Contents lists available at ScienceDirect

Catena

journal homepage: www.elsevier.com/locate/catena

A procedure for quantifying runoff response to spatial and temporal changesof impervious surface in Qinhuai River basin of southeastern China

G.D. Biana,b, J.K. Dua,b,⁎, M.M. Songa,b, Y.P. Xua, S.P. Xiea, W.L. Zhenga,b, C.-Y. Xuc,d

a School of Geographic and Oceanographic Sciences, Nanjing University, Nanjing, Chinab Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application, Nanjing, Chinac State Key Laboratory of Water Resources and Hydropower Engineering Science, Wuhan University, Wuhan, Chinad Department of Geosciences, University of Oslo, PO Box 1047, Blindern, N-0316 Oslo, Norway

A R T I C L E I N F O

Keywords:Annual runoffImpervious surfaceStatistical methodHydrological modelQinhuai River basin

A B S T R A C T

Quantitative assessment of the hydrological response to urbanization has been a major concern in hydrology andwater resources management. In this study, a procedure combining different statistical methods and ahydrological model to quantify annual runoff response to spatial and temporal variations of impervious surfaceareas was proposed and applied to the Qinhuai River basin, an urbanized basin located in southeastern China,over the period from 1986 to 2013. The landscape indicators, such as impervious area and number of imperviouspatches were used to measure the spatial configuration of urbanization and quantify the runoff response tourbanization. Impervious area data were derived from the Landsat images using superior ensemble learningmethod of rotation forest. The Mann–Kendall test, Sen's estimator, Pettitt test and double mass curve methodwere applied to examine gradual trend and abrupt changes for hydro-meteorological data series. A hydrologicalmodel based on stepwise regression analysis was built and used to explore the relation among annual runoff andprecipitation, potential evapotranspiration and landscape indicators. The results showed that annual runoff andrunoff coefficient had significant increasing trends and an abrupt change after year 2001 when the watershedimpervious area reached 8.6%. The average annual runoff increased by 60%, of which, urbanization wasresponsible for 59% of the increase, while precipitation changes were responsible for the remaining 1% in thestudy region. The annual runoff response to impervious area showed a nonlinear relationship, and was moresensitive in dry years than wet years. The changes of impervious area were more remarkable in connecting theexisted impervious patches than in developing the new ones beginning in the early 2000s, which increased thewatershed's drainage capacity and resulted in the abrupt change of runoff response. The study demonstrated thatthe proposed procedure was efficient to quantify the runoff responses to urbanization using landscape metrics ofimpervious surface.

1. Introduction

Urbanization, one of the most widespread land conversions byanthropogenic activities (Alberti, 1999), brings a range of physical andbiochemical changes to hydrological system and processes (Fletcheret al., 2013; Jacobson, 2011).

The main characteristics of urbanization are the increase inimpervious area and construction of artificial drainage systems, whichreduce the infiltration into soils and replace natural drainage pathways.This combination increases hydraulic efficiency and decreases thecapacity of moisture storing, leading to increase in surface runoff(Booth, 1991; Hsu et al., 2000; Zhou et al., 2013), and decrease inbaseflow (Klein, 1979; Smakhtin, 2001). The flood characteristics also

change with decreased lag time and increased peak flow from stormevents (Dunne and Leopold, 1978; Stephan and Tsay, 1988; Kang et al.,1998; Beighley et al., 2003; Wang, 2006). However, some studies foundno significant hydrologic response to urbanization and even oppositeresults (Chang, 2003; Ku et al., 1992; Brandes et al., 2005). The diverseresults require further exploring the hydrological response to urbaniza-tion.

Impervious area has become a key indicator of watershed functionand health over the past several decades (Arnold and Gibbons, 1996).Some studies related impervious area with hydrological and ecologicalresponses to urbanization (Brun and Band, 2000; Alberti et al., 2007;Hamdi et al., 2011; Sun et al., 2014). Impervious area can be classifiedinto total impervious area (IA) and effective impervious area (EIA).

http://dx.doi.org/10.1016/j.catena.2017.05.023Received 31 July 2016; Received in revised form 17 January 2017; Accepted 19 May 2017

⁎ Corresponding author at: School of Geographic and Oceanographic Sciences, Nanjing University, Nanjing, China.E-mail address: jkdu@nju.edu.cn (J.K. Du).

Catena 157 (2017) 268–278

0341-8162/ © 2017 Elsevier B.V. All rights reserved.

MARK

Shuster et al. (2005) defined EIA as the area that is hydraulicallyconnected to a drainage system. Previous studies revealed that EAI hasthe most significant effects on watershed hydrology (Booth andJackson, 1997; Lee and Heaney, 2003; Aichele and Andresen, 2013).EIA, at least conceptually, captures the hydrologic significance ofimperviousness (Booth and Jackson, 1997). Unfortunately, connected-ness of impervious area is difficult to quantify, making EIA mappingdifficult. In order to interpret the hydrologic response to EIA, other EIArelated indexes are substituted as surrogates such as number of roadcrossings (Alberti et al., 2007) and landscape metrics of imperviouspatches, etc. (Olivera and DeFee, 2007; Salavati et al., 2015).

One major concern that has been explored is whether there exists athreshold of impervious area above which the hydrological response ofthe basin is changed (Praskievicz and Chang, 2009). Hamdi et al.(2011) detected a change in the frequency of flood events and annualsurface runoff as well as high flow for the Brussels Capital Region whenIA exceeds 35%. Brun and Band (2000) found a threshold of IA (20%),above which the runoff ratio changed more significantly for upperGwynns Falls watershed, Baltimore. Nirupama and Simonovic (2007)demonstrated that approximately 15% IA may be a threshold abovewhich basin exhibits the typical urban hydrology of flashiness inLondon, Ontario. Olivera and DeFee (2007) found annual runoff depthsand peak flows have increased by 159% and 146% respectively whenthe watershed reached a 10% IA in the Whiteoak Bayou watershed inTexas. Other studies found 3 to 8% IA as the threshold (Yeo andGuldmann, 2006; Booth and Jackson, 1997; Yang et al., 2010).However, Chang (2003) showed no significant increase (less than 2%)in annual runoff in a low-density suburban watershed in southeasternPennsylvania. Chang (2007) also found no significant changes in peakrunoff ratio and annual runoff ratio for an urban watershed and a mixedland-use watershed in the Portland Metropolitan Area of Oregon for theperiod from 1951 to 2000. These studies suggested that the threshold ofimpervious area varied significantly and are dependent on the timescale, the indices, watershed characteristics and climate conditions(Chang, 2007; Beighley and Moglen, 2002; Jones, 1997).

