Urbanization and Water Management Control Stream Water Quality Along a Mountain to Plains Transition

Urbanization can have substantial effects on water quality due to altered hydrology and introduction of constituents to water bodies. In arid and semi‐arid environments, streams are further stressed by dewatering as a result of diversions. We conducted a high‐resolution synoptic survey of two streams in Colorado, USA that transition abruptly from granitic/metamorphic forested mountains to sedimentary urbanized plains and are both highly managed for water supply, yet differ in degree of urbanization. A synoptic mass balance approach developed for mine drainage applications was adapted to elucidate effects of urbanization, geology, and diversions on stream chemistry during baseflow conditions. Urbanization was a more important driver of stream concentrations than geology. The urban area was a strong source of bromide, calcium, chloride, and manganese, while lanthanum and dissolved organic carbon were primarily sourced from the mountains. A majority of streamflow was removed by diversions near the mountains/plains interface. Groundwater accounted for 31% of the subsequent flow increase to the urbanized stream, and delivered at least 33% of chloride loading. Constituents that were primarily urban‐derived (bromide, calcium, chloride, and manganese) were 2–3 times higher in the urban region due to diversions; without diversions, stream water quality would have largely retained characteristics of forested streams through the urban reach. This study provides insights into processes that affect water quality in highly managed streams of the semi‐arid western USA.


Introduction
Urbanization can affect streams in a number of ways, including altered channel morphology, more varied hydrographs, increased specific conductance, elevated concentrations of contaminants and nutrients, and reduced biotic diversity (Conway, 2007;Meyer et al., 2005;Paul & Meyer, 2001).Such changes can create acute, localized water-quality issues and contribute to downstream, regional problems (Gabor et al., 2017;Stets et al., 2020).Increased salinity is commonly observed in urban areas and can contribute to corrosion of infrastructure, pollutant mobilization, and ecosystem degradation (Duan & Kaushal, 2015;Nelson et al., 2009;Stets et al., 2020).Changes to dissolved organic carbon (DOC) in urban areas is of long-standing interest because of its links to oxygen depletion and pollutant transport (Sickman et al., 2007).The multitude of co-occurring changes in these streams is often referred to as the "urban stream syndrome," which emphasizes that these ecosystems often experience multiple physical and chemical stressors simultaneously (Walsh et al., 2005).The myriad and overlapping ways in which urban streams are degraded underscores the need for better understanding of the processes driving water quality in these areas (Booth, 2005).
Although point-source discharges often dominate constituent loading to urban streams, diffuse nonpoint sources can also be important avenues for water-quality degradation (Moore et al., 2017).In contrast to point sources, which are typically understood to be single "end-of-pipe" inputs such as outfalls from wastewater treatment and industrial facilities, nonpoint inputs arise from diffuse runoff into surface waters or through the routing of surface inputs into subsurface flow systems.Impervious surfaces (e.g., roads, sidewalks, buildings) result in more efficient transport of runoff via stormwater diversion (Dunne & Leopold, 1978) and the weathering of these surfaces can be a source of constituents in urban areas (Connor et al., 2014).During storms, rapid runoff from impervious surfaces can result in rapid transfer of water, sediment, metals, nutrients, and mixtures of contaminants that have been called "chemical cocktails" (Booth, 2005;Kaushal et al., 2020;Walsh et al., 2005).Less is known about baseflow (non-storm) conditions, though this is a critical period to characterize stream-groundwater connections and aquatic conditions that dominate most of the time, particularly in arid and semi-arid environments (Gabor et al., 2017;Pilone et al., 2021).
In the Western U.S., water diversions for irrigation and public supply are extensive and can exacerbate urban water-quality degradation by dewatering stream channels and decreasing the amount of water available to dilute urban inputs.This process was recognized as the primary cause of high concentrations of contaminants of emerging concern downstream of wastewater treatment facilities during low-flow conditions in the Western U.S. (Rice & Westerhoff, 2017).However, less is known about how water diversions may affect water quality in areas dominated by nonpoint source inputs.
The primary goal of this study was to elucidate the sources of water-quality degradation in urban streams experiencing both stream channel dewatering from diversions and elevated constituent inputs from an urban corridor.This objective was addressed through a synoptic mass balance approach in which instream concentration and streamflow were combined to develop longitudinal (i.e., upstream to downstream) profiles of constituent load.A similar approach has been used in mountainous streams affected by acid-mine drainage to identify and quantify sources of contamination (Bencala & McKnight, 1987;Byrne et al., 2017;Kimball et al., 2002;Runkel et al., 2009Runkel et al., , 2013)), and a modified approach is used herein to consider the effects of diversions.This approach allows greater elucidation of non-point sources contributing to the load increase.The mass balance also allows investigation into the effects of stream diversions on constituent concentrations through the urban corridor.
This work builds upon an earlier investigation into the sourcing and contribution of nutrients, major ions, trace metals, and organic carbon which demonstrated strong longitudinal patterns in water quality in Boulder Creek, Colorado and attributed increases upstream of a wastewater treatment facility to nonpoint source urban inputs and (or) weathering of underlying sedimentary bedrock (Barber et al., 2006;Murphy et al., 2003).The current study employs much greater spatial resolution to allow more precise identification of the sources of water-quality constituents to Boulder Creek.In addition, Boulder Creek was compared to its less-urbanized tributary South Boulder Creek as a means of comparing the relative effects of urbanization.We focused the analysis on a suite of water-quality constituents-bromide, calcium, chloride, DOC, lanthanum, and manganese-that were selected to (a) serve as source tracers of urbanization or geology (e.g., chloride for road salt, lanthanum present in igneous/ metamorphic geology), (b) differ in biogeochemical reactivity (chloride is conservative, while manganese and DOC are not), and (c) have multiple potential sources (calcium can be derived from weathering of bedrock, concrete, or road salt).

