Purpose and Scope

The purpose of this report is to provide the city of Wildwood, Mo., with context for developing and evaluating measures aimed at mitigating the potential negative effects of future erosion along Caulks Creek through a combined monitoring and modeling study that quantifies recent and long-term rates of geomorphic change and examines the potential effects of climate change and (or) additional storage in the basin. This USGS report advances our understanding of the processes that drive fluvial erosion hazards and provides information to decision makers and the public to help manage resources to meet human and environmental needs (Evenson and others, 2013). Although the results of this study are specific to Caulks Creek, the methods and concepts are broadly applicable for future studies addressing fluvial erosion hazards along small streams. The report is structured in three parts that reflect the specific objectives of this study.

The objective of the “Part 1—Observations of Geomorphic Change in Caulks Creek” section is to quantify and describe long-term mean and recent rates and modes of geomorphic change in six study reaches. Long-term rates of channel widening and bank erosion in the study reaches were determined from analysis of historical aerial photographs. Geomorphic change was monitored through repeat terrestrial light detection and ranging (t-lidar) scans and total station (TS) surveys in six study reaches that were selected based on the locations of concern identified by the city of Wildwood (City of Wildwood, 2020; Hammer, 2020a,b). A t-lidar scan or a TS survey was collected in each of the six reaches twice per year in 2022 and 2023. The repeat surveys were compared to determine rates and patterns of geomorphic change, with a focus on bank erosion rates. Additionally, bank erosion pins were installed throughout the study area and measured several times per year during the 2-year study period.

The objectives of the “Part 2: Hydrologic and Hydraulic Response of Caulks Creek to Design Storms” section are to compare the hydraulic drivers of fluvial erosion in Caulks Creek under current versus predicted future climate scenarios and to assess the effect of additional runoff storage in the Caulks Creek basin on the hydraulic drivers of fluvial erosion. Calibrated hydrologic and two-dimensional (2D) hydraulic models were used to simulate streamflows and hydraulics for 60 design storms including all combinations of the following: (a) storm recurrence intervals (RI), with annual exceedance probability (AEP) in parentheses: 2-year (0.5), 5-year (0.2), 10-year (0.1), 25-year (0.04), 50-year (0.02), and 100-year (0.01); (b) storm durations: 6-hour and 24-hour; and (c) climate scenarios: current climate and projected future climate conditions for years 2050 and 2099 under representative concentration pathways (RCPs) 4.5 and 8.5. RCP 4.5 is considered an intermediate climate-change scenario and RCP 8.5 is considered a “worst-case” climate-change scenario. Hydrologic and 2D hydraulic simulations were completed for all 60 design storms under normal and wet antecedent response conditions, resulting in 120 simulations. An additional 72 simulations were completed to examine the effects of additional storage for design storm scenarios representing current climate and the predicted future climate in 2050 under RCP 4.5.

Lastly, the objective of the “Part 3—Simulations of Bank Retreat at Key Sites in Caulks Creek” section is to examine potential scenarios of bank retreat resulting from fluvial erosion and mass failure processes at five selected locations within the geomorphic change study reaches. A subset of the design storm scenarios was selected for simulating potential bank retreat at these five locations that includes all combinations of the following under normal antecedent response conditions, with no additional storage: (a) storm RI, with AEP in parentheses: 2-year (0.5), 10-year (0.1), and 100-year (0.01); (b) storm duration: 6-hour; and (c) climate scenarios: current climate and predicted future climate in 2099 under RCP 8.5.

Description of the Study Area

This study focuses on the Caulks Creek basin in western St. Louis County, Mo. (fig. 1). The drainage area of Caulks Creek is 17.1 square miles (mi²) at the USGS streamgage 06935830 (Caulks Creek at Chesterfield, Mo.; USGS, 2023) and 19.3 mi² at its confluence with Bonhomme Creek (USGS, 2022b). Prior to development, the Caulks Creek basin was characterized by steep, forested slopes with relief on the order of 150 feet (ft) (Hammer, 2020a). A wave of suburban development occurred in the 1990s, and by 2001, 18.1 percent of the Caulks Creek basin was covered with impervious surfaces (USGS, 2022b). By 2019, the impervious area had increased to 23.4 percent (Dewitz and U.S. Geological Survey, 2021). Development in the Caulks Creek basin was followed by stream incision and widening, in some cases resulting in an increase in channel cross-sectional area from 40 to 100 square feet to more than 800 square feet in less than a decade (Hammer, 2020a). The mean annual precipitation in Missouri Climate Division 2, which includes the city of Wildwood, was 41.35 inches during 1991–2020 (Midwest Regional Climate Center, 2024). Monthly precipitation tends to increase from January to May, then decrease from May to December (Midwest Regional Climate Center, 2024).

The hydrologic and geomorphic character of Caulks Creek gradually changes from upstream to downstream. The upstream-most reaches of Caulks Creek are ephemeral, meaning they are dry most of the year and flow only in response to precipitation. Moving downstream, isolated pools of standing water begin to appear in the deepest scour holes, then become more and more common until they roughly trace the channel thalweg (deepest part of the river channel). The gradual transition into the perennially flowing lower reach of Caulks Creek occurs near the border of the city of Wildwood with the city of Chesterfield, Mo. At USGS streamgage 06935830 (fig. 1), the mean annual flow was 23 cubic feet per second (ft³/s) for water years (WYs) 1997 to 2015 (Granato and others, 2017). Annual peak flows at USGS streamgage 06935830 ranged from 1,120 to 7,940 ft³/s for the period of record (WYs 1972 to 1979 and 1997 to 2022; USGS, 2023), and estimated peak-flow statistics at this streamgage location from USGS StreamStats are given in table 1 (USGS, 2022b). Analysis of annual peak flows at USGS streamgage 06935830 using the Mann-Kendall’s trend test (Kendall, 1975) indicated no significant change during the period of record through 2022 (tau=−0.096, probability value [p-value] = 0.432).

In the ephemeral upper reach of Caulks Creek, relatively straight sections are interspersed with high-sinuosity meander bends. The perennial lower reach is more consistently sinuous. Overall, the channel widens with distance downstream and the bed material gradually fines from a mix of angular chert cobbles, pebbles, and gravel to predominantly pebbles, gravel, and sand. The chert cobbles and pebbles are derived from coarse channel lag deposits in the banks (Hammer, 2020a,b). In some locations, large pieces of riprap have become dislodged from the bank and are located on the channel bed.

Methods for Monitoring Geomorphic Change

Descriptions of historical geomorphic change and present-day geomorphology in the six study reaches were developed based on previous reports provided to the WETF (Hammer, 2020a,b), aerial photographs from the USGS Earth Resources Observation and Science Center archive of the National Aerial Photography Program and High-Resolution Orthoimagery (USGS, 2022a), the 2017 USGS National Elevation Dataset 1-meter (m) digital elevation model (DEM) (USGS, 2018), and visual observations made in the field in 2022 and 2023. Long-term (25-year) bank retreat rates were determined by measuring the distance from the bank to a fixed object (such as the edge of a road, sidewalk, or building) along a fixed angle in aerial photographs taken in 1990 and 2015. Similarly, long-term (25-year) changes in channel width were determined by measuring the distance between the banks in aerial photographs taken in 1990 and 2015 (measured perpendicular to the banks). The 1990 and 2015 aerial photographs were captured during leaf-off conditions, which facilitates identification of the channel banklines, in contrast to more recent imagery such as the 2020 aerial photographs. These long-term rates do not necessarily indicate a constant rate of change over time; it is likely that the change observed from one aerial photograph to the next occurred episodically.

Unless otherwise specified, the bank retreat and widening rate measurements were made at the meander bend apex and may differ from the rates of change at other locations along the bend. These are point measurements of rates of bank retreat or widening and are not reach-averaged values. The bank retreat rate may exceed the widening rate at a particular location if bar deposition is occurring along the opposite bank during the same time period. The method used for the bank retreat and widening rate measurements minimizes potential errors owing to differences in the georectification of the two sets of aerial photographs. However, there is some subjectivity in identifying the bank position, particularly in older images with lower resolution, so these measurements are considered approximate.

Short-term rates of geomorphic change were determined by comparing four repeat surveys of each study reach collected between early spring of 2022 and summer of 2023. Most of the study reaches were surveyed with t-lidar, whereas TS surveys were used in study reach 3 in 2023 to facilitate measurements of bed elevation in areas with standing water. Study reaches were surveyed in early spring 2022 (February/March), summer 2022 (July/August), early spring 2023 (February/March), and summer 2023 (July) (fig. 3, table 2). The 2022 surveys bracketed several moderate flow events and a high-flow event in May 2022 (fig. 3A). Study reach 2 was surveyed after the WY 2022 annual peak flow in August 2022, but the other study reaches were surveyed prior to that event (fig. 3A). The 2023 surveys also bracketed several moderate flow events, including the annual peak flow in April 2023. Additionally, 14 bank erosion pins were installed in February/March 2022 and were re-measured five times, with the final measurements in July 2023. Additional details on the t-lidar scans, TS surveys, and bank erosion pin measurements are provided in the sections below, and these data are available in Hix and LeRoy (2024).

Terrestrial Lidar Scans

During each t-lidar survey, a tripod-mounted FARO Focus3D X-130 scanner (fig. 4A) was used to generate three-dimensional (3D) point clouds representing the bare earth surface of the bank face and portions of the channel bed. The FARO Focus3D X-130 scanner has a stated accuracy of 2 millimeters (mm) at a distance of 25 m (FARO Technologies, Inc., 2015). Each study reach required multiple t-lidar scans to cover the entire area of interest. Target spheres (fig. 4B) were set up throughout the field of view of the scanner to facilitate “stitching” adjacent scans together. Some of the target spheres were placed at known heights above surveyed control points and used to georeference the point cloud (assign real-world coordinates to individual points in the point cloud) during post-processing. The surveyed control points were established to Level II Global Navigation Satellite System (GNSS) quality (Rydlund and Densmore, 2012) using a Trimble R10 or R12 GNSS connected to the Missouri Department of Transportation Virtual Reference System, sometimes in combination with a Trimble M3 TS (Trimble Navigation Limited, 2013).