There are mainly two approaches adopted in the study of hydro-logical response to urbanization. One approach is to apply a hydro-logical model and examine changes of runoff characteristics withdifferent urbanization scenarios (e.g., Bhaduri et al., 2001; Tu, 2009;Zhou et al., 2013; Zhu and Li, 2014; Yan et al., 2016; Yang et al., 2016).This approach enables one to distinguish hydrological effects ofurbanization by controlling other variables, but it is subject to modelingaccuracy and uncertainty (Choi et al., 2016). Another approach is to usestatistical methods to examine long-term trends of runoff characteristicsbetween different periods in a catchment or catchments with differentdegrees of urbanization (e.g., Huo et al., 2008; Rougé and Cai, 2014;Choi et al., 2016). The nonparametric Mann–Kendall test for gradualtrend analysis (Mann, 1945; Kendall, 1975) and Pettitt test for abruptchange detection (Pettitt, 1979) are frequently adopted. This approachis more effective in determining whether hydro-climatic variableschanged significantly over time, than in attributing the changes toparticular causes (Choi et al., 2016). The approach also requires longtime series of hydro-climatic data. In this study, to take advantages ofboth approaches, a procedure is proposed to explore the annual runoffresponse to impervious surface area changes in the Qinhuai River basin,an urbanized basin located in southeast China with monsoon climate(humid subtropical climate). In addition, most studies used imperviousarea as a metric of urbanization to explore the relation betweenhydrological response and urbanization (Bhaduri et al., 2001;DeWalle et al., 2000; Zhu and Li, 2014; Sun et al., 2014). In this study,in addition to impervious area, the number of impervious patches wasalso used to capture the hydrological conveyances of urbanization andinterpret the hydrological response.

The specific objectives of the study are: (a) to test the possibility ofcharacterizing the spatiotemporal changes of impervious surface byusing landscape metrics; (b) to determine whether there exists a

threshold of IA below which the annual runoff response of the basinis unchanged; (c) to explain runoff changes based on landscape metricsof impervious areas if a threshold exists; and (d) to explore therelationship between annual runoff changes and percent imperviousarea under various hydrological conditions. The objectives wereachieved by a procedure consisting of the following steps: changedetection methods (Mann–Kendall test, Sen's estimator, Pettitt test,double mass curve method) were used to examine the temporal changesin the long-term hydro-climatic data series; an annual rainfall-runoffmodel was built by using stepwise regression method to simulate runoffprocess and separate contributions of urbanization and climate varia-bility to runoff response; and the relation between runoff response andurbanization degree (landscape metrics) under varied climate condi-tions was investigated using the annual rainfall-runoff model.

2. Materials and methods

2.1. Study area and data

Qinhuai River, located in the southwest of Jiangsu province, is oneof the tributaries of lower Yangtze River. The basin area is 2631 km2,ranging from 118°39′ to 119°19′ E, and 31°34′ to 32°10′ N. As a typicalwatershed in the Yangtze delta plain, the Qinhuai River basin aboundswith paddy rice and freshwater fish and supports advanced industrialeconomies. With marked advancement of urbanization since thebeginning of the 21st century, significant land use changes haveoccurred in the Qinhuai River basin. It is hypothesized that thehydrologic cycle, even the whole ecological system has changedaccordingly. Therefore, it is of profound significance to quantify theimpervious area changes and hydrological responses of land usepattern.

The annual mean air temperature is 15.4 °C, average annualprecipitation is 1116 mm (1986–2013), and the rainy season is fromApril to October. The measured annual runoff, which mainly occursduring June to August, is about 430 mm. (Hao et al., 2015).

There are 5 types of land use in Qinhuai River basin: paddy land,dry farmland, woodland, impervious surface and water. Paddy land anddry farmland are the dominant land use types.

The basin location, elevation, network, and the distribution of thetwo hydrological stations and seven rain- gauge stations are shown inFig. 1.

The 28-year (1986–2013) daily rainfall data for the seven rain-gauge stations and the daily discharge data of the Inner Qinhuai andWudingmen stations were obtained from the local hydrological bureau.Both of the two runoff gauging stations are located at the outlets of theresearch river. Potential evapotranspiration (PET) estimates wereobtained from Hao et al. (2015) for the period of 1986–2013. Sevencloud-free Landsat images were chosen as the major sources forextracting time series of impervious areas, which were used to estimatethe urbanization level and quantify the changes of land use pattern inthe study area (Table 1).

2.2. The extraction of impervious surface

The preprocessing procedure of remote sensing images includedatmospheric correction and band synthesis. After preprocessing,Rotation Forest, a new classifier ensemble system proposed byRodriguez et al. (2006) was used to extract the impervious surface. Itis based on a decision tree and combines the accuracy of individualclassifiers and the diversity between different classifiers. The bootstrapsamples of the original training set are used to construct a new trainingset. Then the feature set is split to K random subsets and PrincipalComponent Analysis (PCA) is performed on them. As a result, bycombining all the principle components, the feature subset is rebuilt.For the reason that a completely different decision tree can be builtthrough a small rotation of axes, the diversity between the base

G.D. Bian et al. Catena 157 (2017) 268–278

269

classifiers is guaranteed. At last, major vote rule is used to combine theoutputs of all decision trees.

The specific implementation of rotation forest is described asfollows:

Let X be the data set described by n features and Y be thecorresponding class labels, where X contains N training samples in aN × n matrix. Let ω be the set of class values {ω1,···,ωc} where Y takesvalue from ω. The classifiers in the ensemble are denoted by D1, D2,···,DL and the feature set is denoted by Z. In order to construct the trainingset for an individual classifier Di, the following steps are carried out:

Step1: Split Z randomly into K disjointed subsets. Consequently,every subset contains M= n/K features.Step2: The j-th subset of features for training Di is denoted by Zij andthe dataset X for the features in Zij is denoted by Xij. For each subset,select a nonempty subset of classes randomly from Xij. Then, abootstrap of objects is drawn and 75% of the dataset is applied toform a new training set, which is denoted by X′

ij. Afterwards, PCA isrun on X′ij to generate the coefficients a(1)i,j, a(2) i,j, ···, a(Mj

)i,j in a

matrix Ci,j, each of size M× 1.Step3: Organize the sparse rotation matrix Ri with the coefficientsobtained in Ci,j.

⎡

⎣

⎢⎢⎢⎢⎢

⎤

⎦

⎥⎥⎥⎥⎥R

a a

a a a

a a a

=

, ⋯ [0] ⋯ [0]

[0] , , ⋯ ⋯ [0]⋮ ⋮ ⋱ ⋮

[0] [0] ⋯ , , ⋯

i

i iM

i i iM

i K i i KM

,1(1)

,1( )

,2(1)

,2(2)

,2( )

,(1)

,K(2)

,( )k

1

2

According to the original feature sequence, the columns of Ri arerearranged. Assume that Ra

i is the rearranged matrix, then let xRai be the

transformed training set for Di. Note that all the classifiers can betrained in parallel in this way.

