Sediment and phosphorus contributions from eroding banks in a large intensively managed watershed in western Iowa, United States

In this study, a new remote sensing tool was used in conjunction with sampling of river bank sediments to map channel migration patterns and estimate the net contribution of bank erosion to the sediment and phosphorus (P) budget of the Nishnabotna River in southwestern Iowa. Between the years 2009 and 2018, we found that at least 1.81 ± 0.57 × 107 Mg of sediment and 8.26 ± 2.5 × 103 Mg of P entered the Nishnabotna River due to channel migration. This equates to 0.87 Mg of sediment per meter of channel per year and 0.40 kg of P per meter of channel per year. Barring additional deposition elsewhere in the river corridor, these values represent as much as 77% of annual suspended sediment and 46% of the annual P export from the watershed. Our results also indicate that the contribution of net sediment and P volume loss by stream order increases sharply from third to sixth order, even though the total channel length is much smaller in the higher orders. These results suggest that bank erosion is an important source of sediment and P within the watershed and that future attempts to decrease riparian exports of sediment and P should focus on high‐order reaches.

These issues are particularly critical in the Central Lowlands region of the continental United States (U.S.), where patterns of land use change related to production agriculture have led to increases in sediment and P levels in local waterways (Schottler et al., 2014;Simon & Rinaldi, 2006).It has also been shown that sediment and P originating from the region's waterways are major contributors to the development of the hypoxic zone within the Gulf of Mexico (Carpenter et al., 1998).To address these issues, many states have adopted nutrient reduction strategies that seek to reduce nitrogen (N), sediment, and P loading from their watersheds.However, we still do not fully understand the sources of sediment and P within our watersheds, which makes it difficult to create effective nutrient reduction strategies.
An expanding body of research suggests that bank erosion and channel migration represent a significant source of sediment and P. Within the Midwestern U.S., studies have documented eroding bank contributions to annual sediment loads of 25%-96% (Beck et al., 2018;Belmont et al., 2011;Bosch et al., 2008;Hamlett et al., 2013;Mukundan et al., 1999;Odgaard, 1987;Thoma et al., 2005;Wilkin & Hebel, 1982).In a recent study, Schilling et al. (2022) estimated that streambanks contributed approximately one-third of the riverine total phosphorus (TP) export from Iowa.These studies are indicating that sediment and P inputs from eroding banks can be an important component of watershed sediment and P budgets and will need to be considered when seeking to reduce sediment and P loading rates.
Bank erosion is a common and essential process in dynamic alluvial rivers.At a reach scale, bank erosion can deliver zero net sediment and nutrient mass to the river if erosion is completely balanced by deposition.However, mass imbalance resulting in net transfer from banks to river may occur by one or a combination of three processes.The first and simplest process is channel widening, which can be a response to changes in flow regime or an evolutionary step in channel evolution (Simon & Rinaldi, 2006).The second process is channel lengthening (equivalently, an increase in sinuosity or the difference in arc-length between eroding and depositing banks), which causes the channel to occupy a larger area within the floodplain corridor.This would also result in a reduction in longitudinal channel slope and would accompany any re-meandering of previously straightened channels.The final process is floodplain shaving, whereby an eroding cutbank is replaced by accreting floodplain at lower elevation (Lauer & Parker, 2008).The total area of deposition in a reach multiplied by the difference in mean elevation of eroding and depositing banks represents net transfer from floodplain to river by shaving.Any one or more of these processes could result in net transfer of sediment and sediment-associated P to the river.
One major hurdle to quantifying these fluxes is the large spatial variability in the occurrence and rates of bank erosion.Unfortunately, the most common methods used to study bank erosion (pinning, ground-based LiDAR (Light Detection and Ranging), and channel delineation using aerial photography) are prohibitively time-consuming when scaled to large watersheds with more than 1000 km of channel length.In a review by Fox et al. (2016), only one study was reported that analyzed a watershed with more than 1000 km of channel length (Boynton et al., 1995), and this work was completed primarily along coastlines.Also, due to heterogeneity in soil types, erodibility, hydrologic conditions, and land management, it is often inappropriate to extrapolate erosion estimates to reaches not found within the study area (Purvis & Fox, 2016).Wolter et al. (2021) developed a method to estimate streambank erosion at a regional scale using LiDAR elevation data.By calibrating field mapping studies to LiDAR derivative data, the extent of severely eroding streambanks in third-to sixth-order (where order is defined by Strahler stream order, where larger stream orders correspond to larger streams) streams and rivers in Iowa was estimated.Overall, methods are needed for estimating bank erosion that can better incorporate watershed heterogeneity over large spatial extents to allow for a better estimation of eroding bank contributions to sediment and P budgets at regional scales.
To accomplish this, we have developed the aerial imagery migration model (AIMM).As reported in Williams et al. (2020), AIMM has been shown to produce results that are consistent with manual bank delineation studies while also providing an estimate of net eroded volume and greatly increasing the efficiency of analysis.AIMM makes use of channel position change detection techniques found within previous channel migration models (Monegaglia et al., 2018;Rowland et al., 2016), while also optimizing these techniques for use with high-resolution aerial imagery.In addition, AIMM combines its estimation of eroded and depositional area with a digital elevation model (DEM) analysis to estimate volumes of erosion and deposition.
In this study, we seek to quantify sediment and P loading rates of eroding banks within a large watershed by combining AIMM's net eroded volume estimates with watershed-wide soil analyses of erosional and depositional material.This approach differs from other recent studies in the region (e.g., Schilling et al., 2022) by explicitly quantifying sediment and P deposition in addition to bank erosion.Our study takes place within the intensively managed Nishnabotna River watershed, located in SW Iowa (Figure 1), which has a total watershed area of 7252 km 2 and total channel length of 2317 km.Results are compared with estimates of total watershed P export to evaluate the relative importance of streambank contributions.The outputs of this analysis also allow us to identify where the highest rates of net sediment and P input occur within the watershed, allowing us to better prioritize conservation efforts.