Study Area
This study took place in the Boulder Creek watershed in the Colorado Front Range of the Rocky Mountains.The mainstem of Middle Boulder Creek/Boulder Creek was the primary focus of this study; we also evaluated its tributary South Boulder Creek (Figure 1).Boulder Creek is a tributary to the South Platte River and has a total watershed area of roughly 1,160 km 2 (Murphy, 2006).The Boulder Creek watershed includes two physiographic regions: the western portion of the basin is mountainous with a peak elevation of 4,120 m near the Continental Divide while the eastern portion of the basin has much less topographic relief with a bottom elevation of 1,480 m (Kimbrough, 1995).The physiographic regions differ in underlying geology, with the mountainous portion primarily underlain by Precambrian-age granitic and metamorphic rock, whereas the plains portion is primarily composed of Mesozoic-age shale and sandstone overlain by Quaternary-age alluvium and other unconsolidated deposits (Figure S1 and Table S1 in Supporting Information S1) (Cole & Braddock, 2009;Kellogg et al., 2008).
Mean annual precipitation is 400-600 mm in the plains and lower-elevation (<2,700 m) mountains, and increases with elevation in the higher mountains to 1,100 mm near the Continental Divide (Murphy, 2006;Murphy et al., 2015).Streamflow is primarily derived from snowmelt at the highest elevation, and thus peak annual discharge of Boulder Creek and South Boulder Creek occurs from May to July, with low-flow conditions occurring from October to March (Murphy, 2006;Murphy et al., 2003).Boulder Creek and its tributaries are subjected to a complex water management system.Several reservoirs store water throughout the watershed, with the largest reservoirs being Barker Reservoir (elev.2,490 m) on Middle Boulder Creek, and Gross Reservoir (elev.2,220 m) on South Boulder Creek (Figure 1).In addition to snowmelt, Gross Reservoir stores water carried to South Boulder Creek by a transbasin diversion from the western side of the Continental Divide.Downstream of these reservoirs, the streams flow through steep canyons before discharging onto the plains.Here, a complex network of regulated diversions remove varying amounts of water for municipal, industrial, and agricultural use based on water rights and daily streamflow (Crifasi, 2015).As a result, stream discharge varies widely both temporally and spatially along Boulder Creek; at some points nearly all the water in the creek is diverted out of the channel (Murphy, 2006;Murphy et al., 2003).Land cover shifts abruptly from forest to developed urban land at the mountain/plains interface of Boulder Creek; in contrast, after emerging from the mountains, South Boulder Creek primarily flows through grasslands and wetlands before entering urban land and converging with Boulder Creek.The presence of wetlands in the plains reach of the study area (Figure 1) is often a result of water storage or conveyance through the landscape (Crifasi, 2005(Crifasi, , 2015)).The Boulder Creek watershed has 12% urban land cover, whereas the South Boulder Creek watershed has 3.9% urban land cover (both percentages are based on area upstream of their confluence) (Kukkola et al., 2023).The study reach examined here ends upstream of the discharge location for the City of Boulder's wastewater treatment facility on Boulder Creek (Figure 1).

Stream Sampling and Analysis
The study approach was predicated on the assumption that streamflow and constituent concentration were temporally invariant at a given sample location over the course of the 4-day sampling campaign.Sampling occurred in fall when greater hydrologic stability results from typically lower precipitation, decreased evapotranspiration, waning snowmelt inputs, and reduced stream diversions (Murphy, 2006).Streamflow was generally steady during the study period (Kukkola et al., 2023).
Because diel variation in constituent concentration can occur under steady flow conditions due to pH-dependent reactions (Gammons et al., 2015;Nimick et al., 2007), we tested the assumption of hydrochemical stability via a preliminary diel sampling campaign prior to synoptic sampling.This campaign was conducted on 17-18 September 2019, at two locations that represent the range of concentrations along Boulder Creek.The upstream site (B12 in Figure 1) is located at the mouth of Boulder Canyon where the stream transitions from the mountain environment to the plains, immediately upstream of the urban area.The downstream site (B34; 14.4 km downstream of B12) is located at U.S. Geological Survey (USGS) gage 06730200 (U.S. Geological Survey, 2023), after the stream passes through the city of Boulder but upstream of the city's wastewater treatment facility.Ten water samples were collected from the thalweg at each location over a 36-hr period using a peristaltic pump.Temperature, pH, and specific conductivity were measured in situ using an EXO2 Sonde (http://exowater.com), with measurements verified by long-term specific conductivity probes installed at both locations.Data are available in Runkel et al. (2022).
Extensive stream reconnaissance was performed prior (4-14 October 2019) to synoptic sampling to carefully document surface inflows, water diversion structures, and various urban features along the entirety of each study reach (Runkel et al., 2022).Through this reconnaissance we selected 34 sampling locations along the Boulder Creek (BC) mainstem and 17 sampling locations along the SBC mainstem that bracketed potential source areas such as surface inflows (Figure 1).Each pair of adjacent mainstem sites define the upstream and downstream boundaries of a stream "segment" as defined herein.Seventeen surface inflows along the BC study reach and one surface inflow along the SBC reach were also selected, including named tributaries (e.g., Bear Canyon Creek) and small, unnamed culverts (Table S1 in Supporting Information S1; all surface inflows >1 L s 1 into BC between B1 and B32 were sampled, and the SBC study reach was generally devoid of surface inflows, with only one surface inflow identified).All sampling sites were assigned a distance corresponding to the distance downstream from the reservoir at the head of the study reach.
Synoptic sampling campaigns on each study reach spanned 2 or 3 days.The SBC sampling campaign began the morning of 15 October 2019 and concluded the following morning; the BC campaign began midday on 16 October 2019, and concluded October 18.The earliest and latest samples were collected at 08:55 and 18:00, respectively.Sample collection proceeded from downstream to upstream to minimize the effects of instream disturbances during sample collection.Sampling on SBC began near its confluence with BC and terminated near the outlet of Gross Reservoir (Figure 1).Sampling on BC began at the USGS gage near 75th Street and ended near the outlet of Barker Reservoir.Sequential replicates (a second sample collected immediately after the first) were obtained at four locations to assess analytical precision and sampling error (Section 3.3).Further, "overlap" samples were collected at the start of the second, third, and fourth sampling days (i.e., the last mainstem location sampled on one day was resampled at the beginning of the next day) in consideration of the multi-day sampling effort.These overlap samples were used to assess the effects of temporal variability (Section 3.1).