The classification procedure is as follows:Given a text sample x, the probability produced by Di is denoted by

di,j (xRac). Then the confidence uj(x) is calculated for class j by the

following average combination method:

∑u xL

d xR j c= 1 , = 1, …,ji

L

i j ia

=1,

According to the calculated confidence, the test sample x will beassigned to the class with the largest confidence.

2.3. Accuracy evaluation of extracted IA

One thousand sample points were randomly generated in the studyarea for each impervious surface map by ArcGIS, and the information ofthe points in classification results was checked with the reference high-resolution images of Google Earth (2006, 2009, 2013) or originalimages (before 2006). The overall accuracy (Cohen, 1960), the Kappacoefficient (Cohen, 1968) and the F score of impervious surface typewere calculated to evaluate the accuracy of the method. Different fromthe overall accuracy and Kappa coefficient, F score only considers theproblem of missing and wrong identification of one specific class, whichmakes it one of the most useful metrics to evaluate the performance ofimpervious surface extraction. The F score is computed in the followingway:

Fig. 1. Location, elevation, distribution of rainfall and runoff gauging stations, and streams of the study area.

Table 1Information of the remote sensing images.

Year Sensor type Collection time Resolution(m)

1988 Landsat5 TM 1988-07-05 301994 Landsat5 TM 1994-07-06 302001 Landsat7 ETM+ 2001-07-17 302003 Landsat5 TM 2003-02-05 302006 Landsat5 TM 2006-05-20 302009 Landsat7 ETM+ 2009-05-04 302013 Landsat8 OLI 2013-08-11 30

G.D. Bian et al. Catena 157 (2017) 268–278

270

F TPTP FP FN

= 22 + +

where TP (true positive) is the number of impervious surface pixelswhich are classified correctly, FP (false positive) is the number of otherland use pixels which are classified as impervious surface and FN (falsenegative) is the number of impervious pixels which are classified asother land use.

2.4. Landscape metrics selection and estimation

Landscape metrics are measures of the spatial pattern of thedifferent types of the landscape, which can be used to denote thespatial pattern of the impervious surface. There are many landscapemetrics such as total patch area, patch number, mean patch area, patchdensity and edge density (Salavati et al., 2015). Here, two metrics: (1)impervious area (IA) - sum of the areas of the impervious patches, and(2) number of patches (NP) - count of the impervious patches wereselected to reflect amounts and connectivity of the impervious areas.They are expected to provide valuable information for quantifying andexplaining the watershed's hydrologic response to urbanization. Oncethe impervious surface maps were obtained, landscape metrics of eachof them were calculated with Fragstats software (McGarigal and Marks,1994).

2.5. Change detection

Three statistical methods were used to examine the long-termchanges of hydrological and climatic variables: (1) the Mann–Kendalltest was used for gradual trend analysis; (2) Sen's nonparametricmethod was applied to estimate the change per year for an existingtrend (Sen, 1968); and (3) Pettitt test was employed for abrupt changedetection. For the sake of completeness, the version of the statisticalmethods used in this study is briefly introduced herein.

The nonparametric Mann–Kendall (Mann, 1945; Kendall, 1975)(MK) statistical test method has been extensively used for trenddetection in time series due to its robustness and comparable powerto parametric competitors (Yue et al., 2002; Khadr, 2016; Wei et al.,2016). The MK test statistic S and standardized statistic Z can beestimated based on time series, and the significance level can then becomputed for the statistic Z (Zhang et al., 2008). The trend slopes β canbe estimated by a nonparametric index developed by Sen (1968). Anegative value of β indicates a downward trend, whereas a positivevalue represents an upward trend.

However, the MK test is not effective when autocorrelation exists, apreliminary analysis of the autocorrelation of the time series must beconducted before applying the MK test (Zhang et al., 2008). Ifsignificant autocorrelation is detected in the data series, the trend freepre-whitening (MK-TFPW) method proposed by Yue et al. (2002), willneed to be adopted to remove the effect of serial correlation. The MKtest and Sen's estimation together are also called MK-Sen test.

Change point was detected by using the Pettitt test method in thisstudy, which assumes that a sequence of random variables X1, X 2, …,XT, has a change-point at τ; Xt (t= 1,2, …, τ) has a commondistribution function F1(X) and Xt (t= τ+ 1, …, T) has a commondistribution function F2(X), and F1(X) ≠ F2(X)) (Pettitt, 1979). The nullhypothesis of the test H0: no change or τ = T is tested against thealternative hypothesis Ha: change or 1 < τ < T using the statistic Kt:

K = max |U |t T

t Tt1≤ ≤

,

X X

t T

U = U + ∑ sgn( − ),

for = 2,…,

t T t T

T

t, −1,j=1

j

K = max Utt T

t T+

1≤ <,

K = −min Utt T

t T−

1≤ <,

where Kt+ is for downward shift and Kt

− is for upward shift. Thesignificant level is determined by

p T T= 2 exp {−6K + }t2 3 2

The null hypothesis of no change point is rejected when p value issmaller than a specific significance level. Serial correlation affects theeffectiveness of the Pettitt test, and when serial correlation is signifi-cant, a pre-whitening process using MK-TFPW method is needed beforeapplying the Pettitt text.

A double mass curve (DMC) is a plot of the accumulated values ofone variable against another for same period of time. If two variablesare proportional, the graph of the accumulation of them will plot as astraight line, and the slope of this line will represent the relationshipbetween the two variables (Searcy and Hardison, 1960). Any change inslope can provide important information about the time at which achange of relationship occurred. DMC was applied to identify thechanges of hydrological responses to human activities (Huo et al., 2008;Zhang and Lu, 2009).

2.6. Hydrological model building and attribution analysis

Once the hydrologic change point was detected, the total runoffrecord was divided into two periods before and after the changing pointof time. The first period (baseline period) represents the baseline whenno significant climate and/or land cover changes occurred, and thesecond period (change period) represents changed runoff associatedwith climate or/and land cover changes.

A hydrological model relating annual runoff depth to precipitation,PET and spatial metrics of impervious area for baseline and changeperiods was built by using stepwise multiple regression analysis. Thestandard error and efficiency coefficient (Nash and Sutcliffe, 1970)were used to evaluate the accuracy of the built hydrological model. Thestandard error is defined as

nE =

∑ (Q − Q )− 1

in

mi si=12

and the efficiency coefficient as

NS = 1 −∑ (Q − Q )∑ (Q − Q )

in

mi si

in

mi m

=12

=12

where Qmi is the measured annual runoff depth in the ith year, Qsi is thesimulated annual runoff depth in the ith year, Qm is the average annualrunoff of all years, and n is the number of years in the analysis period.The closer the standard error is to zero and the efficiency coefficient toone, the better the estimation is.

