Effects of stream restoration by legacy sediment removal and floodplain reconnection on water quality

Nine sites with watershed areas of 0.33–8.18 km² were studied (figure 1, table 1). All sites were located in the Piedmont physiographic province, Maryland, USA and underlain by silica-rich, metamorphic bedrock. Watershed areas and land use characteristics were determined using US Geological Survey (USGS) StreamStats and land use data primarily coming from the 2011 National Land Cover Database [15]. Six restored sites were studied: three restorations were completed 1–3 years before the beginning of water quality sampling and three restorations occurred during the study period with samples collected pre- and post-restoration (table 1). All sites were restored by Ecotone, Inc. Restoration involved removing 1–2.5 m of sediment to substantially lower the floodplain, which at some sites involved removal of mature riparian forest 10 s of meters in width (more detailed site descriptions in supplementary information including land use maps and pictures, figures S1 and S2 (available online at stacks.iop.org/ERL/16/035009/mmedia)). Following removal of sediment, restored sites were planted with grasses and young trees. Three restored sites have substantial active cultivated crop agriculture in the watersheds (FMB, CBR, and BTR, table 1) with active agriculture within 10–100 m of the restored reaches at two sites (FMB and CBR). A fourth restored site (NSR) is primarily active pasture with a small riparian buffer. The fifth restored site (BCB) is currently suburban with ∼9% impervious surface cover (ISC) and active, ongoing development; previous land use at the restoration site was agricultural, and some agriculture remains in the upper reaches of the watershed. The sixth restored site (PTR) is highly urban (49% ISC).

Zoom In Zoom Out Reset image size

Three additional sites were located at USGS gaging stations. Two served as regional reference sites: a forested watershed with no ISC or agriculture (POBR), which is a sub-watershed of a low-density suburban watershed containing ∼72% forest cover (BARN). The third (PTRG) was used to investigate the detectability of restoration effects at a larger scale; the highly urban restored site Plumtree Run (PTR) comprised a drainage area that was 14.8% of the Plumtree Run Gage (PTRG) watershed and was located ∼4.4 km upstream of PTRG.

At the six restored sites, area-velocity (Ott MF Pro) or salt dilution methods were used to measure discharge when baseflow water samples were collected. At First Mine Branch (FMB), stage-discharge data were measured to establish a rating curve, which combined with barometric and water pressure data collected every five minutes to create a high-frequency discharge record from April 2019 to May 2020. At the two regional reference sites (POBR, BARN) and the site downstream of the highly-urban restored site (PTRG), USGS discharge data were used.

At the three sites restored prior to this study (table 1), baseflow samples were collected approximately monthly from fall 2017 to summer 2019 with sampling increased to every two weeks during summer 2018. For the three sites restored during this study, baseflow samples were collected pre- and post-restoration. Baseflow samples were collected every two weeks for several months preceding restoration. Baseflow sampling ceased during restoration, resumed every two weeks for at least three months following completion of restoration, and then continued on a monthly basis. At the six restored sites, samples were collected upstream and downstream of the restored reach. Samples were collected during several storm events at FMB in summer of 2019 and winter of 2020.

Baseflow samples also were collected at the regional reference sites. Discrete and high-frequency USGS data from PTRG were used to study storm events and calculate daily and cumulative annual loads for nitrate (NO₃ ⁻), total dissolved phosphorus (TDP), and TSS for 2014–2019. Regression models based on discrete samples were used to calculate high-frequency time series (for more details see supplementary information).

Temperature, pH, specific conductance (SC), and dissolved oxygen (DO) were measured in situ using a handheld meter calibrated <24 h previously. Baseflow samples were field-filtered using a pump and Geotech 0.45 µm Dispos-A-Pore filters. Storm samples were collected with an Isco autosampler, retrieved shortly after storms ended, and filtered in the laboratory using syringe filters (0.45 µM). Samples collected for analysis of alkalinity and of total dissolved nitrogen (TDN) and dissolved organic carbon (DOC) were collected with no headspace. Most samples were stored at 4 °C until analysis with some samples frozen until shortly before analysis for N and P speciation.

