Land use control of stream nitrate concentrations in mountainous coastal California watersheds

Authors


Abstract

[1] Concentration versus runoff relationships can reveal how land use and watershed hydrology interactively regulate solute inputs to streams and downstream aquatic ecosystems. In six mountainous southern California coastal watersheds, consistent nitrate-runoff patterns within three broad land use classes exist: dilution in agricultural watersheds, invariance in urban watersheds, and enrichment in an undeveloped watershed. Locally weighted scatterplot smoothing (LOWESS) revealed these patterns in nitrate-runoff relationships. A hyperbolic equation reproduced these relationships, also identifying the most common nitrate concentration (i.e., nitrate mode) observed in stream water during periods of low runoff (i.e., base flow) and high runoff (i.e., stormflow). LOWESS and the hyperbolic equation were also used to reveal and reproduce electrical conductance-runoff relationships, which contrary to nitrate-runoff relationships, demonstrated uniform behavior (significant dilution; p < 0.0001) in all watersheds regardless of land use. Nitrate-electrical conductance plots revealed seasonal shifts in stormflow nitrate modes, indicating nitrate flushing behavior at the beginning of the winter wet season for all watersheds except the two with the highest agricultural land usage. Despite variation in land uses between watersheds, we found a consistent reduction in the variability of stormflow nitrate (∼12%) and electrical conductance (∼21%) modes relative to base flow modes indicating a common water and nitrate source during periods of high runoff. We propose the undeveloped, mountainous upland regions of the watersheds as this source, and suggest that this region plays an important role in determining watershed stream nitrate concentrations and nitrate flux to the Santa Barbara Channel (Pacific Ocean).

1. Introduction

[2] Agriculture and urbanization have increased the amount of biologically available nitrogen in streams through the application of fertilizer, aging and leaky water supply and sewage infrastructure, and creation of impervious landscapes [Galloway et al., 2003; Vitousek et al., 1997]. Biologically available nitrogen exists in three dissolved forms in stream water: ammonium (NH4+), dissolved organic nitrogen (DON), and nitrate (NO3). Stream nitrogen composition is a product of inputs from land uses in the watershed and biogeochemical processes [Lischeid, 2008]. Cation adsorption of ammonium to clay minerals [Dillon and Molot, 1990], nitrification of ammonium to nitrate [Vitousek et al., 1982], and mineralization of DON to ammonium and uptake by plant and microbial communities or subsequent microbial nitrification [Hill and Warwick, 1987; Peterson et al., 2001], are biogeochemical processes that can lead to nitrate being the most abundant form of dissolved nitrogen in streams. In addition, while agricultural and urban land uses increase the concentration of all three forms of dissolved nitrogen, nitrate concentration often has the strongest positive correlations with land use and human population size [Caraco and Cole, 1999].

[3] Comparative studies in streams of concentration versus runoff relationships (i.e., C-Q) among watersheds of varying land use can provide insight into how watershed hydrology and land use interactively regulate nitrogen inputs to stream water and downstream aquatic ecosystems [DeFries and Eshleman, 2004; Jordan et al., 1997; Meixner and Fenn, 2004; Shields et al., 2008]. The simplest models have identified two end-member sources that contribute solutes to streams during times of low runoff (i.e., base flow) and high runoff (i.e., stormflow), and are typically identified as groundwater and precipitation sources, respectively [Burns et al., 2001]. Inclusion of a third end-member, soil water, is often found to be important [Christophersen et al., 1990; Hooper et al., 1990]. Most of the studies that employ mixing models have focused on streams draining predominantly undeveloped watersheds [Poor and McDonnell, 2007]. Watersheds characterized by urban and agricultural land uses can complicate interpretation of mixing models, as these land uses have other potential solute sources, including potable water and sewage from leaking infrastructure [Caraco and Cole, 1999; Lerner, 1986; Silva et al., 2002; Yang et al., 1999], impervious surface runoff from buildings and roads [Driscoll et al., 2003; Jaworski et al., 1997], and agricultural irrigation water [Carpenter et al., 1998; Di and Cameron, 2002]. In such cases, a hyperbolic equation may have the requisite flexibility in fitting C-Q relationships [Barco et al., 2008; Godsey et al., 2009]. A hyperbolic equation generates three parameters: a hyperbolic slope parameter that identifies dilution, invariance, or enrichment behavior, and two solute concentrations, which are base flow and stormflow solute concentration modes. Dilution, invariance, and enrichment occur when the rate of solute mobilization per volume of water input to a stream decreases (dC dQ−1 < 0), remains constant (dC dQ−1 = 0), or increases (dC dQ−1 > 0), respectively, as runoff (Q) increases [Salmon et al., 2001]. The base flow and stormflow modes are estimates of the most frequently observed solute concentration at low runoff and high runoff, respectively. Coupled with additional hydrochemical and land use information, comparative analyses of C-Q behavior and base flow and stormflow solute concentration modes can allow important insights into probable solute sources and biogeochemical mechanisms.

[4] In semi-arid Mediterranean climatic regions, nitrogen mobility in undeveloped watersheds is partly regulated by the timing and magnitude of precipitation [Ávila et al., 1992; Butturini and Sabater, 2002; Poor and McDonnell, 2007]. The dry season, with little to no measurable precipitation, causes depletion of soil water and shallow groundwater stores through streamflow and evapotranspiration [Ávila et al., 1992; Biron et al., 1999]. The arrival of the wet season replenishes soil and groundwater stores, but to a variable extent depending on the number of precipitation events and their magnitude [Beighley et al., 2008; Butturini and Sabater, 2002; Latron et al., 2009]. These pulsed and irregular precipitation dynamics make the timing and amount of water delivery to streams a function of watershed soil moisture deficits [Bernal et al., 2004; Chamran et al., 2002; Latron et al., 2009], which dictate the degree of hydrologic connectedness of streams with their soil water and groundwater solute sources. Soil microbial nitrogen processing, which is coupled to drying and wetting, exerts some control on the timing and amount of nitrogen delivery to streams [Belnap et al., 2005; Fierer and Schimel, 2002; Miller et al., 2005], as does nitrogen build-up during the dry season, which can contribute to nitrogen flushing at the onset of the wet season [Meixner and Fenn, 2004; Sobota et al., 2009].

[5] Streams are not only laterally connected to the landscape, but also are longitudinally connected as they flow from headwater regions to lower reaches. In some situations, headwater streams can contribute an average 70% of the mean annual water flux and 65% of the nitrogen flux to lower reaches and coastal oceans [Alexander et al., 2007]. Hence, alteration of headwater regions can induce downstream ecological and biogeochemical changes such as eutrophication and hypoxia, and impairment of drinking water quality [Freeman et al., 2007; Mitsch et al., 2001]. The presence of urban and agricultural land uses can alter the hydrologic pathways that deliver nitrogen compounds to streams and watershed stores of nitrogen [Basu et al., 2010; Jordan et al., 1997], introducing additional solute sources that can change hydrologic and biogeochemical watershed functioning. It is therefore important to understand the factors that control the degree of hydrological linkage of the headwater regions to their lower receiving water bodies, and the biogeochemical and land use processes that influence solutes that these headwater regions deliver, in order to understand potential ecological alterations and water quality impairments to downstream ecosystems.

