Anthropogenic controls from urban growth on flow regimes

Advances in Water Resources 84 (2015) 125–135

Contents lists available at ScienceDirect

Advances in Water Resources journal homepage: www.elsevier.com/locate/advwatres

Anthropogenic controls from urban growth on flow regimes Alfonso Mejía a, Florian Rossel a,b, Jorge Gironás b,c,d, Tijana Jovanovic a,∗ a

Department of Civil and Environmental Engineering, The Pennsylvania State University, University Park, USA Departamento de Ingeniería Hidráulica y Ambiental, Pontificia Universidad Católica de Chile, Santiago, Chile c Centro de Desarrollo Urbano Sustentable CONICYT/FONDAP/15110020, Santiago, Chile d Centro Interdisciplinario de Cambio Global UC, Pontificia Universidad Católica de Chile, Santiago, Chile b

a r t i c l e

i n f o

Article history: Received 24 February 2015 Revised 10 August 2015 Accepted 17 August 2015 Available online 22 August 2015 Keywords: Stochastic dynamics Hydrologic alteration Urbanization Flow duration curves

a b s t r a c t Streamflow can be drastically perturbed in urban basins with important implications for stream, floodplain, and riparian ecosystems. Normally, the dynamic influence of urbanization on streamflow is studied via spacefor-time substitution. Here we explicitly consider urban growth when determining the flow regime of 14 urban basins. To synthetically represent the flow regime, we employ flow duration curves (FDCs) determined using a stochastic model. The model permits derivation of FDCs that are dependent on few parameters representing climatic, land use, conventional stormwater management, and geomorphological conditions in an urban basin. We use the model, under conditions of urban growth, to assess the influence of urbanization on key model parameters and to determine different indicators of hydrologic alteration. Overall, results indicate consistent changes in the temporal evolution of the perturbed flow regimes, which in this case can largely be explained by the progressive redistribution with urban growth of water from slow subsurface runoff and evapotranspiration to fast urban runoff. © 2015 Elsevier Ltd. All rights reserved.

1. Introduction Environmental flow science deals with the sustainable use of streamflow for meeting both human well-being and ecosystem services [3,53,56]. A keystone of this science is the critical role played by the flow regime as a driver of the diversity and vitality of riverine, riparian, and floodplain ecosystems [41,53,56,57]. In practice, the flow regime is often characterized using indicators of hydrologic alteration or measures of hydrologic perturbation relevant to stream ecology [2,3,33,54]. The utility of the flow regime and hydrologic indicators is widely acknowledged and documented for some streamflow perturbations more than others. For example, perturbations associated with in-stream damming have been widely studied [25,41,42,48,60], while perturbations from urban growth at the basin-scale have been less analyzed [37,52,74], even though there are clear indications that these perturbations can have a strong influence on the flow regime and stream ecological conditions [7,34,37,38,46,66,75]. It is common for studies that quantify the effect of urbanization on abiotic and biotic stream conditions to rely on space-for-time substitution or physical gradients [7,13,21,34,52,76]. For urban basins, space-for-time substitution usually consists of measuring or modeling a hydrologic variable relevant to stream ecology across basins with different levels of urban land-use intensity (i.e., across an urban

∗

Corresponding author. Tel.: +1 814 865 0639. E-mail address: [email protected], [email protected] (T. Jovanovic).

http://dx.doi.org/10.1016/j.advwatres.2015.08.010 0309-1708/© 2015 Elsevier Ltd. All rights reserved.

gradient) [15,52], as well as within time periods where urban growth can be assumed relatively constant. The urban gradient is then used to imply the possible time evolution of the variable [13]. This approach can help overcome challenges related to the lack of temporal urban growth data and, indirectly, nonstationarity. However, there are limitations to the space-for-time substitution approach as discussed, in the context of urban studies, by Carter et al. [15]. In particular, it can obscure the causative link between urban growth and streamflow perturbations. When quantifying the influence of urbanization on streamflow, the emphasis has been on high flows or flood conditions [6,49,62,70] or, to a relatively lesser extent, low flows or baseflow conditions [27,34,61,63]. In contrast, we emphasize in this study the entire temporal range of streamflow conditions by utilizing the flow regime. Nonetheless, relying on the flow regime alone does not allow explicit consideration of the key drivers, e.g., climatic forcing and land use conditions, of the streamflow dynamics [9,10,58]. To circumvent this, we employ here a previously developed stochastic model of streamflow for urban basins whose parameters account for climatic, urban land use, conventional stormwater management, and geomorphologic conditions [44]. Our main goal with this study is to use the flow regime and related indicators of hydrologic alteration, determined using a stochastic model, as tools to investigate and characterize the perturbations induced by urban growth on streamflow at the basin-scale. In contrast with many of the previous studies, we account here explicitly for urban growth when determining the flow regime of an urban basin.

126

A. Mejía et al. / Advances in Water Resources 84 (2015) 125–135

2. Methodology This section is divided into three subsections. First, we describe the stochastic model. Second, we explain the model implementation and how it accounts for urban growth. Third, we define several indicators of hydrologic alteration.

2.1. Modeling approach To determine the flow regime, we use a stochastic model of daily streamflow for urban basins. The model was initially proposed by Botter et al. [10] for heterogeneous natural basins and adapted by Mejía et al. [44] to account for urban conditions. The model assumes that urban basins are comprised of effective pervious areas that allow effective recharge to groundwater, and effective impervious areas that produce fast urban runoff. The effective recharge is used to account for the filtering effect of the soil moisture dynamics and urban sources on rainfall. The effective imperviousness accounts for the fact that not all the impervious areas may contribute to fast urban runoff. Ultimately, both slow subsurface runoff, Q P [L3 /T] (the subscript P denotes effective pervious conditions), and fast urban runoff, Q I [L3 /T] (the subscript I denotes effective impervious conditions), generate daily streamflow at the outlet of the urban basin, such that the total streamflow is Q = Q P + Q I . The model assumes that daily rainfall is a marked Poisson process, with events occurring at a constant mean frequency λR [T−1 ]. Each event carries an amount of water Y [L] drawn from the same exponential probability density function (pdf) h(y), with an average depth γR−1 [L]. However, other distributional forms can be considered [20,69]. Rainfall events generate effective rainfall only if the depth of the event, Y, is larger than a threshold di [L], whose value is different for pervious and impervious areas. Hereafter, the subscript i can be equal to P or I to denote effective pervious or impervious conditions, respectively. The pervious threshold, di = dP , is used to represent the net contribution of rainfall events to effective recharge, while the impervious threshold, di = dI , indicates that small rainfall events may not contribute to the generation of urban runoff. Since the effective daily rainfall is simply equal to the daily rainfall minus the surpassed threshold, it remains a marked Poisson process. Accordingly, with h(y) being exponential, the effective rainfall events occur at a frequency λi = λR exp(−di γR ) [T−1 ] [39,69], where λi = λP and λi = λI denote the frequency of effective recharge and urban runoff events, respectively, and their depths remain exponentially distributed with parameter γ R . Using the effective rainfall, the following stochastic differential equation is used to represent the dynamics of the contributions to streamflow from both pervious and impervious areas

dQi (t ) = −ki Qi (t ) + ki li ξi (t ). dt

where A [L2 ] is the drainage area and U∗ is the fraction of effective imperviousness. The use of the fraction of effective imperviousness U∗ , as opposed to the fraction of total imperviousness U, is an added implementation with respect to the model employed by Mejía et al. [44]. We use U∗ here to represent the ability of an urban basin to generate fast runoff. It accounts for the fact that not all the impervious areas in an urban basin may be directly connected to the stormwater drainage network, i.e. some impervious areas may drain to pervious areas. This is typically the case with urban stormwater runoff [1,8,40,64]. Further, analytical expressions for the statistical moments of the streamflow pdf can be obtained from Eq. (1). For instance, the mean, Q [L3 /T], and variance of the streamflow, var(Q ) [(L3 /T)2 ], can be expressed as [10,44]