TSS concentrations were measured by filtration onto pre-rinsed and ashed filters [16]. Alkalinity is reported as HCO₃ ⁻ concentrations and was determined by Gran titration with titrations performed with a Mettler Toledo G20 with a Rondolino autosampler; data quality were checked using a gravimetrically prepared standard. TDN and DOC (non-purgeable organic carbon) were measured with a Shimadzu TOC-CSN. Major ions and NO₃ ⁻ were measured using a Dionex ICS 5000 ion chromatograph with two columns. Silica, NO₃ ⁻ + NO₂ ⁻, NH₄ ⁺, PO₄ ³⁻ (orthophosphate), and TDP were measured colorimetrically with a Seal AQ1 Discrete Analyzer. NIST-traceable check standards were used during analytical runs for TDN/DOC, major ions, and nutrients plus silica. If concentrations differed by >10% from expected, then results were double-checked via spot-checking with samples re-run as necessary.

All plotting and statistical tests were done in R [17–22]. Water chemistry data (concentrations and fluxes) were generally not normally distributed. Non-parametric Wilcoxon tests were used to determine if differences were statistically significant (p < 0.05) with unpaired tests run for comparisons between sites or years and paired tests runs for upstream–downstream comparisons at a single site.

Baseflow nitrogen concentrations and daily loads were controlled primarily by land use (figures 2(a) and (b)) with restoration showing minimal effects. From the upstream to the downstream ends of the six post-restoration reaches, median baseflow nitrogen concentrations differed by <0.5 mg l⁻¹ following restoration, and baseflow daily loads decreased significantly along the restored reaches only at the two sites where discharge also decreased. At the three sites with pre- and post-restoration data, baseflow nitrogen concentrations did not significantly change following restoration (figures 3(a) and (b)), and baseflow nitrogen daily loads increased significantly at two sites (figures 3(c) and (d)) due to record high precipitation in 2018. These baseflow results indicate little to no change in rates of denitrification or nitrogen removal by other in-stream processes for 1–5 years following restoration.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Land use was the primary driver of baseflow nitrogen concentrations and daily loads, which were lowest at the forested reference site and highest at the agricultural sites (figures 2(a) and (b)). The sites can be classified into four groups based on nitrogen concentrations. First, the lowest concentrations occurred in the forested watershed (POBR): <0.25 mg l⁻¹ for TDN (figure 2(a); table S2). Second, the highly urban (PTR) and mostly forested, suburban reference (BARN) sites had median concentrations of 1.5–2 mg l⁻¹ with nitrogen likely coming predominantly from sewer (PTR) or septic (BARN) systems and atmospheric deposition. Third, the active pasture (NSR), suburban/partially agricultural (BCB), and USGS urban reference (PTRG) sites had median TDN of 2.5–3.0 mg l⁻¹ with likely nitrogen sources including agricultural activity such as fertilizer application (NSR, BCB) and septic (BCB) or sewer (PTRG) systems. Four, the three active cultivated crop agriculture sites (FMB, CBR, BTR) had median TDN of 5.4–6.0 mg l⁻¹ with fertilizer application representing the primary nitrogen source.

Differences in baseflow discharge across the restored sites were minimal in most cases and thus had little effect on nitrogen concentrations or loads. While larger watersheds had higher baseflow discharge (figure 4(a)), specific baseflow discharge (normalized to watershed area) was similar across the restored watersheds with the exception of lower values at the highly urban PTR watershed (figure 4(b)). Baseflow differences did play an important role in controlling baseflow nitrogen loads between the upstream and downstream ends of restored sites and also from 2017 to 2018–19 for the three sites with pre- and post-restoration data (figure 4; tables S4 and S5).

Zoom In Zoom Out Reset image size

Baseflow differences between the upstream and downstream ends within sites were a primary driver of nitrogen loads. The four agricultural sites (FMB, CBR, BTR, NSR) had higher baseflow discharge at the downstream ends of restored reaches (though only significantly higher for FMB, p < 0.001; table S5). For the FMB and Beetree Run (BTR) sites, higher discharge at the downstream end resulted in similar nitrogen loads at the upstream and downstream ends despite small but significant decreases in nitrogen concentrations across the restored reaches (figures 2(a) and (b)). The Cabbage Run (CBR) and North Stirrup Run (NSR) sites also had similar nitrogen daily loads at the upstream and downstream ends due to similar or higher discharge and small but statistically significant increases in nitrogen concentrations from upstream to downstream. At the (sub)urban Bear Cabin Branch (BCB) and PTR sites, discharge decreases from the upstream to downstream ends (table S5), which were not statistically significant, did contribute to a statistically significant decrease in daily baseflow nitrogen loads (figure 2(b)).