Research Impact Statement
We summarize the application of a new geospatial analysis tool, aerial imagery migration model (AIMM), to estimating net sediment and P inputs from riverbank erosion.

| Study area
The Nishnabotna River watershed (Figure 1) is composed of two hydrologic unit code (HUC) 8 watersheds, the East and West Nishnabotna Watersheds, which are located in southwestern Iowa and have total watershed areas of 2975 and 4277 km 2 , respectively.These watersheds are underlain by Pre-Illinoian Till capped with loess that was blown east from the Missouri River floodplain during the retreat of Wisconsinan ice sheets (Yahner, 2016).This loess ranges in thickness from 47 to 62 m and decreases in thickness from west to east (Bettis, 1990).The channel banks of the Nishnabotna watershed are composed primarily of the Holocene-age DeForest Formation, which itself is composed of three main members: the Camp Creek, Roberts Creek, and Gunder Members (Bettis, 1990).The Gunder Member is composed of oxidized silty and loamy material that is relatively cohesive, and often represents the modern channel bottom.The Roberts Creek and Camp Creek members are both silty and loamy alluvium that are highly erodible.Channel networks are young, and both bank and gully erosion are major conservation issues within the watershed (Hadish et al., 1994;Thomas et al., 2004;Tomer & James, 2004).