Water Resources Research
10.1029/2023WR035633 Samples were collected using 2-L HDPE containers which were triple rinsed with sample water prior to collection.Temperature was measured in situ using an alcohol thermometer.Streamflow was measured at all mainstem sites within a half hour of sample collection using Acoustic Doppler Velocimetry (ADV).Sample containers were shuttled to a nearby sample processing station where pH and specific conductivity were measured and various aliquots were prepared for laboratory analysis, generally within 1 hr of sample collection.Filtration was conducted using a peristaltic pump and 0.45-μm AquaPrep disk filters.Unfiltered and filtered aliquots for cation analysis were placed in acid-washed 125-mL HDPE bottles and acidified to pH < 2.0 using concentrated trace-metal grade nitric acid (HNO 3 ).Aliquots for analysis of DOC were prepared by filtering into 40-mL ambercolored borosilicate glass bottles that had been precombusted at 500°C for 4 hr.
Total recoverable concentrations of lanthanum and manganese were used in this analysis because the difference between filtered and unfiltered samples confirmed that a substantial fraction of each constituent existed in colloidal form in Boulder Creek (Runkel et al., 2022).Total recoverable bromide was used because preliminary comparisons showed that total recoverable bromide from the ICP-MS and those from ion chromatography compared favorably, while there were larger discrepancies between filtered bromide from the ICP-MS and those from ion chromatography.Dissolved values were used for calcium, chloride, and DOC in this analysis.
In this paper we focus our analysis on the reach of Boulder Creek between site B4 (downstream of North Boulder Creek) and site B32 (upstream of South Boulder Creek; Figure 1).Samples collected during a temporary flow increase in BC on the afternoon of October 17 (sites B13 through B19; Figure 1) were omitted from loading analyses because they would invalidate the steady-state assumption that underlies the synoptic mass balance approach.A sample collected at site B31 was also omitted due to unusual constituent concentrations which may be attributable to incomplete mixing of a nearby inflow.Total recoverable manganese concentrations at sites B21 and B24 were anomalously high and replaced by spatially interpolated values for the loading analysis.

Mass Balance Calculations
Previous applications of the synoptic mass balance approach have been conducted in headwater streams where the stream gains water over the length of the study reach and there are no substantial decreases in the longitudinal streamflow profile (e.g., Kimball et al., 2002;Runkel et al., 2013).Under those conditions, increases in observed load with distance indicate that a given stream segment is a source of constituent mass, whereas decreases indicate a loss of constituent mass due to biogeochemical reaction (Kimball et al., 2002).For the setting considered here, diversions cause large decreases in streamflow within some stream segments, precluding identification of constituent sources within the segment (i.e., the flow decrease causes a decrease in observed load that masks any increases in load due to sources).We therefore modified the mass balance approach so that constituent sources could be identified along the entire study reach.This modified approach aggregates observed and diverted loads, and the loading analysis is conducted using the aggregate amount of loading.This aggregate quantity is the "constituent load" described below (Section 2.3.1).The modified mass balance approach was used to (a) identify sources of streamflow and constituents, (b) estimate source (inflow) concentrations, and (c) estimate instream concentrations in the absence of diversions.Change in load and several derived calculations were used to provide insight into constituent source/sink behavior within a segment and along the entire study reach.

Source Identification
We calculated the observed load at each sampling site i from the product of streamflow (Q i ) and constituent concentration (C i ).In order to account for the load removed by diversions, we also calculated a diverted load, which is the product of the diverted streamflow and the constituent concentration at the nearest sampling site (C i ).We then summed the observed load for each site i with any diverted load in the segments upstream from that site Water Resources Research 10.1029/2023WR035633 to obtain the constituent load.We also calculated the cumulative load, which is the sum of all increases in constituent load (Kimball et al., 2002).That is, for a given stream segment, the cumulative load: (a) increased if the constituent load increased (if the segment contains a source); or, (b) remained constant if the constituent load decreased (if the segment is subject to biogeochemical reaction).When summed for each study reach (sites B4 through B32 for BC and S1 through S17 for SBC), cumulative load expresses the total amount of constituent loading, whereas the constituent load at the end of the study reach expresses the net amount of loading after the loss of mass due to biogeochemical reactions.
We also calculated the percent contribution of each stream segment to the cumulative load as: where Δ cumulative load is the within-segment cumulative load increase and L end is the cumulative load at the downstream end of the study reach.A related measurement is the cumulative streamflow, which only considers increases in streamflow and thus is indicative of the magnitude of hydrologic inputs to each study reach.
The study reach loading rate (load/km) was calculated as the increase in the cumulative load from the first sampling site to the final sampling site, divided by the length of the study reach, in km.Given the less urbanized nature of SBC (Section 2.1), the ratio of the BC and SBC study reach loading rates may be considered an indicator of the effect urbanization has on a particular water-quality constituent.Loading rate ratios >1 are indicative of increased loading rates within the more urbanized BC study reach relative to SBC (e.g., a loading rate ratio of 3 indicates there is 3 times as much loading for a given constituent in BC per km of stream length).
The inflow load at a segment (IL i ) was calculated from constituent concentration in the identified inflows (IC i ) multiplied by the change in streamflow in that segment of the study reach (ΔQ): The calculation assumes that ΔQ is entirely attributable to the identified inflow.In segments without an identified inflow, IL i was set to zero.The cumulative inflow load (CIL) was calculated by summing IL i over the entire study reach.Similarly, the cumulative inflow was calculated by summing ΔQ in segments with an identified inflow and expresses the identified hydrologic sources to the stream reach.
Comparing the cumulative load to the CIL provided insight into the relative importance of identifiable sources of constituents to BC versus constituent inputs arising from groundwater or biogeochemical reactions within the stream.When cumulative load > CIL, constituents were assumed to be sourced from groundwater or produced through instream reactions.In a similar way, when cumulative streamflow > cumulative inflow, we assumed that this difference was indicative of groundwater inputs to streamflow in the study reach.The difference between the cumulative load and the CIL was used to quantify the percentage of load entering each study reach that is attributable to unsampled input (i.e., the percentage of constituent mass entering via groundwater or produced by reactions, within the study reach): where L start is the constituent load at the start of the study reach.