The three sites with pre- and post-restoration data strongly indicate that restoration did little to change baseflow nitrogen since upstream to downstream differences, or lack of differences, remained the same post-restoration (figure 3). Baseflow nitrogen concentrations did not change following restoration at the upstream and downstream ends of the FMB and PTR sites and increased at BCB (figures 3(a) and (b)). Post-restoration baseflow nitrogen loads increased compared to pre-restoration loads at both the upstream and downstream ends of BCB and at the upstream end of FMB (figures 3(c) and (d)).

Changes in baseflow nitrogen concentrations and loads along the study reaches at these three sites, i.e. between the upstream and downstream ends, were similar before and after restoration. At all three sites, baseflow nitrogen concentrations were slightly lower (0.12–0.51 mg l⁻¹) at the downstream ends both pre- and post-restoration (figures 3(a) and (b), table S3). At the agricultural FMB site, nitrogen loads were significantly different between upstream and downstream before restoration but not significantly different after restoration. At FMB, both concentrations and loads increased post-restoration, which corresponded with higher 2018 and 2019 baseflow discharge (figure 4). Both before and after restoration, daily loads for baseflow nitrogen were significantly lower at the downstream end of the study reaches for the (sub)urban BCB and PTR sites (table S5). Thus it appears that restoration did little to alter in-stream nitrogen cycling at these sites. While changes in denitrification rates may have occurred in the riparian zone and groundwater, they were not detectable in stream chemistry. At other restored Maryland sites, increased denitrification rates and lower nitrogen concentrations (albeit perhaps partially due to hydrology) were observed [11, 23]. Increases in nitrogen concentrations and especially loads due to high precipitation and baseflow have been previously observed in the region [24, 25]. These increases may arise from several sources: (1) an elevated water table flushing nitrogen from normally hydrologically-unsaturated portions of the shallow subsurface [26], restoration-associated changes such exposure of nitrogen-rich soils or other material, or (3) elevated nitrogen in riparian groundwater following tree removal [27].

DOC, DOC:TDN ratios, and nitrogen speciation yield insights about why little denitrification or in-stream nitrogen removal occurs. Median DOC concentrations range between ∼1 and 2.7 mg l⁻¹ and do not closely correspond to land use characteristics or N concentrations (figure 2(c)). Median DOC at the highly urban site is >2 mg l⁻¹; at the other sites, DOC is 0.9–1.8 mg l⁻¹. Median baseflow DOC:TDN ratios are lowest at the three cultivated crop sites, are ∼1 at the pasture and suburban restored sites, and >1 at the highly urban restored site (figure 2(d)). At baseflow, nitrate represents ∼100% of dissolved nitrogen at all sites except the forested reference site (POBR) where NO₃ ⁻ is undetectable (figure S3(a)). Colorimetric analyses of NO₃ ⁻ (or NO₃ ⁻ + NO₂ ⁻) are equivalent to ion chromatography (figure S3(b)).

Carbon availability can limit denitrification, especially in agricultural or urban sites, e.g. when DOC is <2 mg l⁻¹ or DOC to N ratios are <1 [4, 23, 28–31]. The extensive tree removal that is typically part of LSR/FR may further reduce carbon availability at these sites. Thus in-stream denitrification at restored cultivated crop sites is almost certainly limited by carbon availability and likely limited much of the time at the pasture and suburban restored sites. The pasture and (sub)urban sites are where in-stream denitrification may occur with some frequency based on DOC:TDN ratios. At the BCB and PTR sites, similar discharge at the upstream and downstream ends may indicate some groundwater exchange and interaction. By comparison, the four agricultural sites (FMB, CBR, BTR, NSR) appear to be gaining streams with higher discharge at the downstream ends (table S4). At the highly urban site (PTR), denitrification and/or nitrate removal may be indicated by the lowest DO observed at any site (figure S4(a)).