Scans provided by the tripod-mounted t-lidar scanner and control spheres were grouped by study reach and post-processed in CloudCompare version 2.12.4 (CloudCompare, 2022). The individual scans in each study reach were stitched together into point clouds by identifying the same spheres in adjacent scans using the Point-Pair Registration tool. The study reach point clouds were then georeferenced to the Missouri East 2401 state plane horizontal coordinate system and North American Vertical Datum of 1988 (NAVD 88) elevations using the Point-Pair Registration tool to align control point target spheres with the surveyed control points. The root mean square errors (RMSEs) associated with aligning the scans and georeferencing the study reach point clouds are given in table 3. The study reach point clouds were subsampled to a minimum spacing of 0.0328 ft to reduce processing demand using the space method in the Subsample tool.

Table 3.    

Root mean square errors associated with scan-to-scan alignment and georeferencing the study reach point clouds. Study reaches 1, 2, and 6 were divided into sections, labeled alphabetically from upstream to downstream and noted below the study reach number in parentheses.

[All values in table are measured in feet. RMSE, root mean square error; SR, study reach; --, no data]

Table 3.    Root mean square errors associated with scan-to-scan alignment and georeferencing the study reach point clouds. Study reaches 1, 2, and 6 were divided into sections, labeled alphabetically from upstream to downstream and noted below the study reach number in parentheses.

RMSE type	SR1 (A–E)	SR1 (F–G)	SR2 (A)	SR2 (B)	SR2 (C–D)	SR3	SR4	SR5	SR6 (A–C)	SR6 (D–E)		SR6 (F)
February/March 2022 surveys
Scan-to-scan alignment	0.057	0.057	0.059	0.059	0.059	0.030	0.013	0.078	0.010	0.010	0.010
Georeferencing	0.291	0.291	0.293	0.293	0.293	0.084	0.131	0.165	0.178	0.178	0.178
July/August 2022 surveys
Scan-to-scan alignment	0.009	0.009	0.006	0.006	0.006	0.005	0.005	0.005	0.007	0.006		0.009
Georeferencing	0.510	0.510	0.158	0.158	0.158	0.082	0.098	0.152	0.213	0.267		0.058
February/March 2023 surveys
Scan-to-scan alignment	0.032	0.005	0.105	0.105	0.105	--	0.006	0.013	0.132	0.011		0.009
Georeferencing	0.287	0.058	0.131	0.131	0.131	--	0.086	0.108	0.306	0.048		0.088
July 2023 surveys
Scan-to-scan alignment	0.025	0.009	0.017	0.017	0.017	--	0.006	0.019	0.009	0.008		0.005
Georeferencing	0.447	0.160	0.066	0.378	0.628	--	0.066	0.147	0.106	0.127		0.090

The subsampled study reach point clouds initially were classified using the CANUPO tool in CloudCompare (Brodu and Lague, 2012). Two classifiers were created, one for the February/March surveys and one for the July/August surveys, by manually sampling surface and nonsurface (for example, branches, leaves, grass) points within a subsection of a point cloud. The classifiers were then applied to split the study reach point cloud into surface and nonsurface point clouds. The initial classification was manually refined by removing points that were wrongly classified from one cloud and combining them with the correct cloud. Points representing infrastructure (for example, bridges, culverts, and pipes) were included in the surface point clouds, though bridges were later excluded from comparisons of the point clouds. The surface point cloud normals, which describe the orientation of the surface at any given location, were then computed using the CloudCompare Normals tools (Compute Normals with a quadric surface function, Orient Normals with a Minimum Spanning Tree, and if needed, Invert Normals to change the normal orientation).

The surface point clouds from consecutive surveys were compared using the Multiscale Model to Model Cloud Comparison (M3C2) tool (Lague and others, 2013). The M3C2 tool uses surface normals to measure the amount of surface change between point clouds, perpendicular to the local surface, in three dimensions. Therefore, the M3C2 tool can describe erosion and deposition of a complex 3D surface, like a river channel, where other methods may either introduce bias or limit detection of fine details (Lague and others, 2013). For study reaches 1, 3, 4, 5, and 6, the measured change between consecutive surveys was projected onto the surface cloud from the earlier survey for visualization. For study reach 2, the change was projected onto the surface cloud from the later survey owing to the substantial amount of change at that site. Any surface change values less than the maximum RMSE associated with the two surveys (table 3) were within the measurement error. Negative values of determined change in surveys indicated surface retreat outward from the channel (either bank retreat or bed degradation) and positive values of change indicated advance inward toward the channel (either bank advance or bed aggradation).

Study Reach 1 Results

Study reach 1 is a series of meander bends located upstream from the Clayton Road bridge crossing, between Woodlawn Chapel Presbyterian Church and Hope Montessori Academy (fig. 5). Study reach 1 is split into seven sections, A through G (fig. 5H), for use in the further discussion of the historical aerial photographs and the t-lidar survey results. The banks are mostly at the bottom of forested slopes, with the exception of the left bank (as viewed looking downstream) across from the Winding Trails tennis courts (section A in fig. 5H) and the right bank adjacent to Woodlawn Chapel Presbyterian Church (section F in fig. 5H), which are next to flat grassy fields. Bank erosion and bar growth have resulted in increased sinuosity, translation of bends downstream, and widening in this reach since the 1990s (Hammer, 2020b; fig. 5).

Long-term bank retreat rates based on aerial photograph analyses in study reach 1 range from about 1.1 to 4.0 feet per year (ft/yr) and widening rates range from 0.8 to 2.7 ft/yr at the 2015 apices of the meander bends (table 5). The fastest long-term bank retreat rate was observed in section C, where the downstream translation of bends has resulted in a complete change in orientation of the bend since 1990, such that the current (2023) outer bank is the left bank. Bank retreat also has occurred rapidly along the outer banks of the other bends in the reach, particularly along the fine-grained, unforested banks in sections A and F (fig. 5; fig. 6A). Large point bars also developed opposite the cut banks in sections A and F (labeled in fig. 5G). The banks in sections B–E are more heavily vegetated (fig. 6B) and include occasional bedrock outcroppings at the toe, though bank retreat has still occurred in these sections (table 5). In particular, field observations indicate that scour is occurring around stormwater runoff pipes in sections D (fig. 6C) and E (fig. 6D). Riprap boulders were installed along the left bank in the downstream portion of section E and in section G sometime between 1996 and 2002 (fig. 5C), which artificially narrowed the channel in these locations. Some of these boulders have become dislodged and are now located in the channel, though overall the channel bank position in this location has not changed substantially since the riprap was installed.

T-lidar surveys of study reach 1 were completed on February 14 and 16, 2022 (sections F–G and sections A–E, respectively); July 14, 2022; March 1, 2023; and July 25, 2023. Consecutive surveys were compared to quantify bank retreat and bed aggradation/degradation. The comparisons of these t-lidar surveys are shown in the appendix 1 figures. Throughout the study period, bank retreat was observed within sections A–F (including the nonriprapped portion of section E), whereas the riprapped portions of the banks in sections E and G were relatively stable (figs. 1.1–1.7). However, the amount of bank retreat varied substantially throughout the study reach and during the four t-lidar surveys. During the period of t-lidar monitoring, sections A and F showed the greatest amounts of bank retreat in study reach 1 along their outer banks at and downstream from the bend apices (figs. 1.1 and 1.6). In section A, the left bank retreated about 2 to 3 ft (local maximum retreat=4 ft) for February to July 2022, about 1 to 2.5 ft (local maximum retreat=4 ft) for July 2022 to March 2023, and about 1 to 3.5 ft (local maximum retreat=5 ft) for March to July 2023 (fig. 1.1). In section F, the left bank retreated about 2 to 5 ft (local maximum retreat=5.5 ft) for February to July 2022, about 0.5 to 1.5 ft (local maximum retreat=3.5 ft) for July 2022 to March 2023, and about 1 to 5 ft (local maximum retreat=6.3 ft) for March to July 2023 (fig. 1.6). In sections A and F, deposition also occurred along the bank toe or in the thalweg, which was below the area of bank retreat. This pattern may indicate that material from the top of the bank deposited on the bank toe or in the thalweg and had not yet been removed by flow in the channel.

The banks of the meander bends in sections B, C, D, and E are at the bottom of heavily forested slopes, and accumulations of large woody debris and vegetation obscured parts of those banks during the t-lidar surveys. Where the banks were exposed enough for the t-lidar to penetrate, erosion was observed along the outer banks of the meander bends, including retreat and scour at the base of the banks and in the thalweg (section B right bank, fig. 1.2; section C left bank, fig. 1.3; section D right bank, fig. 1.4; and section E left bank, fig. 1.5). In particular, in section E, the t-lidar scans show the progression of scour around a stormwater runoff pipe, resulting in the pipe cracking and collapsing (figs. 1.5E–H, pipe is also pictured in fig. 6D). However, similar to the observations from aerial photographs, the riprapped portion of the left bank of section E (labeled in fig. 1.5A) was relatively stable during the study period.

The changes in the channel bed observed in study reach 1 (figs. 1.1–1.7) can be interpreted to indicate active sediment transport during the flow events that occurred between surveys. In a general sense, the patterns of aggradation (increase of bed elevation) and degradation (decrease of bed elevation) in a river channel are tied to a combination of the spatial and temporal variations in the capacity of the flow to move sediment and the incoming sediment supply. Although measurable bed-elevation changes occurred throughout study reach 1 and in the other study reaches, it is unclear whether these changes indicate longer-term trends or whether they simply reflect the short-term variability in flows and sediment supply. Therefore, detailed descriptions of the bed-elevation changes have been excluded from this report, except where they may be related to specific process-based interpretations. For example, in section A, the deposition along the downstream edge of the point bar between July 2022 and July 2023 (fig. 1.1B and C) is consistent with the long-term observations of point bar growth concurrent with bank retreat and widening determined from the aerial photographs.