| Erosional volume
Because the total length of the Nishnabotna River system is too large to efficiently survey or digitize, the AIMM river monitoring tool (Williams et al., 2020) was used to estimate the net volume of sediment mobilized due to lateral river migration and erosion (code and toolbox available at: https://github.com/forrestfwi lliam s/AIMM_Stable).AIMM is an ArcPro-based tool that combines a change detection analysis of river presence between two time periods and a statistical estimation of bank height to create a spatially explicit estimate of the volume of erosion and deposition caused by river migration (Figure 2).While similar tools exist (e.g., RivMAP, Schwenk et al., 2017), AIMM was chosen because it is optimized for use with high-resolution (<1 m) imagery and includes an estimate of bank height, which allows us to estimate the net volume of erosion.
AIMM has five main inputs: two RGBI (red-blue-green near-infrared) images, a high-resolution DEM (preferably LiDAR-based), a river centerline dataset (with associated Strahler stream orders), and an order-stratified estimation of average channel width.The RGBI images are used to create normalized difference water index (NDWI) images that are used to identify river and non-river areas using Li's entropy thresholding (Li & Lee, 1993), the DEM is used to calculate bank height, and the river centerline dataset is used to constrain the analysis to the local floodplain.AIMM's outputs include a raster that identifies zones of erosion and deposition, as well as polygons for each zone of erosion and deposition that have associated bank height and volume attributes.AIMM utilizes the river centerline data to both constrain the data in the thresholding step and limit analysis to the river corridor.We used a one-width buffer of the centerline dataset in the thresholding step to create a more balanced water/non-water sample set but included all erosional and depositional zones within a two-width buffer.The larger buffer in the second case allowed us to capture movement in areas with high rates of river migration while also excluding spurious zones of erosion and deposition that resulted from misclassification of the aerial images.The river centerline dataset was provided by the Iowa DNR (State of Iowa Open Geospatial Data: https://geoda ta.iowa.gov/)and was sourced from a hydrological analysis of the aforementioned DEM, in combination with manual editing of sites with obvious errors in stream routing.Order-stratified average river width was estimated by measuring river width at the top, middle, and bottom of six reaches per order for the stream orders 1-6, then averaging by order.
Initial tests showed that the Li's entropy NDWI threshold values were not uniform across the study area, which was likely due to differences in atmospheric or ground conditions between the dates of image capture.Consequently, AIMM was run independently for each county mosaic; then, these data were merged to obtain the final estimates of erosional and depositional volume.AIMM classification results were validated by comparing model output with the results of manual bank delineations in 24 test reaches spanning orders 3-6.Pixel classification agreement between AIMM and manual delineation was quantified using Cohen's kappa, following the approach outlined in Williams et al. (2020).This same comparison was also used to quantify model error, as outlined below.

F I G U R E 2 Generalized workflow of AIMM (aerial imagery migration model).
A pair of red-green-blue-infrared aerial images of the same location is classified using the normalized difference water index algorithm.The classified images are then compared to detect temporal changes corresponding to channel migration.Changed pixels are categorized as erosion or deposition and vectorized into eroded or deposited areas.Elevation changes associated with erosion and deposition are then extracted from a Light Detection and Ranging digital elevation model to compute volumes.For a more detailed description of AIMM, the reader is referred to Williams et al. (2020).
Differences in the area (or number of image pixels) occupied by water between two sets of images (W 1 and W 2 ) could arise from either differences in discharge (ΔW Q ) or the effects of real geomorphic change (ΔW G ).More formally, W 2 − W 1 = ΔW Q + ΔW G , where a positive change corresponds to an increase in water area between images 1 and 2 (Figure 3).Real changes in river area unrelated to differences in flow between images could result from widening or lengthening (or their opposites), each of which would manifest as differences between the number of erosion (E) and deposition (D) pixels or area: ΔW G = E − D. We can therefore estimate the relative sizes of discharge-and geomorphologyrelated changes in water area by comparing the observed value of W 2 − W 1 with the observed difference E − D (= − 0.90 × 10 6 m 2 ).

| Field sample collection and processing
To estimate the sediment and P inputs associated with eroding banks, we collected samples of eroding and depositional bank material across the watershed.To ensure a sample well distributed across stream size, three sample sites were randomly selected within the river corridor of each stream order represented in the watershed (1st-6th order), for a total of 18 depositional sample sites and 18 erosional sample sites.Since erosional sites did not necessarily have a corresponding site of deposition, depositional and erosional sites were selected independently (Figure 4).Erosional samples were obtained in the form of sediment cores that were collected using a truck-mounted hydraulic probe (Giddings Machine Co.) at sites no more than 5 m from actively eroding banks.Core lengths ranged from 3 to 5 m and corresponded to the height of the eroding bank from which they were taken.All cores were transported to a lab setting where they were described, classified into the three members of the DeForest Formation, and subsampled at the top, base, and at every 0.5-m interval within each member using a 2-cm diameter cylinder.These samples were then analyzed to identify their bulk density, TP concentration, and particle size distribution.Depositional samples, on the other hand, were collected using 7.62-cm long, 7.62-cm diameter, open-ended, Uhland-type cylinders at the upstream, middle, and downstream portions of each depositional site.Each subsample was independently analyzed to measure its bulk density, TP concentration, and particle size distribution.
We determined the dry weight of each sample by drying the samples at 105°C for a minimum of 24 h, or until the sample's weight stabilized.
We then divided the dry weight by the total sample volume to determine the bulk density.We measured sample TP content using the aqua regia digestion method (Crosland et al., 1995) followed by inductively coupled plasma atomic emission spectroscopy, and measured sample particle size using laser diffractometry (Miller & Schaetzl, 2012).