Baseflow orthophosphate (oP) concentrations were low with the restored sites and forested/mostly forested regional references having median concentrations at or below the detection limit of 0.005 mg l⁻¹ following restoration (figure S5; table S6). A few sites (FMB, NSR, BCB, PTR) occasionally had concentrations of 0.005–0.020 mg l⁻¹. The post-restoration median baseflow concentration at the upstream end of the agricultural FMB site was 0.007 mg l⁻¹, which decreased to a median below detection at the downstream end. By contrast, median stormflow oP concentrations at FMB post-restoration are much higher at 0.461 and 0.254 mg l⁻¹ for upstream and downstream, respectively (see section 3.4; table S6) At the three sites with pre-restoration data, the pre-restoration median oP concentration at FMB was much higher (0.03 mg l⁻¹; table S6) following restoration (<0.005 mg l⁻¹; table S6); concentrations showed no significant post-restoration change at BCB or PTR.

Non-nutrient chemistry reflects land use with the reference forested site (POBR) having much lower values than the other sites. Differences for most non-nutrient parameters typically were larger among sites than within sites. Median SC values and chloride concentrations illustrate cross-site differences (figure 5). SC and chloride median concentrations at the agricultural, suburban, and mostly forested reference sites were about an order of magnitude higher than the forested reference site (figure 5). Median SC and chloride were approximately another order of magnitude higher at the highly urban PTR site (see supplementary information for details on other parameters).

Zoom In Zoom Out Reset image size

Non-nutrient chemistry is relevant since ecological lift is often a stream restoration goal. Elevated SC reflects elevated ion concentrations, a phenomenon sometimes called freshwater salinization syndrome [32]. Low ion concentrations at the forested reference site (POBR) represent regional background conditions to which pre-development aquatic organisms were adapted [33–36]. Agricultural and urban watersheds also have differing ion proportions, e.g. anions shift from being bicarbonate-dominated at the forested reference site to chloride-dominated at all other sites except FMB (figure S4(e)). Elevated, differing, and variable ion concentrations driven by land use represent a significant ecological stressor for aquatic organisms and thus affects post-restoration aquatic communities [37–43].

Storm event concentrations of DOC, ammonia (NH₄ ⁺), orthophosphate, TDP, and TSS at post-restoration FMB were substantially elevated compared to baseflow concentrations (figure 6; table S7). Export of dissolved nitrogen and phosphorus along with TSS were somewhat lower at the downstream end than the upstream end, suggesting that the restored reach may successfully reduce area-normalized loads (figure 7). Storm event concentrations differ substantially from baseflow, including substantial increases in (1) ammonia from below detection to >1 mg l⁻¹ (figure 6(a)), (2) DOC from ∼2.7 to 15 mg l⁻¹ (figure 6(b)), (3) orthophosphate and TDP from close to the detection limit to 1.2–1.35 mg l⁻¹ (figure 6(c)), and (4) TSS from near the detection limit (<10 mg l⁻¹; table S7) to median and peak concentrations of 161 mg l⁻¹ and 3450 mg l⁻¹, respectively (figure 6(d); table S7).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

TDN and NO₃ ⁻ concentrations were diluted during storms, and so storm loads were similar to the high end of daily baseflow load. During storms, TDN and NO₃ ⁻ showed hysteresis with maximum dilution at, or just after peak discharge, and then a slow increase on the falling limb (figure 6(a)). Ammonia concentrations showed less hysteresis and were mostly closely associated with peak, and thus overland, flow. The TDN concentration decrease was smaller than NO₃ ⁻, which is partially explainable by increased NH₄ ⁺ concentrations (figure 6(a)). The mixture of nitrogen species during storms indicates an important role for surface/near-surface event water contributions of NH₄ ⁺ and organic N in storm export in contrast to all nitrogen being present as nitrate in baseflow (figure S3(a)).