The observed runoff change (ΔQ) in the watershed could beassumed to be the result of the sum of change by climate (i.e.,precipitation and PET) (ΔQc) and change by land cover (i.e., urbaniza-tion) (ΔQl):

ΔQ = ΔQ + ΔQc l

Then, the contribution of land cover change to the change in runoffcan be estimated in percent as

ΔQ (%) = (ΔQ − ΔQ ) ΔQ × 100l c

ΔQc is the difference between the observed mean annual runoff forthe baseline period and the calculated runoff for the change period byapplying the hydrological model of the baseline period with inputs ofprecipitation and PET for the change period.

G.D. Bian et al. Catena 157 (2017) 268–278

271

2.7. The relationship between annual runoff depth change and IA for typicalhydrological years

An analysis was conducted for different hydrological years todetermine the relationship between annual runoff change and IA.Three typical hydrological years (dry year with annual precipitationexceedance probability of 90%, normal year with annual precipitationexceedance probability of 50%, and wet year with annual precipitationexceedance probability of 10%) were selected, and regression equationamong annual runoff, precipitation and IA was established in order toestimate annual runoff depth change using IA for typical hydrologicalyears.

3. Results

3.1. The spatial and temporal changes of impervious surface

Table 2 shows the results of accuracy evaluation of derivedimpervious surface maps. Note that the overall accuracy of seven-yearclassification maps exceeded 90%, and the Kappa coefficient attainedvalues over 0.83. In addition, the F scores of impervious surface weremore than 86%. The results met the needs of performing impervioussurface analysis.

Fig. 2 shows the distributions of IA in seven years respectively,whereas the evolutions of IA and NP over time are shown in Fig. 3. Itcan be seen from Figs. 2 and 3 that IA in the Qinhuai River basinincreased from 3.8% in 1986 to 17.5% in 2013, with a net increase of361% over 20 years. The impervious surface coverage in the QinhuaiRiver basin has increased for about four times during the past 28 years.

Note that the NP reached the peak in 2003, and then keptdecreasing from 2003 to 2013. NP is subject to the relative positionbetween newly increased patches and existing patches. If the newlyincreased impervious patches and the existing impervious area patchesare connected, the number of impervious surface patches wouldincrease correspondingly. If the impervious surface extends on thebasis of existing patches, the number of patches would remain constantor decreases with the increase of urban area (Olivera and DeFee, 2007).The change of NP over time suggests that urban land use fragmentationkept rising from 1988 to 2003. After 2003, the impervious patchesexpand together while the impervious area continues to rise. Urbanextension during this period showed a typical “fill-in” growth, whichmade the existing patches more regular and connective. Overall, thechange of NP indicated that the impervious patches, which weredispersedly distributed before the twentieth century, showed anincreasing integrity trend in recent years.

3.2. The trends of precipitation, runoff, runoff coefficient and PET

No significant autocorrelation was found in any of the time series,hence no pre-whitening was required. The results of the MK-Sen test forgradual trends of annual precipitation, PET, runoff, and runoff coeffi-cient series in the basin are shown in Table 3. There was no significantincreasing trend detected at the 0.05 significance level for precipitation.For annual runoff, runoff coefficient, and PET, significant increasingtrends were detected.

The results of the Pettitt test for abrupt changes of annualprecipitation, PET, runoff, and runoff coefficient are shown in

Table 4. At the significance level of 0.05, there was no significantabrupt change of precipitation happening in the basin during the studyperiod. Annual runoff and runoff coefficient had significant abruptupward changes at the significance level of 0.05 and 0.01, thesechanges all appeared in 2001. PET had significant abrupt upwardchanges in 2003 at the significance level of 0.01.

A clear “break point” of total annual runoff around 2001 wasidentified by the DMC analysis (Fig. 4) after removing the data pair ofthe year 1991, when a huge flooding event occurred in the basin with areturn period much longer than the period of the used data series. Theslopes of the regression lines between accumulated precipitation andstream flow increased from 0.26 to 0.49, indicating an increase of 88%in runoff coefficient (Q/P) from 0.26 to 0.49.

The abrupt change point detected by DMC is consistent with theresults by the Pettitt test. The gradual trends of annual runoff, runoffcoefficient and PET before and after their change points were furtherexamined using the Sen-MK test. The results (Table 5) showed that bothannual runoff and runoff coefficient had insignificant decreasing trendsbefore their change points, and had insignificant increasing trends afterthese change points. PET had insignificant increasing trends before andafter the change point. Those results showed that annual runoff, runoffcoefficient and PET all changed insignificantly before and after theirchange points, but changed significantly in the whole study period.Fig. 5 shows the change characteristics of precipitation, PET, runoff,and runoff coefficient during 1986–2013 as a whole.

3.3. The hydrological model based on annual runoff, precipitation andlandscape metrics

Since there existed an abrupt runoff change point, the relation ofannual runoff with precipitation, PET, and landscape matrices would bebetter modeled by two equations for two segments before and after thechange point.

A multiple stepwise regression analysis was conducted for twoperiods (baseline period from 1986 to 2001 and change period from2002 to 2013), and two hydrological models based on annual runoff,precipitation, PET and landscape metrics with a 0.05 level of signifi-cance were obtained after year 1991 was removed:

R P= −515.65 + 0.76 ×i i (1)

R P I= −699.98 + 0.82 × + 25.13 ×i i i (2)

where Ri is the annual runoff (mm) in year i, Pi is the annualprecipitation in year i (mm), Ii is the percent IA in year i, estimatedby linear interpolation method for each year between 1986 and 2013based on data shown in Fig. 3.

Eqs. (1) and (2) were for the baseline period and change period,respectively. The calculated standard error was 50.6 mm and theefficiency coefficient was 0.94 by using two models for the wholeperiod, suggesting a very good representation of the two models for therelationship among the annual runoff, the precipitation, and percent IA.The observed and simulated annual runoff depths were plotted in Fig. 6.

Note that PET and spatial IA and NP were all removed from Eq. (1)due to their insignificance at 0.05 level, suggesting less impact of thosefactors on annual runoff during the baseline period. On the contrary,the IA was included in Eq. (2), meaning that the IA affected theestimated runoff during the change period. This result indicates that,after the early 2000s, the rainfall-runoff relation in the watershedchanged and had to adjust to the new hydrological conditions. The newconditions could be revealed from changes of IA and NP: in a watershedthat experiences urbanization, IA is considered to increase steadily astime goes on, while NP is expected to increase up to a top value andthen decrease. The change of NP indicates that impervious patches aremost dispersed at peak value of NP, and then become integratedgradually. The break point in the annual runoff series in years2001–2002 corresponds to the early 2000s, after which urbanization

Table 2Results of accuracy assessment of impervious surface derivation.