| Estimation of net export and uncertainty
We calculated the net export of sediment (in dimensions of mass, [M]) and P [M] from the watershed using the formulas: for the net sediment export, and: for the net P export, where v e is total volume of erosion, v d is the total volume of deposition, b e is the average bulk density of the erosional samples, b d is the average bulk density of the depositional samples, p e is the average TP concentration of the erosional samples, and p d is the TP concentration of the depositional samples.In the case of both v e and v d , we calculated total volume according to the formula: (1) (2) where a i and h i are the area and height, respectively, of feature i and a feature corresponds to a vectorized pixel region identified as either erosion or deposition.We also used these equations to calculate the sediment and TP contributions of each order individually to facilitate a comparison between the orders.
To constrain the error inherent in our analysis, we estimated the error in each element individually, then propagated these errors through the equations described above.Following the approach of Mount and Louis (2005) and Rowland et al. (2016), we estimated the error in our areas of erosion and deposition by comparing the total area of each feature identified using AIMM to the total area of each feature identified when manually delineating the location of the river in each image within selected reaches.In reaches surrounding each erosion sample site, the banks were hand delineated, and the results of overlaying these hand delineations in a manner similar to Tomer and Van Horn (2018) were compared to AIMM's results.To estimate the error in our measurement of feature height, we used transects derived from the DEM to estimate the heights of erosional and depositional features within each delineated reach.We defined the error in our measurement as the mean absolute error (MAE) between AIMM's result and the result of the manual approaches.We chose to use MAE and not root mean square error (RMSE) because RMSE is more sensitive to large outliers and is less appropriate for propagating error in river change studies (Mount & Louis, 2005).For the bulk density and TP measurements, the standard error of the sample population was used rather than measurement error because the former better represents the inherent uncertainty in the representativeness of these metrics.When we estimated sediment and TP export by order, we also performed our propagation of error analysis by order as well.

| Sediment analyses
In total, 18 eroding bank sediment cores and 18 channel deposit samples were obtained for analysis and were assessed to determine their TP concentration and bulk density.All uncertainties for soil analyses are reported in terms of standard error.According to an ANOVA test, both sets of samples showed significant variation in TP content by stream order.For erosional banks, first-order samples had the lowest mean TP content F I G U R E 4 Sample sites for the order-stratified sampling of erosional and depositional bank material within the Nishnabotna Watershed.The target reach identified as TR is the reach found to have the highest net erosion in the watershed.It is further described in the results section and in 9.
(245 ± 20 mg/kg), and fourth-order samples had the highest (587 ± 66 mg/kg).In general, eroding bank TP content increased from first to fourth order, then decreased from fourth to six (Figure 5).Both fifth-and sixth-order samples had TP concentrations that were less than third order, but greater than second order.Depositional bank samples generally decreased in TP concentration from first (505 ± 72 mg/kg) to sixth order (235 ± 40 mg/kg).There was no significant difference in TP concentration between erosional and depositional samples when averaged across all orders, but depositional samples had higher TP concentration in first-and second-order samples and lower TP concentration in all other orders.Similar to TP concentration, there was a significant difference in bulk density values between erosional and depositional features (Figure 6).Variations in bulk density by stream order were also found to be significant for both feature types.For erosional banks, the highest average bulk densities were found in first and sixth orders (1.29 ± 0.3 and 1.30 ± 0.4 g/cm 3 , respectively), while the lowest average bulk density was found in the fourth order (1.15 ± 0.3 g/cm 3 ).The peak in TP concentration and a low point in bulk density within the fourth-order erosional samples is an interesting result of this analysis.The depositional samples displayed a steady increase in average bulk density with increasing stream order from first order (1.05 ± 0.06 g/cm 3 ) to fifth order (1.61 ± 0.09), but slightly decreased from fifth to sixth order.
Results of the particle size analysis show that the majority of the erosional samples can be classified as either silt loam or loam (Figure 7).
Sixth-order erosional samples had higher sand content than the other orders, and the Gunder member tended to have more samples with sand content greater than 50%.The differences between orders and units however are dwarfed by the differences between erosional and depositional samples.On average, depositional samples had more sand and less silt than erosional samples, with sand content increasing by order.