Q  = Q P  + Q I  =

λP λI ∗ A(1 − U ∗ ) + AU γR γR

(2)

and

λP kP [(1 − U ∗ )A]2 λI kI (AU ∗ )2 + γR2 γR2   2λP kI kP A2U ∗ (1 − U ∗ ) γR dP + 2 + , kI + kP γR2

var(Q ) =

(3)

respectively, where Q P  and Q I  is the mean pervious and impervious contribution, respectively, to streamflow. All the parameters in Eqs. (2) and (3) were already defined. Eq. (1) does not seem to have an analytical solution for the underlying steady state pdf of the total streamflow Q, p(Q) [10,44]. We use instead Monte Carlo simulation to obtain p(Q). Using p(Q), the flow duration curve (FDC) [71], P(Q), is given by

P (Q ) =



∞ Q

p(x)dx.

(4)

P(Q) is the exceedance probability associated with streamflow Q. We use P(Q) here to represent the flow regime. Fig. 1 illustrates the modeling approach. First, a rainfall series is modeled as a marked Poisson process (top horizontal axis in Fig. 1a). This series is then transformed into a streamflow series at the outlet of the urban basin according to Eq. (1) (bottom horizontal axis in Fig. 1a), such that the total streamflow is Q = Q P + Q I (i.e., the addition of the green and red curves in Fig. 1a). Note in Fig. 1a that rainfall produces instantaneous jumps on the streamflow series when the threshold levels dP and dI are surpassed by the rainfall depth and after the jumps occur the pervious and impervious streamflow each decays exponentially at different rates. Fig. 1b and c shows the streamflow pdf and FDC, respectively, obtained from the modeled streamflow series. 2.2. Model implementation

(1)

Eq. (1) says that the time evolution of streamflow follows a deterministic trajectory according to ki Qi perturbed by jumps of random amplitudes given by ki li ξ i (ξi = Yi ), where i can be P (pervious) or I (impervious). Thus, rainfall events with magnitude larger than di generate spikes of daily streamflow that then decrease between effective recharge or runoff events at rates ki [T−1 ] that depend on basin proptakes the meaning of the erties. For the pervious contribution, k−1 P mean response time of a linear groundwater reservoir [18,19,29,43]. reflects In the case of the impervious contribution, the parameter k−1 I the fast response time typical of conventional (connected) stormwater drainage [18,19,29,43]. The frequency λP and λI of events occurrence of the marked Poisson processes ξ P and ξ I , respectively, are obtained using the threshold dP and dI for pervious and impervious areas, respectively. The land use area li [L2 ] is equal to A(1 − U ∗ ) and AU∗ for the pervious and impervious contribution, respectively,

To explicitly account for the influence of urban growth, as well as hydroclimatic variations, on the FDCs, we divide the entire period of analysis of an urban basin into consecutive, non-overlapping time intervals or windows of equal duration. Within each time interval, we apply the stochastic model to simulate streamflow by treating urbanization as being constant. We use for the size of the time interval a multi-year timescale. This is discussed further in the next section. The application of the model at the multi-year timescale, as opposed to seasonal which was the timescale previously used for this model [44], is supported by previous results for regions with weak to moderate rainfall seasonality [11]. Hydroclimatic variations are incorporated into the proposed framework by allowing the rainfall parameters (i.e., λR and γR−1 ) to vary within each time interval [9,55]. The model requires the estimation of eight parameters (A, λR , γR−1 , kP , kI , λP , λI , and U∗ ), which can be obtained a-priori with the exception of kP , kI , and U∗ . We obtain the parameters A, λR , and γR−1 , from

A. Mejía et al. / Advances in Water Resources 84 (2015) 125–135

127

Fig. 1. Illustration of the modeling approach. (a) The top horizontal axis shows the stochastic rainfall series modeled as a marked Poisson process, with exponentially distributed rainfall depths, and the bottom horizontal axis shows the stochastic streamflow series Q/A including the effective pervious, Q P /A, and impervious, Q I /A, contributions to Q/A. (b) Streamflow pdf from which the FDC is determined. (c) FDC including the contribution from Q P /A and Q I /A. To develop this figure, the following parameters values were used: γR−1 = 1 cm, λR = 0.31 days−1 , λP = 0.09 days−1, λI = 0.26 days−1 , kP = 0.02 days−1 , kI = 2.65 days−1 , and U∗ = 29%. (For interpretation of the references to color in the text, the reader is referred to the web version of this article.)

observed data as described by Mejía et al. [44]. The parameters λR and γR−1 can be obtained from precipitation records. The parameter λP is estimated using Eq. (2) [44]. The parameter λI is equal to or less than λR . It is often treated as being less than λR to account for the wetting of urban surfaces (e.g., asphalt and concrete) [24]. Nonetheless, Mejia et al. [44] found that λI does not have a significant influence on the urban FDCs. To determine the value of the parameters kP and kI , one could employed recession analysis but this is particularly challenging in this case because it requires the ability to distinguish a-priori the contribution from effective impervious and pervious areas to streamflow. The challenge in estimating U∗ from observed data is that the required data (e.g., field surveys describing the connectivity of every urban parcel to the stormwater drainage network) are normally not readily available at the basin-scale [28,64]. Since data for U is generally more available, it is often assumed that U∗ is related to U at the basin-scale using the following equation [1,8,40,64]:

U ∗ = ψ + βU φ ,

(5)

where ψ , β , and φ are fitting parameters. In Section 4.1, we use Eq. (5) to account for U∗ in the determination of the FDCs while incorporating, through U, available urban growth information. Notice that the first term in Eq. (5), ψ , accounts for the initial ability of the urban basin to generate fast runoff while the second term represents the fraction of impervious area that contributes to fast urban runoff. Since the parameters U∗ , kP , and kI cannot be obtained a-priori, we estimate their values via calibration, using the optimization approach implemented by Mejía et al. [44]. We then use the calibrated values of U∗ to find the parameters in Eq. (5). In Section 4.1, we compare the performance of the model for the case of the calibrated U∗ values versus the use of Eq. (5). For the calibration, we use the Nash–Sutcliffe efficiency index [47], computed using the log of the streamflow quantiles (LNSE), as the objective function:



N LNSE = 1 −  N

j=1

j=1



2

log (Qo, j ) − log (Q j )



log (Qo, j ) − log (Qo, j )

2 .

(6)

Qo, j and Qj are the observed and estimated quantiles, respectively, for a given value of the exceedance probability, and N is the total number

of quantiles considered. We use N = 40. In Eq. (6), a value of LNSE = 1 indicates a perfect fit. For reference purposes, we also use the classical Nash–Sutcliffe efficiency index (NSE), computed using the streamflow quantiles [17]:



N NSE = 1 −  N

j=1

j=1



Qo, j − Q j



2

Qo, j − Qo, j

2 .

