1 Introduction

Sediment fingerprinting uses a range of physical and biogeochemical properties to differentiate sediment sources (Guzmán et al. 2013; Collins et al. 2020), including an accuracy assessment based on identification of the original source samples (Gellis and Walling 2013). Sediment fingerprinting has been used in basins of 10–100 km2 to apportion streambed-sediment (Williamson et al. 2014; Blount et al. 2022) as well as suspended sediment at event (Cashman et al. 2018), monthly (Fitzpatrick et al. 2019), and seasonal (Boudreault et al. 2019) timescales. Common upland sources of stream sediment include material directly transported from cropland, pasture, and forested parcels as well as that washed from roads that connect the agricultural landscape. Studies in agricultural basins have shown that sediment also commonly includes streambank-derived material, with source contributions varying as a function of physiography, season, and contributing area (Gellis and Noe 2013; Gellis et al. 2014, 2019; Gellis and Gorman Sanisaca 2018; Fitzpatrick et al. 2019; Williamson et al. 2020a).

Measuring streambank erosion in basins where it has been identified by sediment fingerprinting is the next step in guiding resource-management decisions. Geomorphic assessment of the riparian corridor, including both quantification and location of streambank erosion and streambed-sediment storage (Fitzpatrick et al. 2016; Blount et al. 2022), is key to differentiating reaches that provide sediment for transport downstream from those where material accumulates in the channel (Kwan and Swanson 2014; Lamba et al. 2015b; Fitzpatrick et al. 2019; Schilling et al. 2022); together, these provide a sediment budget. Integrating sediment fingerprinting with a sediment budget and fallout radionuclide (FRN)-based chronometers of suspended (Lamba et al. 2015a), streambed (Matisoff et al. 2005), and in-channel exposed (Karwan et al. 2018) sediment provides multiple lines of evidence to guide resource management (Collins et al. 2020).

In the Great Lakes basin, non-cropland land commonly coincides with areas susceptible to erosion, with discontinuous forest marking the irregular valley development at the transition from glacial till to lake-plain sediment (Williamson et al. 2020a; Blount et al. 2022). This glacial till is more erodible than lake-plain sediment (NRCS 2020), resulting in more animal operations along headwater reaches relative to higher-order streams, where cropland is a larger percentage of land use (Homer et al. 2020). Compaction of soil from livestock in pastured uplands reduces plant cover and soil macropore space, causing increased runoff and soil erosion (Foote et al. 2015). Livestock grazing in riparian areas increases streambank erosion (Tufekcioglu et al. 2012), and unmanaged manure is an exposed source of P (Stackpoole et al. 2019).

Fertilizer and manure applications result in dissolved (DP) and particulate P (PP) entering the stream system (Sharpley et al. 1992; Han et al. 2012), with higher potential where soil is easily erodible and when fields are left bare in the early winter and spring (Capel et al. 2018a, b). As P leaves the uplands and is transported through the stream network, it spirals between DP and PP through biotic assimilation and mineralization (Newbold et al. 1983) and is sorbed to(from) suspended and streambed sediment that progressively moves downstream during multiple events (Verhoff et al. 1979). Fine-grained, streambed sediment in agricultural landscapes is commonly saturated with P, increasing the potential for desorption to the water column (Haggard et al. 2007; Palmer-Felgate et al. 2009; Osmond et al. 2012; Kreiling et al. 2019), with in-channel stores of sed-P providing a legacy source on a scale of years to decades (Casillas-Ituarte et al. 2020; King et al. 2021; Williamson et al. 2021a). The proportion of agricultural land use, DP concentrations, and sediment physicochemical characteristics each influence how P is stored or released by streambed sediment (Kreiling et al. 2023). Streambank erosion has been identified as an additional source of sed-P (Collins et al. 2010; Gellis and Noe 2013; Huisman et al. 2013), with concentrations similar to agricultural land (Huisman et al. 2013; Williamson et al. 2014, 2020a; Fitzpatrick et al. 2019), although the cause of these high concentrations may vary among basins.

In recent years, the Maumee River Basin (“the Maumee”; 17,052 km2) was the largest contributor of total phosphorus (TP) to Lake Erie, mostly as non-point sources (Han et al. 2012), despite on-field implementation of conservation management to reduce nutrient runoff related to harmful algal blooms (HABs) and hypoxia in the western Lake Erie Basin (WLEB) (USDA-NRCS 2017; Kast et al. 2021). In recent years, TP in the Maumee has been approximately 20% DP and 80% PP (2009–2013; Scavia et al. 2017). Exchange between DP and PP varies as a function of stream size, substrate, and climatic variability, with DP accounting for 8–42% of TP among tributaries (2015–2018; Williamson et al. 2021b), highlighting that this DP/PP ratio is not static in either space or time. Consequently, management success of upland P is difficult to quantify without understanding riparian-corridor sources and stores, including abundance, transformations, and retention in the stream network. This is also key to prioritizing future conservation management (e.g., Stackpoole et al. 2019; Moriasi et al. 2020).