1988 1994 2001 2003 2006 2009 2013

Overall accuracy 0.93 0.90 0.91 0.92 0.93 0.93 0.90Kappa coefficient 0.86 0.83 0.85 0.88 0.87 0.85 0.88F score 0.90 0.89 0.9 0.91 0.92 0.86 0.89

G.D. Bian et al. Catena 157 (2017) 268–278

272

Fig. 2. Maps of impervious surface at different times.

Fig. 3. Changes of IA and NP over time.

G.D. Bian et al. Catena 157 (2017) 268–278

273

connected more existing impervious patches, rather than created newones, leading to more EIA increase than total impervious area. This newhydrological condition (more connected impervious patches) mademore precipitation convert to runoff, and resulted in the change ofthe watershed response.

3.4. The attributions of precipitation and IA to annual runoff

Eq. (1) represents the runoff response to precipitation without beingaffected by urbanization, when the precipitations in 2002–2013 wereplugged into Eq. (1), the output would be the runoff if the watershedhad not undergone urbanization, hence this runoff was affected only byprecipitation. In other words, the difference between observed andsimulated runoff using Eq. (1) in the change period can be considered asthe change caused by urbanization. On average, the annual runoffdepth in the baseline period was 335 mm, and in the change period was536 mm, the simulated annual runoff depth by Eq. (1) in the changeperiod was 339 mm. The results reveal that the annual runoff depth hasincreased by 60%, of which, precipitation has increased the annualrunoff depth by 1%, and urbanization has increased the annual runoffdepth by 59%. The relative contribution of urbanization to the

increased annual runoff depth is 98%, while the precipitation is 2%.The MK test revealed that the annual runoff depth series in

2002–2013, which showed an insignificant increasing trend as a whole,caused by precipitation indicated a decreasing trend, while caused byurbanization showed a significant increasing trend at the significancelevel of 0.01.

3.5. The relationship between annual runoff change and IA for typicalhydrological years

Three typical hydrological years (dry year, normal year, and wetyear) were selected to be 1992, 1990 and 2003 with annual precipita-tions of 695, 1055 and 1456 mm, respectively. The annual runoff depthseries for each typical hydrological year were calculated by putting IAfrom 1986 to 2013 separately, after plugging annual precipitation ofthe hydrological year into Eqs. (1) and (2). The annual runoff changesof typical hydrological years to IA are shown in Fig. 7. Note that theannual runoff response to impervious area does not show a singlestraight line, when IA is less than 8.6% (~2001), there was no increasein annual runoff. When the percent impervious area reaches 8.6%, therewas an abrupt increase in annual runoff and beyond that a linearrelationship existed between annual runoff changes and IA for alltypical years with steepest slope for the dry year and the gentlest slopefor the wet year. This means that dry years are more sensitive tourbanization than wet years.

4. Discussion

Landscape metrics are able to characterize how urbanizationsprawls and how the impervious patches are connected or concentratedin space, but only few studies used them to assess hydrological responseto urbanization. Olivera and DeFee (2007) used developed patches

Table 3Results of MK-Sen test for annual runoff, precipitation, runoff coefficient and potentialevapotranspiration series. * indicates trend test passes the Sen-MK test at the significancelevel of 0.05, and ** indicates trend test passes the Sen-MK test at the significance level of0.01.

Time series Period n MK-Sen trend test

β S Z Trend

Annual runoff 1986–2013 28 13.6 120 2.35⁎ UpwardAnnual precipitation 1986–2013 28 0.4 4 0.06 UpwardRunoff coefficient 1986–2013 28 0.01 170 3.34⁎⁎ UpwardPET 1986–2013 28 5.23 166 3.26⁎⁎ Upward

Table 4Results of Pettitt test for annual runoff, precipitation, runoff coefficient and PET series.

Time series n Pettitt test for change point

K t Trend p

Annual runoff 28 122 2001 Upward 0.02Annual precipitation 28 38 0.68Runoff coefficient 28 148 2001 Upward 0.003PET 28 136 2003 Upward 0.008

Fig. 4. The relationship between accumulated annual precipitation and runoff for the period of 1986–2013.

Table 5Results of MK-Sen test for annual runoff, runoff coefficient and PET series before and aftertheir change points.

Time series Period n MK-Sen trend test

β S Z Trend

Annual runoff 1986–2001 16 −5.26 −16 0.72 Downward2002–2013 12 3.62 2 0.14 Upward

Runoff coefficient 1986–2001 16 −0.001 −18 0.77 Downward2002–2013 12 0.016 28 1.92 Upward

PET 1986–2001 16 1.85 4 0.14 Upward2002–2013 12 8.4 14 0.89 Upward

G.D. Bian et al. Catena 157 (2017) 268–278

274

areas, number of patches, slope of the number of patches, and edgelength to analyze urbanization on annual runoff depths and peak flowsin the Whiteoak Bayou watershed in Texas. Salavati et al. (2015)related IA, NP, mean patch area, patch density and edge density to theimpact of urbanization on runoff efficiency on 43 catchments in theUnited States. This study showed that IA and NP are effective indicatorsto quantitatively describe urban landscape pattern and represent thehydrological conditions of urbanization in a watershed even thoughthere are many landscape metrics. IA could be used as a covariate toquantify the hydrological response to urbanization, while NP could beused to explain hydrological threshold response to urbanization. It isexpected that the landscape metrics would be widely used in urbanhydrology research with the possibility of extracting impervious surface

fast and accurately from remote sensing images.It should be noted that the results of trend analysis are dependent on

the length of the time series. Previous researches have shown thatclimate variables as well as river flows are fluctuating at multi-decadaltime scales (Willems, 2013; Chen and Grasby, 2009; Taye and Willems,2013; Hannaford et al., 2013). When apparent short-term trends aredetected, these trends may be part of longer-term fluctuations driven bylarge scale climate oscillation. However, the significant increasing trendof annual runoff of the study area during 1986–2013 was not caused bythe increasing flank of a climate oscillation wave, but mainly bychanges in land use/land cover (LULC) dominated by increase of IA.The annual runoff is controlled mainly by precipitation and evapo-transpiration (ET). Since annual precipitation during the period did not

Fig. 5. Changes of runoff (R), precipitation (P), PET and runoff coefficient (Rc) during 1986–2013.

Fig. 6. Comparison of observed and simulated annual runoff (R).