(7)

LNSE and NSE differ in that LNSE tends to de-emphasize somewhat the effect of extreme high streamflow values by placing more weight on the intermediate and low flows. 2.3. Indicators of hydrologic alteration Streamflow perturbations due to urbanization affect stream, riparian, and floodplain ecosystems [74,76]. To assess their potential influence on lotic ecosystems, indicators of hydrologic alteration are employed [26,30,50], given that streamflow data is often more readily available than required biological data. Thus, to further characterize the perturbations induced by urban growth on flow regimes, we employ several indicators of hydrologic alteration, whose relevance for characterizing flow regimes has been indicated before [4,30,38,50]. We use the so-called ecosurplus, ES [L3 /T], and ecodeficit, ED 3 [L /T], indicators which are given by [23,72]



ES =

P∗ 0

 ED =



1 P∗





Q (P ) − Q  (P ) dP, and



Q (P ) − Q  (P ) dP,

(8)

(9)

respectively. In Eqs. (8) and (9), Q and Q are the streamflow for U > 0 and U = 0 (i.e., non-urbanized conditions), respectively. P∗ is the value of the exceedance probability where the FDC for Q and Q cross each other. To determine Q (P), all the model parameters, except U, are the same as the ones used to determine Q(P). In essence, ES and ED quantify the relative gain and loss of streamflow, respectively, from urban growth. Note that this does not consider the impact of other possible human activities on the FDC such as groundwater withdrawals and wastewater treatment effluents. Fig. 2 illustrates

128

A. Mejía et al. / Advances in Water Resources 84 (2015) 125–135

the ES and ED indicators by comparing the FDC for an urban basin with 34% effective imperviousness against the FDC for non-urbanized conditions, U = 0. Fig. 2 shows that with urban growth streamflow may be gained for the lower exceedance probabilities (area shaded in red) while a streamflow loss can occur for the intermediate and higher exceedance probabilities (area shaded in green). To complement the ES and ED indicators, we also use the mean streamflow (Q), variance of streamflow (var(Q )), mode of the streamflow pdf (Qm [L/T]), as well as the percent of streamflow above the mean (PQ ). These four additional indicators are selected because they are frequently employed in environmental flow science [37,50,56]. Here we examine their behavior under conditions of urban growth. 3. Case study Fig. 2. Illustration of the ecosurplus, ES , and ecodeficit, ED , indicators for an urban basin. The red and green lines are the FDCs for urbanized, U > 0, and non-urbanized conditions, U = 0. The areas shaded in red and green represent the value of ES and ED , respectively. They indicate streamflow is gained for the lower exceedance probabilities and loss for the intermediate and higher exceedance probabilities. To compute the FDCs, the following parameter values were used: U∗=34%, ψ = 0.14, β = 0.5, and φ = 1. The remainder parameters values were set equal to the values used in Fig. 1. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

For our case study, we selected 14 basins in the region comprised by the metropolitan areas of the cities of Baltimore and Washington DC, US. We selected this region because it has undergone rapid and fairly intensive urban growth in the last decades [6], the environmental impacts of urbanization are an important regional concern [45,67], and urban growth data are available [5,59]. Fig. 3 illustrates the location of the selected urban basins and Table 1 summarizes their main characteristics. In selecting these basins, we verified that there are no major effluents from wastewater treatment plants or withdrawal

Fig. 3. Map illustrating the location of the 14 urban basins and streamflow gauges used in this study. The urban land use conditions shown are for the year 2001. The basins are all located in the metropolitan areas of the cities of Baltimore and Washington DC, USA.

A. Mejía et al. / Advances in Water Resources 84 (2015) 125–135

129

Table 1 Summary of the main characteristics for the 14 urban basins used in this study. Basin #

A [km2 ]

Q/A [cm/day]a

γR −1 [cm]a

λR  [day−1 ]a

28.6–38.4b 24.9–42.8b 2.1–7.1b 20.3–28.5b 9.2–35.3b 18–30.8b 12.5–29.9b 0.3–18.3b

6.4 14.3 90.1 11.6 9.6 43.3 84.2 54.6

0.13 (0.1–0.16) 0.13 (0.1–0.19) 0.14 (0.1–0.17) 0.14 (0.11–0.19) 0.10 (0.07–0.15) 0.11 (0.09–0.14) 0.11 (0.08–0.15) 0.10 (0.07–0.14)

0.98 (0.9–1.09) 0.99 (0.92–1.09) 1.00 (0.85–1.19) 1.04 (0.94–1.19) 1.01 (0.92–1.12) 1.02 (0.96–1.09) 0.98 (0.9–1.09) 0.99 (0.83–1.12)

0.3 (0.28–0.32) 0.3 (0.28–0.32) 0.33 (0.27–0.39) 0.3 (0.27–0.33) 0.28 (0.25–0.31) 0.29 (0.27–0.31) 0.3 (0.28–0.32) 0.3 (0.25–0.34)

1938–1999 1960–1999 1957–1999 1970–1999

0.6–24.2b 16.5–36b 35.2–40.4b 16–21.2c

98.4 19.7 5.5 188.6

0.11 (0.08–0.15) 0.15 (0.1–0.19) 0.11 (0.09–0.15) 0.13 (0.1–0.15)

0.96 (0.84–1.09) 0.99 (0.85–1.19) 1.01 (0.85–1.19) 0.98 (0.91–1.07)

0.31 (0.28–0.35) 0.32 (0.27–0.39) 0.31 (0.27–0.38) 0.32 (0.3–0.33)

1970–1994

21.5–22.7c

127.9

0.11 (0.09–0.14)

0.98 (0.91–1.07)

0.32 (0.3–0.33)

90.1

0.12 (0.09–0.15)

1.02 (0.94–1.12)

0.3 (0.29–0.32)

Streamflow gauge

Basin name

Period of analysis

Range of U [%]

1 2 3 4 5 6 7 8

01589100 01589330 01581700 01585300 01645200 01653500 01589300 01650500

1957–1989 1960–1987 1967–1999 1960–1989 1957–1987 1952–1978 1957–1999 1938–1999

9 10 11 12

01593500 01585100 01585200 01649500

13

01651000

East Branch Herbert Run Dead Run Winters Run Stemmers Run Watts Branch Henson Creek Gwynns Falls Northwest Branch Anacostia River Little Patuxent River Whitemarsh Run West Branch Herring Run Northeast Branch Anacostia River Northwest Branch Anacostia River Patuxent River

14 a b c

01591000

1970–1990

3.5–4.7

c

The numbers in parenthesis are the ranges associated with using a 5 year time interval. Beighley [5]. Ravirajan [59].

Fig. 4. Illustration of the observed change in the impervious fraction with time for 5 of the selected basins.

points within them [32,68]. However, low flows in the Northwest Branch Anacostia River (basin 8 in Table 1) were occasionally augmented by inter-basin transfers, during the period between 1940 and 1960, for downstream water supply purposes [77]. The key data required to implement the modeling approach are annual urban growth, daily streamflow, and daily rainfall at the basinscale. Annual urban growth data were obtained from Beighley [5] and Ravirajan [59]. The data reported by Beighley [5] were derived from detailed analysis of tax maps, areal imagery, and digital land cover data [31,73], while the data from Ravirajan [59] were derived from poulation data. Fig. 4 illustrates the annual urban growth in 5 of the selected basins. Overall, U ranges from 0.3 to 42.8% and A ranges from 5.5 to 188.6 km2 (Table 1). The pervious land in the selected basins is predominantly agricultural and urban grassed areas. Riparian corridors tend to be forested. The majority of the imperviousness is associated with residential and transportation land use areas, although there are some commercial land use areas in some of the basins. Out of the 14 selected basins, 12 experienced substantial urban growth during their period of analysis, while the remainder two (Winters Run (basin 3) and Patuxent River (basin 14)) are mostly comprised of agricultural and pasture lands, and experienced very little urban growth (Table 1). The period of analysis for each basin (Table 1) was selected on the basis of the available urban growth and streamflow data and they range from 25 to 65 years, with the average being 38 years.