As part of the Great Lakes Restoration Initiative, long-term continuous monitoring was implemented on Black Creek, Indiana (Fig. 1), an agricultural basin with a mix of cropland, active horse pastures that enable farming and transportation for the Amish community, and rural neighborhoods. This 12-digit hydrologic unit code (041000050104) basin was identified as a significant contributor to nutrient, and sediment loads for the WLEB (Robertson and Saad 2011) and individual reaches, most of which were in western tributaries, were identified as impaired for nutrients, biological communities, Escherichia coli, and streambank erosion (SNRT Inc. 2015). Sediment fingerprinting of streambed (August 23–24, 2017) and monthly suspended-sediment composites during the 2018 water year (October 2017–September 2018) indicated that streambank contributions exceeded those from the combination of upland sources (Williamson et al. 2020a, 2021a). Identification of streambank as a primary contributor shifted the conversation about management and conservation in the basin, and raised questions of why/where erosion was occurring and if streambank stabilization should be considered. To answer this question, our objectives were to

  • Fingerprint streambed sediment for apportionment as upland sources and streambank

  • Use rapid geomorphic assessment of environmental setting, to quantify

    1. o

      Streambank erosion occurrence and

    2. o

      Soft, streambed-sediment storage volumes

  • Estimate recency of streambed-sediment accumulation using FRN

  • Quantify phosphorus retention in streambed sediment

Fig. 1
figure 1

Location of 30 reaches described for rapid geomorphic assessment (RGA) overlaid on simplified stream network. Streambed-sediment apportionments are shown for those 13 reaches where this was done in 2019; note that there was no fine, mobile streambed sediment at BC7 when sites were sampled in 2019 so only data from 2017 are provided. Site names include RGA designation (BC1 through BC30), stream order (1 through 4), and generalized land use descriptor (c = crop, p = pasture, f = forest). Each bar is a proportional breakdown (0–100%) for the 2019 sample and that from 2017 (BC1, 7, 8 only). Base map is land cover (Homer et al. 2020) for the Black Creek Basin, including both that area upstream of BC1 (co-located with USGS Site 04183038) and the larger basin for the confluence with the Maumee River

Each of these was evaluated at multiple reaches within the stream network, with contributing areas of 0.1–32 km2, and aggregated for the most downstream reach. We hypothesized that the relation between streambank erosion and streambed-sediment storage would primarily reflect stream size and would vary as a function of adjacent land use and glacial landform setting for similar-sized streams.

2 Materials and methods

2.1 Study site

The U.S. Geological Survey (USGS) streamflow and water-quality monitoring site at Black Creek near Harlan, Indiana (Site ID 04183038; USGS 2023) is discussed here as BC1. Elevation in the basin ranges from 216 to 261 m above mean sea level (amsl), with stream slopes of 0–5%. Harlan (near BC8, 13, and 22 in Fig. 1) is located near the western boundary of the basin on the Ancient Lake Maumee shoreline (Fisher et al. 2015; approximately 235 m amsl) which marks the transition between the Eastern Corn Belt Plain moraines made of glacial till and the Huron/Lake Erie Plains (Fig. 1; Omernik and Griffith 2014). This boundary is commonly marked by forest in what is otherwise an agricultural landscape. Subsurface drainage is common on lake-plain agricultural fields, with less on glacial till (U.S. Department of Agriculture-Natural Resources Conservation Servise [USDA-NRCS] 2016). Glacial till has high runoff risk and is highly erodible (SNRT Inc. 2015; Porter et al. 2018); active horse pastures are common on till. Additional details on basin physiography are summarized by Williamson et al. (2021a).

Mean precipitation is 1014 mm, mostly as rain, with more than half of the precipitation occurring in spring and summer (NOAA 2023). Below-zero soil temperatures (0–2 cm; USGS Site ID 411229084541103) occur between December and March (22–62 d yr−1; 2015–2020) (USGS 2023). Most of the field descriptions and sampling discussed here were done in 2019 (1045 mm), a wet fall and spring when frequent precipitation resulted in fallow of approximately 30% of local agricultural land.

2.2 Sediment sources

Samples were fingerprinted using the same source samples as a previous analysis of Black Creek suspended and streambed sediment (Williamson et al. 2020a; methods and data in Williamson et al. 2020b). Source samples were collected upstream of BC1 between July and September 2017; as a result, source samples for upstream rapid geomorphic assessment (RGA) sites included those from outside of their contributing area. For upland sources, the top 1–2 cm of soil material was composited along a total transect of 250 m (150 m for roads), with one sample collected every 10 m. Sampled cropland included corn, soybean, wheat, and alfalfa fields under conventional and conservation tillage, with a mix of fertilizer and manure nutrient sources. Sampled pastures were limited to those supporting horses; other animals are generally not pastured in this basin. Forests were unmanaged patches, generally smaller than 0.5 km2, and usually included stream channel. Roads were used as a surrogate for development (impervious surface, buildings, etc. in Homer et al. 2020) and included those among agricultural fields and those in neighborhoods; most were paved. Low and middle streambanks were sampled from exposed soil at 10-m intervals along both sides of a stream for a 150-m reach; locations sampled in 2017 were described as part of the RGA. Only the fine-grained (< 63 μm) fraction was analyzed.

2.3 Rapid geomorphic assessment, streambed-sediment storage, and streambank erosion

A combination of stream corridor land-use distribution, valley type, channel slope, stream order (Strahler 1957), and ecoregion (Omernik and Griffith 2014) was used to select 30 RGA reaches using methods of Fitzpatrick et al. (2016); one reach for ~ 3 km of stream length. Roadside ditches were categorized for the land use that ran along the other side of the road. Reaches were categorized as forest if there was forest upstream of the reach and trees separated the stream from agricultural or developed land on both sides. Most reaches included discharge points (pipes) for subsurface agricultural drainage that extended from the streambanks, regardless of adjacent land use and including forested streambanks.