Study Reach 2 Results

Study reach 2 is a series of highly sinuous meander bends located downstream from the Clayton Road bridge crossing, which are referred to as the “triple meander” in city of Wildwood and WETF documents (fig. 7; City of Wildwood, 2020). These bends are characterized by tall, nearly vertical cut banks that are undercut in some locations and prone to mass failures (Hammer, 2020a). Study reach 2 is split into four sections, A through D (fig. 7), for use in the further discussion of the historical aerial photographs and the t-lidar survey results.

The long-term modes of bend migration and rates of bank retreat and meander neck narrowing in study reach 2 were derived from aerial photograph analysis (table 6, fig. 7). Overall, the channel in study reach 2 has widened since the 1990s. The section A bend has rotated in the downstream direction and may be beginning to develop into a double-headed meander (a bend with two local maxima in curvature, labeled in fig. 7E). The long-term mean bank retreat rate of the apex of the bend in section A is 4.4 ft/yr (table 6). Bank retreat in section A resulted in the collapse of a stormwater runoff pipe that empties into Caulks Creek (Hammer, 2020a; fig. 8, approximate location labeled in fig. 7G). The bend in section B has become double-headed during the last few decades. The long-term bank retreat rate at the apex of the bend in section B is 1.2 ft/yr (table 6). Prior to 2007, the bend in section C migrated toward Strecker Road and also rotated downstream. To protect Strecker Road from further erosion, the right cut bank of section C was laid back to a lower angle and the toe was covered with riprap in 2007–08, so bank retreat/widening rates are not reported for this section (fig. 7E). The long-term bank retreat rate at the apex of section D is 0.6 ft/yr. The necks of the meander bends in this reach (Necks A, B, and C, labeled in fig. 7H) have narrowed at rates ranging from 1.3 to 4.2 ft/yr (table 6), prompting concerns about the potential for neck cutoffs in this reach (Hammer, 2020a).

T-lidar surveys of study reach 2 were completed on February 15, 2022; August 9, 2022; February 28, 2023; and July 24, 2023 (figs. 1.8–1.11). Unlike the other study reaches, the second survey of study reach 2 occurred after the WY 2022 peak flow on August 4, 2022 (fig. 3). Widespread bank retreat was observed during the study period along the steep cut banks of sections A, B, and D of study reach 2. Some of the greatest rates of bank retreat observed in the Caulks Creek study area were documented in section A. In section A, bank retreat about 1 to 5 ft and as much as 8 to 16 ft in parts of the bend occurred along the right bank between February and August 2022 (fig. 1.8A). Such rapid bank retreat is likely the result of fluvial erosion destabilizing the bank and resulting in widespread mass failures. An incipient gully feature is also present on the right bank (labeled in fig. 1.8) that eroded throughout the study period, forming a roughly triangular indentation on the bank. Although some aggradation occurred along the edge of the point bar (less than [<]1 ft), there was no major accumulation of sediment at the bank toe between February and August 2022, which may indicate that the eroded bank material was transported out of the bend during the high-flow events. Between August 2022 and February 2023, an isolated mass failure caused bank retreat of as much as 8 ft and deposited a pile of sediment immediately beneath the failure (annotated in fig. 1.8B). Otherwise, bank retreat in section A between August 2022 and February 2023 was generally limited to the lower portion of the bank (approximately 1 to 3 ft). Substantial bank retreat of about 1 to 5 ft in section A occurred once again between February and July 2023, and a new mass failure resulted in as much as 8 to 14 ft in one location (fig. 1.8C). The location of the prior mass failure continued to erode, forming new accumulations of sediment at the bank toe.

In section B, retreat of the left bank of about 1 to 2 ft and as much as 6 ft occurred between February and August 2022 (fig. 1.9A). Minimal change occurred in section B between August 2022 and February 2023, except within some isolated locations of erosion of the bank toe and bed scour (fig. 1.9B). Bank retreat resumed in section B between February 2023 and July 2023, generally about 1 to 2 ft and as much as 4 ft (fig. 1.9C).

As noted above, section C was laid back to a lower angle and the toe was covered with riprap in 2007–08. Brushy vegetation and trees obscured much of the banks in section C during the t-lidar surveys. Along the lower portions of the section C banks where t-lidar data collection was possible, minimal change was observed during the study period (fig. 1.10). Some scour (0.5 to 2 ft) occurred in the thalweg near the section C bend entrance and at the bend apex between February and August 2022 and between February and July 2023.

In section D, between February and August 2022, the lower portion of the left bank in section D retreated approximately 0.5 to 1.5 ft with a local maximum retreat of about 5.5 ft (fig. 1.11A). Minimal change occurred in section D between August 2022 and February 2023 (fig. 1.11B). Between February and July 2023, retreat of the left bank of about 0.25 to 1.5 ft was observed throughout section D (fig. 1.11C). A few isolated locations in section D showed retreat of about 4 ft between February and July 2023; where the retreat is high on the bank, this likely represents mass failures, and where retreat is low on the bank, this likely represents erosion of the bank toe during high-flow events.

Study Reach 3 Results

Study reach 3 surrounds the Strecker Road bridge crossing at the apex of a broad, low-sinuosity meander bend near the intersection of Strecker Road with McBride Pointe Drive (fig. 9). When the bridge was constructed in 2002, riprap was placed at the bridge abutments and downstream from the bridge along the left bank (fig. 9B), and the banks were planted with vegetation intended to help prevent erosion (Hammer, 2020b). This vegetation was subsequently removed (date unknown) and invasive shrubs began growing in its place (Hammer, 2020b). There is a remnant concrete structure in the thalweg of the channel on the upstream side of the bridge (fig. 10A), which dates back at least to 1966 and may have been a low water crossing (J. Vujnich, City of Wildwood Director of Planning and Parks, written commun., 2021). There are two round bridge piers located in the center of the channel; scour holes have formed around these piers and the concrete footings of these piers are partially exposed (Hammer, 2020b; fig. 10B). Aerial imagery did not indicate substantial bank retreat or changes in channel width in study reach 3 (fig. 9).

T-lidar surveys of study reach 3 were completed on February 16, 2022, and July 12, 2022. Standing water was present in the thalweg along the outer bank of the channel around and under the bridge during the t-lidar surveys, so no data are available in those areas for the 2022 t-lidar surveys. Furthermore, the banks on either side of Strecker Road bridge are heavily vegetated, which makes it difficult to detect ground points, resulting in limited data along the banks in the comparison between the two 2022 t-lidar surveys (fig. 1.12A). Downstream from the bridge, the channel bed showed 0.25 to 0.5 ft of degradation between February and July 2022 (fig. 1.12A). To better capture the channel bed around the piers, the 2023 surveys were completed using a TS on February 27, 2023, and July 25, 2023 (fig. 1.12B). The elevations of pairs of points located around the piers from the two surveys were compared (table 7). These comparisons indicated a small amount of aggradation (0.08 to 0.88 ft) around the piers between February and July 2023.

Study Reach 4 Results

Study reach 4 is a short section of channel about 570 m downstream from study reach 3 (fig. 11). Study reach 4 is bounded on the left side by residential property and on the right side by a steep, tree-lined slope and Quaethem Drive. The left bank of study reach 4 has been eroding since at least 1990 (Hammer, 2020b), concurrent with the deposition of a bar on the right side of the channel. As a result, the channel has widened and gradually increased in sinuosity, forming an incipient meander bend. The incipient meander bend has grown in amplitude and translated downstream since the mid-2000s. The bar is separated from the right bank by a small chute channel that is visible in the 2017 DEM (not pictured) and was observed in the field in 2022 and 2023. At the apex of the incipient meander bend, the long-term mean bank retreat rate is 2.2 ft/yr and the channel is widening at a long-term mean rate of 1.8 ft/yr, based on aerial photograph analysis (fig. 11). A bank retreat rate of 4.4 ft/yr for 2007 to 2015 was calculated from the aerial photographs for the left bank at the downstream end of the study reach, where the incipient bend is beginning to translate downstream (distances annotated in fig. 11E, F, and G).

T-lidar surveys of study reach 4 were completed on March 2, 2022; July 13, 2022; March 1, 2023; and July 27, 2023 (fig. 1.13). Similar to observations from aerial photographs, retreat of the left bank of study reach 4 occurred throughout the study period. Between March and July 2022, approximately 0.5 to 1.5 ft of bank retreat occurred along the left cut bank, with some locations showing 2.5 to 3.5 ft of bank retreat (fig. 1.13A). Patches of deposition on the left cut bank may represent deposited blocks of failed bank material, similar to what is pictured in figure 1.13D. Between July 2022 and March 2023 and between March and July 2023, about 4 ft of erosion occurred mainly along the top of the left bank, whereas deposition occurred along the bank toe. This pattern may indicate that material from the top of the bank was deposited on the bank toe and had not yet been removed by flow in the channel.

A single, collapsed tree was present in study reach 4 near the left bank for the duration of the study and prevents comparison of the scans in its immediate vicinity (labeled on fig. 1.13A–C and shown in fig. 1.13D). Although trees or other large woody debris can provide some protection to riverbanks, scour can occur around an isolated, collapsed tree (or other obstacle), owing to increased velocities and turbulence of the flow around the tree, similar to scour around a bridge pier. Farther downstream from the tree, deposition of the scoured material can occur in the low-velocity wake zone. Indeed, 2 to 3 ft of retreat of the left bank occurred near this tree between March and July 2022, with as much as 1.5 ft of bed aggradation downstream from the tree (fig. 1.13A). The bed aggradation downstream from the tree is in contrast to the widespread bed degradation (0.25 to 0.5 ft) throughout the reach from March to July 2022. Between July 2022 and March 2023, erosion of the bank near the tree continued (as much as 3 ft), though any effect of the tree on the pattern of bed deposition is difficult to discern from the widespread bed aggradation in the study reach (fig. 1.13B). From March 2023 to July 2023, there was as much as 3 ft of bank retreat near the tree and an area of bed aggradation downstream from the tree (fig. 1.13C).