G.D. Bian et al. Catena 157 (2017) 268–278

275

show any significant trend, the increase of runoff is attributed to thedecrease of ET. ET is generally controlled by PET, precipitation, andland surface patterns. The annual PET significantly increased, whichmay cause increase in ET, but the stepwise regression analysis for bothbaseline and change periods did not show PET had correlation withannual runoff at the significance level of 0.05. Therefore, the decreaseof ET and increase of runoff can only be attributed to the changes inLULC, which cause the increase of impervious area, especially theincrease of connectivity of impervious area, decrease of retentioncapacity, less water infiltrates into soil for ET and more precipitationconverts into runoff. The relative contribution of urbanization to theincreased annual runoff depth is 98%, while the precipitation is 2%,and the increase of annual runoff caused by urbanization during2002–2013 is significant at the 0.01 level, supporting our conclusion.The increasing trend of annual runoff coefficient by MK test andexistence of break point detected by the DMC analysis indicated thechange in rainfall-runoff relationship of the watershed is mainly causedby land surface changes, which is also supporting our conclusion. Duet al. (2012) and Hao et al. (2015) also reported the increase of annualrunoff in different periods (1988–2009 and 1986–2013) in the basinwas mainly caused by urbanization.

Our results of the non-single linear relationship between annualrunoff and IA are consistent with literatures in Brun and Band (2000)and Olivera and DeFee (2007)). Brun and Band (2000) found two-dimensional form of a logistic expression could describe the relation-ship better between simulated runoff ratio and percent soil saturationand percent impervious cover for upper Gwynns Falls watershed,Baltimore, MD, USA, based on the simulated results of HydrologicSimulation Program Fortran (HSPF). Olivera and DeFee (2007) re-vealed that annual runoff depths had linear relationship with IA in theWhite Oak Bayou watershed in Texas when the watershed reached a10% IA, while there was no effect of IA on annual runoff when it wasless than 10%. But more studies showed linear relationship betweenrunoff and IA (Bhaduri et al., 2001; DeWalle et al., 2000; Zhu and Li,2014; Li and Wang, 2009; Sun et al., 2014). A possible explanation forsuch linear relationship is that change detection, especially abruptchange detection, is not performed due to lack of observed runoff dataseries. The conclusion is then made only on the results of modelsimulation with varied scenarios, such as the results of SCS-CN modelused for long term hydrological impact assessment often reveal a linearrelationship between runoff and IA. In Qinhai River basin, a linearrelationship between runoff and IA could also be obtained if the abruptchange detection was not conducted. But the annual runoff can bebetter described by Eqs. (1) and (2) before and after the abrupt changepoint, a nonlinear relationship between runoff and IA (Fig. 7) was

derived based on the equations. Therefore, the change detection isimportant for more accurate description of hydrological processes andtheir relationship with affecting factors such as climate and LCLUchanges based on observed hydrological data.

The change point of year 2001 in annual runoff with threshold valueof impervious area of 8.6% detected by Pettitt test and DMC in thestudy, can be explained well by the change of NP. This finding isconsistent with those of the threshold response in literatures (Arnoldand Gibbons, 1996; Bledsoe and Watson, 2001; Olivera and DeFee,2007; Yang et al., 2010), suggesting that beyond a certain point, thewatershed has to adjust rainfall-runoff relation to the new conditionsunder urbanization.

Joint use of statistical methods and hydrological models is veryimportant in assessing hydrological response to urbanization. Previousstudies usually applied statistical methods to detect the general trendsfor the data series, and then did the attribution analysis by examininginconsistencies between the variables or catchments in a qualitativeway, or only used a hydrological model to identify relative effects ofurbanization using varied urbanization scenarios. Few studies detectedtrends in runoff due to trends in the impervious areas by combiningstatistical trend analysis and catchment runoff modeling (Crooks andKay, 2015; Mwangi et al., 2016). In this study, both statistical methodsand a hydrological model were adopted. With such a procedure, thehydrological threshold response was detected and nonlinear relation-ship between annual runoff and IA was found. With high accuracy ofthe hydrological model established by using stepwise regressionanalysis for two periods distinguished by statistical methods based onlong time series of data, the results of hydrological impact of urbaniza-tion and relationship between annual runoff and impervious area indifferent climatic conditions based on the model are believed to bemore reliable with less uncertainty.

Our results showed a linear relationship between annual runoff andIA when IA exceeded a threshold of 8.6%. Question remaining iswhether this relationship holds when IA further increases? We willfurther explore the relationship by continuing to collect annual runoffdata and monitoring the impervious area changes for the basin in thefuture.

5. Conclusions

The study proposed a procedure combining different statisticalmethods and a hydrological model to investigate the hydrologicalresponse to urbanization and climate variability, and to explain itsresponse using landscape metrics of impervious surface in the QinhuaiRiver basin, an urbanized basin located in southeast China, over an

Fig. 7. The relationship between annual runoff change and IA (%) for typical hydrological years.

G.D. Bian et al. Catena 157 (2017) 268–278

276

analysis period from 1986 to 2013. It is concluded from the study that:

(1) Annual runoff, runoff coefficient, and PET had significant increas-ing trends, with an abrupt change point in the year 2001 for annualrunoff and runoff coefficient, and in the year 2003 for PET.

(2) The annual runoff was determined by the annual precipitationbefore the year 2001, and by both annual precipitation and IA afterthe year 2001. This suggests that, in the early 2000s, the watershedreached its capacity to incorporate changes and had to adjust to thenew hydrologic conditions.

(3) Since the early 2000s, when the watershed reached an 8.6%impervious area, average annual runoff had increased by 60%, ofwhich urbanization was responsible for 59% of the increase whileprecipitation changes were responsible for the remaining 1%.

(4) The annual runoff response to IA showed non-single linear relation-ships. When the percent IA was less 8.6% (~2001), there was noincrease in annual runoff. When it was greater than 8.6%, there wasan abrupt increase in annual runoff. Beyond this threshold, a linearrelationship existed between annual runoff changes and percentimpervious area. Dry years are more sensitive to urbanization thanwet years.

(5) The change of IA and NP indicated that, starting in the early 2000s,the impervious patches intended to expand together while the IAcontinued to rise, leading to more connection of the existedimpervious areas than to creation of new ones, which increasesthe watershed's conveyance capacity and results in the hydrologicalresponse change.

Acknowledgements

This work was supported by the National Natural ScienceFoundation of China under grants 41371044 and 41371046, whichare greatly appreciated. The authors would like to thank the reviewersfor their valuable comments and suggestions which significantlyimproved the quality of the paper.

References

Aichele, S.S., Andresen, J.A., 2013. Spatial and temporal variations in land developmentand impervious surface creation in Oakland County, Michigan, 1945–2005. J.Hydrol. 485, 96–102.