Effective Inflow Concentrations (EIC)
We also used the EIC (Kimball et al., 2002) to evaluate surface and groundwater contributions to specific stream segments: where the u and d subscripts denote quantities at the upstream and downstream ends of the stream segment, respectively.EIC is the average constituent concentration necessary to result in the observed concentration Water Resources Research 10.1029/2023WR035633 increase or decrease from the beginning of the segment to the end of the segment.EIC is only defined for stream segments exhibiting an increase in constituent load; decreases in constituent load produce negative values of EIC.
Effective inflow concentrations are compared to observed inflow concentrations to assess groundwater inputs in Section 3.6.2.

Estimated Concentrations in the Absence of Diversions
Both study reaches have large managed diversions which decrease instream flow.Instream constituent concentrations downstream of these diversions are more easily influenced by inflows due to decreased dilution capacity.We tested this effect by calculating concentrations for a "no-diversions" scenario using the observed and diverted loads and streamflow estimates.The constituent concentrations that would be observed in the absence of the diversions at each stream sampling location were calculated by dividing the constituent load (observed plus diverted load) by the sum of the observed and diverted streamflow.

Principal Component Analysis (PCA)
A PCA of the concentrations of the six constituents was used to reduce the dimensionality of the survey results, as has been previously done in other studies of water-quality sourcing (Byrne et al., 2017;Christophersen & Hooper, 1992).This analysis allowed greater distinction between groups of samples based on the concentrations of all constituents simultaneously and provided insight into the sourcing, as well as the impacts of diversions.The data were log transformed to standardize the magnitudes of concentrations across constituents.The PCA was computed in R using the prcomp function (R Core Team, 2022).The initial PCA was conducted using the observed mainstem and inflow constituent concentrations.An additional PCA was conducted in which the observed mainstem concentrations were replaced with the estimated concentrations in the absence of diversions (Section 2.3.3) for comparison as if there were no diversions present.

Diel Sampling, Replicates, and Temporal Variability
During the diel sampling campaign, the coefficient of variation (CV) of all constituents was <25%, with calcium and DOC each being <5%, and bromide, chloride, and manganese being slightly higher (Table S2 in Supporting Information S1).Lanthanum had the highest CV at 22% at segment B12% and 14% at segment B34.These results generally support the validity of the assumption of steady-state conditions for these constituents during the synoptic sampling.
With the exception of one outlier, relative percent differences for the sequential replicate samples ranged from 0.2% to 5.9% (Table S3 in Supporting Information S1).These data suggest a relatively low amount of analytical and sampling error, lending credence to the loading analysis.
For the synoptic campaign, relative percent differences for bromide, calcium, chloride, and DOC in the overlap samples ranged from 0.3% to 2.7%, with larger differences observed for manganese and lanthanum (7%-21%; Table S3 in Supporting Information S1).These results generally support the validity of the assumption of steadystate conditions during the synoptic sampling.

Streamflow
Streamflow in BC averaged 0.45 m 3 s 1 in the mountain reach and decreased substantially in the upper urban reach when a diversion between sites B19 and B20 (24.8 km downstream of Barker Reservoir) diverted 85% of the stream water out of the channel (Figures 1 and 2a; Figure S2 in Supporting Information S1; Runkel et al., 2022).Streamflow in the urban area then gradually increased by 42% from 0.10 m 3 s 1 at B20 to 0.17 m 3 s 1 at B32, partly due to inputs from Bear Canyon Creek, Goose Creek, and numerous small inflows (Figure 1).
Streamflow in SBC at the first two sites downstream of Gross Reservoir averaged 2.5 m 3 s 1 (substantially higher than BC due to greater reservoir release) before a diversion between sites S2 and S3 (10 km downstream from Gross Reservoir) diverted 90% of the stream water out of the channel (Figures 1 and 2b; Figure S2 in Supporting Information S1; Runkel et al., 2022).Streamflow in SBC between sites S3 and S13 was 0.26 m 3 s 1 before decreasing to 0.063 m 3 s 1 at site S14 and 0.024 m 3 s 1 at site S16 due to additional diversions.

Water Resources Research
10.1029/2023WR035633 MURPHY ET AL.
Cumulative streamflow and cumulative inflow were similar along both streams in the mountain reaches.In the plains reaches, cumulative streamflow generally exceeded cumulative inflow (Figures 2a and 2b), which we attribute to subsurface inflows (i.e., shallow groundwater), and estimate that such inflows contributed 31% and 98% to the increases in streamflow for BC and SBC, respectively (Table 1).

Constituent Concentration Profiles
Concentrations of calcium, chloride, bromide, and manganese were consistently low in the mountain reach of Boulder Creek and were about 3-6 times higher in the urban area.Mean concentrations of calcium increased from 7.0 mg L 1 in the mountain reach to 20 mg L 1 in the urban reach; chloride from 5.8 to 35 mg L 1 ; bromide from 8.4 to 39 μg L 1 ; and manganese from 3.0 to 10 μg L 1 (Figures 2c and 2e; Runkel et al., 2022).Concentrations of these constituents began to increase noticeably in BC at 24.8 km, near the upstream end of the urban area and immediately downstream of the large diversion, and were highest at the most downstream site, B32 (Figures 1 and 2).Mean concentrations of DOC were 2.0 mg L 1 in both the mountain and urban reaches of BC, but showed more variation in the urban reach, initially decreasing in the urban area before reaching the highest DOC concentration (2.6 mg L 1 ) at the most downstream site (Figure 2c).Lanthanum concentrations in BC generally decreased  Water Resources Research 10.1029/2023WR035633 downstream, with mean concentrations higher in the mountain reach (0.10 μg L 1 ) than in the urban reach (0.07 μg L 1 ) (Figure 2e).
Concentrations of calcium, chloride, and bromide were also consistently low in the mountain reach of South Boulder Creek; concentrations were similar or slightly higher in the non-urban plains reach, and then were about 3-6 times higher in the urban reach.Mean concentrations of calcium increased from 4.2 mg L 1 in the mountain reach to 13 mg L 1 in the urban reach; chloride from 3.2 to 18 mg L 1 ; and bromide from 9.9 to 32 μg L 1 (Figures 2d and 2f; Runkel et al., 2022).Mean manganese concentrations decreased from the mountain reach (12 μg L 1 ) to the non-urban plains reach (6.7 μg L 1 ), but then increased in the urban reach (24 μg L 1 ) (Figure 2f).Concentrations of these four constituents increased substantially at the final SBC site (S17), located downstream of a small on-channel reservoir (Figures 1, 2d, and 2f).Mean DOC concentrations increased slightly from the mountain reach (3.0 mg L 1 ) (Figure 2d) to the non-urban plains reach (3.2 mg L 1 ), but were substantially higher in the urban reach (4.5 mg L 1 ).Lanthanum concentrations decreased downstream from the mountain reach (0.24 μg L 1 ) (Figure 2f) to the non-urban plains (0.16 μg L 1 ) and urban plains (0.12 μg L 1 ).
Mean concentrations of calcium, and chloride were higher in BC than in SBC, while those of manganese, DOC, and lanthanum were lower in BC than in SBC, whether all sites are included or sites are separated into mountain and plains reaches (Figure 2; Runkel et al., 2022).Mean bromide concentration was slightly lower in the forest reach of BC than in SBC, but substantially higher in the plains reach of BC than in SBC (Figure 2, Runkel et al., 2022).