Streamflow data were obtained from the United States Geological Survey and the rainfall data from the National Oceanic and Atmospheric Administration’s National Climatic Data Center. The location of the streamflow gauges is illustrated in Fig. 3 and the gauge numbers are summarized in Table 1. The selected basins have an average γR−1 , λR , and Q/A approximately equal to 1 cm, 0.31 days−1 , and 0.125 cm/days, respectively. The average and range of these parameter values are summarized for each basin in Table 1. Fig. 5a–d illustrates the temporal variability of γR−1 , λR , annual rainfall, and Q/A, respectively, for 5 of the selected basins, using consecutive, non-overlapping 5 year time intervals. As previously indicated, all the parameter values can be obtained a-priori from the available data with the exception of U∗ , kP , and kI . Additionally, based on the available literature [24], the value of dI was set equal to 0.17 cm, which corresponds to a value of λI equal to 0.26 days−1 . In the absence of any other information, we use this value for all the basins. Most of the selected basins are located in the Piedmont physiographic province. This province has a mild sloped terrain and is underlain primarily by crystalline bedrock [22,65]. Henson Creek (basin 6) is entirely located in the Atlantic Coastal Plain physiographic province, which is underlain by unconsolidated layers of sand, gravel, silt, and clay [65]. Stemmers Run (basin 4), Whitemarsh Run (basin 10), and Northeast Brach Anacostia River (basin 12) are partially located in both the Piedmont and Atlantic Coastal Plain province. Overall, climatic and physiographic differences among the selected basins are relatively mild. Thus, the emphasis of this study is not on regional differences, instead we keep these conditions similar to underscore differences due to urban growth. 4. Results and discussion This section is divided into 3 subsections. In the first subsection, we assess the performance of the model. Then, in the second subsection, we examine the behavior of key model parameters. In the last subsection, we investigate the influence of urban growth on selected indicators of hydrologic alteration. To summarize and present our results, throughout this section, we group the results from all the basins analyzed according to prescribed or selected ranges of U. For the selected ranges, we employed 9 bins of equal length, i.e. 5%, for values of U between 0 and 45%. We selected these ranges to ensure that each bin contains at least 4 basins or data points. We use these ranges to determine the average value of the model performance measures, model parameters, and hydrologic indicators associated with each bin. All the results for the FDCs were obtained using a 5 year time

130

A. Mejía et al. / Advances in Water Resources 84 (2015) 125–135

Fig. 5. Illustration of the observed variability of (a) γR−1 , (b) λR , (c) annual rainfall, and (d) Q/A with time for 5 of the selected basins. To determine γR−1 , λR , the annual rainfall, and Q/A, the daily observed rainfall and streamflow data in each basin were divided into consecutive, non-overlapping intervals of five years. The dot symbol and dashed line indicate the local (i.e., within each time interval) and global average from all the basins.

interval since other time intervals (e.g., 1, 2, and 10 years) yielded qualitatively similar results. 4.1. Model application and performance To apply and assess the performance of the model, we worked with two different modeling scenarios denoted as S1 and S2. S1 employs the calibrated values of U∗ while S2 relies on the use of Eq. (5). The goal with S2 is to incorporate the available of urban growth data (see, e.g., Fig. 4) while accounting for effective imperviousness. For S1, we calibrated U∗, kP , and kI for each, non-overlapping 5 year time interval in the period of analysis of each basin. From Fig. 6, it is seen that the average performance of S1 is quite satisfactory; it reproduces the observed or empirical FDCs across values of U, as indicated by the high values of LNSE and NSE in Fig. 6a and b, respectively. As expected, it is slightly better for LNSE than NSE since LNSE was used as the objective function when calibrating the parameter values. Note that in Fig. 6 we show the average performance for each scenario. The average was determined by grouping the performance results from all the basins according to the 9 selected ranges of U. In the case of S2, we used Eq. (5) to find the values of U∗ and then calibrated the parameters kP and kI for each time interval in a basin. To use Eq. (5), we first determined the values of ψ , β , and φ by fitting the values of U∗ from S1 to the observed values of U using the information from all the basins. From this fitting, we found that ψ = 0.14, β = 0.5, and φ was set equal to 1 for simplicity since its value was very close to it (∼0.98). The overall performance of the fitted equation (5) is satisfactory with R2 = 0.8. In the absence of information to directly assess the quality of our U∗ estimates, we note that the good performance of the fitted equation (5) and the similarity of the fitted parameter values with previous results [1,64] is indicative that our U∗ values are reasonable. Additionally, the use of Eq. (5) does not seem to have a strong influence on the performance of S2. Indeed, from Fig. 6a, the performance of S2 is quite similar to that of S1 for both the low and high values of U. For the intermediate values of U, the performance of S2 is somewhat lower than that of S1. This is more visible in Fig. 6b than in Fig. 6a, thereby suggesting that the loss in performance between S1 and S2 may be partly due to the lesser ability of S2 to reproduce the extreme high streamflow values.

Fig. 6. Average model performance versus the impervious fraction for S1 and S2, using (a) LNSE and (b) NSE.

Overall, the model performance for S1 and S2 is satisfactory. Both LNSE (Fig. 6a) and NSE (Fig. 6b) tend to be greater than or equal to 0.9 across values of U. 4.2. Behavior of key model parameters To explore the behavior of key model parameters, we focus our attention on the calibrated parameters kI and kP , and the parameter for the frequency of effective recharge events λP . In Fig. 7a–c, we show the average values of the parameters kI , kP , and λP , respectively, as a

A. Mejía et al. / Advances in Water Resources 84 (2015) 125–135

131

Fig. 7. Average estimates of the parameters (a) kI , (b) kP , and (c) λP versus the impervious fraction. The blue lines are the average empirical estimates of the parameter values. The blue error bars are the 90% range of variability of the empirical estimates. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

function of U (i.e., by grouping the parameter values from all the time periods in the 14 basins analyzed according to the selected ranges of U). In Fig. 7a, kI varies over a restricted range of values (approximately from 3 to 7.6 days−1 ) and tends to fluctuate around its mean with no clear dependence on U. The fluctuations are more pronounced for S2 than S1, thereby suggesting that the use of Eq. (5) tends to increase the variability of kI . Nonetheless, the differences in the values of kI between S1 and S2 are relatively small, except perhaps for the intermediate values of U. In Fig. 7b, kP behaves similar to kI . It tends to vary over a restricted range of values (approximately from 0.015 to 0.055 days−1 ) and it does not seem to depend on U. However, kP is two orders of magnitude less than kI , thus reflecting the large difference in the response time of the effective impervious and pervious contributions. The values obtained for both kI and kP are within the range of variability of previously reported values [9,12,18,44]. We also show in Fig. 7b an average estimate of kP obtained directly from the streamflow data. This average empirical estimate was obtained using the approach of Brutsaert and Nieber [14] and by assuming that the effect of kI on streamflow becomes negligible after several days (from 4 to 7 days, depending on the basin) from the occurrence of a streamflow jump or peak. However, in implementing

the latter assumption, we found that the observed recession parameters for the different time intervals in a basin can exhibit large fluctuations as the recession time (i.e., the time after a streamflow peak) increases or may not converge to a stable estimate, which makes in this case the determination of kP for each time interval in a basin unreliable. Thus, we only use the empirical estimates of kP for comparison purposes and, as indicated previously in Section 4.1, we relied on calibration to determine kP for each time interval. Fig. 8 shows the average empirical estimates of the recession parameter for the selected ranges of U. It also shows that, approximately 5 days after the streamflow peak occurs, the recession parameter tends to become fairly constant and independent of U. At any rate, the comparison in Fig. 7b between the average empirical and the calibrated recession parameter values is satisfactory for both S1 and S2. The calibrated values tend to be contained within the 90% range of variability of the empirical estimates. It is interesting to note that S1 (i.e., the scenario for which U∗ is determined by calibration) tends to match more closely the average empirical estimates, thereby suggesting that the variability of U∗ may be an important feature of urban basins. In Fig. 7c, we show the average values of λP for both S1 and S2. λP varies little between S1 and S2, thereby indicating that λP is not