Dissolved carbon and phosphorus appeared to be contributed both from overland flow and quickflow subsurface pathways (e.g. interflow) while TSS was likely associated predominantly with overland flow. DOC and phosphorus concentrations displayed hysteresis with an abrupt increase until peak discharge or just after and then subsequent slow decrease (figures 6(b) and (c)). The slow decrease on the falling limb suggests contributions from interflow or other (shallow) subsurface pathways that reach the stream more slowly than overland flow. In contrast, TSS showed little hysteresis with the highest concentrations, and thus fluxes, associated with peak, or near peak discharge (figure 6(d)).

Paired comparisons of area-normalized export for nitrogen, phosphorus, and TSS at the upstream and downstream end of FMB across three storm events indicate small load reductions along the restored reach (figure 7). The peak reduction in nitrogen load (300–350 g km⁻²) was <5% of integrated storm load and ~5% of the daily baseflow load (6343 g km⁻²) at the downstream end of FMB. Similarly, peak stormflow reductions for dissolved phosphorus were relatively small. Integrated across the storms, the reductions ranged from 10% to 37% between the upstream and downstream ends, which was similar to the observed reductions in specific (or area-normalized) discharge. These results are positive but only suggestive without pre-restoration storm data or additional post-restoration data.

A few factors may contribute to smaller loads at the downstream end. Peak phosphorus and TSS are higher at the upstream end for the three storm events while peak discharge is similar between the upstream and downstream ends (figures 6 and S6). With similar peak discharge between upstream and downstream but a 32% larger watershed area at the downstream end (table 1), area-normalized exports would be expected to decrease.

More substantial reductions at FMB might be achieved during larger storm events. The engineered bankful capacity at the restored site was 2.83 m³ s⁻¹, and though the upper end of the rating curve was not well constrained, the high-frequency discharge data suggest bankful capacity was exceeded six times from late April 2019 to mid-June 2020, which is undoubtedly more frequent inundation than occurred pre-restoration. Unfortunately, none of these storms were sampled.

The USGS site (PTRG) downstream of the highly urban restored site (PTR) was used to investigate the larger-scale effects of the mid-2017 PTR restoration. Similar to FMB, NO₃ ⁻ is diluted as discharge increases while TSS and TDP show flushing behavior (figure S7). At PTRG, TSS concentrations and discharge peak at about the same time. Based on high-frequency time series, annual loads for NO₃ ⁻ varied by <2× while annual loads for TSS varied by >7.5× (figure 8; table S8); TDP behaved similarly to TSS (figure S8(a)). Annual load rankings for NO₃ ⁻, TDP, and TSS are the same (table S8 and are highly correlated with median annual discharge (R² = 0.94, 0.84, and 0.82, respectively). NO₃ ⁻ export increases relatively consistently through the year while TSS and TDP export shows substantial increases over brief periods indicating the importance of storm events (figure 8). Predominantly baseflow-driven NO₃ ⁻ export and event-driven TSS and phosphorus export also is observed elsewhere in the region [2, 44–46].

Zoom In Zoom Out Reset image size

Additionally, six paired pre- and post-restoration storm events were investigated based on similar precipitation intensities. Five of the six post-restoration events had higher loads for NO₃ ⁻, TSS, and TDP with three post-restoration events having TSS and TDP loads that were 3.8–17 times higher than similar pre-restoration events (table S9). Higher post-restoration event loads are likely due to 2018 and 2019 being high precipitation and discharge years. It is not surprising that the effects of the PTR restoration were not seen downstream given the small reductions observed at FMB. Though the restored reach for the highly urban PTR site represented ∼15% of the watershed draining to PTRG, high variability in annual export obscured the larger-scale effects of stream restoration.

Assessing effectiveness of stream restoration projects is challenging: (1) pre-restoration data are commonly unavailable, (2) collecting sufficient data to accurately characterize changes in fluxes and loads is time-intensive and expensive, (3) many years may elapse before the desired outcomes of restoration are observable due to legacy inputs and/or the time needed to establish key biogeochemical processes, and (4) stream restoration responses may be obscured by weather/climate variability [25, 47]. For example, high retention and removal rates at one Mid-Atlantic restoration resulted in decreased flow-weighted mean concentrations of most constituents, but loads did not significantly decrease for NO₃ ⁻ and TSS [48]. Additionally, for constituents with reduced loads, the effects of restoration in that same study were not detectable ∼600 m downstream except for orthophosphate.