During July 22–25, 2019 (streamflow at BC1 was 0.038–0.048 m3 s−1), the thickness and spatial extent of soft, mobile, fine-grained (mainly silt and clay) streambed sediment was inventoried along 150-m transects (data and methods in Blount et al. 2021). Streambank erosion was also inventoried (USDA-NRCS Wisconsin 2015); erosion along a shallow waterway in a horse enclosure (BC27) was estimated from an upstream reach (BC26). The ratio of streambed-sediment storage to streambank erosion (S:E; after Fitzpatrick et al. 2019) differentiated erosional (S:E < 1) and depositional (S:E > 1) environments; S:E is discussed in years: Mg km−1/Mg km−1 yr−1. Reach-based descriptions of streambank erosion and streambed-sediment storage and sed-P were extrapolated to the remaining network, including both ephemeral and perennial reaches (Fig. 1) (USGS 2020; Blount et al. 2021).

A subset of 13 RGAs that spanned the range of land use, ecoregion, and stream characteristics were selected for streambed-sediment sampling and FRN analysis and sediment fingerprinting. Soft, mobile, fine-grained, streambed-sediment deposits were sampled with plastic tools during summer low-flow conditions, when the potential for sorption/desorption of P is higher due to the longer contact time of water with the sediment (Withers and Jarvie 2008).

2.4 Fingerprinting of fine-grained sediment

Fingerprinting was focused on the fine-grained (< 63 μm) portion of sediment because this is the fraction most likely to be eroded and transported from uplands as well as (re)suspended by streamflow (Lowrance et al. 2006); metals, including P, are concentrated in finer particles (Sharpley et al. 1992). Indicators included 42 metals as well as C and N forms (SI1); only the fine-grained portion was analyzed. Streambed samples from 2019 (data and methods provided in Williamson and Hoefling 2023) underwent the same analyses as those from the earlier study with the exception that C and N were run at the U.S. Geological Survey Upper Midwest Environmental Sciences Center (UMESC) on a Variomax CN analyzer (Bremner and Tabatabai 1971).

Statistical methods (Gorman Sanisaca et al. 2017) were the same as those from the 2018 study (Williamson et al. 2020a) and account for differences in particle size distribution (median particle diameter [d50]) and reactivity (surface area and organic carbon [OrgC] concentrations) between sources and stream sediment. Chemical and physical data for source samples were used to apportion streambed-sediment samples as a combination of upland sources and streambank. One source sample (321848_c; Williamson et al. 2020b) was removed as an outlier for all samples, reducing the number of source samples to 70: 13 cropland, 14 pasture, 15 forest, 13 road, and 15 streambank samples. Streambed collected in August 2017 (0.018 m3 s−1) was compared to that from 2019 (BC1,7,8; Williamson et al. 2020c).

Source apportionment and an accompanying source verification were run individually for each target with a 1000-iteration Monte Carlo analysis; allocations of less than 0.5% were considered zero. Mean values from the 1000 iterations are reported here. Source-verification goodness of fit ranged from 72 to 92%, with the exception of one sample (BC22; 61%). Overall, 18 of 70 individual source samples were mis-classified for 1–11 targets; no forest samples were mis-classified, and nothing was mis-classified as road (SI2).

2.5 Proportion of new sediment in bulk streambed using fallout radionuclides

Beryllium-7 (7Be) and lead-210 (210Pbxs) are primarily delivered during rain events to exposed surface soil or sediment. Their different half-lives (7Be 53.3 days, 210Pb 22.3 years) provide a mechanism to differentiate material that has moved within weeks to months versus that moving on the order of decades (Matisoff et al. 2005). Excess 210Pb (210Pbxs) accounts for the 210Pb that exceeds that from Radium-226 (226Ra) decay and is attributed to atmospheric deposition of the 210Pb decay product of Radon-222, which diffuses from the subsurface to the atmosphere.

Unsieved, bulk streambed from 12 RGA reaches (omitted BC10) was analyzed at the University of Minnesota (UMN) FRN laboratory according to standard gamma counting methods (Karwan et al. 2016; method details and data available as Karwan and Williamson 2023). 210Pbxs was computed as the difference of total 210Pb and supported 210Pb based on the daughters of 226Ra–214Pb and Bismuth-214 (Mabit et al. 2014). Contact with metal was avoided so that the same samples could later be sieved and analyzed for trace elements for sediment fingerprinting. Streambed 7Be/210Pbxs was evaluated as the fraction of new sediment after Matisoff et al. (2005):

$$\text{Proportion of new sediment = }\frac{ sediment \nicefrac{{7}_{{\text{Be}}}}{{210}_{{\text{Pb}}-{\text{xs}}}}}{rain \nicefrac{{7}_{{\text{Be}}}}{{210}_{{\text{Pb}}-{\text{xs}}}}}$$
(1)

A precipitation ratio (denominator) of 4 was based upon regionally (Minnesota and Wisconsin) collected, summer precipitation from 2019 (Hankins 2023). While the 7Be signature can vary among and within individual events, this 7Be/210Pbxs ratio varies less (Matisoff et al. 2005), enabling relative comparison of sediment among sites that may reflect accumulation of days to years. This method quantifies “time since rainfall contact”; the ratio translates to a site-specific proportion of sediment recently in contact with precipitation (Matisoff et al. 2005). Based on the range of 7Be activity in these streambed samples, 7Be would remain detectable for between 5 months (median) and 1 year (maximum value) after collection. Hence, we assign the timeframe of months using this method with our data.