[6] Our goal is to identify variation in nitrate-runoff relationships in coastal California watersheds with varying amounts of agricultural, urban/suburban, and undeveloped area (Figure 1 and Table 1). Nitrate is the focus of our investigation because it comprises the largest fraction of total dissolved nitrogen. We generate hyperbolic equation parameters for nitrate-runoff relationships for each watershed to derive their C-Q type (i.e., dilution, invariance, and enrichment) and base flow and stormflow nitrate modes. To assess source water mixing dynamics, we also use the hyperbolic equation to generate C-Q types and base flow and stormflow electrical conductance (EC) modes for EC-runoff relationships. In light of the high inter- and intra-annual variability in precipitation characteristic of the Mediterranean climate in coastal California, we assess the suitability of using only one stormflow nitrate mode to characterize stream nitrate concentrations during high runoff using nitrate-EC plots. We determine if a comparison of inter-watershed variation in nitrate-runoff and EC-runoff relationships, and nitrate-EC plots, allows us to identify the most probable sources of water and nitrate to streams given our knowledge of the spatial variation of land uses.

Figure 1.

Locations of watersheds included in study. Land use is indicated by color: urban is black, agricultural is green, and undeveloped is white. Note that bare rock and urban/suburban land use have undifferentiated spectral signatures, hence the apparent urban/suburban land use in the upland (northern) region of the watersheds (black speckling pattern). This land use is in actuality bare rock (i.e., undeveloped).

Table 1. Watershed Topography and Land Use
 RattlesnakeMissionArroyo BurroCarpinteriaBell CanyonFranklin
Land use classUndevelopedUrbanUrbanAgriculturalAgriculturalAgricultural
Area (km2)8.2930.0425.4439.2115.8211.62
Relief (m)933120111881405928504
Mean slope (deg)21141521179
Percent urban023201120
Percent agriculture03891921
Percent undeveloped1007472908058

2. Methods

2.1. Physical Setting and Climate Characteristics

[7] The watersheds we sampled are located on the south-facing slopes of the coastal Santa Ynez Mountains (California) and are characterized by mountainous headwaters that drain across mildly sloping, narrow coastal plains and outlet into the Santa Barbara Channel (Pacific Ocean) (Figure 1). Over 80% of annual precipitation typically falls between December and March (Figure 2). The relief (Table 1), southern orientation of the watersheds, and the predominant south-southwest winds with the passage of storm systems during the rainy season cause substantial orographic enhancement of rainfall [Beighley et al., 2003]. Mean annual precipitation at a ridgeline gauge (1000 m asl) during our period of study from 2002 to 2008 was 210% greater than that of a low elevation (30 m asl) gauge (Figure 2). Inter-annual rainfall variability was high; from 2002 to 2008 mean annual rainfall ranged from a low in the 2nd percentile in 2007 to a high in the 96th percentile in 2005 of the long-term (1868–2008) precipitation record (Table 2).

Figure 2.

Mean monthly precipitation for three Santa Barbara precipitation gauges (water year ranges) located at elevations of 30 m (1868–2008), 364 m (1951–2008), and 1,000 m (1967–2008).

Table 2. The 2002–2008 Total Annual Precipitation for the City of Santa Barbaraa
Water YearAnnual Total (cm)Percentileb
  • a

    Elevation of Santa Barbara is 30 m asl.

  • b

    Percentile is calculated for water years 1868–2008.

200222.98.8
200363.482.5
200427.214.6
200593.896.4
200657.074.5
200716.32.2
200844.859.1

2.2. Field Measurements

[8] Stream water level was measured at a 5-minute frequency near the watershed outlet of each stream using a pressure transducer (Solinst Canada Ltd., Georgetown, Ontario) corrected for atmospheric pressure. These water level records were converted to stream discharge using geomorphic profiles surveyed for each stream and a hydraulic computation program, Hydrologic Engineering Center-River Analysis Program (HEC-RAS) [U.S. Army Corps of Engineers, 2005]. Runoff was calculated as discharge divided by watershed area. Base flow and stormflow measurements were averaged over 12-h and 1-h periods, respectively. Rain was collected during storm events in a sampler composed of a polyethylene funnel and 2-L bottle mounted 2 m above the ground. A 10 cm diameter manually read rain gauge was used to measure rainfall in association with sampling for rain chemistry.

[9] Base flow samples were collected weekly from November through May and biweekly from June through October. These samples were filtered in the field through Gelman A/E glass fiber filters (1 μm nominal pore size) into 60 mL high density polyethylene (HDPE) bottles triple-rinsed with filtered stream water before sample collection. Samples were stored at 4°C upon return to the laboratory.

[10] Stormflow samples were collected hourly on the rising limb of the hydrograph, and every two to four hours on the falling limb. Stormflow samples were collected either manually in triple-rinsed 500 mL HDPE bottles or automatically using ISCO 6712C portable water samplers (Teledyne Isco, Inc., Lincoln, Nebraska). The 500 mL ISCO sample collection bottles were triple rinsed with deionized water prior to deployment. Autosampler deployments typically lasted 20–24 h, and samples were iced during this period and stored at 4°C after retrieval.

2.3. Analytical Measurements

[11] Dissolved ammonium and nitrate concentrations were determined on a Lachat flow injection autoanalyzer (Hach Company, Loveland, Colorado). Ammonium was measured by adding base to the sample stream, which converted NH4+ to NH3. Nitrate was measured using the Griess-Ilosvay reaction after cadmium reduction. The detection limit for ammonium and nitrate was 0.5μM, sensitivity was ±0.2 μM and accuracy was ±5%. Filtered samples were assayed for total dissolved nitrogen (TDN) by alkaline persulfate digestion in an autoclave for 25 min followed by determination of nitrate as above. The detection limit for TDN was 1 μM, sensitivity was ±0.5 μM and accuracy was ±10%. Dissolved organic nitrogen (DON) was calculated as the difference between TDN and the sum of nitrate and ammonium. Specific conductance of unfiltered water was measured with a conductivity bridge (cell constant = 1.0) and readings were corrected to 25°C. A 1,400 μS cm−1 standard was used for calibration.

2.4. Land Use Analysis

[12] Land use classification was performed using data obtained by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS), which is an optical sensor that captures reflected spectral radiances from the Earth's surface in 224 contiguous spectral channels with wavelengths from 400 to 2,500 nm. Two AVIRIS flightlines were flown on August 6, 2004 for the Santa Barbara region. Imagery from these two flights was processed and merged to form one scene. Spectral libraries were generated, and an iterative selection of the spectra was performed to identify spectral signatures for individual land use classes of interest. Multiple end-member spectral analysis was then used to produce a land use and land cover map [Roberts et al., 1998].

2.5. Data Analysis

[13] Volume-weighted mean (VWM) concentrations were calculated for each watershed for their respective period of sampling (Table 3) using the following equation:

equation image

where Ci = stream water concentration (μM), Qi = watershed discharge (L hr−1), and N = total number of data points. Median absolute deviations (MAD) were calculated since frequency distributions of stream water solute concentrations often exhibited skewness and had outlier values that biased a measure such as the standard deviation [Dennis and Hirsch, 1993]. MAD is a robust measure of the variability of a uni-variate data set, and is defined as the median of the absolute deviations from the data set's median.

Table 3. Stream and Precipitation Volume-Weighted Means and Median Absolute Deviationsa
 Rattlesnake (2002–2008)Mission (2002–2008)Arroyo Burro (2002–2006; 2008)Carpinteria (2002–2005)Bell Canyon (2005–2008)Franklin (2002–2005)Precipitation (2002–2008)
VWMnVWMnVWMnVWMnVWMnVWMnVWMn
  • a

    VWM, volume-weighted means; MAD, median absolute deviations.