| Exported sediment volume
Our initial investigations of the AIMM model outputs within the first-and second-order reaches of the Nishnabotna River system suggested that the 0.61-meter resolution of the imagery was too coarse to reliably resolve channel change in these small, often-wooded stream corridors.
Consequently, we decided to exclude these reaches from our analysis and will focus on stream orders three and above for the remainder of this analysis.Uncertainties for depositional and erosional volume estimates are reported in terms of standard error.Validation results indicated good agreement between AIMM and manual classifications (85% agreement, Cohen's κ = 0.71).
AIMM identified approximately 71.9 × 10 6 pixels as water in the 2009 imagery and 61.5 × 10 6 pixels as water in 2018 imagery along thirdorder and larger reaches in the watershed.This difference, corresponding to 14.5% less water and 3.88 × 10 6 m 2 more land in the later image, can be partly explained by the lower flows during the second imagery campaign, as discussed below.Approximately 12.90 × 10 6 pixels were identified as erosion and 15.33 × 10 6 pixels as deposition basin-wide, resulting in depositional area exceeding erosional area by 0.90 × 10 6 m 2 .This increase in depositional area was distributed throughout third-through fifth-order reaches, while erosion pixels outnumbered deposition pixels in the sixth-(and short seventh-) order reaches.The mean bank height for eroded banks was 3.23 m (±2.43 m standard deviation), while depositional banks averaged 0.68 m (±0.84 m standard deviation).Both bank height distributions were positively skewed.Total eroded sediment volume exceeded deposited volume within the watershed, but the imbalance varied with stream order.Total depositional volume increased from third-to fourth-order reaches, but then decreased in orders 5 and 6 (Figure 8).Depositional volume was greatest in fourth-order reaches, and deposition had the largest proportional impact in third-order reaches, where depositional volume was 92% of erosional volume.Conversely, erosional volume was smallest in third-order reaches, was relatively similar in fourth-and fifth-order reaches, and largest in sixth-order reaches.Overall, sixth-order reaches were by far the largest contributors of sediment by volume, having both the largest volume of erosion, and lowest volume of deposition.This resulted in sixth-order reaches contributing 67% of the net sediment F I G U R E 6 Sediment bulk density by stream order.Bulk density generally increased with stream order for depositional samples, but bulk density for erosional samples followed the inverse trend to the one displayed in TP content.Bulk density decreased from first to fourth order, then increased from fourth to six.volume export even though they only represent 11% of the total channel length for stream orders 3 and higher (Table 1).In fact, one 11.5 km sixth-order stretch of the Nishnabotna River contained 6.7% of the total erosion in the watershed (Figure 9).This only represents 0.1% of the watershed total channel length and is slightly more than all the erosion identified in third-order reaches.

| Exported sediment and phosphorus mass
Based on a comparison of manual analysis with AIMM within our uncertainty analysis, we estimate the MAE in our depositional/erosional area and bank height estimates to be 35% and 41%, respectively, of their total values.Conversely, the standard errors of our bulk density and TP measurements were 2% and 7%, respectively, of their average values.This clearly indicates that most of the error in our analysis is derived from our calculation of the net volume of sediment loss.Consequently, variations between the stream orders in exported sediment mass and P are largely explained by the variations in the exported volume of sediment (Figure 10).
During the 9-year study period from 2009 to 2018, we estimate that river migration led to the export of 1.17 ± 0.47 × 10 7 m 3 of sediment from the Nishnabotna River system.When combined with our soil analyses, this corresponds to a net sediment mass export of 1.81 ± 0.57 × 10 7 Mg and 8.26 ± 2.5 × 10 3 Mg of P. When these results are framed in terms of sediment/P contribution per meter of channel per year, we find that channels in the Nishnabotna Watershed exported an average of 0.87 Mg of sediment per meter per year and 0.40 kg of P per meter per year between 2009 and 2018.