2.6 Phosphorus retention potential of bulk sediment

Additional streambed sediment was collected July 8, 2019 (0.076 m3 s−1) and July 13, 2021 (1.10 m3 s−1 up from 0.10 m3 s−1 the previous day) at four RGA sites (BC1, 7, 8, 18) to characterize the potential to retain P (Kreiling et al. 2022). Unsieved, bulk samples were analyzed at UMESC for equilibrium P concentration (EPC0; Kreiling et al. 2019; Kreiling et al. 2023), total elemental P concentration (sed-PBulk; USEPA 1996), particle size distribution (Durner et al. 2017), and d50Bulk. Water samples were also collected at each site and analyzed for DP (Standard Methods 2021). The EPC0 is the concentration of DP in the water column where there is no net sorption or desorption of P to the sediment and can be considered an index of the sediment P-buffering capacity (Simpson et al. 2021), with finer material generally having a higher buffering capacity (Sharpley et al. 1992). Streambed sed-PBulk provided a complement to sed-P of the fine-grained, RGA streambed sediment. Bulk streambed sediment collected in 2019 was analyzed for Mehlich-3 (Meh) P, aluminum (AlMeh), and iron (FeMeh) concentrations at the UMN Research Analytical Laboratory (Mehlich 1984); these were compared to values for bulk source samples and streambed sediment sampled as part of sediment fingerprinting in 2017 (data and methods in Perner et al. 2022). The P saturation ratio (PSR) of sediment, another index of the sediment P-retention potential, was calculated using molar concentrations of the PMeh, AlMeh, and FeMeh. The potential for the streambed sediment to adsorb additional P is indicated by PSR < 0.10; PSR > 0.15 indicates a potential for the sediment to be a source of P to the water column (Nair 2014).

2.7 Statistical methods

Statistical analyses were done in R (R Core Team 2022). Spearman rank-based correlations were used to compare watershed attributes to streambank erosion, streambed-sediment storage, apportionment of streambed sediment, and proportion of new sediment as well as influence of land use and ecoregion. Tukey box plots were used to compare the sed-P and sed-PMeh among sources and streambed sediment; these populations were also evaluated using pairwise Wilcoxon rank-sum tests (Benjamini and Hochberg 1995). Statements of significance indicate a p value ≤ 0.05. Letters included on graphs indicate significantly different groups: a differs from b but ab cannot be differentiated from either.

3 Results

3.1 Land use as a function of ecoregion

Land use was similar among the 30 described RGA reaches and the 13 reaches for which streambed sediment was fingerprinted (Fig. 1, Table 1). Land use differed between glacial till and lake plain, with more pasture upstream of glacial-till reaches (p value = 0.0033) and more cropland upstream of lake-plain reaches (p value = 0.0294). More developed area was upstream of lake-plain reaches (p value < 0.0001), highlighting the density of roads in the rural community centered on the ancient shoreline. In contrast, forest was similarly distributed in both ecoregions (7% median area), mostly in the riparian corridor or near the ancient shoreline.

Table 1 Description of land use distribution for 30 rapid-geomorphic-assessment reaches (Homer et al. 2020) and for reaches at which sediment was sampled for fingerprinting and fallout radionuclide analysis (in parentheses). Fingerprinting apportionment for 13 reaches is shown in italics

3.2 Source apportionment of streamed sediment

Apportionment of streambed sediment as cropland was consistently less than the percentage of the contributing area (Fig. 1, Table 1, SI3a). In contrast, pasture apportionment exceeded land use percentage at three reaches (BC18, 25, 26; SI3b), and road apportionment exceeded developed area at two sites (BC22, 25; SI3d). Road-sediment apportionment correlated to developed area (rho = 0.66, p value = 0.0160). Reaches near or downstream of the transition from till to lake plain generally had a forest apportionment that exceeded the land-use percentage (SI3c).

Apportionment of streambed sediment as streambank increased as contributing area increased (SI3e), with contributing areas > 2.7 km2 having 55–96% streambank, regardless of ecoregion. The small lake-plain site below this threshold (BC22) exceeded the streambank apportionment for small glacial-till sites (BC18, 25, 26). More sediment was apportioned as streambank in 2017 than in 2019 for the two lake-plain sites (BC1, 8), with ≥ 55% streambank in each year (Fig. 1).

3.3 Streambank erosion and streambed-sediment storage

Evidence of streambank erosion from field measurements along RGA reaches increased with increasing contributing area and stream order (Fig. 2a, b), with a general inverse correlation with stream slope (Fig. 2c) but no correlation to streambed-sediment apportionment as streambank. Soft, streambed-sediment storage had a general inverse correlation to stream order (Fig. 2e), with highest storage where slope was < 1% (Fig. 2f). Stream slope inversely correlated to drainage area (Fig. 2g; rho =  − 0.37, p value = 0.0226). Fluvial erosion of streambanks was the predominant mechanism, but slumping of valley sides or terraces into the channel was observed at the two sites with the highest erosion (Figs. 1 and 2; BC1, 21).

Fig. 2
figure 2

Annual bank erosion (left column; ac) and soft-sediment storage (right column, df) as a function of basin area (a, d), stream order (b, e), and stream slope (c, f) at 30 rapid geomorphic assessment (RGA) reaches. The relation between area and slope is shown inset (g). Reaches were categorized as a function of nearby majority land use, either cropland, pasture, or forest. Numbers represent RGA designation of the site names from Fig. 1 that are referenced in the text

The transition from second- to third-order tributaries generally coincided with the shoreline feature separating glacial till from lake plain (Figs. 1 and 2b, e). Streambank erosion (Figs. 2 and 3) was higher downstream of this feature (p value = 0.0468), which was commonly marked by forested patches (Figs. 1 and 2b). Median streambank erosion along forest reaches (92.6 Mg km−1 yr−1) exceeded that for cropland and pasture reaches (4.4 and 1.5 Mg km−1 yr−1; p value ≤ 0.0151). Soft, streambed-sediment storage was highest along reaches with < 0.7 km2 contributing area (Fig. 2d; p value < 0.0001) but did not vary by land use.