Ammonium (μM)0.69862.911302.68481.98151.03564.17884.175
Ammonium MAD0.5 7.3 4.6 8.8 0.7 5.8 7.1 
Nitrate (μM)68.298767.9113192.0852110.6815228.6356407.07886.475
Nitrate MAD31.8 32.2 37.5 215.3 507.7 584.8 8.7 
DON (μM)19.297834.8111740.284240.481238.9355127.57838.175
DON MAD10.0 25.9 20.8 61.6 159.9 177.0 24.0 
TDN (uM)88.1978105.61117134.7842153.0812268.6355538.678318.675
TDN MAD35.6 51.7 47.2 255.9 566.2 613.7 33.8 

[14] Electrical conductance has been used by researchers as a conservative tracer to identify the contribution of pre-event water (also referred to as base flow or groundwater) and event water (also referred to as precipitation) to stream water during storms in both undeveloped [Laudon and Slaymaker, 1997; Pilgrim et al., 1979] and urban watersheds [Pellerin et al., 2008]. Studies using EC to distinguish these two water sources have reported relatively small differences (<50 μS cm−1) between groundwater and precipitation values in temperate, forested watersheds, creating uncertainty and ambiguity in assigning relative contributions of the two sources [Laudon and Slaymaker, 1997; Matsubayashi et al., 1993; McDonnell et al., 1991]. In watersheds impacted by urban and agricultural land uses, base flow EC values are often elevated by 1–2 orders of magnitude relative to precipitation EC values, reducing the uncertainty in two-source separations, but complicating identification of potential water and solute sources, as there are often a greater number of these sources and delivery pathways in urban and agricultural than in undeveloped watersheds. However, the mathematical form of EC-runoff relationships within individual watersheds and the identification of potential base flow and stormflow EC modes, together with additional hydrochemical information, can help determine potential solute sources during periods of low and high runoff.

[15] Nitrate-runoff and EC-runoff relationships were investigated using locally weighted scatterplot smoothing (LOWESS). LOWESS is an exploratory data analysis technique that assumes no pre-determined functional form of these relationships [Cleveland, 1979; Cleveland and Devlin, 1988]. The main difference between the LOWESS technique and other polynomial regression techniques is that local, rather than global, first degree polynomial fits are computed, allowing potential nonlinearity of relationships to emerge. A bandwidth parameter (f) is set by the data analyst, from 0 to 1, which controls the “smoothness” of the LOWESS fit. For each runoff value in the data set, xi, a local first order polynomial is fit to [fn] of the data points whose abscissas are closest to xi, where n is the total number of x-y pairs. Each of these local fittings is performed by weighted least squares in which data points close to xireceive large weight and those farther away receive lower weight. LOWESS fits allow for the selection of appropriate parametric models that can reproduce the form of the fits, aiding mechanistic interpretation of bi-variate patterns. Bandwidths of 0.5 were used to generate LOWESS fits for nitrate-runoff and EC-runoff scatterplots (Figures 3a and 4a). This bandwidth was chosen based on plotting three bandwidths (0.25, 0.50, and 0.75) followed by visual inspection of the LOWESS fit for each bandwidth. A bandwidth of 0.5 adequately represented the nonlinearity of the solute-runoff patterns without over-smoothing of the data or distortion of the trend [Cleveland, 1979]. LOWESS procedures were implemented in MATLAB (The MathWorks, Inc., Natick, MA).

Figure 3.

(a) Locally weighted scatterplot smooths (LOWESS) of nitrate concentration (μM) versus runoff (mm hr−1) for watersheds delineated by land use class; undeveloped (dash-dotted), urban (solid), and agricultural (dashed) watersheds. From top to bottom, the smooths correspond to Franklin, Bell Canyon, and Carpinteria. (b–d) Scatterplots of nitrate versus runoff, LOWESS smooths (dashed), and hyperbolic equation fits (solid) for Rattlesnake (Figure 3b), Mission (Figure 3c), and Bell Canyon (Figure 3d) watersheds. Note that the apparent large deviation between the LOWESS smooth and hyperbolic equation fit for lower nitrate-runoff data points for Rattlesnake watershed is a function of the log-log scale, as the magnitude deviation in y-intercept nitrate concentrations is only ∼7μM.

Figure 4.

(a) Locally weighted scatterplot smooths (LOWESS) of electrical conductance (μS cm−1) versus runoff (mm hr−1) for watersheds delineated by land use class; undeveloped (dash-dotted), urban (solid), and agricultural (dashed) watersheds. (b–d) Scatterplots of electrical conductance versus runoff, LOWESS smooths (dashed), and hyperbolic equation fits (solid) for Rattlesnake (Figure 4b), Mission (Figure 4c), and Bell Canyon (Figure 4d) watersheds.

[16] A hyperbolic equation was used to fit nitrate-runoff and EC-runoff relationships (equation (2)). This equation was selected based on examination of the shape of the LOWESS fits (Figures 3a and 4a), which were log-sigmoidal as opposed to log-linear, indicating the appropriateness of a hyperbolic function rather than a power function in fitting both nitrate-runoff and EC-runoff relationships [Godsey et al., 2009]. In addition, the flexibility of a hyperbolic equation in fitting bi-variate relationships affords the ability to examine intra- and inter-watershed variation in generated parameters such as slope, base flow mode (i.e., low runoff), and stormflow mode (i.e., high runoff)

equation image

In equation (2), CS (CS-EC) = stormflow nitrate or EC mode (μM or μS cm−1), δ (δEC) = base flow nitrate or EC mode (CB or CB-EC; μM or μS cm−1) minus CS (CS-EC), β (βEC) = hyperbolic fitting parameter, and Q is watershed runoff (mm hr−1). Equation parameters were derived using nonlinear least squares fitting procedures implemented in MATLAB's Curve Fitting Toolbox.

[17] Nonlinear least squares fits and 95% confidence intervals (CI) were computed for the three hyperbolic parameters, δ (δEC), CS (CS-EC), and β (βEC) (Tables 4 and 5). The F-statistic was used to test the validity of using the hyperbolic equation [Johnson et al., 1969]. To accomplish this, the equation was made linear by taking 1/1 + βQ as the independent variable, and nitrate concentration or EC value as the dependent variable. The slopes of the relationships are equal to δ (δEC), and significant slope effects were defined as δ (δEC) > or < 0 at the p < 0.05 level of significance (Tables 4 and 5).

Table 4. Hyperbolic Model Fits for Nitrate-Runoff Relationships
 RattlesnakeMissionArroyo BurroCarpinteriaBell CanyonFranklin
CS85.265.490.293.1245.8289.8
CS 95% CI lower78.460.383.859.7175.7182.3
CS 95% CI upper92.070.596.7126.4316.0397.2
β8.472.0100.769.8140.436.9
β 95% CI lower5.9−961.2−201.836.783.022.7
β 95% CI upper10.91105.0403.3103.0197.751.1
δ−85.02.4−18.9396.11374.01454.0
δ 95% CI lower−91.4−5.1−35.7352.11240.01322.0
δ 95% CI upper−78.59.9−2.1440.11509.01586.0
R20.483.7E-040.010.280.600.39
F-statistic925.10.48.9313.4520.0497.7
p-value<0.00010.52<0.01<0.0001<0.0001<0.0001
n9871131852815356788
Table 5. Hyperbolic Model Fits for Electrical Conductance-Runoff Relationships
 RattlesnakeMissionArroyo BurroCarpinteriaBell CanyonFranklin
CS-EC264.5392.0513.2443.8607.6352.8
CS-EC 95% CI lower242.1361.8442.2412.1497.2249.4
CS-EC 95% CI upper287.0422.2584.3475.6717.9456.2
βEC24.5254.358.9292.782.922.3
βEC 95% CI lower19.8172.938.5216.451.514.1
βEC 95% CI upper29.1335.779.3369.1114.230.6
δEC598.71016.01496.01161.02005.01269.0
δEC 95% CI lower572.9928.31357.01081.01832.01160.0
δEC 95% CI upper624.51105.01635.01240.02178.01377.0
R20.690.490.440.620.620.43
F-statistic2074.0672.2597.51222.6570.5468.1
p-value<0.0001<0.0001<0.0001<0.0001<0.0001<0.0001
n9151034806763399742