Study Reach 5 Results

Study reach 5 surrounds the Strecker Road bridge crossing near the intersection of Strecker Road with Quaethem Drive and includes the meander bend upstream from the bridge (fig. 12). Study reach 5 is split into two sections, A and B (fig. 12H), for further discussion of the historical aerial photographs and the t-lidar survey results. The downstream limb of the study reach 5 meander bend (section A) has widened at a long-term rate of 3.7 ft/yr since 1990 owing to bank erosion and the concurrent development of a midchannel bar. Though erosion has occurred on both sides of the channel, the left bank is unforested and consists of fine-grained sediment and therefore has retreated rapidly at a long-term mean rate of 3.6 ft/yr, according to aerial photograph analysis. Erosion of the left bank was likely accelerated by gullying associated with the outflow from a stormwater runoff pipe (labeled “Pipe outlet” in fig. 12H). Although an isolated pile of riprap was placed along the bank where the overland flow from the pipe enters the channel (labeled “Riprap pile” in fig. 12H and shown in fig. 13), this bank continues to actively erode (Hammer, 2020b). According to Hammer (2020b), the riprap may have accelerated erosion downstream owing to increased turbulence around the rock pile.

Erosion and channel widening at and downstream from the Strecker Road bridge in study reach 5 (section B) also was noted as a location of concern owing to potential erosion effects on the bridge and to Dresser Hill Road (Hammer, 2020b). At some point between 2018 and 2019, vegetation was cleared and riprap was placed on the upstream and downstream sides of the bridge and directly on the bridge abutments of both banks, as well as in two isolated piles farther downstream along the left bank (Hammer, 2020b). Although it is not possible to assess changes occurring under the bridge from aerial photographs, the channel has widened upstream and downstream from the bridge since 1990 (fig. 12). In particular, the channel immediately downstream from the bridge has widened at a long-term rate of 1.1 ft/yr, largely owing to retreat of the left bank toward Dresser Hill Road.

T-lidar surveys of study reach 5 were completed on February 28, 2022; July 15, 2022; March 2, 2023; and July 27, 2023. Between February and July 2022, the left bank in section A indicated generalized retreat of about 0.5 to 3 ft, as well as a local mass failure that resulted in 2 to 5 ft of bank retreat and deposited about 3 ft of sediment on the bank toe (fig. 1.14A). Between July 2022 and March 2023, there was widespread retreat of the left bank of section A of about 0.5 to 2.5 ft, with a local maximum retreat of about 4 ft (fig. 1.14B). Additionally, some of the sediment deposited by the prior mass failure was removed. Bank retreat between March 2023 and July 2023 was largely limited to the toe of the downstream half of the bank (0.5 to 2 ft), though removal of the sediment deposited by the mass failure in 2022 continued (fig. 1.14C). The t-lidar surveys indicated minimal change in section B of study reach 5, with the exception of some retreat of the left bank downstream from the Strecker Road bridge (about 0.5 to 2 ft each consecutive pair of surveys; fig. 1.15) and a localized mass failure of about 4 ft in the right bank upstream of the bridge (fig. 1.15C). The retreat of the left bank, downstream from the bridge, is consistent with the aerial photograph observations.

Study Reach 6 Results

Study reach 6 is a large meander bend, followed by a comparatively straight section that ends at the Strecker Road bridge crossing near the intersection of Strecker Road and Church Road (fig. 14). Study reach 6 is split into six sections, A through F (fig. 14I), for the purpose of further discussion of the historical aerial photographs and the t-lidar survey results.

Erosion of the left bank has occurred in section A, resulting in a long-term mean bank retreat rate of 1.2 ft/yr and a long-term mean widening rate of 0.7 ft/yr at the widest point of the 2015 channel (labeled “1” in fig. 14G), as determined from aerial photographs. In section B, riprap was placed along the right cut-bank sometime prior to 2002 (labeled “Riprap” in fig. 14C), which initially narrowed the channel by about 20 ft, after which there was minimal change in the position of that bank. An access trail was built across the channel in section B sometime between 2015 and 2016 (labeled “Access trail” in fig. 14H). A large pile of riprap is present along the right bank at the location where the trail meets the channel, as well as a partially exposed utility hole (labeled “Riprap” in fig. 14I and shown in fig. 15A). Farther downstream, the left bank in section C also shows signs of active bank erosion, such as a near-vertical bank face, live exposed roots overhanging and emerging from the bank, and mass failures of bank material (Hammer, 2020b; and observed in the field, labeled “2” in fig. 14G). The long-term mean widening rate in this location is 1.2 ft/yr.

There are two points of interest on the downstream limb of the large meander bend in study reach 6. A bank breach has developed on the left bank in section D, where flow has intermittently exited the main channel and deposited sand, gravel, and cobbles on the adjacent floodplain since at least 2014 (Hammer, 2020b; labeled “Breach” in fig. 14I). The apex of the right cut bank in section E (labeled “3” in fig. 14G) has eroded at a long-term mean rate of 2.3 ft/yr, according to aerial photograph analysis. The bank material in this location consists of a lower layer of coarse, noncohesive, cross-bedded channel lag deposits overlain by an upper layer of cohesive fine material (fig. 15B). Visual observations indicated a tendency for erosion of the coarse lower layer to undercut the bank.

Hammer (2020b) identified the Strecker Road bridge crossing in section F of study reach 6 as another location of concern (fig. 14I). A new bridge was constructed in spring 2010 in this area and riprap was placed along both banks surrounding and underneath the bridge. The riprap placement on the right bank, upstream from the new bridge, narrowed the channel somewhat and therefore was not smoothly aligned with the bankline immediately upstream from the riprap placement (Hammer, 2020b). As a result, there is evidence of active bank erosion immediately upstream from the riprap on the right bank in this area, in the form of a near-vertical bank face with overhanging live exposed roots (“4” labeled in fig. 14G and shown in fig. 15C). The bridge has two wall piers (fig. 15D). Large woody debris was piled on the upstream side of the right pier during the field visits for this study.

T-lidar surveys of study reach 6 were completed on March 1, 2022 (sections A–C); March 2, 2022 (sections D–F); July 12, 2022 (sections A–C); July 13, 2022 (sections D–F); February 27, 2023; and July 26, 2023. Water in the thalweg in study reach 6 prevented t-lidar acquisition of the deepest part of the channel and the bank toe. Brushy vegetation also obscured the banks in parts of study reach 6. Even so, bank retreat was measurable with the t-lidar scans in all sections of the reach. In section A, there was widespread retreat of the left bank between March and July 2022, about 0.25 to 1.5 ft, with a local area of retreat of about 2 to 4.5 ft at the location labeled “1” in figures 14G and 1.16A. There was also an area of retreat observed on the right bank (2 to 5 ft) between March and July 2022 (fig. 1.16A), though much of the right bank was obscured by brushy vegetation. Between July 2022 and February 2023, there was 0.25 to 1.5 ft of retreat on both banks in section A (fig. 1.16B), but minimal retreat occurred between February 2023 and July 2023 (fig. 1.16C).

In section B, there is an actively eroding section of the right bank, upstream from the two riprapped areas. Between March 2022 and July 2022, 0.25 to 1.5 ft of erosion occurred on the lower part of the bank, whereas 0.25 to 1.5 ft of deposition occurred on the upper part of the bank (fig. 1.17A). Close inspection of the lidar data indicated that the bank advance observed on the upper part of the bank was mostly debris (leaves and sticks). That debris was subsequently removed between July 2022 and February 2023 (0.25 to 1.5 ft), but more debris accumulated on the bank between February and July 2023 (0.25 to 1 ft) (fig. 1.17B and C). Small amounts of change were observed along the riprapped portions of the right bank, which is again attributed to the accumulation and removal of debris.

The left bank of section C includes a location noted as showing field evidence of active erosion (labeled “2” in fig. 14G and fig. 1.18A) as well as a heavily vegetated portion toward the middle of the section (fig. 1.18). The t-lidar surveys indicated active erosion upstream and downstream from the heavily vegetated portion between March 2022 and July 2022 (generally 0.25 to 1 ft, and as much as 3.5 ft), between July 2022 and February 2023 (generally 0.5 to 3 ft, and as much as 4 ft), and between February 2023 and July 2023 (generally 0.5 to 1.5 ft) (fig. 1.18). The patchy areas of bank advance are interpreted to be accumulations of debris (leaves and sticks) based on close inspection of the lidar data.

Minimal bank erosion occurred in section D between March 2022 and July 2022, except for about 0.5 ft at the bank breach (fig. 1.19). Between July 2022 and February 2023, about 0.5 ft of bank retreat occurred (local maximum retreat=1 ft) and between February 2023 and July 2023, about 0.2 to 0.4 ft of bank retreat occurred (local maximum retreat=1 ft) in section D.

Section E contains an actively eroding meander cut bank (fig. 1.20). Widespread bank retreat of about 0.5 to 1.5 ft (some patches of 2 to 3 ft) occurred in section E between March 2022 and July 2022, with local bank failures resulting in 5 to 12 ft of retreat. Some accumulation of sediment is visible below the large bank failure on the downstream end of the bend but most of the bank toe is hidden by standing water. Between July 2022 and February 2023, 0.5 to 2.5 ft of bank retreat occurred throughout the bend, and the local bank failure in the downstream end of the bend resulted in 5 to 8 ft of retreat. Lastly, between February 2023 and July 2023, 0.2 to 2.5 ft of bank retreat occurred throughout the bend, with 4 to 5 ft of retreat in the local bank failure in the downstream end of the bend.