F I G U R E 8
Volume of sediment contribution by stream order.Erosional volume generally increased from third to sixth order, and depositional volume peaked in the fourth order.Overall, sixth order reaches contributed the majority of sediment volume even though they represent a small portion of the watershed.

TA B L E 1
Total channel length and net sediment volume loss grouped by Strahler stream order for the Nishnabotna River system.While orders 3 and 4 contain the majority of the total channel length, orders 5 and 6 are responsible for the majority of the net volume loss.

| DISCUSS ION
This study has demonstrated the application of AIMM for efficiently estimating net streambank erosion in a large river watershed in the intensively farmed Midwestern U.S. AIMM indicates that the Nishnabotna River system is dynamic, and appears to have introduced more sediment and P to the river by cutbank erosion than was redeposited along channel margins between 2009 and 2018.The sixth-order mainstem reaches were particularly important for sediment and P contributions.In this section, we place these results in a regional context and compare with relevant prior work.We also review the limitations of AIMM and suggest avenues for improvement.
In general, we found that our measurements of the properties of DeForest Formation members were similar to a recent study in central Iowa (Beck et al., 2018), but we did find less variation between the individual members than was found in that study.This could be attributed to both differing sampling methodologies and differences in local geology (western vs. central Iowa).
In erosional sites, TP concentrations peak in the fourth-or fifth-order reaches before diminishing slightly in the sixth-order reaches.Elevated sand content within the upper portions of sixth-order cores suggests that this trend could be due to the higher number of overbank events that deposit sand-rich layers there.Overbank events are more common in these downstream reaches, and thus, low-density and low-P sand deposits make up a larger proportion of the sixth-order bank materials.These differences in bulk density and P content are, however, dwarfed by the differences in erosional volume, which makes sixth-order reaches the dominant source of sediment and P within the stream orders studied.
F I G U R E 9 This 11.5 km stretch of the East Nishnabotna River accounts for 6.7% of the total erosion in the watershed, but only represents 0.1% of the watershed total channel length.The left panel displays an aerial view of the river reach (imagery provided by Google Maps), and the right panel displays the areas of deposition and erosion as identified by AIMM.
Depositional samples generally indicated increasing sand concentrations in higher order reaches, and a corresponding decline in TP concentration.This is at least partly related to the increasing abundance of well-developed point bars with coarse lateral accretion deposits in the dynamic and sinuous high-order reaches.The extent and height of depositional banks was slightly higher in fourth-order reaches than fifth and sixth order, though, leading to a peak in depositional volume in fourth order.Even so, depositional sediment volume and TP mass were far smaller than corresponding erosional contributions, causing a reach-scale sediment and TP mass imbalance that produced net transfer from banks to the river.
Our AIMM results can help to evaluate whether widening, lengthening, or shaving contribute most to the net sediment and P transfer.
First, however, we must address the extent to which differences in streamflow on the imagery dates influence our channel migration results.
Following the logic of Figure 3, we find that the change in water area between images due to differences in discharge ΔW Q is − 2.98 × 10 6 m 2 , or nearly 77% of the observed change in water area.Since depositional banks (point bars) tend to have lower slopes than erosional banks (cutbanks), we infer that the lower discharge and corresponding decrease in water area between the 2009 and 2018 imagery is reflected primarily in larger exposed point bars in 2018.This would lead to AIMM overestimating depositional area as the main consequence of the differing discharges.
Both channel widening and channel lengthening should manifest as net increases in water area if erosional area exceeds depositional area.
Our finding that depositional area exceeded erosional area over the whole basin is not consistent with this, though apparent overestimation of depositional area allows for some possibility that these processes contribute.Such a contribution is particularly likely in the highest order reaches (6 and 7) where erosion was much greater than deposition.However, the more than 2.5-m difference between the average eroded bank height and average deposited bank height points to floodplain shaving as likely the dominant process accounting for net sediment transfer.
Since most previous studies have focused on the erosional yield associated with channel migration, we emphasize the inclusion of deposition and storage of sediment and P that is also a component of channel migration.Accounting for in-channel deposition during the channel migration process can provide a more complete accounting of the role that bank erosion plays in the export of sediment/P from watersheds.