[18] Nitrate-EC plots were examined to identify potential seasonal shifts in base flow and stormflow nitrate modes, which if present could invalidate using fixed CB and CSvalues derived using hyperbolic nitrate-runoff relationships alone (Figures 3b–3d and 4b–4d). For each watershed, hyperbolic equation-derived base flow and stormflow EC modes (i.e., CB-EC and CS-EC) and their lower and upper 95% CI values were used to calculate nitrate VWM concentrations and MADs for nitrate samples that had EC values contained within the respective bounds (i.e., lower and upper 95% CI values) (Table 6). These nitrate VWMs were used to validate the hyperbolic base flow and stormflow nitrate values (i.e., CB and CS), and the MADs, in combination with qualitative assessment of nitrate-EC mixing behavior, were used to evaluate potential seasonal variability in base flow and stormflow nitrate modes. The use of MADs instead of standard deviations of nitrate concentrations contained within the upper and lower 95% CI bounds reduces the potential bias of assuming normality of the base flow (i.e., samples contained within the lower and upper CB-EC 95% CI) and stormflow (i.e., samples contained within the CS-EC lower and upper 95% CI) nitrate concentration distribution [Dennis and Hirsch, 1993].

Table 6. Nitrate VWMs and MADs Calculated Using Lower and Upper CS-CE and CB-CE 95% Confidence Intervals
 RattlesnakeMissionArroyo BurroCarpinteriaBell CanyonFranklin
  • a

    Here |(Nitrate VWMStormflow − CS) * 100| and |Nitrate VWMBase flow − CB * 100|.

Stormflow      
    Nitrate VWM92.568.7112.495.0231.0206.9
    Nitrate MAD32.839.141.7141.149.894.9
    Nitrate VWM − MAD59.729.670.7−46.1181.2112.0
    Nitrate VWM + MAD125.3107.8154.1236.1280.8301.8
    CS (hyperbolic fit)85.265.490.293.1245.8289.8
    Percent difference (absolute)a7.94.819.82.06.440.1
Base flow      
    Nitrate VWM23.359.582.0410.51315.81220.0
    Nitrate MAD10.025.121.9134.2309.2310.7
    Nitrate VWM − MAD13.334.460.1276.31006.6909.3
    Nitrate VWM + MAD33.384.6103.9544.71625.01530.7
    CB (hyperbolic fit)0.267.971.4489.21619.81743.8
    Percent difference (absolute)a99.114.112.919.223.142.9

[19] A two end-member mixing model was used to estimate lowland stream nitrate concentrations

equation image
equation image

In equations (3) and (4), CS = stormflow nitrate mode (μM), CU = upland stormflow nitrate mode (μM), CL = lowland stormflow nitrate mode (μM), PU = proportion of water export from the upland (from 0–1), and PL = proportion of water export from the lowland (from 0–1).

3. Results

[20] Ammonium, nitrate, and DON VWM concentrations (μM) ranged from 0.6 to 4.1, 67.9 to 407.0, and 19.2 to 127.5, respectively (Table 3). Rattlesnake Creek had the lowest VWM concentrations for all dissolved species except nitrate, which was only slightly lower in Mission Creek. Franklin Creek had the highest VWM concentrations. Nitrate was the predominant form of dissolved nitrogen in the streams, comprising 73.8 ± 7.3% (±1 SD) of total dissolved nitrogen (TDN) concentrations. DON was second highest, comprising 24.9 ± 6.5%. Ammonium was a minor fraction, comprising 1.3 ± 0.9%. Rain VWM concentrations of ammonium, nitrate, and DON were 4.1 μM, 6.4 μM, and 8.1 μM, respectively (Table 3).

[21] Based on MADs, nitrate was the most temporally variable dissolved nitrogen species, DON was the second most variable, and ammonium was the least variable (Table 3). Ratios of nitrate MAD to DON MAD ranged from about 1.2 to about 3.5. The two urban watersheds, Mission and Arroyo Burro, had the lowest nitrate MAD:DON MAD ratios of 1.2 and 1.8, respectively, while the agricultural and undeveloped watersheds had the highest nitrate MAD:DON MAD ratios (mean of 3.3 ± 0.2). This pattern can be seen in the LOWESS plot (Figure 3a), with nitrate concentrations for the two urban watersheds having lower variance over the runoff range than the agricultural and undeveloped watersheds.

[22] LOWESS fits and hyperbolic equation slopes (δ) indicated that nitrate-runoff relationships for five of the six watersheds were hyperbolic (Figure 3a and Table 4). The three agricultural watersheds—Franklin, Bell Canyon, and Carpinteria—had significant dilution (δ > 0). The undeveloped watershed, Rattlesnake, and one of the two urban watersheds, Arroyo Burro, had significant enrichment (δ < 0). The other urban watershed, Mission, was invariant (δ= 0). Arroyo Burro watershed had a lower p-value than four of the watersheds (Table 4), with a nitrate-runoff pattern more similar to Mission watershed's invariance rather than enrichment. In contrast to nitrate-runoff relationships, which had all three C-Q types (dilution, enrichment, and invariance), EC-runoff relationships for all six watersheds had only dilution (δEC > 0) and were highly significant (Table 5). The CS-EC values are a minimum of 16.8 to a maximum of 38.7 times higher than the precipitation EC VWM value of 15.7 μS cm−1 (MAD of 13.6 μS cm−1), as the CS-EC values represent a mixture of water sources during rainstorms.

[23] Nitrate VWMs calculated using CB-EC 95% CI bounds (i.e., base flow) and CS-EC95% CI bounds (i.e., stormflow) showed close agreement with the hyperbolic nitrate-runoff CB and CS parameters (Table 6). The base flow nitrate VWMs had a higher mean % difference relative to CB values (35.2%) than the stormflow nitrate VWMs relative to CS values (13.5%), reflecting mainly the large % differences for Rattlesnake Creek. Most of the base flow nitrate VWMs are lower than the CB values, which is a function of the hyperbolic equation identifying CB values at runoff = 0, whereas the nitrate VWMs are calculated for a range of EC values that occur at lowest runoff.

[24] Nitrate-EC plots had strong linear relationships for Franklin and Bell Canyon creeks (Figures 5a and 5b), and strong seasonal linear relationships for Rattlesnake Creek (Figure 6a). Therefore, in addition to stormflow CS (Table 4) and stormflow nitrate VWMs (Table 6), an additional estimate of the stormflow nitrate mode can be calculated using the linear relationships (Figures 5a, 5b, and 6a). Nitrate-EC plots for Mission, Arroyo Burro, and Carpinteria creeks did not have clear linear relationships, but did have seasonal effects evident in the triangular shape of the data points, indicating a third nitrate mode in the beginning storms of the season, shown as the dark blue (i.e., “Fall”) points on the nitrate-EC plots (Figures 6b, 7a, and 7b). Overall, nitrate-EC plots for all watersheds had patterns consistent with their nitrate-runoff relationships; dilution in agricultural watersheds (although a relatively less clear pattern for Carpinteria, discussed further below), invariance in urban watersheds, and enrichment in an undeveloped watershed (Figure 8).

Figure 5.