In section F, the t-lidar surveys indicate continued retreat of a portion of the right bank located upstream from the riprap placement (labeled “4” in fig. 14G), of about 0.25 to 0.5 ft from March to July 2022, 0.25 to 0.75 ft from July 2022 to February 2023, and 0.25 to 1.5 ft from February 2023 to July 2023 (fig. 1.21). The t-lidar surveys also indicated retreat of the riprapped part of the right bank (about 0.25 ft); however, the riprap was covered with brushy vegetation that increases the uncertainty of the comparison between the t-lidar surveys. The channel bed in study reach 6 went through a cycle of mostly degradation from March 2022 to February 2023 followed by aggradation from February 2023 to July 2023.

Bank Erosion Pin Observations

The bank erosion pin observations generally indicated bank retreat throughout the study area (fig. 16). Bank retreat may be caused by removal of sediment from the bank by flowing water or mass failures of the bank. The bank retreat measured at the bank pins during the whole study period (February/March 2022 to July 2023) corresponded to annual rates ranging from 0.002 to 0.75 ft/yr and averaging 0.28 ft/yr (excludes pins 3 and 12, which showed net bank advance, and pin 4, which is missing data for part of the study period). The greatest bank retreat was observed at bank pins 6 (located within study reach 4), 8, and 9 for the interval between July 2022 and October 2022 (0.63 to 0.75 ft over 3.54 months, corresponding to a rate of 2.14 to 2.54 ft/yr). This interval included several high-flow events, including the WY 2022 annual peak flow (fig. 3). Although most bank pins showed net retreat during the whole study period, many of the bank pins indicated bank advance for some of the measurement intervals. The bank faces in which the bank pins were installed were fairly steep and it is unlikely that fluvial deposition is occurring on these surfaces. Deposition of sediment on the bank may occur as a result of slumping or the mass failure of an overlying block of sediment; therefore, the observed bank advance may be tied to erosional processes. An alternate explanation for bank advance is swelling owing to the absorption of water in the sediment.

Spatial and Temporal Variability of Morphologic Change in Caulks Creek

The analyses of historical aerial photographs indicated that bank retreat and channel widening has occurred since the 1990s in all the study reaches except study reach 3 (figs. 5, 7, 9, 11, 12, 14) at long-term mean rates ranging from 0.6 to 4.4 ft/yr. The t-lidar surveys, however, indicated that local, short-term bank retreat rates vary considerably in space and time. The amount of bank retreat observed with the t-lidar surveys not only varied substantially among the study reaches but also varied along individual bank faces. Furthermore, bank retreat rates in any particular location differed between the consecutive surveys. These results highlight the episodic nature of bank retreat. In other words, bank retreat at a given location likely occurs in short, rapid bursts interspersed with periods of minimal change, rather than occurring at a consistent, steady pace. As a result, the local, short-term bank retreat can differ substantially from the long-term mean rates.

The measurements made at the bank pins provide some additional context to the t-lidar survey, though such point measurements may not be representative of the surrounding reach, or even the local bank face. As indicated by the t-lidar scans, morphologic change along a bank face can be highly spatially variable. However, given the consistency of observed bank retreat across the bank pin locations, the results may indicate that bank erosion during the period from February/March 2022 to July 2023 was widespread throughout the study area, albeit at an overall slower rate (0.002 to 0.75 ft/yr) than the historical rate within the study reaches (0.6 to 4.4 ft/yr). Thus, bank retreat in the study area may be systemic and tied to the watershed-scale balance between the driving and resisting factors of erosion, whereas the differences in the local rates of bank erosion arise from local variations in this balance. The drivers of erosion include the hydraulic forces that entrain and transport sediment from the bank during flow events (referred to as “fluvial erosion”), which are affected by hydrologic conditions of the watershed, as well as gravitational forces acting on an exposed bank face that can result in instability and collapse (referred to as “mass failures”). The resistance to fluvial erosion is largely dependent on the grain size and cohesion of the bank material, whereas resistance to mass failure is determined by the shear strength of the bank material (Knighton, 1998). Multiple factors affect shear strength, including the bank material characteristics (grain size and cohesion), pore-water pressure, the presence or absence of tension cracks, and strength added by riparian vegetation (Langendoen and Simon, 2008). Although roots can increase the shear strength of a bank, the weight of large trees may contribute toward bank destabilization; however, the phenomenon is complex and dependent on the relative depth of the roots compared to the bank height and the bank material cohesion (Abernethy and Rutherfurd, 2000a,b; Pollen 2007; Török and Parker, 2022).

Observations from the six study reaches indicated that relatively high rates of bank retreat occurred at the outer banks of meander bends and where banks are unforested and the riparian vegetation is limited to shallow-rooted grasses. Typical bank retreat along actively eroding outer banks of meander bends in the study reaches was about 0.5 to 3 ft between consecutive surveys (approximately 5 to 8 months), though much greater values were measured at some locations (as much as 16 ft) that are likely due to mass failures. During the period of study, rapid bank retreat occurred on unforested banks in study reach 1, sections A and F; study reach 4; and study reach 5. However, bank retreat also occurred along banks located within heavily forested slopes (that is, study reach 1, sections B through E; study reach 2; and study reach 6), indicating that the processes that drive bank erosion are sufficient to overcome the resistance of the bank material in those forested sections as well. In particular, the greatest amount of bank retreat observed with the t-lidar scans occurred along the heavily forested outer bank of study reach 2, section A. The rapid retreat of that bank is hypothesized to be related to this bank being very tall and nearly vertical and the tight curvature of the meander bend. Even though the top of the bank is forested, the depth of the roots relative to the total height of the bank may be less than other locations where the banks are not as tall. It is possible the weight of the trees contributed to destabilizing the banks, but a detailed stability analysis is beyond the scope of this study. The tight curvature of the bends in study reach 2, section A may also promote rapid bank retreat, owing to curvature-induced secondary circulation during flow events, which tends to concentrate the highest velocities of a flow toward the outer bank toe (Dietrich, 1987).

Wetting-drying and freeze-thaw cycles also have likely contributed to bank retreat in Caulks Creek. During early spring surveys, as air temperatures rose above freezing and the water in the bank sediment began to melt, small amounts of material were heard and seen falling off banks and mud was observed oozing down the banks. Wetting-drying and freeze-thaw cycles also can reduce the shear strength of bank material by contributing to the formation and growth of tension cracks (Knighton, 1998).

Engineered bank protection, like riprap, is intended to increase the resistance of the banks to erosion. Riprap placements resulted in relatively stable banks in study reach 1, section E; study reach 2, section C (with adjustments to the bank angle); and study reach 3. However, riprap in study reach 5, section A has been largely ineffective as a bank protection measure and may even have exacerbated erosion along the surrounding bank, as the relatively small pile of riprap acts as a turbulence-generating obstacle to the flow. Similarly, although the riprapped portions of study reach 6, sections B and F have remained relatively stable, bank erosion has remained an issue immediately upstream from the riprap placements, which may be due to a combination of the steep banks in this area and (or) issues aligning the riprap with the local channel curvature. Furthermore, even in locations where the riprapped banks appear stable, pieces of riprap were present in the bed of the channel, indicating that it is possible for the concrete pieces to be mobilized by flows. The results of this study indicate that artificially stabilizing a bank with riprap may, in some cases, just transfer the erosion to another location.

The combination of fluvial erosion and mass failure processes likely contributes to the episodic nature of bank retreat that was previously noted. Although the amount of fluvial erosion resulting from a given flow event is expected to roughly scale with the magnitude of the flow event, any flow or sequence of flows that generates sufficient bank toe erosion to destabilize the bank could trigger a mass failure that causes rapid bank retreat. Thus, a flow or sequence of flows that are not considered exceptional or extreme could lead to rapid bank retreat if the bank is near a stability threshold. The substantial bank retreat observed in study reach 2, section A in 2022 provides an example. Several flow events occurred between the two t-lidar surveys of study reach 2 (fig. 3), including the annual peak flow of WY 2022. However, the annual peak flow of WY 2022 (August 4, 2022; peak flow=2,350 ft³/s) was not an exceptional event; the annual peak flow was higher in 13 of the previous 26 WYs (since 1997; USGS, 2023). Despite this, erosion occurred along the entirety of the outer bank of study reach 2, section A during this period, including large mass failures that contributed to up to 16 ft of bank retreat.

Methods for Estimating Precipitation-Derived Streamflow

The USGS streamgage Caulks Creek at Chesterfield, Mo., (06935830) is located near the downstream end of the Caulks Creek basin and the downstream end of the spatial extent of the hydrologic and hydraulic models (fig. 17). A precipitation-driven hydrologic model was developed to provide accurate streamflow hydrographs within the study reach to be used as inflows for the hydraulic model.

Hydrologic Model Calibration

The hydrologic model was manually (no automatic functions used) calibrated to observed streamflow data for four high-flow events between 2008 and 2022 and ranging from 1,890 to 6,520 ft³/s. The parameters CN, Tc, R, and base-flow recession were the primary parameters used in calibrating the simulated hydrograph peak and shape and were adjusted to match the corresponding hydrograph shape, magnitude, and timing for USGS streamgage 06935830. Final calibration results indicated similarity in simulated and observed hydrographs (fig. 18). The Nash-Sutcliffe efficiency (NSE) coefficient (Nash and Sutcliffe, 1970) and percentage bias (PBIAS; Gupta and others, 1999) statistics were used to assess model fit. Values of NSE can vary from negative infinity to 1. Values of 1 correspond to a perfect match between simulated and observed time series, whereas values <0 indicate the observed mean is a better predictor than the simulated values. The NSE of the four calibrated models ranged from 0.86 to 0.98, indicating that the simulated values were a good to excellent predictor of the observed hydrograph (table 11). The PBIAS is a measure of mean tendency of the simulated data to be larger or smaller than observed values. Values of PBIAS can vary from negative infinity to infinity with an optimum value of 0. The calibrated models yielded PBIAS values of 0.01 to −27.5, indicating an excellent to fair model fit (table 11). A comparison of relevant hydrograph characteristics also indicated similarities, in general, between simulated and observed peak flow and time of peak (table 11).

Table 11.    

Simulated and observed hydrograph characteristics of the calibration and validation high-flow events at the U.S. Geological Survey streamgage Caulks Creek at Chesterfield, Missouri (06935830).

