different flood magnitudes.

2.Methodology

2.1.Stream network connectivity assessment

To examine the impact of stream network connectivity on storage capacity and flood control,it is necessary to rationally assess stream network connectivity.In this study,a stream network connectivity assessment model was developed,based on water flow resistance and hydrologic processes,to consider the features of stream networks,the water diversion capacity of waterways,and hydrologic processes.The water flow resistance and graph theory were used to define waterway connectivity,and the water level differences between stations was used to define hydrological connectivity.Stream network connectivity was calculated as the weighted average of waterway connectivity and hydrological connectivity.Fig.1 provides a schematic diagram of stream network connectivity assessment.

Fig.1.Schematic diagram of stream network connectivity assessment process.

2.1.1.Waterway connectivity

A graph modelG（V，E）was established with the river confluences as vertices and rivers as edges,whereVis the vertex set andEis the edge set.The adjacency matrix(R)demonstrates the degree of water conveyance capacity of a vertex with other vertices.Fig.2 demonstrates the diagram of a sample graph mode with five vertices and the adjacency matrix,whererijrefers to the degree of water conveyance capacity between verticesviandvj,andωekrefers to the weight of edgeek.In the case of no direct edge connection between verticesviandvj,rijequals zero.When multiple edges are connected betweenviandvj,rijis calculated to be the sum of the weight(ω)of each edge.

Different types of waterways have different water flow resistances.The weight of each edge is determined according to water flow resistances.Flow resistance represents friction force and inertial force,and it is closely related to river section,roughness,river length,and other factors(Kang,1993;Xu et al.,2012).For waterways in a trapezoidal cross-section,the flow resistance(RH)is calculated according to the findings of Xu et al.(2012),which is expressed by

wherelis the river length,nis the roughness coefficient,bis the average river bottom width,his the average water depth,andmis the side slope coefficient.Low water flow resistance indicates greater waterway connectivity.Thus,the reciprocal value of water flow resistance is used to represent the edge weight:

When at least one edge exists betweenviandvj,rijis expressed as follows:

wheremijis the number of total edges between verticesviandvj,andis the water flow resistance of edgeq.In the case of no edge betweenviandvjori=j,rijis expressed as follows:

Based on the adjacency matrix and matrix multiplication,the product matrix(R(k))is calculated:

Due to the criss-crossing of river channels,a river is often divided by nodes into multiple river segments with different lengths,when river networks are generalized into a graph model.To address edge weight calculation,the value of edge weight(ω)tends to be large for short rivers,thereby leading to extremely high water conveyance capacity.To tackle this problem,many calculation runs were performed.It was found that with a river length shorter than 150 m,theωvalue was higher than 0.3,leading to a high water conveyance capacity.Therefore,in river network generalization,the river lengths of the river were first screened.Afterward,theωvalues were filtered.Finally,the rivers with lengths shorter than 150 m and ωvalues higher than 0.3 were merged with their neighboring rivers(Fig.3).

Fig.2.Diagram of a stream network model and adjacency matrix.

Fig.3.Diagram of river channel modification.

According to the adjacency matrix of river networks,the water conveyance capacity matrix(F)is calculated:

wherefijis the maximum conveyance capacity betweenviandvj.The water conveyance capacity of vertexvi(Di)and the waterway connectivity(D)are as follows:

2.1.2.Hydrological connectivity

Hydrological connectivity refers to the ability of mass and energy(water,nutrients,sediment,heat,etc.)to flow and be transported through water.It can be quantitatively calculated by such indicators as duration of water runs,changes in flow,changes in water level,and changes in water temperature(Xu,2012).Water level change in plains is consistent with the longitudinal connectivity of stream networks,and water levels at nearby stations in a river basin have a close connection(Deng et al.,2018a,2018b).Therefore,water level difference between stations is used to calculate hydrological connectivity.To avoid a situation of zero water level difference and negative hydrological connectivity with negative water level difference,hydrological connectivity is defined according to the water level difference between stations:

whereChdenotes hydrological connectivity,andΔZis water level difference at adjacent stations.