Nitrate versus electrical conductance plots for agricultural watersheds (a) Bell Canyon and (b) Franklin. Diameter of data points are proportional to runoff magnitude, except for the smallest diameter points, which have been enlarged for visibility. Linear regression equations are [Nitrate] = 0.56[EC] − 97.5 (R2 = 0.87) for Bell Canyon and [Nitrate] = 0.98[EC] − 78.75 (R2 = 0.88) for Franklin.

Figure 6.

Nitrate versus electrical conductance plots for (a) the undeveloped watershed Rattlesnake and (b) an agricultural watershed Carpinteria. Diameter of data points are proportional to runoff magnitude, except for the smallest diameter points, which have been enlarged for visibility. Seasonal time of year of sample collection is denoted by data point color. Linear regression equations for Rattlesnake are [Nitrate] = −0.16[EC] + 154.0 (R2 = 0.86) for the upper regression line and [Nitrate] = −0.16[EC] + 91.2 (R2 = 0.76) for the lower regression line.

Figure 7.

Nitrate versus electrical conductance plots for urban watersheds (a) Arroyo Burro and (b) Mission. Diameter of data points are proportional to runoff magnitude, except for the smallest diameter points, which have been enlarged for visibility. Seasonal time of year of sample collection is denoted by data point color.

Figure 8.

Nitrate versus electrical conductance plots for all watersheds grouped by land use: white data points with black outline are the undeveloped watershed (Rattlesnake), black data points are urban watersheds (Arroyo Burro and Mission), and gray data points are agricultural watersheds (Carpinteria, Bell Canyon, and Franklin).

4. Discussion

[25] Nitrate-runoff relationships were consistent within land use types; dilution in agricultural watersheds, enrichment in an undeveloped watershed, and invariance in urban watersheds. The consistency of these nitrate-runoff relationships within land use types implicates nitrogen inputs related to land use as a primary driver of these patterns. The percentage ratios of the standard deviation of CS to CB parameters and stormflow nitrate VWMs to base flow nitrate VWMs are ∼12%. The percentage ratio of the standard deviations of CS-EC to CB-EC values are ∼21%. These observed 4.8 to 8.8 fold reductions in variability, which are similar for both nitrate and EC, in spite of different land uses, implies different nitrate sources to streams during base flow (i.e., low runoff), but similar nitrate sources during stormflow (i.e., high runoff). Further, unlike humid temperate regions with small differences between base flow and stormflow EC values [Laudon and Slaymaker, 1997; Matsubayashi et al., 1993; McDonnell et al., 1991], these coastal California watersheds have base flow/stormflow EC differences of a minimum of 599 μS cm−1 to a maximum of 2005 μS cm−1, and all EC-runoff relationships demonstrate significant dilution during stormflow (Table 5). The strength of these hyperbolic EC-runoff relationships implies bimodal EC behavior, allowing us to examine nitrate-EC plots to identify probable land use and biogeochemical mechanisms underlying seasonal shifts in nitrate modes during periods of high runoff (i.e., stormflow). The following discussion addresses two fundamental questions that arise from the observed nitrate-runoff patterns with land use: how is the variability of stormflow nitrate modes reduced during high runoff despite divergent land uses, and how does seasonality (i.e., pronounced dry/wet dynamics characteristic of a Mediterranean climate) potentially alter nitrate-runoff patterns and stormflow nitrate modes?

4.1. Base Flow Nitrate Variability

[26] Base flow nitrate concentrations (i.e., CB) in the three agricultural watersheds were roughly 5 to 20 times greater than the two urban watersheds, and 67 to 241 times greater than the undeveloped watershed (conservative estimate using CB lower 95% CI values for agricultural watersheds and CB upper 95% CI values for urban and undeveloped watersheds). The elevation of agricultural CB values relative to urban and undeveloped watersheds indicates nitrogen fertilizers as the likely source of elevated base flow nitrate concentrations in agricultural streams. Nitrate is a common groundwater pollutant in agricultural regions, largely resulting from application of nitrogen fertilizer [Almasri and Kaluarachchi, 2004; Nolan et al., 1997; Spalding and Exner, 1993]. A proportion of these fertilizers either leaches through soils into groundwater [Di and Cameron, 2002], the primary source of stream water during base flow periods, or is directly discharged to streams through storm and/or tile drains. Estimates of mean nitrogen fertilizer application rates for agricultural land use in the eastern portion of our study region, which encompasses Franklin and Carpinteria watersheds, are 2,200 kg N km−2 year−1 for young (<4 years old) avocado orchards, 5,100 kg N km−2 year−1 for mature (>8 years old) avocado orchards, 36,400 kg N km−2 year−1 for greenhouses, and 39,000 kg N km−2 year−1 for nurseries [Robinson et al., 2005]. These fertilizer application rates are a minimum of 5 to a maximum of 574 times higher than estimates of other nitrogen inputs to watersheds, such as wet deposition (68 kg N km−2 year−1), dry deposition and plant foliar leaching (430 kg N km−2 year−1) [Schlesinger et al., 1982], and nitrogen fixation by Ceanothus communities (110 kg N km−2 year−1) [Kummerow et al., 1978], which are abundant members of plant communities in the upper watersheds.

[27] In the undeveloped watershed, Rattlesnake, base flow nitrate concentrations are the lowest of all watersheds (Table 4). There are two non-mutually exclusive explanations for these low nitrate values, the importance of each depending on the hydrologic pathways that feed streams during base flow periods. The first explanation is that nitrogen removal mechanisms (e.g., denitrification, plant and microbial uptake) reduce soil water and groundwater nitrate concentrations before percolation into the stream. The stormflow nitrate modes for Rattlesnake Creek (112.5 and 49.6μM) are a significant enrichment relative to the precipitation nitrate VWM of 6.4 μM and the base flow nitrate mode of 0.2 μM, indicating nitrate removal along the soil/groundwater flowpaths to the stream. The second potential explanation for the low base flow nitrate concentration in Rattlesnake watershed is that there are hydrologic pathways that bypass nitrate-enrichment in the soil. The Santa Ynez Mountains are composed of highly fractured arkosic sandstones and shales, with bedding planes that are nearly vertical due to intense faulting and folding [Rademacher et al., 2003]. These bedding planes act as groundwater conduits within the mountains, directing groundwater flow vertically and laterally. If recharge to these bedrock aquifers mainly bypasses upper organic soil layers, then we might expect nitrate concentrations of bedrock groundwater to be similar to precipitation nitrate values in the absence of appreciable microbial transformation, uptake, and/or organic matter mineralization within the aquifer. For our 2002–2008 study period, the nitrate VWM concentration of precipitation for 75 recorded storms was 6.4 μM, with a MAD of 8.7 μM. This value lies within Rattlesnake watershed's CB upper 95% CI value of 6.7 μM, suggesting that if a hydrologic connection exists between bedrock groundwater and the stream, then stream water nitrate during base flow periods may be sourced from bedrock groundwater inputs that have bypassed nitrate-enrichment in organic soil layers.

[28] The CB values for the two urban watersheds are intermediate between the agricultural and undeveloped values. It is difficult to estimate nitrate sources in these watersheds because urban land use is characterized by a diversity of activities and landscape properties that mobilize multiple nitrate sources [Spalding and Exner, 1993; Wakida and Lerner, 2005]. Dry weather flows that entrain atmospherically deposited nitrogen compounds from impervious surfaces [Driscoll et al., 2003; Jaworski et al., 1997], leaking sewage infrastructure [Caraco and Cole, 1999; Silva et al., 2002], and lawn and golf course fertilizers [Shuman, 2001; Wong et al., 1998] likely all contribute to elevated nitrate concentrations in streams during base flow periods. Storm drains are thought to discharge human waste to our urban watershed streams based on elevated fecal indicator bacteria and human-specificBacteroides marker counts, implicating leaky sanitary sewer lines [Sercu et al., 2009].