Fig. 3
figure 3

Streambed-sediment storage (S) and streambank erosion (E) at each described reach. The border for each plot is color coded by land use at the reach. Note that all graphs have the same axes and scale, including the second Y-axis which shows the S:E ratio (circles) over a scale of days to months, indicating an erosional environment, or years to decades, indicating a depositional environment. Site descriptions include the rapid geomorphic assessment designation together with stream order and a letter indicating land use along the reach: cropland (c), pasture (p), and forest (f)

3.4 Streambed-sediment characteristics and comparison to source samples

The fine-grained portion of streambed-sediment and source samples was compared to determine how sediment varied as a function of transport from source environments and through the stream network. Streambed-sediment d50 (3–20 μm) was coarser (p value = 0.0002) than source samples (1–26 μm). Sed-P in streambed-sediment (870–1520 mg P kg−1) exceeded that of most sources (134–1670 mg P kg−1), with the exception of pasture samples (Fig. 4a). Streambed-sediment d50 differed as a function of ecoregion (p value = 0.0391), not land use, reflecting the steeper stream slope (p value < 0.0001) on glacial till (median = 0.84%) relative to lake plain (0.54%). Streambed-sediment d50 from glacial-till reaches, mostly first- and second-order streams, was finer (6.6–11.5 μm) than that from lake plain reaches (3.0–20.1 μm), mostly third- or fourth-order streams. This particle-size difference did not translate to a difference in streambed-sediment sed-P, nor was sed-P distinguishable by land use at individual reaches. For comparison, sed-P in non-road source samples was lower for lake-plain relative to glacial till (p value = 0.0210; median 905 versus 1030 ppm) although d50 was not differentiable (p value = 0.2148). When road samples were included, sed-P was not differentiable between ecoregions (p value = 0.2095).

Fig. 4
figure 4

Evaluation of phosphorus (P) availability. a Fine-grained sediment-bound P (sed-P) in source samples, suspended sediment from water-year 2018 at BC1, and streambed from 2019. b Mehlich-3 P (PMeh) for bulk source and streambed samples. c P saturation ratio (PSR) of bulk-source and -streambed samples. Coloring of boxplots matches that in other figures for land-cover categories; in-channel sediment is shown as brown and black. Values for individual streambed-sediment samples are shown for comparison. Letters indicate significantly different groups: a differs from b but ab cannot be differentiated from either

Extractable elements were measured in bulk samples for both sediment and source samples. Pasture PMeh was higher than other source samples and streambed sediment (Fig. 4b), with higher sed-P in fines (Fig. 4a inset) that was most similar to streambank samples. Pasture samples were more saturated than other sources and indicated potential release of P to the dissolved phase, with PSR above the 0.15 threshold (Fig. 4c). In contrast, road samples had lower P relative to other sources, with high PSR that reflected low FeMeh + AlMeh availability (< 11 mmol compared to 5–48 mmol among other land use types). Most cropland and forest samples showed potential to sorb additional P. Median streambank-sample PSR was in the transition between the potential to adsorb and the potential to release P.

Comparison of sed-P with d50 for bulk and fine-grained streambed sediment underscores the higher concentrations in finer material (Fig. 4a and Table 2). All years had similar PMeh (Fig. 4b; median = 41 ppm), accounting for 5–15% of total sed-P for streambed sediment (2019 and 2021). Based on PSR for bulk sediment, streambed at these four sites was at or close to being saturated with P (Table 2 and Fig. 4c). In 2019, when streamflow was the highest of the years sampled, the three larger sites indicated potential release of P from streambed sediment to the water column. PSR from 2021 and 2017 showed only one of three sites above the threshold, with ranges that were similar to streambank source samples. At three of the four sites, EPC0 values were less than the measured water column DP (Table 2), indicating streambed sediment as a potential P sink when sampled. Only BC7 was a potential P source and only in 2019.

Table 2 Data from bulk bed sediment cores collected at four sites (2019 and 2021) for equilibrium phosphorus concentration (EPC0), P saturation ratio (PSR), sediment total P (sed-PBulk), and median particle size (d50Bulk) of streambed and water column dissolved phosphorus (DP) concentrations. Samples are listed from downstream to upstream order. Bolded numbers identify those where values indicated potential release of P to the water column (source); note that in most cases, EPC0 values indicate sites as storing P (sink), but that PSR does not. Sed-P and d50 of the fines for July 2019 are shown in brackets. Note that mg P kg−1 is the same as parts per million but is used here for comparison to EPC0 and DP. NA, not available

3.5 Differentiating erosional and depositional environments in stream network

3.5.1 Bed-sediment accumulation as a function of streambank erosion

There was no consistent relation between RGA reach-specific streambank erosion and streambed-sediment storage (Fig. 3, SI4). Median streambed-sediment storage was 6 Mg km−1 (mean = 29 Mg km−1). Median streambank erosion was 10 Mg km−1 yr−1 (39 Mg km−1 yr−1). Two described reaches had no measurable bed sediment, four reaches were stabilized as grassed waterways, and two were grassed, roadside ditches. Some grassed waterways and roadside ditches had enough channelized flow to result in active streambank erosion, and one pasture site had a poorly defined channel that precluded estimation of streambank erosion. For three reaches, the S:E was 0.01–0.03 yr (Fig. 3 circles), indicating an erosional environment where material is transported out of the reach and downstream on the order of days. Five additional reaches had an S:E < 0.3 yr, indicating an erosional environment but with seasonal accumulation of sediment. Four reaches indicated a depositional environment and storage of sediment from upstream sources, with years (S:E = 3–5 yr) to decades (S:E = 12–35 yr) of material stored in the stream channel. One pasture reach had evidence of long-term sediment storage (S:E > 100 yr). Mean S:E (14 yr) exceeded median S:E (0.2 yr) by two orders of magnitude; storage and erosion showed similar patterns.