[ft³/s, cubic foot per second; hh, hour; mm, minute; NSE, Nash-Sutcliffe Efficiency; PBIAS, percentage bas; CNII, Soil Conservation Service runoff curve number representing a “normal” antecedent response condition; CNI, Soil Conservation Service runoff curve number representing a “dry” antecedent response condition; CNIII, Soil Conservation Service runoff curve number representing a “wet” antecedent response condition]

Table 11.    Simulated and observed hydrograph characteristics of the calibration and validation high-flow events at the U.S. Geological Survey streamgage Caulks Creek at Chesterfield, Missouri (06935830).

Peak flow (ft3/s)			Time of peak (hh:mm)			Model performance metric
Simulated	Observed	Percentage difference	Simulated	Observed	Difference	NSE	PBIAS
September 14, 2008, CNII calibration event
6,520	6,370	2.4	09:20	09:10	00:10	0.86	−27.5
September 10, 2016, CNI calibration event
3,350	3,350	0.0	04:50	04:55	00:00	0.98	5.2
July 22, 2019, CNIII calibration event
5,130	5,090	0.8	07:20	07:40	00:20	0.94	−17.2
July 28, 2022, CNIII calibration event
1,890	1,850	2.2	16:55	16:50	00:05	0.86	0.01
June 14, 2015, CNII validation event
2,640	2,600	1.5	20:35	20:45	00:10	0.98	-8.4
August 9, 2020, CNIII validation event
4,140	3,980	4.0	3:20	3:15	00:05	0.99	3.9
August 4, 2022, CNIII validation event
2,440	2,350	4.0	1:05	1:40	00:35	0.81	−0.02

Hydrologic Model Sensitivity

The primary hydrologic model parameters that were modified for calibration were Tc, R, and CN values. The sensitivity of the Caulks Creek simulated streamflows to these parameters in addition to precipitation, reservoir storage capacity, and HEC–RAS roughness coefficients (Manning’s n-values) were assessed through sensitivity analyses. The sensitivity was analyzed by modifying each of these parameters/components by +10 percent and −10 percent and determining the resulting effect on peak flow (determined as the difference in peak flow from the base scenario) at the downstream end of the study area. All sensitivity analyses were performed using a common baseline CNII, 10-year, 6-hour runoff event. Results indicated that the generated peak flow from the HEC–HMS model was most sensitive to the CN parameter (−33.7 to 48.8 percent change in peak flow as a result of the 10-percent change in CN) followed by precipitation (−11.8- to 19.2-percent change in peak flow), with 10-percent changes in the remaining model parameters affecting peak flow by <1.8 percent (fig. 19).

Methods for Computation of Water-Surface Profiles and Hydraulic Properties

A 2D HEC–RAS model was developed for Caulks Creek based on a provided one-dimensional HEC–RAS model of Caulks Creek and Bonhomme Creek (Alecia Williams, WSP, Inc., written commun., 2021). An unsteady-state flow computation method was used in developing the scenarios for this study.

Methods for Analysis of Hydraulic Property Raster Layers

Development of Channel Centerline and Transect Geometries

The centerline of Caulks Creek was hand-digitized from just upstream from Study Reach 1 to just downstream from the Kehrs Mill Road bridge crossing in Esri ArcGIS Pro (Esri, 2022), based on the 2017 USGS National Elevation Dataset 1-m DEM and channel banklines exported from HEC–RAS. The hand-digitized centerline was then smoothed and resampled to a 3.28-ft vertex spacing. The following metrics were computed at each centerline vertex: the streamwise distance along the channel from the upstream end of the centerline (referred to herein as the “streamwise coordinate”), the channel width (based on the banklines exported from HEC–RAS), and the centerline curvature (fig. 20). The longitudinal profiles of channel width and the absolute value of centerline curvature were smoothed using a centered moving mean with a window size of 50 vertices (164 ft) for plotting. High values of curvature indicate relatively tight bends, and low values indicate relatively straight reaches. Finally, perpendicular transects were generated at each centerline vertex and assigned the streamwise coordinate of that centerline vertex. Each transect is the length of the local width of the channel, with a transect vertex spacing of 1 ft.

Hydrologic and Hydraulic Modeling Results

Velocity and Shear Stress in Caulks Creek under Current Climate and Storage Conditions

The smoothed longitudinal profiles of the transect-mean and transect-maximum velocity and shear stress at the time of peak flow at study reach 3 are shown in figures 21 and 22 for scenarios with current climate and storage conditions. The time of peak flow at study reach 3 was selected because that location is near the midpoint of the study area. Note that these profiles represent a single point in time and do not address how the hydraulic properties vary with time. The critical shear stress required to mobilize sediment with a 0.5-inch and 5-inch diameter (0.26 and 2.6 pounds per square feet, respectively) are plotted for context.

Overall, the velocity and shear stress throughout the channel tend to be greater for the design storms with long RIs owing to the greater precipitation magnitude and intensity of those storms. There are some locations where the longer RI (greater intensity) storms result in lower velocity and shear stress than the shorter RI storms (for example, near the streamwise coordinates: 2,000 ft; 11,000 ft; and 14,750 ft), owing to flow spilling overbank in these locations and expanding into a greater cross-sectional area. The CNIII scenarios result in greater velocity and shear stress in the channel than the CNII scenarios, because a greater proportion of the input precipitation reaches the channel to become streamflow under the CNIII condition. Lastly, the 24-hour storms result in slightly higher velocity and shear stress than the 6-hour storms, particularly for the shorter RI storms, but the greater total precipitation amounts of the 24-hour storms relative to the 6-hour storms are somewhat mitigated by lower precipitation intensities associated with the 24-hour events. It should be noted that the precipitation-derived peak flows for current climate conditions (table 2.1) far exceed the peak-flow estimates derived from historical streamgage data (table 15), which may indicate that the precipitation-derived peak flows, and therefore the velocity and shear stress associated with those flows, may be conservatively high.

Table 15.    

Comparison of peak-flow statistics for Caulks Creek from U.S. Geological Survey StreamStats (based on an impervious area coverage of 23.4 percent) to simulated peak flows for design storms under current climate conditions (U.S. Geological Survey, 2022b).

[AEP, annual exceedance probability; RI, recurrence interval; USGS, U.S. Geological Survey; ft³/s, cubic feet per second; 6-hr, 6-hour storm duration; 24-hr, 24-hour storm duration; CNII, Soil Conservation Service runoff curve number representing a “normal” antecedent response condition; CNIII, Soil Conservation Service runoff curve number representing a “wet” antecedent response condition]

Table 15.    Comparison of peak-flow statistics for Caulks Creek from U.S. Geological Survey StreamStats (based on an impervious area coverage of 23.4 percent) to simulated peak flows for design storms under current climate conditions (U.S. Geological Survey, 2022b).

AEP (percent)	RI (years)	USGS StreamStats peak flow (ft3/s)	Simulated peak flows (ft3/s)				Percent difference from USGS StreamStats peak flow (percent)
			6-hr storm, CNII	24-hr storm, CNII	6-hr storm, CNIII	24-hr storm, CNIII	6-hr storm, CNII	24-hr storm, CNII	6-hr storm, CNIII	24-hr storm, CNIII
50	2	2,200	4,170	5,730	8,430	9,080	90	160	283	313
20	5	3,430	6,300	7,700	11,000	11,600	84	124	221	238
10	10	4,500	7,890	10,200	13,200	13,600	75	127	193	202
4	25	5,680	11,300	13,300	15,600	16,000	99	134	175	182
2	50	7,090	13,500	15,300	17,600	17,900	90	116	148	152
1	100	8,150	15,700	17,600	20,100	20,400	93	116	147	150

The transect-mean and transect-maximum values of velocity and shear stress vary substantially along the stream (figs. 21 and 22). The velocity and shear stress at a location for a given flow will depend on a variety of factors, including slope, the size and geometry of the inundated cross-section, local roughness (affected by sediment grain size, bed and bar forms, vegetation, channel curvature), and backwater effects, which may result from structures or changes in the hydraulic character of the channel itself. Given the similarity in the overall pattern of variability in velocity and shear stress among the scenarios, the CNII 10-year, 6-hour scenario is used to examine how this variability relates to the local channel width and centerline curvature. The transect-mean and transect-maximum values of velocity and shear stress are negatively and significantly (p-value<0.01) correlated with the local channel width and the absolute value of centerline curvature, though the negative correlation is stronger for width (table 16). Therefore, faster velocities and greater shear stress occur where the channel is relatively narrow and straight. These results match expectations, as water must flow faster to move the same volume through a narrow cross-section than a wide cross-section. Furthermore, straight reaches represent a steeper path downslope than a curved reach and do not generate as much resistance to flow.

Table 16.    

Results of Pearson correlation tests for local channel width and the absolute value of centerline curvature with the transect-mean and transect-maximum velocity and shear stress for the normal antecedent response condition (CNII), 10-year, 6-hour scenario under current climate and storage conditions.

[p-value, probability value; <, less than]

Table 16.    Results of Pearson correlation tests for local channel width and the absolute value of centerline curvature with the transect-mean and transect-maximum velocity and shear stress for the normal antecedent response condition (CNII), 10-year, 6-hour scenario under current climate and storage conditions.