2.1.3.Stream network connectivity

Given that waterway connectivity and hydrological connectivity are calculated in different units and often not in the same magnitude,they are normalized as follows:

whereD′,Dmin,andDmaxare the normalized,minimum,and maximum waterway connectivity,respectively;andC′h,Chmin,andChmaxare the normalized,minimum,and maximum hydrological connectivity,respectively.Stream network connectivity(E)is calculated as a weighted average of waterway connectivity and hydrological connectivity:

wherew1andw2are the weights for waterway connectivity and hydrological connectivity,respectively.These weights can be estimated according to river properties and the importance of regional flood control and drainage.

2.2.One-dimensional hydraulic model

The rainfall-runoff(RR)module and hydrodynamic(HD)module of the one-dimensional hydraulic model(MIKE 11)were coupled to simulate flood processes and explore the dynamic water level change under different stream network conditions.

2.2.1.RR module of MIKE11

In the RR module,the NAM model is adopted to simulate the rainfall-runoff processes at catchment scales.The RR module can either be applied independently or used to represent one or more contributing catchments that generate lateral inflows to a river network.The NAM model simulates the regional runoff concentration process by calculating the water content of four different and mutually interrelated storages.These storages are snow storage,surface storage,lower or root zone storage,and groundwater storage.The main parameters in the NAM model are maximum water content in surface storage(Umax),maximum water content in root zone storage(Lmax),the overland flow runoff coefficient(CQOF),the time constant for interflow(CKIF),the time constant for overland flow concentration(CK1,2),the root zone threshold value for overland flow(TOF),the root zone threshold value for interflow(TIF),the time constant for baseflow concentration(CKBF),and the root zone threshold value for groundwater recharge(TG).Each parameter value in the NAM model represents the average condition of a basin.All the parameters require continuous adjustment to achieve a better fit between the simulated and observed discharge.The runoff calculated by the NAM model is defined as lateral inflow entering the river network for the HD module.

2.2.2.HD module of MIKE11

The HD module calculates water flow movement in river channels by solving the Saint-Venant equations(Yi et al.,2014;Daneshmand et al.,2019;Lin et al.,2019).The Abbott-Ionescu six-point finite difference format was used to solve the Saint-Venant equations by alternating the calculation points for water level and discharge,and riverbed roughness was adjusted to achieve a better fit between the simulated and observed water level and discharge.

2.3.Stream network storage capacity

Stream network storage capacity includes static and dynamic storage capacities.Static storage capacity refers to the average regulation and storage capacity of a river network at a certain time or within a certain period.Dynamic storage capacity denotes the dynamic change of mitigation capacity during a flood event.

2.3.1.Static stream network storage capacity

Static storage capacity generally includes channel storage capacity(C)and mitigation capacity(CM)(Wang and Liang,2003;Yuan et al.,2005;Shen,2015).Channel storage capacity reflects the water storage capacity of stream networks at a normal water level,and mitigation capacity represents the ability to mitigate floods at the warning water level.To analyze static storage capacity independent of regions,this study used the unit area channel storage capacity(CR)and unit area mitigation capacity(CMR),which denote channel storage capacity and mitigation capacity per unit area,respectively.These values can provide a comparison of the static storage capacity of stream networks in different regions or in different places within the same region.Therefore,the channel storage capacity,mitigation capacity,unit area channel storage capacity,and unit area mitigation capacity were selected as the indices to measure static storage capacity.These four indices are calculated as follows:

whereAis the area of a river cross-section,C0is the total volume at the normal water level,CJis the total volume at the warning water level,andAzis the drainage area of the watershed.