Inflow Constituent Concentrations
Concentrations of constituents in the identified mountain inflows into BC were generally similar to concentrations in the mainstem, while concentrations in urban inflows varied widely (Figure 3; Runkel et al., 2022).Concentrations of bromide, calcium, and chloride in urban inflows were often considerably higher than in the mainstem.The longitudinal changes in chloride inputs exemplify this pattern, with inflows in the mountains all <30 mg chloride L 1 while all but three of the urban inflows were >30 mg chloride L 1 (Figure 3c, Table S1 in Supporting Information S1).Two inflows draining highly urbanized areas (IB16 [Bear Canyon Creek] and IB17 [Cottonwood Pond, which received inflow from Goose and Wonderland Creeks during the sampling event; it has since been converted to a stream channel with wetlands]) had concentrations of 148 and 166 mg chloride L 1 , respectively, and five other urban inflows (IB6, IB9, IB11, IB14, IB15) had chloride concentrations above 100 mg chloride L 1 (Figure 3c; Runkel et al., 2022), resulting in progressive enrichment of chloride in mainstem Boulder Creek.
The constituents DOC, lanthanum, and manganese, on the other hand, showed more complex behavior, with low/ diluting inputs prevalent in both the mountain and plains reaches in BC and a relatively small number of enriching inputs in the urban part of the study reach (Figure 3).Inflows with the highest DOC concentrations were IB16 and IB17 (Figure 3d), which have large urban drainage areas that include wetlands and open water (Figure 1, Table S1 in Supporting Information S1).Manganese concentrations were highest in IB17 and IB14 (Figure 3f).Lanthanum concentrations in urban inflows were generally similar to or lower than stream concentrations, with the exception of two inflows (Figure 3e): IB10 and IB16 (Bear Canyon Creek).(The EIC, also displayed in Figure 3, is described further in Section 3.6.)

Loading Analysis
Initially, observed loads of all constituents decreased substantially within the urban area of BC (Figure 4) due to 85% of streamflow being removed by the large diversion at 24.8 km (Figure 2).All observed loads then increased noticeably in the downstream urban reach (28-32 km, between sites B29 and B32), which receives input from Bear Canyon Creek and Goose Creek/Cottonwood Pond (Figure 1).The cumulative load plot emphasizes the downstream, urban sources of bromide, calcium, chloride, and manganese, while lanthanum was sourced primarily from the upstream mountainous reach.Dissolved organic carbon had mixed sourcing, with much of the load originating in the mountainous watershed and a secondary source within the urban reach (Figure 4).
Additional information about the source of constituents can be gained by comparing the cumulative load and CIL (Figure 4).When these values are similar, most of the observed loading can be explained by the sampled inflows, whereas large departures suggest the presence of other sources.These potential sources are collectively known as "unsampled input" (Equation 3, Table 2) and include unsampled surface inflows, biogeochemical reactions, and (or) groundwater inflow.For the case considered here, unsampled surface inflows are negligible given our Water Resources Research 10.1029/2023WR035633 detailed stream reconnaissance (all observed surface inflows between B1 and B32 were sampled; Section 2.2).Constituent mass added by biogeochemical reactions is also likely to be negligible given the short residence time within the discrete stream segments and the slow rates of weathering associated with the studied constituents.The difference in cumulative load and CIL is therefore attributable to groundwater inflow.In BC, bromide and calcium loading can be explained by the sampled surface inflows (unsampled inputs 20% or less), while chloride, DOC, lanthanum, and manganese loading from groundwater is substantial (unsampled inputs 33%-42%; Figure 4, Table 2).In contrast, almost all of the loading to SBC is derived from groundwater inflow (75%-97%; Table 2).
Loading rate ratios (Section 2.3.1)indicated approximately 9, 6, and 4 times as much constituent loading for chloride, bromide, and calcium, respectively, in BC versus the less urbanized SBC (Table 2).If the final urbaninfluenced segment of the SBC study reach (which includes a small on-channel reservoir) was excluded from the analysis, BC/SBC ratios under this scenario increased to 14, 8.7, and 4.5.These loading rate ratios exceed the BC/ SBC urban area ratio (12%/3.9%= 3.1; Section 2.1).

Percent Contributions
Greater than 70% of streamflow and loads of lanthanum and DOC in BC originated from upstream of the urban area, with >50% of each originating from above the study reach altogether (Figure 5).In contrast, most of the  4), which are not plotted.Length of each EIC line equals the segment length (Table S1 in Supporting Information S1); segments B12 through B20 were combined due to the temporary flow increase and B30 through B32 were combined due to exclusion of sample at site B31.
loads of bromide, calcium, chloride, and manganese originated in the urban area.Notable inputs were observed in urban segments B29 through B32, which drain subcatchments Bear Canyon Creek (IB16) and Goose and Wonderland Creeks (IB17) and include wetlands and open water (Figure 1).All of the constituents also gained mass loading from additional, numerous small inputs, emphasizing the nonpoint nature of urban inputs to BC (Figure 5).