Despite this improvement, vertical sediment accretion and P transfer to storage on the floodplain is not included in the analysis.This may not be a major drawback, however, since there were few out-of-bank events in the Nishnabotna River during the study period.Nevertheless, developing observational methods to account for floodplain sediment and P storage in large river networks remains an important area of future research need.
When only erosive input is considered, our approach estimates total P inputs to be 0.42 kg of P per year per meter of channel length.This result is comparable to the values reported in earlier studies (Table 2), despite the order of magnitude larger watershed size and channel length F I G U R E 1 0 Mass of sediment and P contributed by stream order for the 9 years included in this study.For both graphs, trends in sediment contribution by order are similar to those found in Figure 8, but the influence of fourth-order streams increases in the P export graph due to the relatively high concentration of P in fourth-order sediments.
studied herein.When we account for the re-deposition of P-laden sediment within the channel, we find that loading to be reduced by just 0.02 kg per meter per year, or 5%.and 33% (though this may still be too large; cf.Streeter et al., 2021), we can roughly estimate that mean annual upland sediment supply to the river during that period lies between 0.99 and 3.28 × 10 6 Mg.Even when the lower SDR value is used, the sum of net bank and upland erosion inputs (3.0 × 10 6 Mg) exceeds but falls within one standard deviation of export estimates.
Recent studies of riverine P export can also be leveraged to contextualize our results.15% and 143%.Hence, the estimated contribution of streambanks to P export from the Nishnabotna River above Hamburg is consistent with long-term regional estimates.
While AIMM allows us to effectively scale bank delineation studies, it is still the dominant source of error within our analysis.Limited channel migration in first-and second-order reaches (compared to image pixel size) necessitated eliminating these reaches from analysis.Increased availability of high-resolution (<1 m pixel size) imagery could improve the minimum detectable change in principle, but would require consistent and unobscured water levels between images.AIMM is also ill-suited for use within river reaches that have over-hanging tree cover; in such cases, manual delineation may still be the best existing option for analyzing aerial imagery.Additionally, AIMM's estimation of bank height is dependent on the quality of the underlying DEM, which in some cases led to poor results.As LiDAR and high-resolution imagery technology improves, however, the accuracy of approaches similar to AIMM will improve as well.Despite these challenges, AIMM performs well in medium to large rivers where lateral channel changes exceed several times the pixel size during the interval between images.
Neither AIMM, hand delineation nor erosion-pin methods are effective tools for monitoring in-channel sources of sediment, such as vertical incision and gully erosion, but repeat aerial LiDAR surveys offer a way to quantify these aspects of the erosional system.Another feature that would greatly improve the utility of AIMM would be the introduction of a spatially dependent NDWI thresholding approach.Selection of a single NDWI threshold value to distinguish water from land in a large study area results in classification errors when environmental conditions varied across image acquisition dates in the region of interest.Our use of different thresholds for each county image mosaic was an acceptable solution to the presence of non-stationary threshold values, but a fully automated approach to selecting optimal thresholds with more specificity would lead to better results.
The most important findings of this study are that eroding banks are a major source of sediment and P within the Nishnabotna River system, and that the highest order reaches contributed the most sediment, even though they represent the smallest proportion of the total channel length.Wolter et al. (2021) reported that bank heights and the percentage of bank length eroding increased substantially with order in the ecoregion containing the Nishnabotna River.These results suggest that bank erosion mitigation efforts in higher order reaches may be the most cost-effective way to reduce the sediment/P export rates of rivers.It is important to note, however, that the sediment/P inputs TA B L E 2 Channel length observed and reported phosphorus (P) load as a bulk value and as a channel length normalized value.P load per meter of channel length per year ranges from 0.02 to 3.57, with the results from this study falling in the middle of this distribution at 0.42 P kg/m/year.addressed here are related to channel migration only and neglect the effects of other geomorphic processes such as overbank sedimentation, channel incision, and gully head migration that may be important in headwater reaches.