2.3.2.Dynamic stream network storage capacity

Dynamic storage capacity represents the dynamic change of mitigation capacity during a flood event.However,the indices of static stream network storage capacity are not feasible for the characterization of dynamic processes.The process of a stream network's contribution to floods can be examined by comparing peak flood levels and pre-flood level changes.Therefore,the pre-flood mitigation capacity(CMq)and peak mitigation capacity(CMf)were selected to analyze the dynamic storage capacity.The pre-flood mitigation capacity is the total volume of river storage between the warning water level and pre-flood water level.It reflects the capability of stream networks to mitigate floods.The peak mitigation capacity is the total volume of river storage between the peak flood level and the highest historical flood level.It measures the capability of stream networks to mitigate floods after flood peaks appear.These indices are expressed as follows:

whereCqis the total volume at the pre-flood level,Ctis the total volume at the highest historical flood level,andCfis the storage capacity when the peak of a flood event appears.

3.Study area and data processing

3.1.Study area

The Qinhuai River Basin in the southwest of Jiangsu Province of China was selected as the study area(Fig.4).The Qinhuai River Basin covers a total drainage area of 2 631 km2,with an annual average precipitation of 1 047.8 mm and an annual average runoff volume of 695×106m3.The river flows through Lishui District of Nanjing City,Jurong City,and the downtown area of Nanjing City(Gao et al.,2018b).The watershed has been rapidly urbanized in the last several decades.The river originates from the Jurong River in the Baohua Mountain of Jurong City and the Lishui River in the Donglu Mountain area.These two rivers converge at Xibei Village in Nanjing to form the Qinhuai River.In Dongshan Township of Nanjing,the Qinhuai River mainstream is divided into two waterways with their river mouths at Wudingmen Gate and Qinhuaixinhe Gate.Both waterways flow into the Yangtze River in the northwest of the Qinhuai River Basin.

Before 1990,the Qinhuai River Basin was undergoing a slow urbanization process,and most streams and rivers were not altered by urbanization and other human activities.In the period of 1990-2010,the basin experienced rapid urbanization,thereby resulting in the loss of streams and alteration of stream network connectivity.Since 2010,the pace of urbanization in the watershed has been remarkably slowed down with stable land uses.Meanwhile,environmental protection has gained attention.Preservation and protection of streams and rivers has become a critical factor for land resource utilization and other economic development activities.Therefore,the years of 1990,2000,2010,and 2015 were selected as representative years for pre-,in-,mid-late,and posturbanization,respectively.

Fig.4.Schematic diagram of Qinhuai River Basin and hydrometeorogical stations.

3.2.Data

According to their channel width and main functions(Table 1),stream channels were classified into such three categories:large,medium,and small channels.The stream network in 1990 was obtained by digitizing the printed stream network map with corrections.The stream network in 2000,2010,and 2015 were derived using the digital elevation model(DEM)data from Computer Network Information Center of Chinese Academy of Sciences.The Shuttle Radar Topography Mission DEM with a resolution of 90 m×90 m,released in 2003,was adopted to derive the stream network in 2000.The advanced spaceborne thermal emission and reflection radiometer global DEM with a resolution of 30 m×30 m,published in 2009,was used for the 2010 and 2015 stream networks.Hydrologic analysis was conducted using ArcGIS software,and the corrections of the derived stream network maps were performed by comparing the historical images from Google Earth.Fig.5 shows the stream network maps of the Qinhuai River Basin for 1990,2000,2010,and 2015.

Table 1 Stream channel classification.

The required data from the Qinhuai River Basin,including precipitation,discharge,water level,and pan evaporation from 1986 to 2017,were collected from the hydrological yearbooks of the Yangtze River Basin.The precipitation data include the daily precipitation records at five rain gauges:Zhaocun Reservoir,Tianshengqiao Gate,Wudingmen Gate,Qianhancun,and Jurong stations(Fig.4).The Thiessen polygon method was used to transform the observed precipitation data at rain gauges into basin-averaged precipitation data.The discharge data consist of the daily discharge records at the Qinhuaixinhe Gate and Wudingmen Gate stations(Fig.4).The water level data include the daily water level records at the Dongshan and Qianhancun stations(Fig.4).Evaporation data

consisted of the daily pan evaporation data at the Nanjing and Dongshan stations(Fig.4).