4.2. Reduction in Inter-watershed Stormflow Nitrate Variability

[29] The reduced variability of stormflow nitrate and EC modes relative to base flow nitrate and EC modes implies a common nitrate and water delivery mechanism for all watersheds despite their differing land uses, muting the high inter-watershed nitrate variability evident during base flow (Tables 4 and 5). Hydrological studies have identified the spatial variation in runoff-generating regions of our watersheds under varying land use scenarios.Beighley et al. [2008]examined storm runoff responses of a Santa Barbara watershed (Atascadero, urban classification under our scheme) over historical and projected ranges of land use conditions (1929, 1998, 2050) for a 14-year precipitation record (October 1, 1988 through September 30, 2002). They examined these storm runoff responses at three different scales: whole watershed, lowland regions (low elevation coastal plain), and upland regions (steep, mountainous headwaters). For the 1929 conditions, with relatively low urban land use (5% versus 39% and 50% for 1998 and 2050 conditions, respectively), 78% of mean annual watershed storm runoff originated from the uplands region, dropping to 51% under 2050 conditions. They concluded that the majority of runoff originating from the uplands is a product of (1) the orographic enhancement of precipitation along the Santa Ynez Mountains and (2) the geomorphology of the uplands region of Santa Barbara watersheds, characterized by steep slopes, thin soils, and rock outcroppings, which promotes saturated overland excess flow and shallow soil interflow that rapidly delivers water from hillslopes to upland streams [Latron et al., 2009].

4.3. Upland, Undeveloped Watershed Nitrate-Runoff Dynamics

[30] Rattlesnake Creek is an upland watershed (Figure 1). Unlike the other watersheds, Rattlesnake's nitrate-runoff behavior shows an enrichment (Table 4). Examination of the nitrate-EC plot for Rattlesnake Creek reveals two linear relationships, with nitrate-EC pairs along the upper regression line occurring earlier in the wet season than the pairs along the lower regression line, which occur later (Figure 6a). The slopes of the two regression lines are not significantly different from one another, but their intercepts are significantly different with the difference equal to 62.8 μM. This difference is likely the result of nitrate flushing, whereby soil nitrate that has accumulated in upper soil layers during the dry season is flushed during the first storms as water enters the previously dry, organic-rich soil horizons [Meixner and Fenn, 2004; Sobota et al., 2009]. A possible source of this nitrate is nitrogen-containing osmolyte compounds that soil microbial communities synthesize during the dry season to prevent desiccation [Fierer and Schimel, 2002; Miller et al., 2005]. Upon rewetting, the microbial cells release the compounds to the infiltrating soil water to maintain osmotic potential, as the hypertonicity of their cellular fluids would otherwise cause rapid water influx and cell bursting. Once released to the soil water, these substrates are available for mineralization and nitrification, increasing the nitrate concentration of soil water which is transported to the stream. Additionally, the wetting of the upper soil layers could mobilize non-microbial nitrogenous soil organic matter, making it increasingly accessible to microbial mineralization and nitrification and transport to the stream [Bernal et al., 2005; Butturini and Sabater, 2002].

[31] The distinct linear nitrate-EC relationships observed for Rattlesnake Creek suggests that the timing and magnitude of precipitation events interacts with the mountainous geomorphology of the upland watersheds to control hydrologic connectivity and soil microbial activity, which in turn regulates earlier versus later wet season nitrate flushing. Therefore, defining only one stormflow nitrate mode (85.2μM using the hyperbolic nitrate-runoff equation and 92.5μM using the lower and upper CS-EC95% CI bounds) appears to be a simplification of upland watershed nitrate-runoff behavior. Instead, two nitrate modes can be estimated using the two linear equations derived for the Rattlesnake nitrate-EC relationship and the stormflow CS-EC value (264.5 μS cm−1) (Table 6), yielding 112.5 μM for the non-supply limited (i.e., higher intercept) regression line and 49.6μM for the supply limited (i.e., lower intercept) regression line (Figure 6a). The supply limited regression line occurs later in the wet season, when the more labile and/or abundant soil nitrogen pool that supported enhanced nitrate delivery to the stream is likely exhausted.

4.4. Lowland, Urban Watershed Nitrate-Runoff Dynamics

[32] For the urban watersheds, Mission and Arroyo Burro, it appears that the nitrate-EC plots show early wet season nitrate flushing, with progressive decreases (for a given EC value) of nitrate concentrations with later season storms. Urban watersheds are characterized by a diversity of activities and landscape properties that mobilize multiple temporally variable nitrate sources over the course of storm events. The source of the early season nitrate flushing could be atmospherically deposited nitrogen compounds from impervious surfaces that accumulated during the dry season [Driscoll et al., 2003; Jaworski et al., 1997], nitrate from leaking sewage pipes and septic tanks [Caraco and Cole, 1999; Silva et al., 2002], or upland nitrate. The seasonal flushing evident in the urban watersheds appears to be well approximated by using the nitrate VWM ± MAD limits as stormflow nitrate modes, which correspond to 107.8 and 154.1 μM in Mission and Arroyo Burro watersheds, respectively, for earlier wet season storms, and 29.6 and 70.7 μM for later wet season storms. Mission watershed has higher impervious surface coverage than Arroyo Burro watershed, 12.7% versus 8.6% [Peters et al., 2005], which increases the contribution of surface runoff from the urban lowland region to the stream [Beighley et al., 2003; Beighley et al., 2008]. Arroyo Burro watershed has a higher percentage of agricultural land use than Mission watershed, introducing a potential nitrogen fertilizer source (Table 6), although this agricultural area is located to the north (upstream) of urban land use (Figure 1), and is likely diluted by impervious surface runoff before discharge to the ocean. Roughly 7 and 2 percent of Mission and Arroyo Burro watersheds, respectively, have septic systems instead of sewage connections [Hantszche et al., 2003]. While roughly a third the size in area, the density of septic systems in Arroyo Burro are twice that of Mission, and are situated closer to the watershed outlet to the ocean (i.e., on the coastal plain versus the foothills lowland-upland transition). The lower impervious surface area, higher percentage of agricultural land use, higher density of septic systems, and closer proximity of the septic region to the watershed outlet is the likely cause of a stormflow nitrate mode that is higher by 46.3μM (154.1 − 107.8 μM) for earlier wet season storms, and 41.1 μM (70.7 − 29.6 μM) for later wet season storms for Arroyo Burro watershed relative to Mission watershed. For both watersheds, the lowest EC values converge toward a nitrate mode of ∼25 μM, likely indicating an impervious surface contribution after the initial nitrate flushing, where nitrate and other solute sources are exhausted and stream nitrate and EC are influenced by mixing with a dilute water source (e.g., precipitation from impervious surfaces).