EIC
Most previous applications of the synoptic mass balance approach have been conducted in small headwater streams where tracer-dilution methods are used to accurately determine the longitudinal streamflow profile.Given these accurate estimates of streamflow, uncertainty in the calculated values of the EIC (Equation 4) should  a Unsampled input is the difference between the cumulative load and the cumulative inflow load (Table S4 in Supporting Information S1), divided by the change in cumulative load over the length of the study reach.This input is assumed to be sourced from groundwater, as described in Section 3.5.b Study reach loading rate is the change in cumulative load over the length of the study reach divided by the length of the study reach.
be low, provided that laboratory analysis of the constituent concentrations is also accurate.Implementation of tracer-dilution methods in this study was impractical due to the spatial extent of the study reaches considered.Streamflow estimates provided by ADV are therefore used to calculate EIC, and these values are likely to be more uncertain than those calculated using tracer-dilution estimates, given the difficulties of ADV streamflow measurement in mountain streams.
Despite this potential added uncertainty, EICs present a useful means to estimate potential groundwater inflow to stream segments.Two specific cases are of interest.First, if the stream segment of interest includes a sampled inflow, the EIC can be compared to the observed inflow concentration to assess whether or not additional inflow waters are entering the segment.If the EIC exceeds the observed inflow concentration, diffuse groundwater with a concentration higher than the observed inflow concentration is likely to be entering the segment.When EIC is less than the observed inflow concentration, more dilute groundwater may be entering the segment and (or) some of the mass added by the sampled surface inflow has been lost due to instream reactions (i.e., causing the numerator of Equation 4 to be less than it would be without reaction).Bromide, calcium, chloride, and manganese EICs were much higher than observed inflow concentration at the most downstream urban segment of BC (sites B30-B32, at 28.9-30.9km, Figure 3).Although some of this discrepancy may be attributed to uncertainty in the ADV streamflow measurements, this uncertainty may be small given the relatively close correspondence between the EIC and the sampled inflow concentration for DOC (Figure 3d).Diffuse groundwater inputs of bromide, calcium, chloride and manganese are therefore likely along this 2-km segment.As a second case of interest, because the EIC represents the average concentration of all waters entering a stream segment, it provides an estimate of groundwater input to segments without sampled surface inflows.For example, in the last mountainous segment of BC, there were no sampled inflows and EICs for lanthanum and chloride are elevated (20.0-22.2km, Figure 3e).As such, the EICs for lanthanum and chloride provide estimates of the concentrations of inflowing groundwater in this area.

PCA
The first principal component (PC1) explained 53% of the variation and indicated that observed concentrations of bromide, calcium, and chloride in BC and SBC had strong positive correlations, lanthanum was negatively correlated with these ions, and DOC and manganese had weaker correlations to them (Figure 6a).The second principal component (PC2) was less predictive, explaining 28% of the variation, and demonstrated a tendency for manganese, lanthanum, and DOC to behave similarly.
The urban inflows generally had high values on PC1 and low values on PC2, while the mountain inflows were the opposite (Figure 6c).Likewise, the mainstem BC samples progressed toward resembling the urban inflows.For example, the most downstream segment of the study reach, B30-B32, resembled the urban inflow IB8.Sites along SBC showed a similar pattern, but were clustered more closely and had overall lower values along PC2, indicative of higher lanthanum, manganese, and DOC.The outlet of a small reservoir at the downstream end of SBC, S17, resembled the urban inputs to BC from Goose Creek, IB16, which also discharged from a small pond (Figures 1  and 6c).

No-Diversion Scenario
The "no-diversions" loading calculations indicated that concentrations of bromide, calcium, chloride, and manganese in the mainstem of BC were 2-3 times higher in the urban reach due to diversions during this sampling event (Figure 7).In contrast, lanthanum concentrations decreased because of diversions.The "no-diversions" scenario indicated that DOC concentrations would have remained more stable.The effect of diversions on concentrations will likely vary under different flow regimes.The PCA analysis also showed that without diversions, constituent concentrations in mainstem BC and SBC sites in the urban area would be much more similar to upstream conditions (Figure 6d).For example, all of the mainstem sites on SBC clustered together, indicating minimal changes in water quality throughout the study reach.Similarly, without diversions, the most downstream reach of BC would have a similar water-quality profile as inflow IB3, Fourmile Creek, which is a mountainous catchment with moderate development (Figures 1  and 6d).

Discussion
The processes causing water-quality degradation in urban streams often interact in complex ways (Kaushal et al., 2020), and likewise Boulder Creek displayed the combined effects of increased constituent input from urbanization and exacerbation of these effects by diversion of water out of the stream (Figures 2,4,6,and 7).
Behavior differed among the six water-quality constituents evaluated here, indicating a variety of processes determining their distribution throughout the study reach.The high-resolution synoptic sampling and subsequent analytical techniques enabled a detailed analysis of the sources of these constituents.