Cumulative streambank erosion and streambed-sediment storage upstream of each RGA reach give an indication of the overall variability in erosion and storage and how it potentially changes through the basin (SI5). At BC1, the most downstream reach, this totaled 1300 Mg km−1 yr−1 and 2200 Mg km−1, respectively, indicating 1–2 yr worth of upstream streambank erosion accumulated in the upstream network. Twelve lake-plain reaches had 1–5 yr, and three had < 1 yr of cumulative storage. Among lake-plain reaches, the most storage (6 yr) was upstream of a small (0.4 km2), second-order stream reach. Three glacial-till pasture reaches, each first or second-order streams, had > 10 yr worth of stored sediment. Among other glacial-till sites, four had accumulated 1–3 yr of upstream erosion and five had < 1 yr worth, with less accumulation in streams near the shoreline feature. One grassed waterway (glacial till) and one roadside ditch (lake plain) had no upstream storage.

3.5.2 Proportion of new sediment in bulk streambed sediment using fallout radionuclides

Relative age for twelve RGA bulk streambed-sediment samples ranged from all old to all new (0–100% new sediment). One lake-plain, cropland site had no discernable new sediment (BC17). Two forested sites on the ancient-shoreline transition from glacial till to lake plain had no discernable old sediment (BC18, 21). The median (29% new) was bounded by the largest contributing area and a site on the shoreline feature. Those sites with a mix of sediment age (18–57% new) spanned the range of contributing area, land use, stream order, and ecoregion (Fig. 5).

Fig. 5
figure 5

Evaluation of proportion of new sediment using fallout radionuclides (FRNs) as a function of stream characteristics. Sites are color-coded for land use at the site, and sites fingerprinted as > 50% streambank are underlined. a Comparison of 7Be and 210Pbxs in streambed sediment at twelve reaches and b translation to proportion of new sediment using a rain ratio of 4. Comparison of proportion of new sediment to c stream slope, d streambank erosion, e streambed-sediment storage, f the storage to erosion (S:E) ratio, g sediment-bound P (sed-P), and h median grain diameter (d50) in fine-grained fraction (< 63 μm). BC25 is shown with a horizontal line for S:E because this site had no discernable streambank erosion

Streambed-sediment samples identified as mostly sourced from streambank were in the lower range of activity for both FRN (Fig. 5a, underlined sites), with a consistent slope between 7Be and 210Pbxs activity. The proportion of new sediment (Fig. 5b) did not correlate with either stream slope (Fig. 5c) or in-channel storage of streambed sediment (Fig. 5e). There was a general positive correlation to streambank erosion (Fig. 5d), especially for those sites that were mostly streambank-derived sediment, and a negative relation to S:E (Fig. 5f). The proportion of new sediment did not correlate to either the sed-P concentration of the soft, fine-grained streambed sediment (Fig. 5g) or median grain diameter (Fig. 5h).

4 Discussion

After multiple decades focused on field management to mitigate HABs (Annex 4 Task Team 2015; Baker et al. 2019), research has shown in-stream sources of P as an important piece of the puzzle (Jarvie et al. 2013; McDowell et al. 2020; Williamson et al. 2020a; King et al. 2021; Kreiling et al. 2023). This inventory and recency assessment of Black Creek in-channel sediment and associated P differentiated upland and streambank contributions to streambed sediment (Fig. 1), with a mix of recently mobilized sediment and sediment not recently exposed to rainfall (Fig. 5). Together, these data indicate that in-stream accumulation of streambed sediment, even in small watersheds (Fig. 3), has the potential to affect downstream sediment and P loads (Fig. 4) in the Maumee River on a management timescale, delaying successful reductions to Lake Erie.

4.1 Streambed-sediment source, storage, and implications for legacy phosphorus source

Fine-grained material (< 63 μm) was the focus of this sediment inventory because it includes particles most readily transported by streamflow as suspended sediment (Lowrance et al. 2006). Suspended sediment fingerprinted for Black Creek was ≈ 90% fines, with d50 ≈ 4 μm (Williamson et al. 2021a). In contrast, streambed sediment from both glacial till (median = 7 μm) and lake-plain (14 μm) reaches was coarser than previously sampled suspended sediment.

Streambed sediment was significantly coarser than upland sources, and sed-P exceeded that of cropland, forest, and road source samples, but was similar to pasture, streambank, and suspended sediment (Williamson et al. 2020a). This contrasts a similar analysis from a Fox River tributary, another Great-Lakes priority basin, where sed-P from suspended sediment and manure-treated cropland exceeded that of streambank and streambed sediment (Blount et al. 2022). The implication is that the Black Creek in-channel environment, with median base-flow DP = 0.068 mg L−1 (Williamson et al. 2021a), drives similar sed-P for streambed sediment, suspended sediment, and streambank. In 2019, DP concentrations at BC8 exceeded concentrations measured at all other sites in the Maumee River, and BC1, 7, 8 had the three highest DP concentrations in 2021 (78 sites total; Kreiling et al. 2023); these sites were among those with the smallest contributing area. This indicates potential for ongoing sorption and storage of P in streambed sediment and value in understanding headwater-tributary environments.