132

A. Mejía et al. / Advances in Water Resources 84 (2015) 125–135

Fig. 8. Average value of the recession parameter obtained from the observed streamflow data as a function of both the time after the streamflow peak (jump) and the impervious fraction. The curves shown are best fit estimates of the data points and they are only included to highlight the different trajectories of the data points.

strongly affected by the use of Eq. (5) in S2. Also from Fig. 7c, λP seems to depend weakly on U, as it tends to decrease slightly with U. This means that recharge events become less frequent as the basins become more urbanized. A possible explanation for this behavior is that soils allow less infiltration as urbanization increases, e.g., due to increasing soil compaction [16,51], but this hypothesis will need to be further verified using field data. As done before for kP , we also obtained average empirical estimates of λP (Fig. 7c). To obtain these empirical estimates, we set Q P  in Eq. (2), after assuming that the contribution from the effective impervious areas is negligible, equal to the average baseflow values implied by the empirical estimates of kP and solved for λP . We then averaged the empirical λP estimates from all the basins according to the selected ranges of U. Note that the approach followed to estimate λP is different when performing the model calibration for S1 and S2. In that approach, Eq. (2) is used to obtain λP using both the observed value of Q and the estimated value of U∗. Nonetheless, as shown in Fig. 7c, the average empirical values of λP compare well against the estimated values. The estimated values tend for the most part to be within the 90% range of variability of the average empirical estimates of λP , thus suggesting that the estimated values are within plausible ranges for both S1 and S2. In summary, the calibrated parameters values for kI and kP are similar for S1 and S2, thus suggesting that differences between the calibrated values of U∗ versus the estimates from Eq. (5) are not pronounced. Additionally, the parameters values for kP and λP , in both S1 and S2, tend to fall within the observed or empirical range, reinforcing the ability of the parameter estimation approach to produce reasonable parameter estimates. 4.3. Performance and behavior of the hydrologic indicators We employ in this section the hydrologic indicators described in Section 2.3 (i.e., ES /A, ED /A, Q/A, var(Q )/A2 , Qm /A, and PQ ) to examine and quantify the influence of urban growth on flow regimes. To determine ES /A and ED /A for each time interval in a basin, we used the modeled FDC for the case of U > 0 (Q(P) in Eqs. (8) and (9)) together with the modeled FDC for U = 0 (Q (P) in Eqs. (8) and (9)). We determined ES /A and ED /A only for S2 since for S1 the case of U = 0 is not defined. In Fig. 9a, we show the average values of ES /A for S2 as a function of U. Same as before, the average values were obtained by grouping the ES /A values from all the basins according to the selected ranges of U. In Fig. 9a, ES /A tends to increase with U. This means that for a given exceedance probability of streamflow the high flows tend to increase with U, i.e., the high flows become more frequent. We also show in Fig. 9a an observed or empirical estimate of ES /A. To determine this empirical estimate, we used the empirical

Fig. 9. Average modeled estimates of (a) ES /A and (b) ED /A for S2 versus the impervious fraction. The blue lines represent the empirical estimates obtained from the streamflow data. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

(urbanized) FDCs from each time interval together with an average empirical (non-urbanized) FDC. The non-urbanized FDC was determined as the average from all the observed FDCs with urbanization levels of 0–5% since many of the urban basins considered are already urbanized (i.e., U> 5%) at the start of their period of analysis (see Table 1). In Fig. 9a, the modeled values of ES /A compare well against the empirical estimates. In Fig. 9b, we show the results for the average values of ED /A. In this figure, the tendency is for ED /A to increase with U. Thus, for a given exceedance probability of streamflow, the flows tend to decrease with U. However, the modeled values in Fig. 9b tend to underpredict the observed or empirical estimates. This bias is mainly due to the observed low flows being larger than the modeled estimates in the first interval in Fig. 9b, comprised of basins with impervious fractions between 0 and 0.05. These large, observed low flow values are caused by basin 8 (Table 1). Low flows in basin 8 were augmented, through inter-basin transfers, during part of its gauging period (from 1940 to 1960) [77], when the impervious fraction in basin 8 was less than 0.05. This flow augmentation is not accounted for by the modeled estimates. Also, note that only 4 out of the 14 basins considered have impervious fractions within the range 0–0.05, thus sampling uncertainty is another factor affecting the empirical estimates of ED /A. Additionally, this bias may partly reflect the stronger sensitivity of the FDCs to variability in the low flow estimates [23,44,77]. In Fig. 10, we show the average observed or empirical FDCs from all the basins, grouped according to the selected ranges of U. Fig. 10 provides additional support to the results shown in Fig. 9. The empirical FDCs in Fig. 10 tend to shift upward with increasing U for the low exceedance probabilities and vice-versa for the high exceedance probabilities, as implied by Fig. 9a and b, respectively. A common view emerges from Figs. 9 and 10, which indicates that urban growth progressively redistributes water from slow subsurface runoff and evapotranspiration to fast urban runoff. A similar finding is reported by Konrad and Booth [36] for urbanizing basins in the Puget Sound

A. Mejía et al. / Advances in Water Resources 84 (2015) 125–135

Fig. 10. Average empirical FDCs obtained from the streamflow data for different ranges of the impervious fraction U.

region, USA. Nonetheless, we recognize that as these basins continue to experience urban growth and/or as their infrastructure ages, this redistribution of water could be further modified by interactions between the urban water drainage infrastructure and the surrounding environment. Moreover, other future human activities such as groundwater exploitation and wastewater effluents could influence these flow regimes. Fig. 11 illustrates the average values for the other indicators of hydrologic alteration (i.e., Q/A, var(Q )/A2 , Qm /A, and PQ ) as a function of U. Note that all the indicators in Fig. 11 compare relatively well against the observed or empirical estimates. In Fig. 11a, Q/A increases somewhat with U. The weak dependence of Q/A on U suggests that the increase in fast urban runoff with increasing U tends to be counteracted by a decrease in slow flow contributions with increasing U. This interpretation of Fig. 11a is in correspondence with