Effects of Urbanization on Water Chemistry
The urban reach of BC was free from the influence of major identified point sources such as industrial outfalls, wastewater treatment facility inputs, or agricultural irrigation returns, and yet had dramatic changes in concentration and load, with 70% of chloride and bromide and 60% of calcium being contributed in the urban region (Figure 5), despite only a 20% contribution of discharge.Concentrations of chloride and bromide increased by more than an order of magnitude in the urban reach (Figure 2).Prior work has attributed elevated chloride concentrations and loads to road salt, particularly in snow-dominated regions, and to leaky pipe infrastructure and atmospheric deposition (Bird et al., 2018;Corsi et al., 2015;Fanelli et al., 2019;Gabor et al., 2017;Kaushal & Belt, 2012;Kaushal et al., 2021;Moore et al., 2017;Stets et al., 2018Stets et al., , 2020;;Welty et al., 2023).All concentrations of chloride measured in the mainstems of BC and SBC were below the U.S. Environmental Protection Agency's secondary standard of 250 mg L 1 , but recent work suggests that lower concentrations may affect aquatic life, especially in dilute waters (Bird et al., 2018;Elphick et al., 2011).Bromide is a minor contaminant of halite, which is commonly used as road salt (Davis et al., 1998;Soltermann et al., 2016); the chloride/bromide mass ratios in BC increased from <700 to >1,100 at the most downstream site (Runkel et al., 2022), approaching the levels that have been identified for waters affected by road salt (Dailey et al., 2014).Concrete weathering is a source of calcium to urban surface waters (Connor et al., 2014;Moore et al., 2017), and dissolution of calcium from soils and concrete can also be enhanced through cation exchange in the presence of high sodium concentration due to road salt accumulation (Cooper et al., 2014).In the study area, applied road salt contains not only sodium but also calcium, magnesium, and potassium (Boulder County, 2023a; City of Boulder, 2023), and thus road salt itself is a potential source of calcium.
Shallow subsurface flow is likely an important vector of delivery for urban-associated constituents to BC.The alluvial aquifer underlying much of the urban area is closely connected with the stream (Kimbrough, 1995;Robson et al., 2000), and previous work has reported that water moves from the aquifer into the stream during low-flow periods (Babcock, 2007;Bruce & O'Riley, 1997;Hall et al., 1980;Kimbrough, 1995).Approximately 33%, 20%, and 12% of chloride, bromide, and calcium, respectively, contributed along the BC reach is attributed to groundwater (Table 2; Section 3.5).These values are likely underestimated, given the possibility that some of the sampled surface inflows include contributions from shallow groundwater.The infiltration of urban-derived constituents into the subsurface has been recognized in other areas as an important contributing factor to surface water contamination and can result in persistent, legacy behavior in these systems (Cooper et al., 2014;Corsi et al., 2015;Novotny et al., 2008).
While increased DOC in urban streams has often been linked to wastewater discharge (Barber et al., 2006;Edmonds & Grimm, 2011;Hosen et al., 2020;Murphy et al., 2003;Sickman et al., 2007;Westerhoff & Anning, 2000), findings on the effect of nonpoint urban sources on DOC concentrations have been inconsistent (Aitkenhead-Peterson et al., 2009;Hosen et al., 2020;Kaushal & Belt, 2012;Parr et al., 2015;Sickman et al., 2007).Concentrations of DOC in this study were remarkably constant along much of BC, with higher concentrations in the lowest urban reaches (Figure 2).Other studies have also found that DOC concentrations remain stable, but reported changes in DOC composition that reflect an increase in bioavailability and (or) a switch toward more autochthonously-produced material (Hosen et al., 2020;Parr et al., 2015).While we did not consider DOC composition in this analysis, it is plausible that such a transition occurred in BC.After BC exits the canyon, it is exposed to lower channel slope, slower velocity, and fringing reservoirs and wetlands (Figure 1), all of which promote in-stream production (Bernhardt et al., 2018;Hosen et al., 2019).Likewise, primary production and respiration, which are key drivers of DOC composition, were found to be comparatively high in BC (Reed et al., 2021).Increases in DOC in urban areas have been attributed to soil disruption by land use change, irrigation of turf grass, and accumulation of organic matter in storm drains, gutters, and other urban infrastructure (Aitkenhead-Peterson et al., 2009;Fork et al., 2018;Kaushal & Belt, 2012;Sickman et al., 2007).However, in this study, the predominance of wetlands at the most downstream urban sites is likely an important factor in increased DOC; wetlands often lead to higher downstream DOC concentrations (Aitkenhead et al., 1999).
Comparison of a highly urbanized stream (BC) to a much less urbanized stream (SBC) showed that land cover was a much greater controlling factor over water chemistry than was geology.An earlier synoptic study of Boulder Creek attributed increasing downstream concentrations of many inorganic constituents to urban inputs and (or) contributions from weathering of underlying sedimentary bedrock (Barber et al., 2006;Murphy et al., 2003), but did not discern between the two factors.Both streams transition abruptly from granitic bedrock in the mountains to alluvium overlying sedimentary bedrock in the plains (Figure S1 in Supporting Information S1; Murphy et al., 2003), but while BC transitions almost immediately to urban land, SBC traverses largely through grassland and wetlands for an additional ∼11 km before entering primarily urban land (Figure 1).Instream lanthanum concentrations, derived most commonly from granitic and related metamorphic rocks (Tyler, 2004), generally decreased downstream in both streams.Within the urban reach of BC, the greatest contribution of lanthanum was from segment B29-B30 (Figure 5), which includes inflow IB16 (Bear Canyon Creek), a tributary with headwaters in granitic foothills (Figure S1 in Supporting Information S1) that also conveys water diverted from BC near the mouth of the canyon (Holleran, 2000).The other geogenic constituents evaluated here (chloride, bromide, calcium, and manganese) increased substantially in the plains reach of BC, but remained relatively constant in the plains reach of SBC until the final urbanized segments (Figures 2 and 4).Loading rate ratios for chloride, bromide, and calcium in BC/SBC were 9, 6, and 4, exceeding the BC/SBC urban area ratio (3.1), and similar to a BC/SBC ratio for microplastics (4.2), another indicator of urbanization (Kukkola et al., 2023).In contrast, the loading rate ratio for DOC is close to unity, suggesting similar sourcing of this constituent within the BC and SBC watersheds (Table 2).