The Chesapeake Bay Program grants regulatory credit to stream restoration projects upon completion via three relevant protocols with each being annual and additive [49]. Protocol 1 Prevented Sediment during Storm Flow credits sediment and nutrient reduction based on prevented bank or channel erosion. Protocol 2 Instream and Riparian Nutrient Processing During Baseflow credits nitrogen reductions based on design features promoting denitrification during baseflow. Protocol 3 Floodplain Reconnection Volume credits sediment and nutrient reduction for reconnecting stream channels to floodplains.

Removal of erodible bank and floodplain deposits that were 1–2.5 m thick at the study sites via LSR/FR restoration likely resulted in decreased TSS loads and associated particulate nitrogen and phosphorus, suggesting that Protocol 1 was addressed. In mid-Atlantic Piedmont watersheds, bank sediments supply much of the suspended load [12], e.g. 6%–90% (average 57%) in Baltimore County, MD and 30%–65% in Big Spring Run [14, 50]. Substantial decreases in TSS concentrations and a 69% decrease in TSS loads with accompanying total phosphorus decreases were observed at Big Spring Run, ∼50 km northeast of the study region [50]. Among the first LSR restorations, Big Spring Run has a high-frequency data record from multiple USGS gages for 3 years pre-restoration and four years post-restoration. At another Piedmont LSR/FR site, TSS fluxes increased during restoration with the restoration period extending across the 2018–2019 high discharge years (no post-restoration data available, 51).

We found little evidence for decreased post-restoration nitrogen concentrations or fluxes resulting from denitrification or other biogeochemical processes (Protocol 2). Results were similar at the Big Spring Run LSR/FR restoration where NO₃ ⁻ concentrations increased (though not significantly) and orthophosphate remained unchanged [50]; groundwater NO₃ ⁻ concentrations began decreasing 4–5 years post-restoration with carbon accumulation as an important factor [52]. At a Piedmont site with a similar study period (2017–19), NO₃ ⁻ concentrations decreased slightly but fluxes increased during restoration; orthophosphate did not change [51]. A Piedmont site with observed post-restoration nitrogen reductions had initially low nitrogen concentrations (similar to the (sub)urban BCB and PTR sites), and reductions may have resulted from a combination of denitrification and lower nitrogen inputs to restored reaches [11]. At Mid-Atlantic sites with substantial post-restoration denitrification, large amounts of carbon were added during restoration and multiple lines of evidence pointed to denitrification, e.g. nitrate decreases accompanied by ammonia and dissolved organic nitrogen increases and low DO at times [2]. By contrast, baseflow nitrogen in this study was entirely nitrate and remained so following restoration.

Other post-restoration factors may contribute to increased denitrification with time. Denitrification may increase in the lower post-restoration floodplains due to increased carbon inputs from more frequent inundation or by a water table closer to the surface. Terrestrial vegetation biomass, and thus carbon inputs, on restored floodplains should increase with time.

Lags between restoration completion and improved water quality are common, especially in agricultural watersheds. Time is needed to establish sufficient rates of biogeochemical processing. Legacy nutrients in soils and groundwater may take years to decades to reach streams [4, 47, 53, 54]. Legacy contributions (older than 1 year) represent 50% of nitrogen exported by the Susquehanna River, and 18% is older than 10 years [55].

We have insufficient data to determine the success of changes related to Protocol 3 for FR. Achieving success related to this protocol may be challenging, particularly in urban watersheds. In five urban Piedmont watersheds, much of the water accessing the floodplain during overbank events occurred in during large events (50% of floodplain flow during 2–3 events per year), resulting in 0.2%–1% of nitrogen retention or removal by Protocol 3 [56].

Results of this study suggest some positive outcomes of LSR/FR stream restoration. Future research on LSR/FR or similar projects should likely focus on data collection related to sediment erosion prevention and FR (Protocols 1 and 3). Pre- and post-restoration high-frequency data collection for discharge and sediment transport would greatly facilitate answering questions related whether these protocols are successfully addressed. Groundwater sampling would be useful for addressing Protocols 2 and 3, but changes may take several years to emerge, especially related to baseflow nutrient processing.