133

our previous results in Figs. 9 and 10, where we highlighted the tendency of urban growth to redistribute water from slow subsurface to fast urban runoff. Similar trends to the one in Fig. 11a have been reported for other urban basins [36]. As it was the case with Q/A (Fig. 11a), var(Q )/A2 also increases with U (Fig. 11b). However, var(Q )/A2 shows a stronger sensitive or response to U than Q/A, which makes var(Q )/A2 a more useful metric than Q/A for quantifying the impact of urbanization on streamflow. The increase of var(Q )/A2 with U highlights the role played by conventional stormwater drainage in the alteration of streamflow. With increasing urbanization less of the rainfall forcing is filtered by the basin due to both increased imperviousness and the ability of conventional stormwater drainage to rapidly convey urban runoff to the basin outlet. This rapid conveyance of runoff increases the variability of the streamflow dynamics, which is here reflected by the increase of var(Q )/A2 with U. Fig. 11c highlights that changes in the first two statistical moments of streamflow, Q/A and var(Q )/A2 , with U are accompanied by alterations in the streamflow pdf. In particular, the mode of the streamflow pdf, Qm /A, tends to decrease with increasing U. This means that the most likely streamflow state decreases with increasing U. Thus, streams in heavily urbanized basins are more likely to find themselves in a lower flow state than streams in basins with little urbanization. This is relevant to management efforts aimed at maintaining the ecological integrity of urban streams, since the surpassing of a given low flow threshold can have detrimental effects on stream ecosystems [50]. Additionally, note that the behavior of Q/A (Fig. 11a) and var(Q )/A2 (Fig. 11b) is similar; they both tend to increase with U but seem to level off for values of U approximately greater than 25%, suggesting that the influence of urban growth on streamflow may potentially reach a saturation level. However, this saturation effect is not confirmed by the other two indicators. Indeed, Qm /A (Fig. 11c) and PQ (Fig. 11d) tend both to decrease continuously with U. Thus, it is not possible to determine from Fig. 11 a level of urbanization

Fig. 11. Average modeled estimates of the (a) mean streamflow (Q/A), (b) variance of streamflow (var(Q )/A2 ), (c) mode of the streamflow pdf (Qm /A), and (d) percent of streamflow above the mean streamflow (PQ ) for S1 and S2 versus the impervious fraction. The blue lines represent the empirical estimates obtained from the streamflow data. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

134

A. Mejía et al. / Advances in Water Resources 84 (2015) 125–135

after which changes in hydrologic impact are minimal. This has implications for urban planning. For instance, a planning approach based solely on the results from Fig. 11a and b could lead to indiscriminately supporting further urban development on the most urbanized basins by reasoning that little additional impact on stream hydrology will be caused. However, this reasoning is not supported by the other metrics in Fig. 11c and d. Overall, Fig. 11 highlights the dependence of the streamflow dynamics in urban basins on urban growth, e.g., the statistical moments and the pdf of streamflow depend on U. It also highlights the need to determine different indicators of hydrologic alteration when assessing the influence of urban growth on flow regimes as indicators can provide contrasting information. 5. Conclusions We used FDCs to characterize the flow regime of basins influenced by urban growth. The FDCs were obtained using a stochastic model of daily streamflow that depends on a few key parameters representing climatic, land use, conventional stormwater, and geomorphologic conditions in an urban basin. We applied the stochastic model to 14 urban basins in the metropolitan areas of the cities of Baltimore and Washington DC, US. As part of the model application, we assessed the influence of effective imperviousness on the FDCs and determined different indicators of hydrologic alteration. On the basis of our results, we emphasize the following conclusions: - The performance of the stochastic model of daily streamflow under conditions of urban growth is satisfactory. We find that the use of a regional (i.e., across the set of urban basins used here) expression relating the effective imperviousness to the actual imperviousness has only a relatively small impact on the modeled FDCs. This suggests that regional urban growth information may be helpful for modeling the FDCs of basins with little urban growth information. - The values of key model parameters (the frequency of effective recharge and the mean response time of effective pervious areas) compare well against the observed or empirical estimates. The approach followed to obtain these empirical estimates has potential as an a-priori tool for parameter estimation. The approach will need to be evaluated further using different regions and number of basins. - The ecosurplus and ecodeficit indicators suggest consistent changes in the flow regime with urban growth; high flows tend to increase with urban growth for a given low exceedance probability and vice-versa for the low flows. This behavior may be explained by the progressive redistribution with urban growth of slow subsurface runoff and evapotranspiration to fast urban runoff. - The mean and variance of streamflow suggest that the perturbations induced by urban growth on streamflow may saturate for the higher levels of urbanization considered. However, the mode of the streamflow pdf as well as the percent of streamflow above the mean do not support this saturation hypothesis, instead they indicate that the perturbations tend to increase continuously with urban growth. Acknowledgments We are thankful to the three reviewers for their criticisms and suggestions which helped improve the quality of the original manuscript. The funding support provided by the Pennsylvania Sea Grant as well as the Department of Civil and Environmental Engineering at the Pennsylvania State University is gratefully acknowledged. The support from project FONDECYT 1131131 and CEDEUSConicyt/Fondap/15110020 is also acknowledged. The present work

was partially developed within the framework of the Panta Rhei Research Initiative of the International Association of Hydrological Sciences. References [1] Alley W, Veenhuis J. Effective impervious area in urban runoff modeling. J Hydraul Eng 1983;109(2):313–19. http://dx.doi.org/10.1061/(ASCE)07339429(1983)109:2(313). [2] Angus Webb J, Miller KA, King EL, de Little SC, Stewardson MJ, Zimmerman JKH, et al. Squeezing the most out of existing literature: a systematic re-analysis of published evidence on ecological responses to altered flows. Freshwater Biol 2013;58(12):2439–51. http://dx.doi.org/10.1111/fwb.12234. [3] Arthington AH, Naiman RJ, McClain ME, Nilsson C. Preserving the biodiversity and ecological services of rivers: new challenges and research opportunities. Freshwater Biol 2010;55(1):1–16. http://dx.doi.org/10.1111/j.1365-2427.2009.02340.x. [4] Baker DB, Richards RP, Loftus TT, Kramer JW. A new flashiness index: characteristics and applications to Midwestern rivers and streams. J Am Water Resour Assoc 2004;40(2):503–22. http://dx.doi.org/10.1111/j.1752-1688.2004.tb01046.x. [5] Beighley RE. GIS adjustment of measured streamflow data from urbanized watersheds. University of Maryland; 2001. Ph.D. dissertation p. 262. [6] Beighley RE, Moglen GE. Adjusting measured peak discharges from an urbanizing watershed to reflect a stationary land use signal. Water Resour Res 2003;39(4):1093. http://dx.doi.org/10.1029/2002WR001846. [7] Booth D, Karr J, Schauman S, Konrad C, Morley S, Larson M, et al. Reviving urban streams: land use, hydrology, biology, and human behavior. J Am Water Resour Assoc 2004;40(5):1351–64. http://dx.doi.org/10.1111/j.17521688.2004.tb01591.x. [8] Booth DB, Jackson CR. Urbanization of aquatic systems: degradation thresholds, stormwater detection, and the limits of mitigation. J Am Water Resour Assoc 1997;33(5):1077–90. http://dx.doi.org/10.1111/j.1752-1688.1997.tb04126.x. [9] Botter G, Basso S, Rodriguez-Iturbe I, Rinaldo A. Resilience of river flow regimes. Proc Natl Acad Sci 2013;110(32):12925–30. http://dx.doi.org/10.1073/ pnas.1311920110. [10] Botter G, Porporato A, Daly E, Rodriguez-Iturbe I, Rinaldo A. Probabilistic characterization of base flows in river basins: roles of soil, vegetation, and geomorphology. Water Resour Res 2007;43(6):W06404. http://dx.doi.org/10.1029/ 2006WR005397. [11] Botter G, Zanardo S, Porporato A, Rodriguez-Iturbe I, Rinaldo A. Ecohydrological model of flow duration curves and annual minima. Water Resour Res 2008;44(8):W08418. http://dx.doi.org/10.1029/2008WR006814. [12] Botter G, Basso S, Porporato A, Rodriguez-Iturbe I, Rinaldo A. Natural streamflow regime alterations: damming of the Piave river basin (Italy). Water Resour Res 2010;46(6):W06522. http://dx.doi.org/10.1029/2009WR008523. [13] Brown LR, Cuffney TF, Coles JF, Fitzpatrick F, McMahon G, Steuer J, et al. Urban streams across the USA: lessons learned from studies in 9 metropolitan areas. J North Am Benthol Soc 2009;28(4):1051–69. http://dx.doi.org/10.1899/08-153.1. [14] Brutsaert W, Nieber JL. Regionalized drought flow hydrographs from a mature glaciated plateau. Water Resour Res 1977;13(3):637–43. http://dx.doi.org/ 10.1029/WR013i003p00637. [15] Carter T, Jackson CR, Rosemond A, Pringle C, Radcliffe D, Tollner W, et al. Beyond the urban gradient: barriers and opportunities for timely studies of urbanization effects on aquatic ecosystems. J North Am Benthol Soc 2009;28(4):1038–50. http://dx.doi.org/10.1899/08-169.1. [16] Chen Y, Day SD, Wick AF, McGuire KJ. Influence of urban land development and subsequent soil rehabilitation on soil aggregates, carbon, and hydraulic conductivity. Sci Total Environ 2014;494–495:329–36. http://dx.doi.org/ 10.1016/j.scitotenv.2014.06.099. [17] Claps P, Giordano A, Laio F. Advances in shot noise modeling of daily streamflows. Adv Water Resour 2005;28(9):992–1000. http://dx.doi.org/10.1016/ j.advwatres.2005.03.008. [18] Coutu S, Del Giudice D, Rossi L, Barry DA. Parsimonious hydrological modeling of urban sewer and river catchments. J Hydrol 2012;464–465:477–84. http://dx.doi.org/10.1016/j.jhydrol.2012.07.039. [19] Coutu S, Del Giudice D, Rossi L, Barry DA. Modeling of facade leaching in urban catchments. Water Resour Res 2012;48(12):W12503. http://dx.doi.org/10.1029/ 2012WR012359. [20] Daly E, Porporato A. Effect of different jump distributions on the dynamics of jump processes. Phys Rev E 2010;81(6). http://dx.doi.org/10.1103/ PhysRevE.81.061133. [21] DeGasperi CL, Berge HB, Whiting KR, Burkey JJ, Cassin JL, Fuerstenberg RR. Linking hydrologic alteration to biological impairment in urbanizing streams of the Puget lowland, Washington, USA. J Am Water Resour Assoc 2009;45(2):512–33. http://dx.doi.org/10.1111/j.1752-1688.2009.00306.x. [22] Fenneman NM. Physiography of eastern United States. 1st. New York, NY: McGraw-Hill, Inc.; 1938. p. 714. [23] Gao Y, Vogel RM, Kroll CN, Poff NL, Olden JD. Development of representative indicators of hydrologic alteration. J Hydrol 2009;374(1–2):136–47. http://dx.doi.org/10.1016/j.jhydrol.2009.06.009. [24] Gironás J, Roesner LA, Davis J, Rossman L. Storm water management model applications manual. National Risk Management Research Laboratory, Office of Research and Development, USEPA; 2009. [25] Graf WL. Downstream hydrologic and geomorphic effects of large dams on American rivers. Geomorphology 2006;79(3–4):336–60. http://dx.doi.org/10.1016/ j.geomorph.2006.06.022.