4.5. Lowland, Agricultural Watershed Nitrate-Runoff Dynamics

[33] In the agricultural watersheds, Franklin, Bell Canyon, and Carpinteria, linear nitrate-EC relationships are evident for Franklin and Bell Canyon watersheds (Figures 5a and 5b), but not for Carpinteria watershed (Figure 6b). This difference is likely a result of lower agricultural land use in Carpinteria (9%) versus Franklin (21%) and Bell Canyon (19%) watersheds. In the nitrate-EC plot for Carpinteria watershed, early season flushing is apparent with nitrate concentrations in the 750–1,500μM range, which is approximately 1.5- to 3.5-times above the CB value (489.2 μM) and base flow nitrate VWM (410.5 μM). This early season flushing is similar to that in urban watersheds, although the nitrate concentrations are higher. The seasonal difference may be the result of riparian and/or hyporheic denitrification removing groundwater nitrate before it enters the stream during low runoff [Ostrom et al., 2002; Seitzinger et al., 2006]. During the initial storms of the wet season, hydraulic gradients between the lowland groundwater and stream steepen, as groundwater mixes with oxygenated infiltrating soil water and residence times in the riparian and hyporheic zones decrease, causing reduced denitrification capacity and thus high nitrate concentrations in the streams. The flushing of high nitrate groundwater represents a shift from a high EC-high nitrate mode (∼1,500μS cm−1, 400 μM) in the beginning of the wet season to a low EC-low nitrate mode (∼700μS cm−1, 40 μM) as the wet season progresses. The seasonal behavior of the stormflow nitrate mode appears to be well approximated by using the nitrate VWM ± MAD limits, which correspond to 236.1 μM for earlier wet season storms, and 34.4 μM for later wet season storms. The later wet season nitrate mode is the minimum of nitrate concentrations within the CS-EC 95% CI bounds, as the nitrate VWM minus the MAD would equal a negative value.

[34] The linear nitrate-EC relationships for Franklin and Bell Canyon watersheds indicate little influence of seasonal drying/wetting dynamics on stream nitrate concentrations (Figures 5a and 5b). This is characteristic of intensively managed agricultural watersheds, whereby accumulated stores of nitrogen due to fertilizer use creates (1) transport versus supply limitation of nitrate delivery to streams and (2) delivery of water with relatively high and temporally similar nitrate concentrations that damps smaller magnitude nitrate variations due to other processes [Basu et al., 2010; Kemp and Dodds, 2002]. In contrast to the early wet season flushing observed in Carpinteria watershed, which elevates nitrate concentrations well above the base flow nitrate mode, Franklin and Bell Canyon watersheds do not show this behavior, indicating the limited extent of nitrate removal pathways (e.g., denitrification) due to nitrate saturation in the lowland riparian and hyporheic zones during low runoff periods [Bernot and Dodds, 2005; Mulholland et al., 2008]. Franklin watershed, which has 20% urban land use, has earlier wet season flushing of impervious surfaces, which is the reason that the EC-runoff relationship is not as strong as in Bell Canyon (Table 5). Using the nitrate-EC linear equations and CS-EC values for each watershed to determine their stormflow nitrate modes produces values of 265.7 and 240.9 μM for Franklin and Bell Canyon watersheds, respectively. These values agree well with CS values of 289.8 and 245.8 μM (Table 6).

[35] Overall, our examination of nitrate-EC plots for all watersheds indicates caution in using a single nitrate concentration (i.e., CS) to characterize stormflow nitrate modes, as high intra-annual variation in precipitation can seasonally alter these nitrate modes for undeveloped (Rattlesnake Creek), urban (Mission Creek and Arroyo Burro Creek), and lower intensity agricultural (Carpinteria Creek) watersheds. The hyperbolic nitrate (i.e., CS) and nitrate VWMs (Table 6) can be considered mean wet season nitrate modes, which in the case of Franklin and Bell Canyon, appear to sufficiently capture the dominant stormflow nitrate concentrations. For the other watersheds, accounting for seasonal nitrate flushing provides a more accurate depiction of dominant nitrate concentrations. Electrical conductance, despite potentially reflecting a variety of differing solute sources under varying land use, demonstrates significant dilution behavior in all watersheds (Figure 4 and Table 5) and a pattern of reduced variability in stormflow values versus base flow values similar to nitrate-runoff relationships (Tables 4 and 5). The 22% to 28% reduction in stormflow relative to base flow EC modes for the five lowland watersheds indicates that during storms, the predominant water and nitrate source to lowland streams is not lowland base flow. If base flow water sources were dominant, then dilution in all watersheds under variable land uses would not be evident (Figure 4 and Table 5), consistent with studies conducted in many humid temperate watersheds that show little variation between stormflow and base flow EC values [Laudon and Slaymaker, 1997; Matsubayashi et al., 1993; McDonnell et al., 1991]. Further, we would expect comparable variability in both stormflow and base flow EC modes between watersheds characterized by different land use mixtures, as base flow water sources often comprise a large proportion of storm runoff volume [Buttle, 1994; Kirchner, 2003]. In these coastal California watersheds, the magnitude of reduction in inter-watershed variability during storms is similar for both EC (SD-CS-EC: SD-CB-EC, 20.6%), a relatively conservative water source tracer, and nitrate (SDStormflow nitrate VWM: SDBase flow nitrate VWM, 11.3% and SD-CS: SD-CB, 12.0%), a biologically reactive ion.

4.6. Upland-Lowland Watershed Connectivity

[36] In light of the above considerations, and hydrological modeling results of Beighley et al. [2008], we hypothesize that the upland, undeveloped watershed is important in regulating whole watershed (i.e., combined upland and lowland) stream nitrate dynamics (Figure 1). Using seasonally refined stormflow modes, in concert with the hydrological modeling of Beighley et al. [2008], the potential role of the upland, undeveloped watershed regions in modifying stream nitrate concentrations during high runoff periods can be examined. We assume that the two stormflow nitrate modes for Rattlesnake Creek are representative of the undeveloped, upland watershed regions in the other five study watersheds during the earlier and later wet season (i.e., 112.5 and 49.6 μM). The stormflow nitrate modes for these five watersheds represent a mixture of both upland and lowland watershed nitrate sources, allowing identification of the lowland nitrate modes. We can examine the seasonal variability in these estimates for Mission, Arroyo Burro, and Carpinteria watersheds using the identified range in nitrate VWMs for the urban watersheds and the lower intensity agricultural watershed, where the high concentration (nitrate VWM + nitrate MAD) equals the earlier wet season nitrate mode and the low concentration (nitrate VWM − nitrate MAD) equals the later wet season nitrate mode.

[37] The results of the hydrologic modeling by Beighley et al. [2008]indicates that a minimum of 51% to a maximum of 78% of mean annual watershed runoff (i.e., over a 14-year period) is derived from the upland watershed regions for a range of historic, contemporary, and projected land use conditions in this coastal mountainous California region. The variance in mean annual upland water estimates is a function of urban land use, as increases in impervious surface cover in the lowland watershed regions progressively decreases the mean annual percentage contribution of the upland regions to whole watershed runoff. Using Beighley et al.'s 1929 and 2050 estimates of land use and mean annual upland runoff contributions, and assuming a linear relationship between urban land use percentage and mean annual upland contribution, we calculated mean annual upland contributions of 67% and 69% for Mission and Arroyo Burro watersheds, respectively, using AVIRIS-derived urban land use values (Table 1). For Mission watershed, the equation-derived value of 67% is similar to an empirical mean value of 62% for water years 2003, 2004, and 2005, during which annual precipitation was in the 83rd, 15th, and 96th percentiles, respectively, of the long-term (1868–2008) precipitation record for Santa Barbara. For agricultural watersheds, Franklin, Bell Canyon, and Carpinteria, we calculated upland contributions of 69% for Franklin (20% urban land use, 21% agricultural land use) and 78% for Bell Canyon and Carpinteria (both 1% urban land use, which is below Beighley et al.'s 5% 1929 urban land use, and therefore likely a slightly lower estimate of upland contribution).