Alberti, M., 1999. Urban patterns and environmental performance: what do we know? J.Plan. Educ. Res. 19, 151–163.

Alberti, M., Booth, D., Hill, K., Coburn, B., Avolio, C., Coe, S., Spirandelli, D., 2007. Theimpact of urban patterns on aquatic ecosystems: an empirical analysis in Pugetlowland sub-basins. Landsc. Urban Plan. 80, 345–361.

Arnold, C.L., Gibbons, C.J., 1996. Impervious surface coverage. J. Am. Plan. Assoc. 62 (2),243–258.

Beighley, R.E., Elack, J.M., Dunne, T., 2003. Impacts of climatic regimes and urbanizationon streamflow in California coastal watersheds. J. Am. Water Resour. Assoc. 39 (6),1419–1433.

Beighley, R.E., Moglen, G.E., 2002. Trend assessment in rainfall-runoff behavior inurbanizing watersheds. J. Hydrol. Eng. 7 (1), 27–34.

Bhaduri, B., Minner, M., Tatalovich, S., Harbor, J., 2001. Long-term hydrologic impact ofurbanization: a tale of two models. J. Water Resour. Plan. Manag. 127, 13–19.

Bledsoe, B.P., Watson, C.C., 2001. Effects of urbanization on channel instability. J. Am.Water Resour. Assoc. 37, 255–270.

Booth, D.B., 1991. Urbanization and the natural drainage system. Northwest Environ. J. 7(1), 93–118.

Booth, D.B., Jackson, C.R., 1997. Urbanization of aquatic systems: degradationthresholds, stormwater detection, and the limits of mitigation. J. Am. Water Resour.Assoc. 33 (5), 1077–1089.

Brandes, D., Cavallo, G.J., Nilson, M.L., 2005. Base flow trends in urbanizing watershedsof the Delaware river basin. J. Am. Water Resour. Assoc. 41 (6), 1377–1391.

Brun, S.E., Band, L.E., 2000. Simulating runoff behavior in an urbanizing watershed.Comput. Environ. Urban. Syst. 24, 5–22.

Chang, H.J., 2003. Basin hydrologic response to changes in climate and land use: theConestoga River basin, Pennsylvania. Phys. Geogr. 24 (3), 222–247.

Chang, H.J., 2007. Comparative streamflow characteristics in urbanizing basins in thePortland Metropolitan Area, Oregon, USA. Hydrol. Process 21 (2), 211–222.

Chen, Z., Grasby, S.E., 2009. Impact of decadal and century-scale oscillations onhydroclimate trend analyses. J. Hydrol. 365, 122–133.

Choi, W., Nauth, K., Choi, J., Becker, S., 2016. Urbanization and rainfall—runoffrelationships in the Milwaukee River Basin. Prof. Geogr. 68 (1), 14–25.

Cohen, J., 1960. A coefficient of agreement for nominal scales. Educ. Psychol. Meas. 20(1), 37–46.

Cohen, J., 1968. Weighted Kappa: nominal scale agreement with provision for scaleddisagreement or partial credit. Psychol. Bull. 70 (4), 213–220.

Crooks, S.M., Kay, A.L., 2015. Simulation of river flow in the Thames over 120 years:evidence of change in rainfall-runoff response? J. Hydrol. Reg. Stud. 4, 172–195.

DeWalle, D., Swistock, B., Johnson, T., McGuire, K., 2000. Potential effects of climatechange and urbanization on mean annual streamflow in the United States. WaterResour. Res. 36, 2655–2664.

Du, J.K., Li, Q., Rui, H.Y., Zuo, T.H., Zheng, D.P., Xu, Y.P., Xu, C.-Y., 2012. Assessing theeffects of urbanization on annual runoff and flood events using an integratedhydrological modeling system for Qinhuai River basin, China. J. Hydrol. 464–465,127–139.

Dunne, T., Leopold, L.B., 1978. Water in Environmental Planning. W. H. Freeman and Co,New York, pp. 818.

Fletcher, T.D., Andrieu, H., Hamel, P., 2013. Understanding, management and modellingof urban hydrology and its consequences for receiving waters; a state of the artreview. Adv. Water Resour. 51, 261–279.

Hamdi, R., Termonia, P., Baguis, P., 2011. Effects of urbanization and climate change onsurface runoff of the Brussels Capital Region: a case study using an urban soil-vegetation-atmosphere-transfer model. Int. J. Climatol. 31 (13), 1959–1974.

Hannaford, J., Buys, G., Stahl, K., Tallaksen, L.M., 2013. The influence of decadal-scalevariability on trends in long European streamflow records. Hydrol. Earth Syst. Sci. 17,2717–2733.

Hao, L., Sun, G., Liu, Y., Wan, J., Qin, M., Qian, H., Liu, C., Zheng, J., John, R., Fan, P.,Chen, J., 2015. Urbanization dramatically altered the water balances of a paddy field-dominated basin in southern China. Hydrol. Earth Syst. Sci. 19, 3319–3331.

Hsu, M.H., Chen, S.H., Chang, T.J., 2000. Inundation simulation for urban drainage basinwith storm sewer system. J. Hydrol. 234 (1), 21–37.

Huo, Z.L., Feng, S.Y., Kang, S.Z., Li, W.C., Chen, S.J., 2008. Effect of climate changes andwater-related human activities on annual stream flows of the Shiyang river basin inarid north-west China. Hydrol. Process. 22, 3155–3167.

Jacobson, C.R., 2011. Identification and quantification of the hydrological impacts ofimperviousness in urban catchments: a review. J. Environ. Manag. 92, 1438–1448.

Jones, J.A.A., 1997. Global Hydrology: Processes, Resources and EnvironmentalManagement. Prentice Hall, New York.

Kang, S., Park, J.I., Singh, V., 1998. Effect of urbanization on runoff characteristics of theOn-Cheon stream watershed in Pusan, Korea. Hydrol. Process. 25, 351–363.

Kendall, M.G., 1975. Rank Correlation Methods. Griffin, London, UK.Khadr, A., 2016. Temporal and spatial analysis of meteorological drought characteristics

in the upper Blue Nile river region. Hydrol. Res. http://dx.doi.org/10.2166/nh.2016.194.

Klein, R.D., 1979. Urbanization and stream water quality impairment. Water Resour. Bull.15 (4), 948–963.

Ku, H.F.H., Hagelin, N.W., Buxton, H.T., 1992. Effects of urban storm-runoff control onground-water recharge in Nassau County, New York. Ground Water 30 (4), 507–514.

Lee, G.L., Heaney, J.P., 2003. Estimation of urban imperviousness and its impacts onstorm water systems. J. Water Res. Pl-Asce. 129, 419–426.