A. Mejía et al. / Advances in Water Resources 84 (2015) 125–135 [26] Hamel P, Daly E, Fletcher TD. Which baseflow metrics should be used in assessing flow regimes of urban streams? Hydrol Process 2015. http://dx.doi.org/10.1002/hyp.10475. [27] Hamel P, Daly E, Fletcher TD. Source-control stormwater management for mitigating the impacts of urbanisation on baseflow: a review. J Hydrol 2013;485:201–11. http://dx.doi.org/10.1016/j.jhydrol.2013.01.001. [28] Han W, Burian S. Determining effective impervious area for urban hydrologic modeling. J Hydrol Eng 2009;14(2):111–20. http://dx.doi.org/10.1061/ (ASCE)1084-0699(2009)14:2(111). [29] Hejazi MI, Moglen GE. The effect of climate and land use change on flow duration in the Maryland Piedmont region. Hydrol Process 2008;22(24):4710–22. http://dx.doi.org/10.1002/hyp.7080. [30] Homa ES, Brown C, McGarigal K, Compton BW, Jackson SD. Estimating hydrologic alteration from basin characteristics in Massachusetts. J Hydrol 2013;503:196– 208. http://dx.doi.org/10.1016/j.jhydrol.2013.09.008. [31] Homer C, Dewitz J, Fry J, Coan M, Hossain N, Larson C, et al. Completion of the 2001 National Land Cover Database for the conterminous United States. Photogramm Eng Remote Sens 2007;73(4):337–41. [32] Kenny JF, Barber NL, Hutson SS, Linsey KS, Lovelace JK, Maupin MA. Estimated use of water in the United States in 2005: U.S. Geological Survey circular 1344; 2009. p. 52. [33] King J, Brown C. Integrated basin flow assessments: concepts and method development in Africa and South-east Asia. Freshwater Biol 2010;55(1):127–46. http://dx.doi.org/10.1111/j.1365-2427.2009.02316.x. [34] Klein RD. Urbanization and stream quality impairment. J Am Water Resour Assoc 1979;15(4):948–63. http://dx.doi.org/10.1111/j.1752-1688.1979.tb01074.x. [36] Konrad CP, Booth, DB. Hydrologic trends associated with urban development for selected streams in the Puget Sound basin. Western Washington. U.S. Geological Survey, Tacoma, Washington; 2002. p. 40. [37] Konrad CP, Booth D. Hydrologic changes in urban streams and their ecological significance. The American Fisheries Society, American Fisheries Society Symposium 47; 2005. pp. 157-177. [38] Konrad CP, Booth DB, Burges SJ. Effects of urban development in the Puget lowland, Washington, on interannual streamflow patterns: consequences for channel form and streambed disturbance. Water Resour Res 2005;41(7):W07009. http://dx.doi.org/10.1029/2005WR004097. [39] Laio F. A vertically extended stochastic model of soil moisture in the root zone. Water Resour Res 2006;42(2):W02406. http://dx.doi.org/10.1029/ 2005WR004502. [40] Lee J, Heaney J. Estimation of urban imperviousness and its impacts on storm water systems. J Water Resour Plann Manage 2003;129(5):419–26. http://dx.doi.org/10.1061/(ASCE)0733-9496(2003)129:5(419). [41] Lytle DA, Poff NL. Adaptation to natural flow regimes. Trends Ecol Evol 2004;19(2):94–100. http://dx.doi.org/10.1016/j.tree.2003.10.002. [42] Magilligan FJ, Nislow KH. Changes in hydrologic regime by dams. Geomorphology 2005;71(1–2):61–78. http://dx.doi.org/10.1016/j.geomorph.2004.08.017. [43] McCuen RH, Snyder WM. Hydrologic modeling: statistical methods and applications. Englewood Cliffs, NJ: Prentice-Hall; 1986. [44] Mejía A, Daly E, Rossel F, Jovanovic T, Gironás J. A stochastic model of streamflow for urbanized basins. Water Resour Res 2014;50(3):1984–2001. http://dx.doi.org/10.1002/2013WR014834. [45] Mejía AI, Moglen GE. Spatial distribution of imperviousness and the spacetime variability of rainfall, runoff generation, and routing. Water Resour Res 2010;46(7):W07509. http://dx.doi.org/10.1029/2009WR008568. [46] Morley SA, Karr JR. Assessing and restoring the health of urban streams in the Puget Sound basin. Conserv Biol 2002;16(6):1498–509. http://dx.doi.org/10.1046/j.1523-1739.2002.01067.x. [47] Nash JE, Sutcliffe JV. River flow forecasting through conceptual models part I—a discussion of principles. J Hydrol 1970;10(3):282–90. http://dx.doi.org/10.1016/0022-1694(70)90255-6. [48] Nilsson C, Reidy CA, Dynesius M, Revenga C. Fragmentation and flow regulation of the world’s large river systems. Science 2005;308(5720):405–8. http://dx.doi.org/10.1126/science.1107887. [49] Ogden FL, Raj Pradhan N, Downer CW, Zahner JA. Relative importance of impervious area, drainage density, width function, and subsurface storm drainage on flood runoff from an urbanized catchment. Water Resour Res 2011;47(12). http://dx.doi.org/10.1029/2011WR010550. [50] Olden JD, Poff NL. Redundancy and the choice of hydrologic indices for characterizing streamflow regimes. River Res Appl 2003;19(2):101–21. http://dx.doi.org/10.1002/rra.700. [51] Pitt R, Chen S, Clark S, Swenson J, Ong C. Compaction’s impacts on urban storm-water infiltration. J Irrig Drain Eng 2008;134(5):652–8. http://dx.doi.org/10.1061/(ASCE)0733-9437(2008)134:5(652). [52] Poff NL, Bledsoe BP, Cuhaciyan CO. Hydrologic variation with land use across the contiguous United States: geomorphic and ecological consequences for stream ecosystems. Geomorphology 2006;79(3–4):264–85. http://dx.doi.org/10.1016/j.geomorph.2006.06.032.