Parameter 1	Parameter 2	Pearson correlation coefficient	p-value
Channel width	Transect-maximum velocity	−0.55	<0.01
	Transect-mean velocity	−0.63	<0.01
	Transect-maximum shear stress	−0.40	<0.01
	Transect-mean shear stress	−0.48	<0.01
Absolute value of centerline curvature	Transect-maximum velocity	−0.26	<0.01
	Transect-mean velocity	−0.24	<0.01
	Transect-maximum shear stress	−0.25	<0.01
	Transect-mean shear stress	−0.28	<0.01

Although the results in table 16 do not address all the factors that affect the hydraulics of a reach, they lend some insight into the spatial patterns of velocity and shear stress produced by the HEC–RAS model. There are notable locations of peaks in velocity and shear stress between the study reaches where the channel is relatively narrow and straight (for example, between study reaches 1 and 2). In contrast, study reaches 1, 2, 4, 5, and 6 all contain local maxima in channel width that align with local zones of relatively low velocity and shear stress (figs. 20, 21, and 22). Substantial widening has been observed in study reaches 1, 2, 4, 5, and 6 since the 1990s (figs. 5, 7, 11, 12, 14), which is why they were identified by the city of Wildwood as areas of concern. Furthermore, bank erosion was observed with repeat t-lidar surveys in these reaches in 2022 and 2023. This raises the question of why the known areas of widening/bank erosion do not align with areas of high velocity and shear stress.

There are several factors that contribute to the complexity of the patterns of widening/bank erosion observed in Caulks Creek beyond what can be captured by a 2D hydraulic model. First, a 2D hydraulic model cannot fully capture the 3D hydrodynamics that occur in tight meander bends. In particular, the effect of curvature-induced secondary circulation on the distribution of velocity across the channel is not addressed. Second, the outputs of the hydraulic model only address the factors that drive fluvial erosion (that is, the removal of sediment from the bank by flowing water) and do not address the role of mass failures (that is, collapse or slumping of a bank under its own weight) driving bank retreat or the factors that resist erosion. The resistance to erosion and the stability of a bank is affected by multiple factors, including the bank geometry, bank material/soil characteristics and stratification, pore-water pressure, the confining pressure of stream water, and riparian vegetation (Langendoen and Simon, 2008). Finally, as the channel and landscape evolve, the flow hydraulics evolve in response. Any hydraulic model is based on a terrain and other parameters (that is, roughness or the locations of structures) that represent conditions at a single point in time. Therefore, the amount of change that has occurred in Caulks Creek in recent decades further complicates any attempts to use a hydraulic model representing recent static conditions to understand past or future change.

Effects of Projected Climate Change on the Response of Caulks Creek to Design Storms

The effects of projected climate change on the hydrologic response of Caulks Creek to design storms was evaluated by comparing the peak flows and total runoff volumes for the climate-change scenarios to the current climate scenarios (appendix 2). For all simulated scenarios, the projected climate conditions resulted in higher peak flows at the downstream end of the study area compared to current conditions. Peak flows increased by 6 to 21 percent for the year 2050 and 10 to 42 percent for the year 2099 (table 2.1). Similarly, the total runoff volumes were greater for the projected climate conditions compared to current conditions. Total runoff increased by 6 to 15 percent for the year 2050 and 10 to 35 percent for the year 2099 (table 2.2). The percentage increase in peak flow and total runoff volume relative to current conditions was greater under RCP 8.5 conditions compared to RCP 4.5 conditions and for normal (CNII) compared to wet (CNIII) antecedent response conditions for a given design storm.

The CNII 10-year, 6-hour scenario was used as a representative example for an initial, qualitative evaluation of the effect of projected climate change on the velocity and shear stress generated by the design storms. The longitudinal profiles of transect-mean velocity and shear stress for the current climate condition and the RCP 4.5–2050 and RCP 8.5–2099 climate conditions are shown in figure 23. The RCP 4.5–2050 and RCP 8.5–2099 climate conditions were selected because they provide lower and upper bounds, respectively, on the increase in precipitation and peak flow associated with the projected climate conditions. Overall, the projected climate-change scenarios have slightly higher transect-mean velocity and shear stress than the current climate condition but show the same general pattern over the length the of the study area (fig. 23). The greatest increases in velocity and shear stress for the projected climate scenarios compared to the current climate scenario tend to be located at local maxima in their longitudinal profiles.

Descriptive statistics computed over the whole study area to quantify the effect of projected climate change on the velocity and shear stress in the Caulks Creek channel are shown in appendix 3. The study area median values of transect-mean velocity (table 3.1) and shear stress (table 3.2), and transect-maximum velocity (table 3.3) and shear stress (table 3.4) all increase with the peak flow (at the downstream end of the study area) associated with the particular design storm (fig. 24). The standard deviation of these parameters over the study area also increases with peak flow (fig. 25). Therefore, for a given design storm (RI and duration) and antecedent response condition, projected climate change is predicted to result in faster flows with greater shear stress, as well as more along-stream variability in velocity and shear stress.

The projected climate scenarios resulted in higher transect-mean and transect-maximum velocities and shear stress (tables 3.5–3.8, summarized in table 17) for all projected climate scenarios except the CNII, 5-year, 6-hour, RCP 4.5–2050 scenario. The study area median increase in transect-mean velocity was as much as 6 percent for 2050 and 12 percent for 2099 (as much as 5 percent for 2050 and 11 percent for 2099 for transect-maximum velocity). The study area median increase in transect-mean shear stress was as much as 10 percent for 2050 and 20 percent for 2099 (as much as 9 percent for 2050 and 20 percent for 2099 for transect-maximum shear stress). The CNII, 5-year, 6-hour, RCP 4.5–2050 scenario was unique in that it resulted in an overall study area median decrease in the transect-mean and transect-maximum velocity and shear stress (a 1-percent decrease in transect-mean velocity, a 2-percent decrease in transect-mean shear stress, a 6-percent decrease in transect-maximum velocity, and an 8-percent decrease in transect-maximum shear stress, tables 3.5–3.8). These results are due to the slightly greater flow width and more even distribution of velocities across the width for the CNII, 5-year, 6-hour, RCP 4.5–2050 scenario compared to the CNII, 5-year, 6-hour, current climate scenario.

Table 17.    

Ranges in study area medians and standard deviations of the local percent differences between projected future climate conditions and current climate conditions for scenarios with existing storage. The ranges represent the different scenarios of recurrence interval, storm duration, and climate conditions listed in detail in tables 3.5–3.8.

[CNII, Soil Conservation Service runoff curve number representing a “normal” antecedent response condition; CNIII, Soil Conservation Service runoff curve number representing a “wet” antecedent response condition; SD, standard deviation]

Table 17.    Ranges in study area medians and standard deviations of the local percent differences between projected future climate conditions and current climate conditions for scenarios with existing storage. The ranges represent the different scenarios of recurrence interval, storm duration, and climate conditions listed in detail in tables 3.5–3.8.

Year of climate projection	Study area statistic	Transect-mean percent difference		Transect-maximum percent difference
		CNII	CNIII	CNII	CNIII
Velocity
2050	Median	−1 to 6	1 to 3	−6 to 5	1 to 3
	SD	2 to 5	1 to 3	3 to 24	3 to 6
2099	Median	2 to 12	1 to 6	2 to 11	1 to 6
	SD	2 to 9	2 to 6	5 to 23	4 to 12
Shear stress
2050	Median	−2 to 10	1 to 5	−8 to 9	1 to 5
	SD	2 to 9	1 to 3	4 to 49	3 to 13
2099	Median	3 to 20	2 to 11	3 to 20	2 to 11
	SD	2 to 10	2 to 6	5 to 48	4 to 19

Methods for Bank Retreat Simulations

The bank stability and toe erosion model (BSTEM) version 3.3.1 simulates the retreat of a bank profile owing to a combination of fluvial and geotechnical processes (Ursic and Langendoen, 2021). BSTEM first simulates fluvial erosion of the bank toe resulting from a flow event, then computes the geotechnical stability of the bank along potential failure planes. BSTEM simulates planar and cantilever failures but not rotational failures (Langendoen and Simon, 2008). Inputs to BSTEM include a bank elevation profile, bank layer thicknesses, bank material properties, and time series of water-surface elevation and shear stress for a particular flow event. Bank material properties include friction angle, cohesion, saturated unit weight, suction angle, hydraulic conductivity, critical shear stress, erodibility coefficient, and Manning’s roughness (n-value). There are two options for simulations—deterministic and stochastic. In deterministic simulations, a single set of bank material properties are entered by the user and the result is a single bank profile. In stochastic simulations, the user supplies probability distributions for each bank material property and specifies a number of realizations. A Monte Carlo analysis then generates realizations representing different combinations of the bank material properties sampled from the provided distributions. A BSTEM simulation is performed for each realization, and the results are combined to create probabilistic bank profiles for the following percentiles: 25, 50, 75, 90, 95, and 99 (that is, the 99th-percentile bank profile has a 1-percent chance of showing greater bank retreat for the supplied inputs). Thus, the stochastic simulations allow for an assessment of the range of possible outcomes when there is uncertainty or variability in the bank material properties. BSTEM also includes functionality for computing sediment loads based on grain size distributions as well as adding bank protection measures and vegetation effects to the simulations, but these options were beyond the scope of the present study.

Five locations at cut banks within study reaches 1, 2, 4, 5 and 6 (fig. 2) were selected for analysis with BSTEM. Bank profiles were extracted from the first t-lidar scan at each location, simplified to 23 data points, and used as the initial conditions for BSTEM simulations. The vertical elevations of the bank profiles are given in feet above NAVD 88, whereas the horizontal positioning (herein referred to as “station”) is given in feet along an arbitrary horizontal axis that is perpendicular to the bank face. For four of the five locations (study reaches 1, 4, 5, and 6), sediment samples were collected at three locations spaced vertically down the bank. Dry sieving and pipet analysis were used to determine grain size distributions for these samples (summarized in table 19, full distributions in tables 5.1–5.4). No grain size data are available for the BSTEM location in study reach 2 because it was added to the BSTEM analysis scope after field work was completed. Study reach 3 was not included because minimal bank retreat has been observed in that reach (fig. 9).