Li, Y.k., Wang, C.Z., 2009. Impacts of urbanization on surface runoff of the DardenneCreek watershed, St. Charles County, Missouri. Phys. Geogr. 30 (6), 556–573.

Mann, H.B., 1945. Nonparametric tests against trend. Econometrica 13, 245–259.McGarigal, K., Marks, B.J., 1994. FRAGSTATS. Spatial Pattern Analysis Program for

Quantifying Landscape Structure, General Technical Report PNW-GTR-351. USDAForest Service, Pacific Northwest Research Station, Portland, OR.

Mwangi, H.M., Julich, S., Patil, S.D., Mc Donald, M.A., Feger, K.-H., 2016. Relativecontribution of land use change and climate variability on discharge of upper MaraRiver, Kenya. J. Hydrol. Reg. Stud. 5, 244–260.

Nash, J.E., Sutcliffe, J.E., 1970. River flow forecasting through conceptual models-part 1-A: discussion of principles. J. Hydrol. 10 (82), 282–290.

Nirupama, N., Simonovic, S., 2007. Increase of flood risk due to urbanisation: a Canadianexample. Nat. Hazards 40, 25–41.

Olivera, F., DeFee, B., 2007. Urbanization and its effect on runoff in the Whiteoak Bayouwatershed, Texas. J. Am. Water Resour. Assoc. 43 (1), 170–182.

Pettitt, A.N., 1979. A non-parametric approach to the change-point problem. Appl. Stat.28, 126–135.

Praskievicz, S., Chang, H., 2009. A review of hydrological modelling of basin-scaleclimate change and urban development impacts. Prog. Phys. Geogr. 33 (5), 650–671.

Rodriguez, J.J., Kuncheva, L.I., Alonso, C.J., 2006. Rotation forest: a new classifierensemble method. IEEE Trans. Pattern Anal. 28 (10), 1619–1630.

Rougé, C., Cai, X., 2014. Crossing-scale hydrological impacts of urbanization and climatevariability in the Greater Chicago Area. J. Hydrol. 517, 13–27.

Salavati, B., Oudin, L., Furusho, C., Ribstein, P., 2015. Analysing the impact of urbanareas patterns on the mean annual flow of 43 urbanized catchments. In: Proc. IAHS.370. pp. 29–32.

Searcy, J.K., Hardison, C.H., 1960. Double-mass curves. U. S. Geol. Surv. Water SupplyPap. 1541-B, 31–66.

Sen, P.K., 1968. Estimates of the regression coefficient based on Kendall's tau. J. Am. Stat.Assoc. 63, 1379–1389.

Shuster, W.D., Bonta, J., Thurston, H., Warnemuende, E., Smith, D.R., 2005. Impacts ofimpervious surface on watershed hydrology: a review. Urban Water J. 2 (4),263–275.

Smakhtin, V.U., 2001. Low flow hydrology—a review. J. Hydrol. 240 (2), 147–186.Stephan, J.N., Tsay, T.K., 1988. Alternative stategies for stormwater detention. J. Am.

Water Resour. Assoc. Bull. 24 (3), 609–614.Sun, Z.C., Li, X.W., Fu, W.X., Li, Y.K., Tang, D.S., 2014. Long-term effects of land use/land

cover change on surface runoff in urban areas of Beijing, China. J. Appl. Remote.

G.D. Bian et al. Catena 157 (2017) 268–278

277

Sens. 8 (1), 1354–1365.Taye, M.T., Willems, P., 2013. Identifying sources of temporal variability in hydrological

extremes of the upper Blue Nile basin. J. Hydrol. 499, 61–70.Tu, J., 2009. Combined impact of climate and land use changes on streamflow and water

quality in eastern Massachusetts, USA. J. Hydrol. 379, 268–283.Wang, L., 2006. Investigating Impacts of Natural and Human-induced Environmental

Changes on Hydrological Processes and Flood Hazards Using a GIS-basedHydrological/Hydraulic Model and Remote Sensing Data. (PhD dissertation)Department of Geography, Texas A &M University.

Wei, Q., Sun, C.H., Wu, G.H., Pan, L., 2016. Haihe River discharge to Bohai Bay, NorthChina: trends, climate, and human activities. Hydrol. Res. http://dx.doi.org/10.2166/nh.2016.142.

Willems, P., 2013. Multidecadal oscillatory behaviour of rainfall extremes in Europe.Clim. Chang. 120, 931–944.

Yan, R.H., Huang, J.C., Wang, Y., Gao, J.F., Qi, L.Y., 2016. Modeling the combined impactof future climate and land use changes on streamflow of Xinjiang Basin, China.Hydrol. Res. 47 (2), 356–372.

Yang, G., Bowling, L.C., Cherkauer, K.A., Pijanowski, B.C., Niyogi, D., 2010.Hydroclimatic response of watersheds to urban intensity—an observational andmodeling based analysis for the White River basin, Indiana. J. Hydrometeorol. 11 (1),

122–138.Yang, X.L., Chen, H.L., Wang, Y.L., Xu, C.-Y., 2016. Evaluation of the effect of land use/

cover change on flood characteristics using an integrated approach coupling land andflood analysis. Hydrol. Res. http://dx.doi.org/10.2166/nh.2016.108.

Yeo, I.Y., Guldmann, J.M., 2006. Land-use optimization for controlling peak flowdischarge and nonpoint source water pollution. Environ. Plan. 33B, 903–921.

Yue, S., Pilon, P., Phinney, B., Cavadias, G., 2002. The influence of autocorrelation on theability to detect trend in hydrological series. Hydrol. Process. 16, 1807–1829.

Zhang, S.R., Lu, X.X., 2009. Hydrological responses to precipitation variation and diversehuman activities in a mountainous tributary of the lower Xijiang, China. Catena 77,130–142.

Zhang, S.R., Lu, X.X., Higgit, D.L., Chen, C.T.A., Han, J.T., Sun, H.G., 2008. Recentchanges of water discharge and sediment load in the Zhujiang (Pearl River) Basin,China. Glob. Planet. Chang. 60, 365–380.

Zhou, F., Xu, Y.P., Chen, Y., Xu, C.-Y., Gao, Y.Q., Du, J.K., 2013. Hydrological response tourbanization at different spatio-temporal scales simulated by coupling of CLUE-S andthe SWAT model in the Yangtze River Delta region. J. Hydrol. 485, 113–125.

Zhu, C.H., Li, Y.K., 2014. Long-term hydrological impacts of land use/land cover changefrom 1984 to 2010 in the Little River watershed, Tennessee. Int. Soil Water Conserv.Res. 2, 11–22.

G.D. Bian et al. Catena 157 (2017) 268–278

278