The potential for streambed sediment to sorb and store P from the water column at 3 of 4 sites is supported by EPC0 and PSR (Table 2); streambank PSR was similar to that for streambed sediment (Fig. 4). While sed-P was distinguishable both as a function of land use and ecoregion in Black Creek source samples, streambed-sediment sed-P did not consistently vary as a function of proximal land use, ecoregion, or d50, even though particle size increased with stream order. This further indicates homogenization of sed-P during sediment transport.

4.2 Recency and environments of streambed-sediment accumulation

Additional evidence of homogenization of sed-P concentrations is the mixed age (18–57% new) of streambed among reaches with similar storage (10–20 Mg km−1; Fig. 5). The consistent relation between 7Be and 210Pbxs reflects a longitudinally consistent process (Matisoff et al. 2005), regardless of stream order and sediment source, with material progressively moving downstream (Verhoff et al. 1979). There was evidence of correlation between proportion of new sediment and streambank erosion, with an inverse relation to S:E, affirming FRN-based relative “age” as time since contact with rain, and that contact with rain can occur during in-network mobilization and transport.

Mixed-age streambed-sediment environments were bracketed by three sites with streambed-sediment storage similar to the median but with proportions of new sediment of either 0 (time since contact with precipitation exceeded 5–12 months) or 1 (moved by most recent event). Among these three sites, streambed sediment was apportioned as a mix of streambank (11–84%), forest (16–38%), and pasture (1–51%); no cropland or road influence was identifiable. At the site with no discernable new sediment (BC17), this indicates that forest-sourced sediment (38%) takes multiple seasons to move through the system, while streambank-sourced sediment (60%), contributed as erosion of the steep-walled channel, is less likely to come into contact with precipitation. This fits with the location of this reach in the stream network: downstream of the ancient shoreline and at the transition to lower stream slope, enabling accumulation, with forested land upstream and a drainage area (4.4 km2) that provides sufficient energy to drive streambank erosion. In contrast, no discernable old sediment indicates recent transport of upland source material, which corresponds to the relatively low storage and small apportionment of streambank (11%) at BC18, reflecting a small contributing area (1.6 km2) and erosional environment upstream from the ancient shoreline. However, at BC21 (2.8 km2), which was similarly situated relative to the shoreline and had no discernable old sediment, 86% of streambed was apportioned as streambank, reflecting streambank erosion that exceeded all but the most downstream site. The implication is that collapsed streambank along smaller streams may be exposed to rainfall when streamflow is below bankfull.

4.3 Differentiating erosional and depositional reaches to guide management

Mean storage and erosion were skewed by individual reaches with high values; median values were lower (6 Mg km−1, 9.8 Mg km−1 yr−1), highlighting the potential of identifying and managing vulnerable sites to reduce transport. Of 25 reaches with visible streambank erosion, 17 indicated an erosional environment (S:E ≤ 1 yr) and eight a depositional environment, where streambed sediment generally accumulated from upstream erosion (S:E > 3 yr). Most sites had evidence of streambed-sediment accumulation from recent upland transport mixed with channel-derived sediment, as has been seen in similar studies (Matisoff et al. 2005; Lamba et al. 2015a; Gellis et al. 2019) and underscoring that suspended and streambed sediment interact. This included the most-downstream reaches (BC1, 3; 33 and 41% new, respectively), where sed-P was relatively low (870–1070 ppm; Fig. 2), reflecting the potential for higher streamflow to limit interaction between the water column DP and accumulated streambed sediment. These fourth-order, erosional reaches had low storage and high erosion, with > 75% of streambed sediment apportioned as streambank (Fig. 1), similar to what was seen previously at BC1 (Williamson et al. 2020a) and in Green Bay tributaries (Fitzpatrick et al. 2019; Blount et al. 2022).

The proximity of streambank to the channel facilitates sediment contribution to streambed, even during summer events, when upland sources are likely less. Upland sediment was discernable at each site; however, sources varied. Only two sites (BC1, 25) had > 3% of streambed sediment apportioned as cropland (23% and 17%, respectively). These sites are at opposite ends of the network, with remaining streambed sediment mostly apportioned as streambank at BC1, but as upland sources (48% pasture, 8% road) at BC25.

Streambed-sediment apportionment as cropland was less than the proportion of contributing area at all sites. This corroborates results from 2017 (BC1, 7, 8; Williamson et al. 2020a) and indicates that cropland-derived sediment is unlikely to settle to streambed but remains in suspension for transport to the receiving stream (the Maumee). Identification of cropland material in suspended sediment further supports this interpretation, although apportionments were consistently exceeded by streambank contributions (BC1, 7, 8; Williamson et al. 2020a).

Some combination of road, pasture, and/or forest was identified in each 2019 streambed-sediment sample, with a general inverse relation to drainage area (Fig. 1). Smaller sites had significantly less streambank and reflected contributing-area land use, similar to studies in other Midwest basins (Williamson et al. 2014; Lamba et al. 2015b; Fitzpatrick et al. 2019). Thus, upland material from non-cropland sources is likely to accumulate as streambed sediment, indicating that management of non-cropland landscape may be integral to addressing long-term sediment and P loads. Road-derived sediment was distinguishable at half of the sites, but had significantly lower P and coarser d50, indicating less potential to contribute to the P load. In contrast, pasture was distinguishable for four of five glacial-till sites and was the source with the highest sed-P and PMeh.