Fig.5.Stream network maps of Qinhuai River Basin in 1990,2000,2010,and 2015.

4.Results

4.1.Changes of stream network connectivity

According to the Flood Control Plan of the Qinhuai River Basin and the related hydrological data,the average water depths of large,medium,and small channels are 3.5,2.2,and 1.8 m,respectively.The side slope coefficients of river channels in the Qinhuai River Basin are between 1:1.5 and 1:3,and the soil texture in the basin is mainly clay and silt clay(JWCSDI,2008;HBMWR,2017).Based on the methods for estimating channel side slope coefficient and roughness coefficient(Li,2006)in combination with the average water depth,river characteristics,and soil characteristics of the Qinhuai River Basin,the roughness coefficients for large,medium,and small channels were estimated to be 0.022 5,0.025 0,and 0.027 5,respectively;and the side slope coefficients were defined as 1:3,1:2,and 1:1.5,respectively.The weights for waterway connectivity and hydrological connectivity(w1andw2)were estimated based on the natural and functional attributes of the river channels and the importance of regional flood control and drainage.For the regions with large river level differences,a high proportion of main river channels,and a high demand for flood control and drainage,stream network connectivity is mainly influenced by large channels,and the flood control capacity of large channels is mainly related to the resistance of water flow.In this situation,a largew1value should be adopted to calculate stream network connectivity in these regions.For regions with small river level differences and low flood control and drainage requirements,hydrological processes have more significant impacts on stream network connectivity than on water flow resistance.Therefore,a largew2value should be used to compute stream network connectivity.The Qinhuai River Basin varies remarkably in terms of channel scales.The flood discharge and drainage functions of large channels are more important in the basin.Flood discharge and drainage capacity are closely related to flow resistance.Given that the basin is largely a plain with a low amount of water level variation,w1andw2were respectively set to be 0.7 and 0.3 using the expert survey method combined with relevant expert advice.According to the aforementioned stream network connectivity calculation method,waterway connectivity,hydrological connectivity,and stream network connectivity were computed for 1990,2000,2010,and 2015(Table 2).

Table 2Stream network connectivity in Qinhuai River Basin.

As shown in Table 2,the waterway connectivity decreased by 9.58%from 1990 to 2000,56.04%from 2000 to 2010,and 50.77% from 2010 to 2015.This demonstrates that urbanization altered the physical features of stream networks in the watershed as many medium and small channels were shortened,straightened,or removed(Fig.5).The hydrological connectivity decreased by 0.39%,8.61%,and 6.34% in the three periods,respectively.Compared with the magnitude of waterway connectivity change in the same period,the change in hydrological connectivity was minimal.The stream network connectivity in the Qinhuai River Basin also presented a declining trend,with reduction ratios of 6.34%,38.25%,and 26.10% in the periods of 1990-2000,2000-2010,and 2010-2015,respectively.This indicates that the water level is minimally impacted by urbanization when waterways are still physically connected.

4.2.Static storage capacity of stream network

In this study,the normal water level was defined as the median perennial average daily water level at the Dongshan Station.The normal water level in 1990 was 6.85 m,which was the daily water level median in 1990.The normal water levels in 2000,2010,and 2015 were defined as the median daily water levels in the periods of 1991-1999,2000-2009,and 2010-2015,which were 6.91,7.28,and 7.505 m.The warning flood level was 8.5 m.Table 3 shows the calculated static storage capacities in 2000,2010,and 2015.In the period of 1990-2015,all four indices(channel storage capacity,mitigation capacity,unit area channel storage capacity,and unit area mitigation capacity)decreased remarkably.TheCRvalues decreased by 6.93×103m3/km2from 1990 to 2000,by 3.84×103m3/km2from 2000 to 2010,and by 1.34×103m3/km2from 2010 to 2015.TheCMRvalues showed a reduction of 4.82×103,7.08×103,and 3.14×103m3/km2in these three periods.The results show that the static storage capacity of a stream network is significantly impacted by a change in the stream network.