[38] On an intra-annual timescale, per-storm upland water contributions are a function of the timing and amounts of antecedent precipitation, which in turn regulates watershed soil saturation and runoff-generation. This is demonstrated by increasing and plateauing upland water contributions for storm sequences in “wet” years 2003 and 2005, when annual precipitation was in the 83rd and 96th percentiles, respectively, of the long-term (1868–2008) precipitation record for Santa Barbara (Figure 9a). However, during “dry” years such as 2004, where annual precipitation was in the 15th percentile of the long-term precipitation record, upland water contributions remain low (Figure 9a). Using the range in upland water contributions for our three-year storm runoff series in Mission watershed, in concert with our seasonal stormflow nitrate modes (i.e., earlier and later wet season), we can identify the variation in lowland stormflow nitrate modes over the range in upland water contribution for our five developed watersheds. We represent these lowland nitrate modes as percentage changes relative to the whole watershed nitrate mode for each watershed (i.e., whole watershed nitrate modes which include the upland water and nitrate contribution) (Figure 9b). Doing so removes the upland water and nitrate contribution, revealing that lowland stream nitrate modes would be 3% lower to 6% higher for Mission, 2 to 219% higher for Arroyo Burro, 3% lower to 75% higher for Carpinteria, 5 to 582% higher for Bell Canyon, and 5 to 596% higher for Franklin than whole watershed nitrate modes. Using mean annual upland contributions from Beighley et al.'s model, calculated as a function of lowland urban land use, longer-term mean annual nitrate modes would be 3% lower for Mission, 64% higher for Arroyo Burro, 74% higher for Carpinteria, 270% higher for Bell Canyon, and 278% higher for Franklin. Mean annual lowland stormflow nitrate modes would also be 470% more variable (SDLowland nitrate modes: SDWhole watershed nitrate modes) with the removal of the upland contribution. The later wet season nitrate mode estimates for Mission and Carpinteria watersheds resulted in negative values for mean annual lowland nitrate mode calculations and much of the range in upland water contribution for intra-annual lowland nitrate mode calculations. Examination of nitrate results for Mission and Carpinteria watersheds revealed that later wet season nitrate VWM concentrations were on average 50μM and 49.4 μM, respectively, which we chose to use as the later wet season nitrate modes instead of the nitrate VWM − MAD estimates.

Figure 9.

(a) Upland water contributions for storm series in 2003, 2004, and 2005 for Mission watershed. (b) Lowland nitrate modes for five developed watersheds as a function of upland water contribution over the observed 2003 through 2005 range (8% to 88%). Lowland nitrate modes are presented as percentage changes relative to the whole watershed (i.e., upland and lowland combined) nitrate mode for given upland water contributions.

[39] As the percentage water contribution from the upland regions increases during the wet season, the lowland nitrate modes for all watersheds except Mission increase, off-setting the progressive decline in nitrate concentration of upland water during stormflow (Figure 9b). However, the lowland nitrate mode for Carpinteria plateaus around 75% upland water contribution, sharply declining thereafter, indicating supply limitation of nitrate to the lowland stream when the watershed is most saturated. Arroyo Burro and Carpinteria have roughly the same percentage of agricultural land use, and Arroyo Burro does not show a supply limitation of nitrate, but rather an enhancement of lowland nitrate at the highest upland contribution levels. This difference may be due to enhanced nitrate input from leaking septic tanks, which are located near the watershed outlet to the ocean in Arroyo Burro [Hantszche et al., 2003], and thus would be relatively un-diluted by impervious runoff or upland water during storms. Regardless of this difference in lowland nitrate behavior at highest upland water contributions, all watersheds except Mission have appreciable lowland nitrate enhancement as the watersheds become progressively saturated with successive storms through the wet season (Figure 9b). We attribute this effect to enhanced soil saturation and rising shallow water tables in the lowland watersheds, which would enhance mineralization and nitrification of previously dry soil organic matter, and enhance transport of nitrate originating from fertilizers and leaking sanitary infrastructure. The minimal lowland nitrate enhancement in Mission watershed may be a function of high impervious surface cover and stream channelization in the lowland region, which reduces soil and groundwater nitrate inputs by restricting soil infiltration, lowering water tables, physically blocking soil and groundwater stream input, and delivering a higher fraction of dilute water from impervious runoff.

[40] Our examination of nitrate-runoff relationships in six mountainous coastal California watersheds has revealed the following:

[41] 1. The stream nitrate-runoff relationships for six mountainous coastal California watersheds of varying land use were consistent within predominant land use class: enrichment in an undeveloped watershed (Rattlesnake), invariance in urban watersheds (Mission and Arroyo Burro), and dilution in agricultural watersheds (Bell Canyon, Franklin, and Carpinteria).

[42] 2. Nitrate-EC plots revealed seasonality in nitrate modes for the undeveloped (Rattlesnake), two urban (Mission and Arroyo Burro), and lower intensity agricultural watershed (Carpinteria), showing decreases in nitrate modes with the progression from the earlier to later wet season.

[43] 3. The variability of stormflow nitrate and electrical conductance (EC) modal concentrations and values, identified using a hyperbolic model, for these six watersheds were ∼12% and ∼21%, respectively, of base flow nitrate and EC modes, implying a common water and nitrate source during storms, which previous hydrological modeling studies and a three-year empirical runoff series for one of the watersheds (Mission) indicated was the undeveloped, upland watershed regions.

[44] 4. Removal of the water and nitrate contribution from the upland watershed regions using a two end-member mixing model revealed the lowlands stream nitrate modes for five developed watersheds on both multi-annual (i.e., 14-year mean annual upland water contribution) and intra-annual (i.e., empirical range in upland water contributions from three-year runoff series) timescales. In the absence of the moderating influence of the upland water and nitrate contribution, lowland mean annual nitrate modes would be 470% more variable than the whole watershed nitrate modes (i.e., combined lowland and upland), and would be 3% lower for Mission, 64% higher for Arroyo Burro, 74% higher for Carpinteria, 270% higher for Bell Canyon, and 278% higher for Franklin than the whole watershed nitrate modes. Lowland intra-annual nitrate modes, relative to the whole watershed nitrate modes, would be 3% lower to 6% higher for Mission, 2 to 219% higher for Arroyo Burro, 3% lower to 75% higher for Carpinteria, 5 to 582% higher for Bell Canyon, and 5 to 596% higher for Franklin. The lowest percentage changes correspond to the earlier wet season when the upland does not contribute much to the whole watershed runoff, and the highest percentage changes correspond to the later wet season, except in dry years, when the upland becomes a dominant contributor to whole watershed runoff and the lowland nitrate sources become increasingly activated and mobilized.

[45] The upland regions of the watersheds thus appear to play an important role in regulating stream nitrate concentrations by reducing inter-watershed variability in nitrate concentrations during storms, thereby making whole watershed nitrate concentration an integrated product of upland watershed hydrology and nitrogen cycling and lowland watershed nitrogen land use subsidies. However, the strength of the upland-lowland linkage is sensitive to climate in our semi-arid Mediterranean region, weakening in drier years and strengthening in wetter years, regulating the relative importance of the lowland versus the upland to watershed nitrogen cycling on intra-annual timescales. Nonetheless, this linkage may mitigate ecological and water quality impairments to downstream aquatic and marine ecosystems due to excessive lowland agricultural and urban nitrogen inputs during periods of highest water export (i.e., storms). In other mountainous watersheds of the world, the controls on watershed stream nitrate concentrations may similarly shift from downstream human land use during dry periods to the undeveloped, upland headwaters during storms.

Acknowledgments

[46] This work was supported by the Santa Barbara Coastal Long-Term Ecological Research project, funded by the National Science Foundation (OCE-9982105 and OCE-0620276). We thank F. Setaro, A. Doyle, and undergraduate assistants for laboratory analyses, A. Leydecker, S. Coombs, T. Robinson, and undergraduate assistants for sample collection, S. Peterson for AVIRIS land use classification, and C. Tague, S. Sadro, and A. Pischedda for their editorial and scientific comments. B. Goodridge received additional financial support from the Henry Luce Foundation.