Pasture-derived streambed sediment equates to storage of sed-P and bioavailable P. Pasture also had the highest PSR, in a range indicating release of bioavailable P to the water column. Of the 90-km total Black Creek stream network, 20% traverses active horse pastures (Williamson et al. 2020a) that support an Amish community like those growing in 22 other states (Donnermeyer et al. 2013). Most of these pastures are on highly erodible soils (SNRT Inc. 2015; Porter et al. 2018). Expanding the management strategy for pastures and associated streams could address sediment with the potential to supply both total and bioavailable P to Black Creek, the Maumee, and the WLEB.

In contrast, both cropland and forest samples had relatively low PMeh and PSRs, indicating a potential to sorb DP during transport. This further supports research in the Maumee that P sorption to sediment is slowing bioavailable-P delivery to downstream water bodies (King et al. 2021; Williamson et al. 2021a, b). Black Creek tributaries transition from first- and second-order to third-order streams near the transition from glacial till to lake plain along the ancient Lake Maumee shoreline; similar features occur in other WLEB and Great Lakes basins (Fisher et al. 2015). Forest preservation is common along these transitions, and erosion was higher in forested reaches relative to cropland and pasture reaches (Fig. 3), underscoring that this is likely why these locations are not managed as cropland; this includes reaches in the community of Harlan. The position of forested land in the basin may present an opportunity to develop water-quality related infrastructure or conservation management to mitigate headwater sediment and P, without interfering with agricultural production.

Identifying areas where management of streambed-sediment storage and streambank erosion could be focused may facilitate successful mitigation. This analysis indicates the importance of supplementing basin-wide monitoring with reach-level surveys. Streambank erosion significantly increased with contributing area and stream order, potentially indicating a relation with the accumulation of flow or erosional potential as a function of reach-specific combinations of streamflow, slope, and width of flow in the channel. However, a third-order stream with a small drainage area had one of the highest erosion rates, exceeded only by the most downstream reach, indicating the importance of identifying reaches where the channel intersects geomorphic features that are susceptible to erosion. None of these factors had a significant correlation with in-stream storage, indicating that soil properties, stream characteristics, and land management combine to drive accumulation of streambed sediment.

Fine-grained streambed sediment stored in the Black Creek network (≈ 2200 Mg) potentially holds ≈ 2300 kg sed-P, with the estimate using the median streambed sed-P (2367 kg) similar to that estimated by integrating the source apportionment and median sed-P of individual sources (2248 kg). For context, the annual PP load (TP-DP) was 1293–10,771 kg (water years 2016–2021; median = 4951); the annual suspended-sediment load was 533–7616 Mg (median = 3230 Mg) (USGS 2023). This contrasts with what was measured in a Green Bay tributary, where fine, streambed-sediment storage amounted to 3 yr of modeled suspended-sediment load, but only 20% of annual TP (Blount et al. 2022). That Black Creek streambed-sediment storage equates to two-thirds of the annual suspended-sediment load and half of the annual PP load (20–40% of TP) makes it relevant to WLEB goals of reducing TP by 40% (Annex 4 Task Team 2015). However, these in-channel sources are commonly overlooked or are inadequately represented in basin models of P (Lamba et al. 2016).

Total bioavailable P was more difficult to estimate with available data. We estimated storage of 90 kg PMeh using the median bulk streambed concentrations from all years and 168 kg PMeh using the source apportionment at 13 sites and median source PMeh from bulk source samples. This estimate applies bulk PMeh values to the fine-grained storage volume, with the interpretation that bioavailable P is concentrated in finer particles (Sharpley et al. 1992). These estimates are an order of magnitude less than the annual DP load for Black Creek (1513–3735 kg; median = 2671 kg), almost all of which has the potential to be bioavailable (Baker et al. 2014). However, this underscores that 4–7% of streambed-stored P may be bioavailable.

5 Conclusions

This inventory of in-channel sediment and sed-P sources highlights that managing nutrients that sustain HABs could benefit from improved consideration of the complexities of streambed-sediment storage and addressing streambank erosion as a source of sediment and sed-P. Streambed-sediment sed-P concentrations exceeded those of most upland sources, indicating sorption of DP from the water column to sediment during transport which fosters increased PP through the stream network. This highlights that in-channel sediment storage has implications for the longevity of both DP and PP in this basin and downstream. Streambed-sediment storage equates to storage of sed-P and bioavailable P.

A combination of road, pasture, and/or forest was identified in each 2019 streambed sample, with a general inverse relation to drainage area; first- and second-order streams, most of which were on glacial till, more reflected contributing-area land use. Pasture was distinguishable in most glacial-till streambed sediment and had the highest sed-P and PMeh. Cropland, which was 8–60% of contributing areas, was only identified in three of 13 streambed samples (3–23%); combined with the presence of cropland in suspended sediment analyzed in WY 2018, this indicates that cropland-derived sediment remains in suspension at upstream reaches, with delivery to larger, more shallow-sloped streams. Forest source samples had similar sed-P and PMeh to cropland, and PSR was the lowest of the source types, indicating that this sediment has the potential to sorb P from the water column. These forested reaches commonly marked the transition of glacial till to lake plain, a geomorphic feature that is susceptible to erosion. Consequently, these locations may provide an opportunity to manage sediment, P, and the hydraulic energy that results as contributing area accumulates and the network adjusts to the lake-plain environment.

Larger contributing areas, that included the transition from till to lake plain, had more streambank-derived streambed sediment. Streambed sediment near this transition had no differentiable old sediment, indicating that the low S:E at these sites reflects reworking of sediment and exposure to rainfall when streamflow recedes. Otherwise, streambed sediment was generally a mix of “new” sediment that was recently in contact with rainfall and older streambed sediment, highlighting that this material moves progressively downstream during multiple events.