To further explore the variation of static storage capacity for large,medium,and small channels,CRandCMRwere calculated for these three types of channels.As shown in Fig.6,theCRvalues for large,medium,and small channels were reduced by 27.65%,30.90%,and 48.36%,respectively,in the period of 1990-2015,andCMRdropped 56.92%,74.40%,and 72.01%,respectively.This demonstrates that stream network change has more significant impacts on the static storage capacity of small and medium channels than on that of large channels.In addition,theCRfor large channels declined quickly from 1990 to 2000,whereas for medium and small channels it showed a rapid decrease from 2000 to 2010.The declining trend ofCRfor large channels was similar to the overall trend ofCRin the Qinhuai River Basin.This indicates thatCRis mainly affected by large channels.As forCMR,its decline rate for large channels remained stable in different periods.By contrast,CMRfor medium and small channels declined more quickly from 2000 to 2010,and the trend of medium and small channels was similar to that of overallCMRin the Qinhuai River Basin.This indicates thatCMRis mainly influenced by medium and small channels.

Table 3Static storage capacity of stream network in Qinhuai River Basin in different years.

4.3.Dynamic stream network storage capacity

The MIKE11 model was used to simulate the flood processes and calculate the flood levels under different stream network conditions.The NAM model in MIKE 11 allows for automatic calibration,and it usually requires a long series of observed hydrological and meteorological data for a period of 3-5 years.Therefore,the observed precipitation,discharge,and pan evaporation data from 1988 to 1992 were used for automatic calibration of the NAM model using the stream network condition in 1990.Meanwhile,the NAM model was calibrated for the three periods of 1998-2002,2008-2012,and 2013-2017 using the stream network conditions in 2000,2010,and 2015,respectively.Table 4 demonstrates the calibrated model parameters under different stream network conditions.Fig.7 compares the simulated and observed daily hydrographs.It shows that the coefficient of determination(R2)values were higher than 0.75,and the overall water balance error(EWB)was lower than 15%.This indicates that the NAM model can be used for rainfall-runoff simulation in this study area.

To improve simulation efficiency,the stream network was generalized by only keeping large channels and important medium channels in the MIKE 11 model.This generalizationdid not change the water transport capacity and stream network storage capacity(Wang et al.,1999;Zhou,2013).The hydrodynamic models were constructed using the stream network conditions in 1990,2000,2010,and 2015,and the runoff simulated by the NAM model was used as the lateral inflow for the HD module.The roughness of each simulated river was calibrated within a range from 0.02 to 0.03.Table 5 and Fig.8 demonstrate the performance of MIKE 11 in simulating daily water levels at the Dongshan Station.The simulated water levels agree with the observations,with the correlation coefficient(R)and Nash-Sutcliffe model efficiency coefficient(NSE)being close to 1.This shows that the model can be used for the Qinhuai River Basin.

Fig.6.Channel storage capacities and mitigation capacities for large,medium,and small channels in different periods in Qinhuai River Basin.

Table 4Calibrated parameters of NAM model using different stream network conditions.

Fig.7.Simulated and observed daily hydrographs in periods of 1988-1992,1998-2002,2008-2012,and 2013-2017 in Qinhuai River Basin.