| CON CLUS ION
In this study, we used the AIMM model in conjunction with laboratory analyses of erosional and depositional sediments to estimate the effects of channel migration on sediment and P contributions to surface waters in the Nishnabotna River system in southwestern Iowa.We found that channel migration contributed 0.87 tons of sediment and 0.40 kg of P to the Nishnabotna River per year per meter of channel length between the years of 2009 and 2018 in stream orders of three and higher.These sediment and P contributions amount to approximately 77% and 43% of estimated exports from the Nishnabotna Watershed, respectively.Due to the directionality of expected errors, and the omission of headwater reaches that would likely be net sediment and P sources rather than sinks, these numbers likely represent conservative values.Our results further indicate that the sediment and P contributions from streambank erosion are dominated by the highest order reaches (Strahler stream order 6), even though these reaches represent a small portion of the total channel length.These results suggest that bank erosion is an important source of sediment and P within the watershed and that future attempts to decrease riparian exports of sediment and P should focus on managing bank erosion in reaches that are larger than those that are commonly considered.
In this analysis, two color-infrared aerial imagery sets of western Iowa from the spring of 2009 and 2018 (State of Iowa Open Geospatial Data: https://geoda ta.iowa.gov/)were used to perform the analysis.The 2009 (March 14) and 2018 (April 19) imagery were selected because F I G U R E 1 Map of the Nishnabotna watershed.The Nishnabotna watershed is composed of the East and West Nishnabotna River systems and has a total watershed area of 7252 km 2 .they are the two most recent surveys that were conducted during leaf-off conditions.Leaf-off conditions were desired because riparian tree cover limits the accuracy of both hand delineation and AIMM methods.The 2009 survey was conducted by the Iowa DNR during the spring of 2009, with a flight height of 6100 m above ground level.Images were georectified, cut into a tiled grid, and then converted to county mosaics with 0.61 m (2 ft) spatial resolution.The measured positional horizontal accuracy of these images is 3 m at a 95% confidence level.The 2018 survey was conducted by Surdex on the behalf of the Iowa Department of Administrative Services between April 19 and May 5 with a flight altitude of 3750 m above ground level.The imagery was orthorectified and mosaiced into county mosaics at a 0.30-m (1 ft) spatial resolution.The measured positional horizontal accuracy of these images is 0.61 m at a 95% confidence level.The elevation data were sourced from a 2-m resolution LiDAR-derived DEM that has been hydro-conditioned (State of Iowa Open Geospatial Data: https://geoda ta.iowa.gov/).The LiDAR survey was conducted in the summer of 2009 and most closely reflects the conditions found in the 2009 aerial surveys.Discharge of the Nishnabotna River above Hamburg, IA (USGS gage number 06810000) was 2160 cfs on the image date in 2009 and 1710 cfs in 2018, representing a stage difference of 0.7 ft.

F
I G U R E 3 Schematic illustration of the changes in river sub-reach area due to geomorphic changes and changes in stage.Symbols are defined in the main text.

F
I G U R E 5 Total phosphorus (TP) concentration of erosional and depositional banks by stream order.TP concentration generally decreased from first to sixth order for depositional samples, and erosional TP concentration peaked in the fourth-order samples.Error bars display the standard errors.INTENSIVELY MANAGED WATERSHED IN WESTERN IOWA, UNITED STATES

F
Sediment particle size content for erosional samples within the Camp Creek, Robert's Creek, and Gunder stratigraphic members, as well as the depositional samples.The black square and black circle correspond to the average deposition and erosion values, respectively.The right portion of the figure displays the USDA soil texture classification for each region of the figure.
(Gelder et al., 2018ent and P watershed budgets are beyond the scope of this analysis, we can gain some context by examining our results in a budget context alongside other studies relevant to the area.Model simulations of Nishnabotna River total suspended solids loads calibrated with direct sampling over the period 2011-2018 suggest an average annual suspended sediment export of 2.59 ± 1.33 × 10 6 Mg(Anderson, 2022).Our estimate of mean annual net sediment loading from channel migration (2.01 ± 0.63 × 10 6 Mg) is more than 77% of this export.The Daily Erosion Project (DEP)(Gelder et al., 2018) is an event-based implementation of the Water Erosion Prediction Project model for Iowa and can be used to estimate upland soil erosion inputs to streams.For the period of our study, the DEP model estimates that the total mass of surface erosion within the Nishnabotna River basin was 8.95 × 10 7 Mg.If the sediment delivery ratio (SDR) is taken to lie between 10% Table adapted from Fox et al. (2016).