135

[53] Poff NL, Allan JD, Bain MB, Karr JR, Prestegaard KL, Richter BD, et al. The natural flow regime. Bioscience 1997;47(11):769–84. http://dx.doi.org/10.2307/1313099. [54] Poff NL, Richter BD, Arthington AH, Bunn SE, Naiman RJ, Kendy E, et al. The ecological limits of hydrologic alteration (ELOHA): a new framework for developing regional environmental flow standards. Freshwater Biol 2010;55(1):147–70. http://dx.doi.org/10.1111/j.1365-2427.2009.02204.x. [55] Porporato A, Vico G, Fay PA. Superstatistics of hydro-climatic fluctuations and interannual ecosystem productivity. Geophys Res Lett 2006;33(15):L15402. http://dx.doi.org/10.1029/2006GL026412. [56] Postel S, Richter B. Rivers for life: managing water for people and nature. Washington D.C.: Island Press; 2003. p. 253. [57] Power ME, Sun A, Parker G, Dietrich WE, Wootton JT. Hydraulic food-chain models. Bioscience 1995;45(3):159–67. http://dx.doi.org/10.2307/1312555. [58] Pumo D, Noto L, Viola F. Ecohydrological modelling of flow duration curve in Mediterranean river basins. Adv Water Resour 2013;52(0):314–27. http://dx.doi.org/10.1016/j.advwatres.2012.05.010. [59] Ravirajan K. Development and application of a stream flashiness index based on imperviousness and climate using GIS. University of Maryland; 2007. M.S. thesis p. 275. [60] Richter BD, Baumgartner JV, Powell J, Braun DP. A method for assessing hydrologic alteration within ecosystems. Conserv Biol 1996;10(4):1163–74. http://dx.doi.org/10.1046/j.1523-1739.1996.10041163.x. [61] Roesner L, Bledsoe B, Brashear R. Are best-management-practice criteria really environmentally friendly? J Water Resour Plann Manage 2001;127(3):150–4. http://dx.doi.org/10.1061/(ASCE)0733-9496(2001)127:3(150). [62] Rogers GO, DeFee Ii BB. Long-term impact of development on a watershed: early indicators of future problems. Landscape Urban Plann 2005;73(2–3):215– 33. http://dx.doi.org/10.1016/j.landurbplan.2004.11.007. [63] Rose S, Peters NE. Effects of urbanization on streamflow in the Atlanta area (Georgia, USA): a comparative hydrological approach. Hydrol Process 2001;15(8):1441– 57. http://dx.doi.org/10.1002/hyp.218. [64] Roy AH, Shuster WD. Assessing impervious surface connectivity and applications for watershed management. J Am Water Resour Assoc 2009;45(1):198–209. http://dx.doi.org/10.1111/j.1752-1688.2008.00271.x. [65] Rutledge, A.T., Mesko, T.O.. Estimated hydrologic characteristics of shallow aquifer systems in the Valley and Ridge, the Blue Ridge, and the Piedmont physiographic provinces based on analysis of streamflow recession and base flow. U.S. Geological Survey, Washington, D.C.; 1996. 58 pp. [66] Schoonover JE, Lockaby BG, Helms BS. Impacts of land cover on stream hydrology in the West Georgia Piedmont, USA. J Environ Qual 2006;35(6):2123–31. http://dx.doi.org/10.2134/jeq2006.0113. [67] USEPA. ICLUS V1.2 user’s manual: ArcGIS tools and datasets for modeling US housing density growth. Global Change Research Program, National Center for Environmental Assessment, Office of Research and development, Washington D.C.; 2009. [68] USEPA. http://www.epa.gov/enviro/html/frs_demo/geospatial_data/geo_data_ state_combined; 2013. [69] Verma P, Yeates J, Daly E. A stochastic model describing the impact of daily rainfall depth distribution on the soil water balance. Adv Water Resour 2011;34(8):1039– 48. http://dx.doi.org/10.1016/j.advwatres.2011.05.013. [70] Villarini G, Smith JA, Serinaldi F, Bales J, Bates PD, Krajewski WF. Flood frequency analysis for nonstationary annual peak records in an urban drainage basin. Adv Water Resour 2009;32(8):1255–66. http://dx.doi.org/10.1016/ j.advwatres.2009.05.003. [71] Vogel R, Fennessey N. Flow-duration curves. I: new interpretation and confidence intervals. J Water Resour Plann Manage 1994;120(4):485–504. http://dx.doi.org/10.1061/(ASCE)0733-9496(1994)120:4(485). [72] Vogel RM, Sieber J, Archfield SA, Smith MP, Apse CD, Huber-Lee A. Relations among storage, yield, and instream flow. Water Resour Res 2007;43(5):W05403. http://dx.doi.org/10.1029/2006WR005226. [73] Vogelmann JE, Howard SM, Yang L, Larson CR, Wylie BK, Van Driel JN. Completion of the 1990 s National Land Cover Data Set for the conterminous United States. Photogramm Eng Remote Sens 2001;67:650–62. [74] Walsh CJ, Fletcher TD, Burns MJ. Urban stormwater runoff: a new class of environmental flow problem. PLoS One 2012;7(9):e45814. http://dx.doi.org/10.1371/journal.pone.0045814. [75] Wang L, Lyons J, Kanehi P, Bannerman R, Emmons E. Watershed urbanization and changes in fish communities in southeastern Wisconsin streams. J Am Water Resour Assoc 2000;36(5):1173–89. http://dx.doi.org/10.1111/j.17521688.2000.tb05719.x. [76] Wang S-Y, Sudduth EB, Wallenstein MD, Wright JP, Bernhardt ES. Watershed urbanization alters the composition and function of stream bacterial communities. PLoS One 2011;6(8):e22972. http://dx.doi.org/10.1371/journal.pone.0022972. [77] Yorke, T.H., Herb, W.J.. Effects of urbanization on streamflow and sediment transport in the Rock Creek and Anacostia River basins, Montgomery County, Maryland, 1962-74. USGS Professional Paper 1003. Washington, D.C.;1978.