To understand dynamic storage capacity,it is necessary to explore the entire processes of floods of different magnitudes.This study simulated three types of flood processes:large,medium,and small flood events.According to the flood records of the Qinhuai River Basin from 1986 to 2017,flood event No.20150625 was selected as the large flood,flood event No.20120801 as the medium flood,and flood event No.20141123 as the small flood.Fig.9 shows the simulated water levels at the Dongshan Station at different flood levels under various stream network conditions.As the stream network changed,the flood peak water levels for large,medium,and small floods increased by 0.334 m(from 10.303 to 10.637 m),0.162 m(from 9.107 to 9.269 m),and 0.007 m(from 8.580 to 8.587 m),respectively.This indicates that stream network connectivity change has certain impacts on regional flood control.With the decrease of flood magnitude,the effect of stream network connectivity on flood peak water levels gradually decreases.

To further investigate the changes in dynamic storage capacity under different stream network conditions,CMqandCMfwere calculated to characterize the dynamic storage capacity of stream networks.Fig.10 displays the calculated dynamic stream network storage capacity for large,medium,and small floods under four stream network conditions.As the stream network changed,the pre-flood and flood peak levels at different flood severity conditions showed an increasing trend.As a result,bothCMqandCMfdecreased,and the dynamic storage capacity of stream networks tended to continuously decrease.Stream network change had a significant correlation with dynamic storage capacity.Notably,CMqdecreased by37.95%,31.85%,and 33.66%,respectively,for large,medium,and small floods,andCMfdecreased by 69.89%,39.74%,and 34.21%.At different flood severity levels,the magnitudes of the change inCMqwere quite similar,whereas the change inCMfwas quite different.As floods became less severe,the magnitude of the change inCMfgradually decreased.This implies that the change inCMqis less affected by flood severity.The change inCMfis closely related to flood peak level.With the increase of flood severity levels,the effect of stream network change onCMfbecomes more pronounced.

Table 5 Performance of daily river level simulations at Dongshan Station.

5.Discussion

In this study,the change rate of static stream network storage capacity varied significantly in different periods,with higher change rates ofCRbefore 2000 and higher change rates ofCMRafter 2000(Table 3).The total amount of normally available water in the stream network decreased rapidly in the first period and then changed slowly.By contrast,the mitigation storage capacity of the stream network,which measures the capacity of flood regulation,did not present a rapid decrease in the first period but showed a clear change in the second period.These changes,in relation to the trend of stream network connectivity(Table 2),were investigated.It was found thatCMRand the stream network connectivity both showed a rapid declining tendency after 2000.As a result,a significant correlation betweenCMRand stream network connectivity was found.As the stream network connectivity decreased,CMRfor the static stream network storage capacity changed,and both were largely influenced by medium and small channels(Fig.6).As stream network connectivity decreased,CRdecreased as well.However,the change inCRwas primarily influenced by primary channels,and the change in stream network connectivity was primarily influenced by secondary and tertiary channels(Fig.6).Although these trends were slightly different,the overall reduction trend was similar.It can be concluded that changes in stream network connectivity significantly impact static storage capacity.With the decrease of stream network connectivity,static storage capacity decreases.The change rates ofCRandCMRare affected by changes in stream network connectivity.If the decrease in stream network connectivity is primarily due to large channels,CRdeclines faster thanCMR.Conversely,if the decrease in stream network connectivity is mainly caused by medium and small channels,CMRshows in a faster decline rate thanCR.

Fig.8.Simulated and observed daily water levels at Dongshan Station in 1990,2000,2010,and 2015.

Fig.9.Simulated water levels at Dongshan Station for large,medium,and small floods under different stream network conditions.