Exacerbation of Urbanization Impacts on Water Quality by Water Management
Water diversions exacerbated water-quality impacts related to urbanization.Previous work (Sprague et al., 2006a(Sprague et al., , 2006b) ) found that in contrast to other regions, urbanized streams in the Colorado Front Range did not show a strong link between urbanization and stream chemistry and ecosystem health; they suggested that water management was likely complicating the interpretation.The current study demonstrates that water diversions removed 85% of the water from BC in the urban area during this baseflow period (Figure 2a), which reduces the ability of the stream to dilute subsequently introduced constituents.We estimate diversions resulted in increased concentrations of urban-sourced pollutants by 2-3 times, while slightly reducing the concentration of lanthanum (Figure 7).The PCA analysis shows that if no diversions were present on BC or SBC, the sites would cluster more tightly, indicating less chemical alteration along the study reach (Figure 6, Section 3.7).
In addition to removing dilution capacity from the streams, the conveyance and storage of water in and through urban areas imparts a substantial change in watershed hydrology, with implications for surface and ground water chemistry.In arid and semi-arid regions, urbanization often increases the presence of surface water compared to undeveloped land (Steele et al., 2014).Groundwater levels in urban areas may be elevated by the introduction of water from other regions (Oswald et al., 2023;Paliwal & Baghela, 2007), affecting flux of subusurface water.
Water storage in small ponds and reservoirs (which are often remnants of gravel mining; Crifasi, 2005Crifasi, , 2015) ) at or near the mouths of tributaries that enter BC in the urban area (Figure 1) capture sediment from runoff and increase biogeochemical reactions between sediment and water and in the water column.These reservoirs increase local ground water level, often leading to wetlands (Figure 1).In addition, wetlands in the plains section of SBC are largely related to diversions carrying water through areas that were previously dry grassland (Crifasi, 2005(Crifasi, , 2015)).
Ponds and wetlands can alter biogeochemical cycling.Manganese cycles between reduced and oxidized phases, with the former being much more mobile in aquatic settings (Dean & Schwalb, 2002); reducing conditions in pond sediments can lead to very high Mn concentrations in surface water (Dean et al., 2003).In this study, the downstream urban segment between sites B30 and B32, which included a pond, contributed only ∼1% of streamflow, but 25% of manganese loading to BC (Figure 5).Ecosystem metabolism, and the associated cycles in oxygen concentration and reducing conditions, is high in BC (Reed et al., 2021), supporting the notion that oxidation-reduction reactions are important to elemental cycling in the study reach.Finally, stormwater control measures that are designed to reduce peak flows and pollutant flushes during storms can increase infiltration and increase solute concentrations in the alluvial aquifer, potentially leading to prolonged delivery of conservative elements (including chloride and bromide) to surface water during baseflow (Fanelli et al., 2019;Gabor et al., 2017;Welty et al., 2023).(This work did not evaluate the effectiveness of stormwater control measures in this watershed, of which there are many; City of Boulder, 2022).
A complex array of hydrologic and geochemical processes affected water quality of streams in an urbanized region of the Colorado Front Range.The Boulder Creek watershed, like many others in the western U.S., shows substantial annual and seasonal variation in climate and hydrologic regime.Our study occurred during baseflow conditions, which occurs the majority of time in arid and semi-arid regions.During high-flow periods, such as spring runoff, urbanization will play less of a role in stream chemistry and biogeochemical processing in BC due to greater dilution from snowmelt (Barber et al., 2006;Murphy et al., 2003;Reed et al., 2021); during this time, stream water moves into the alluvial aquifer (Babcock, 2007;Hall et al., 1980), providing dilution to groundwater.Additional spatial and temporal sampling would be needed to determine the stability of the water chemistry observed here (Abbott et al., 2018;Dupas et al., 2019).Future work could include stronger identification of the role of groundwater and reservoir storage in the introduction of metals via weathering and hyporheic exchange.A more nuanced understanding of the role of different local and regional attributes is needed to understand water quality in this region and elsewhere (Booth, 2005).

Conclusions
Urbanization is a well-known cause of water-quality degradation through inputs of pollutants from point and nonpoint sources.In regions where water supply is limited, water diversions can exacerbate effects of urbanization by decreasing dilution of incoming pollutants.Water management also affects hydrology and chemistry in other ways, by adding standing surface water, increasing groundwater levels, and adding wetlands in previously dry regions.While the processes leading to water-quality degradation in Boulder Creek, Colorado were complex, a similar combination of factors is likely to exist in many arid and semi-arid regions.Water conveyance structures meant to store and divert snowmelt for agricultural irrigation and drinking water are common not only in the Colorado Front Range but across the Western U.S. Therefore, many of the rivers and streams draining this region likely experience a similar sequence of alterations as Boulder Creek: stream channel dewatering followed by intensified inputs from developed areas in plains reaches.

Figure 1 .
Figure 1.Map of study area, showing (a) land cover in 2019 (Dewitz & U.S. Geological Survey, 2021), outlines of Boulder Creek and South Boulder Creek watersheds upstream of their confluence (U.S. Geological Survey, 2019), streams and main ditches (Boulder County, 2023b), and sampling locations; (b) detail for inset shown in a, with major roads (Boulder County, 2023b).U.S. Geological Survey streamgaging station 06730200 is located at site B34, and the City of Boulder's Wastewater Treatment Facility discharges to Boulder Creek 100 m downstream of the gage.See TableS1in Supporting Information S1 for more detailed information about the sampling sites and FigureS1in Supporting Information S1 for maps of land cover and geology of the Boulder Creek watershed.

Figure 2 .
Figure 2. Streamflow in (a) Boulder Creek and (b) South Boulder Creek; concentrations of chloride (Cl), calcium (Ca), and dissolved organic carbon in (c) Boulder Creek and (d) South Boulder Creek; concentrations of bromide (Br), manganese (Mn), and lanthanum (La) in (e) Boulder Creek and (f) South Boulder Creek.Dashed line indicates transition from mountains to plains; gray shading indicates urban area.

Figure 3 .
Figure 3. Observed constituent concentrations in the mainstem and inflows, and effective inflow concentrations (EICs) in Boulder Creek.(a) bromide (Br), (b) calcium (Ca), (c) chloride (Cl), (d) dissolved organic carbon, (e) lanthanum (La), (f) manganese (Mn).Gray shading indicates urban area.EIC is the average constituent concentration necessary to result in the observed concentration increase or decrease from the beginning of the segment to the end of the segment.EIC is only defined for stream segments exhibiting an increase in constituent load; decreases in constituent load produce negative values of EIC (Equation4), which are not plotted.Length of each EIC line equals the segment length (TableS1in Supporting Information S1); segments B12 through B20 were combined due to the temporary flow increase and B30 through B32 were combined due to exclusion of sample at site B31.

Figure 5 .
Figure 5. Percent contribution of streamflow and constituents to Boulder Creek identified by stream segment.Percentages were calculated in reference to cumulative flow or load at site B32, the last site upstream of South Boulder Creek (Figure1).Green colors represent inputs in upstream, forested stream segments while orange/tan colors represent inputs in downstream, urban stream segments.Segments that contributed >5% of any constituent are shown.Segments B12 through B20 were combined due to the temporary flow increase (Section 2.2); B30 through B32 were combined due to exclusion of sample at site B31.*B4 is treated as the first segment for this analysis.DOC, dissolved organic carbon.

Figure 6 .
Figure 6.Principal component analysis vector under (a and c) observed conditions and (b and d) "no diversion" scenario.Arrows represent direction of increasing urban influence.

Table 1
Cumulative Flow Increases in Boulder Creek and South Boulder Creek, Calculated as the Difference in Cumulative Streamflow at the Beginning and Ending of the Study Reach Note.Groundwater contributions were calculated as the difference between cumulative streamflow and cumulative inflow.

Table 2
Loading Metrics for Boulder Creek and South Boulder Creek