The grain size data and field observations (Hix and LeRoy, 2024) were used to make an initial estimate of the bank material properties based on recommended values within BSTEM (Ursic and Langendoen, 2021). The bank was divided into two layers for the locations in study reaches 4 and 5 owing to the strong contrast in grain size in the upper and lower portions of the banks (tables 5.2 and 5.3). An initial set of deterministic BSTEM simulations were run using water-surface elevation and shear stress data from the HEC–RAS simulations of the July 28, 2022, and August 4, 2022, flow events (table 9). The July 28, 2022, event was selected as a proxy for the smaller flow events that occurred between the first two t-lidar surveys of study reaches 1, 4, 5, and 6, and the August 4, 2022, event was the peak-flow event between the first two t-lidar surveys of study reach 2. Vegetation effects and bank protection treatments were not included. The final simulated bank profiles from these simulations were compared to a bank profile extracted from the second set of t-lidar surveys (July 2022 for study reaches 1, 4, 5, and 6, and August 2022 for study reach 2; table 2). The bank material properties were adjusted manually until the final simulated bank profiles approximated the bank profiles extracted from the second set of t-lidar surveys (table 20). There was minimal change in the bank profiles between the first and second t-lidar surveys for study reaches 4, 5, and 6, so a variety of parameter values were tested and values were selected based on the operator’s best judgement. This is not a true calibration, as multiple flow events occurred between the first and second t-lidar surveys but is rather an attempt to roughly constrain reasonable values for the bank material properties. The resulting bank material properties served as starting points for selecting ranges for each parameter to be used in stochastic BSTEM simulations.

Stochastic BSTEM simulations were run for each location for six design storm scenarios using a set of 250 bank material realizations that were randomly sampled from uniform distributions with the minima and maxima given in table 20. The design storm scenarios included all combinations of the following, with no additional storage and CNII conditions: (a) storm RI, with AEP in parentheses: 2-year (0.5), 10-year (0.1), and 100-year (0.01); (b) storm duration: 6-hour; and (c) climate scenarios: current climate and predicted future climate in year 2099 under RCP 8.5. The time series of water-surface elevation and shear stress near the bank were extracted from HEC–RAS for these design storm scenarios for use in the BSTEM simulations (fig. 26). Vegetation effects and bank protection treatments were not included. All BSTEM simulations, including full inputs and outputs, are available in LeRoy and Hix (2024).

Results of Bank Retreat Simulations

The bank profiles representing the 25th, 50th, and 75th percentiles of the stochastic BSTEM simulations are provided herein for all six design storm scenarios at each of the five BSTEM locations (within study reaches 1, 2, 4, 5, and 6). The complete outputs of the stochastic BSTEM simulations are provided in LeRoy and Hix (2024). It should be noted that although the stochastic simulations allow for some characterization of the uncertainty associated with the bank material properties, the bank material properties are poorly constrained by limited field data. Therefore, the results should be considered qualitative rather than precise predictions of expected bank retreat resulting from design storms. The bank profiles extracted from the second t-lidar surveys also are included for reference. However, comparisons between the t-lidar survey data and the stochastic BSTEM simulation results for the design storms are complicated by the occurrence of multiple flow events between the t-lidar surveys (fig. 3).

At the BSTEM location within study reach 1, the design storm scenarios result in as much as 1–2 ft of erosion at the bank toe, depending on the intensity of the storm (fig. 27). The 100-year storms result in roughly double the toe erosion as the 2-year storms. The RCP 8.5–2099 scenarios result in slightly more toe erosion than the corresponding current climate scenarios. The upper bank face—between 577 and 582 ft above NAVD 88—is predicted to be unstable for the selected bank material properties and fails along a near-vertical plane. According to the HEC–RAS simulations, the shear stress exerted by the flow at this location has roughly the same peak for the different design storms, but the more intense storms resulted in a longer duration of relatively high shear stress (fig. 26).

Comparing the bank profiles from the first and second t-lidar surveys at the BSTEM location in study reach 1 shows toe erosion similar to what which was predicted for the 10-year design storms (fig. 27), which indicates that there is potential for several smaller flow events to result in bank retreat that is comparable to what may be generated by a larger flow event. Furthermore, the second t-lidar survey indicates undercutting of the bank (centered around 576 ft above NAVD 88) that is not captured by the BSTEM simulations. Although a single bank layer was used in the BSTEM simulations for simplicity, the photograph of the bank face at the BSTEM location (fig. 27C) shows there are layers of coarser, but less cohesive, sediment within the bank face that appear to be preferentially eroded. The undercutting observed in the second t-lidar survey is likely related to the preferential erosion of one of these sediment layers.

The toe erosion simulated at the BSTEM location within study reach 2 ranges from 1 to 2 ft, depending on the design storm (fig. 28). Similar to study reach 1, the 100-year storms result in roughly double the toe erosion as the 2-year storms and the RCP 8.5–2099 scenarios result in slightly more toe erosion than the corresponding current climate scenarios. Additionally, the bank is predicted to fail along a near-vertical plane between about 557 and 563 ft above NAVD 88 and at the overhang between 568 and 572 ft above NAVD 88. However, the observed bank retreat between the first two t-lidar surveys is an order of magnitude greater than what either the deterministic BSTEM simulation for the August 4, 2022, event or the stochastic BSTEM simulations for the design storms predict (approximately 10 ft). Further discussion of the discrepancy between the BSTEM simulations and real-world observations is provided at the end of this section.

The BSTEM simulations for study reaches 4 and 5 (figs. 29 and 30) incorporated two distinct layers of bank material: a relatively fine and cohesive upper layer and a relatively coarse and less cohesive lower layer (table 20). The relatively high critical shear stress of the coarse sediment in the lower layer of the bank limits toe erosion at both of these locations. The upper layer of the bank shows <1 ft of bank retreat in response to the 10-year and 100-year design storm scenarios in study reach 4. In study reach 5, the upper layer of the bank shows as much as 2 ft of bank retreat in response to the 100-year design storm scenario, but the 2-year and 10-year design storms scenarios result in minimal change to the bank.

Comparing the first two t-lidar surveys at the BSTEM locations in study reaches 4 and 5 indicated that the BSTEM simulations may be underpredicting bank retreat at these locations. The simulated design storms scenarios generate higher shear stresses than the flows that occurred between the first two t-lidar surveys (fig. 26), yet the design storm scenarios generally resulted in less bank retreat than was observed during this time period. Additionally, the second t-lidar survey at the BSTEM location in study reach 5 shows evidence that failed material from the upper portion of the bank can deposit at the bank toe and may not be immediately removed. The presence of failed upper bank material deposited on the bank face and bank toe was also observed in the field. This failed material can serve to protect the bank toe from erosion during subsequent flows until it is removed by hydraulic action (Wood and others, 2001). Deposition of failed material on the bank toe is not modeled by BSTEM.

The BSTEM location in study reach 6 was not split into two bank material layers. The BSTEM simulations resulted in as much as 1 to 2.5 ft of toe erosion in response to the design storm scenarios (fig. 31). The upper portion of the bank was not predicted to fail under the simulated conditions. Minimal change was observed at this location between the first and second t-lidar surveys, but this does not necessarily indicate an issue with the BSTEM simulations, as the shear stress generated by the flows during this time period was less than that of the design storm scenarios (fig. 26).

Overall, the BSTEM simulations in this study provide a qualitative view of the relative difference in the amount of bank retreat resulting from design storms of varying magnitudes. As expected, the lower-frequency, higher-magnitude design storms resulted in more bank retreat than the higher-frequency, lower-magnitude design storms. Although scenarios with additional storage were not directly simulated in BSTEM, it is likely that the additional storage would result in reduced bank retreat compared to the same design storm with existing storage.

The BSTEM results from individual design storm scenarios are not directly comparable to long-term rates of bank retreat/widening or the recent monitoring observations from this study because multiple flow events of different magnitudes have occurred during the study period. However, those observations can help put the BSTEM results into context. The analysis of historical aerial photographs indicated long-term rates of bank retreat and widening of at least 0.6 ft/yr and as much as 4.4 ft/yr since 1990 at the BSTEM locations in study reaches 1, 2, 4, 5, and 6 (figs. 6, 7, 11, 12, and 15). The recent monitoring observations from this study highlight the spatial variability of bank retreat on annual timescales but also indicated that bank erosion is actively occurring at the BSTEM simulation locations, in some cases by several feet in less than a year (appendix 1). The flow events that occurred during the study period all had peak flows that were less than any of the design storm scenarios, though multiple flow events did occur between the t-lidar surveys. Overall, the results indicated that the BSTEM simulations may be underpredicting bank retreat, but this cannot be stated definitively.

If BSTEM simulations are underpredicting the absolute quantity of bank retreat, then there are several possible explanations that are not necessarily mutually exclusive. The bank material properties used in the BSTEM simulations could be overly conservative and do not fully describe the vertical layering that was present at the selected sites. In the case of the BSTEM location in study reach 2, additional BSTEM testing with a wide range of bank properties failed to replicate the extensive observed bank retreat. The stochastic simulations allow for some characterization of the uncertainty associated with the bank material properties; however, the bank material properties are poorly constrained by limited field data. In addition to poorly constrained bank material properties, there are important processes that are not captured by the BSTEM simulations, such as the effects of wetting-drying and freeze-thaw cycles, gullying caused by surface runoff, subsurface flow and piping effects, the potential for rotational mass failures, or enhanced toe erosion owing to impacts from transported bed material. In particular, wetting-drying and freeze-thaw cycles can generate tension cracks that substantially reduce the shear strength of bank material. Additionally, freeze-thaw cycles can generate downslope movement of material as freezing pushes material outward, which then slumps downward upon thawing. Visual field observations during spring surveys indicated that small amounts of material sloughed off the banks and mud oozed down the banks as air temperatures rose above freezing and melting occurred.

In addition to the limitations of the BSTEM model, the hydraulics simulated in HEC–RAS will never fully capture the complexity of the real world. First, a 2D model cannot fully capture the 3D processes that occur in tight meander bends, such as strong secondary circulation of the flow and submergence of the high velocity core toward the outer bank toe (Dietrich, 1987). Second, the hydraulics simulated herein are not necessarily representative of conditions in Caulks Creek during the period of record of aerial photographs. The aerial photographs indicated that the channel was previously narrower and would therefore be expected to have greater velocities and shear stress for a given flow than the wider, present-day channel (assuming all else is equal).