Under different stream network connectivity conditions,the changes of peak flood levels with different flood severities followed a similar trend to that of stream network connectivity.After 2000,the change rate of stream network connectivity significantly increased(Table 2),as did the increase in peak flood levels(Fig.9).This indicated that the changes in stream network connectivity significantly affected peak flood levels.During the large flood,the dynamic storage capacities ofCMqandCMfsignificantly changed over the period of 2000-2010(Table 6),with the same trend as that of stream network connectivity.Stream network connectivity and dynamic storage capacity had a close correlation.However,with the decrease of flood severity,the magnitude of increase in the peak flood level declined.The change rate of the dynamic storage capacity varied in different periods.The change rates ofCMqandCMfwere higher in the first period than those in the second stage.This pattern indicates that the effect of stream network connectivity on dynamic storage capacity gradually decreased,and the effect was mainly achieved by the changed peak flood levels.As flood severity decreased,the effect of stream network connectivity on peak flood levels became less pronounced,with an alleviated effect on dynamic storage capacity.Under low flood severity conditions,dynamic storage capacity was mainly influenced by the stream network,and the effect of stream network connectivity was not significant.Therefore,with the decrease of stream network connectivity,dynamic storage capacity gradually decreased.However,the extent of the contribution of stream network connectivity to dynamic storage capacity was largely influenced by flood severities.With more severe flooding,the impact of stream network connectivity on dynamic storage capacity was greater.

Fig.10.Dynamic storage capacity for large,medium,and small floods under different stream network conditions.

Table 6Change rates of dynamic stream network storage capacities in different periods.

6.Conclusions

In this study,a stream network connectivity assessment model was developed on the basis of the graph theory,water flow resistance,and water level difference,and the variation of stream network connectivity was analyzed.This model not only reflects the characteristics of channel structure change,but also considers the watershed rainfall-runoff processes.Therefore,this model has a high level of operability and a wide range of applicability.Changes in static storage capacity(average storage capacity and mitigation capacity of the stream network in certain periods)were quantified.Changes in dynamic storage capacity were investigated as well,with a focus on the changes in stream network mitigation and storage capacity during flood events.Furthermore,the effects of stream network connectivity on storage capacity were analyzed from both static and dynamic perspectives,and a one-dimensional hydrodynamic model was used to simulate flood water level processes for quantitative analysis of changes in dynamic storage capacity under different stream network conditions.The main conclusions are as follows:

(1)Increasing urbanization has led to a decrease in stream network connectivity over the past 25 years.Urbanization has resulted in the disappearance of many small and medium river channels with trunked river networks,blocked water circulation,and reduced stream network connectivity.

(2)As stream network connectivity decreased,the static storage capacity of the stream network declined.Static storage capacity was significantly correlated with stream network connectivity.The change rates of the unit area channel storage capacity and unit area mitigation capacity were affected by the change in stream network connectivity.The unit area channel storage capacity was mainly influenced by large channels,and the unit area mitigation capacity was primarily affected by medium and small channels.

(3)The decreased stream network connectivity tended to increase peak flood levels.The change in stream network connectivity significantly increased regional flood control pressure and reduced dynamic storage capacity.There was a certain correlation between the change rate of stream network connectivity and flood severities.With more severe flooding,the impact of stream network connectivity on dynamic storage capacity was greater.

Stream network connectivity significantly correlates with stream network regulation and storage capacity.A decrease in stream network connectivity leads to a reduction in stream network regulation and storage capacity and challenges regional flood control.Therefore,it is necessary to improve the connectivity between large rivers,between large and medium-small rivers,and between rivers and floodplains.Meanwhile,the protection of medium and small rivers should be strengthened to improve the storage capacity of the stream network.In this way,the pressure of urban flood control can be relieved,and flood risk can be reduced.

In this study,only stream network data from 1990,2000,2010,and 2015 were available.Therefore,the statistical correlation analysis between the stream network and storage capacity was not carried out,owing to the low sample size.Future work will focus on utilizing remote sensing techniques and the Google Earth tool to increase the sample size of stream network data.Statistical correlation will be conducted between the stream network and storage capacity.In addition,the present work only considered water level in the hydrological connectivity evaluation.Future work will entail using hydrological models to simulate the changes in flow velocity and discharge for hydrological connectivity evaluation.

Declaration of competing interest

The authors declare no conflicts of interest.