Sediment resuspension mechanisms and their contributions to high‐turbidity events in a large lake

High‐resolution field data, collected during April to October of 2008–2009, were analyzed to investigate the quantitative contribution of sediment resuspension to high‐turbidity events in central Lake Erie. Resuspension events were distinguished within high‐turbidity events according to turbidity, fluorescence and acoustic backscatter timeseries, as well as satellite images. We observed 16 high‐turbidity events, causing a total duration of ∼20 d (out of 344 d) with elevated nearbed turbidity (> 10 NTU). Of these events, 64% were correlated with algal biomass, with the remaining 18%, 5%, and 4% being attributed to sediment resuspension by surface waves, storm‐generated currents and enhanced nearbed turbulence induced by high‐frequency internal waves, respectively. This is the first time that resuspension by enhanced nearbed turbulence from high‐frequency linear internal wave degeneration has been observed in a large lake. Resuspension was parameterized as a function of the instantaneous critical bottom velocity, bottom shear stress and the Shields parameter. From the in situ measurements, we suggest an extended Shields diagram for silty bed material that can be used to predict resuspension in other aquatic systems with similar sediment composition (∼20% cohesive sediment).

Sediments in aquatic systems are resuspended into the watercolumn when the bottom shear stress becomes greater than the critical shear stress for the initiation of suspension (Van Rijn 1993). During resuspension, turbulent eddies overcome the settling velocity and lift sediments above the bottom (Bagnold 1966). Resuspended sediments influence aquatic biogeochemistry by increasing the turbidity and thereby changing the rate of photosynthesis, as well as the vertical distributions of biomass, nutrients and contaminants (Fr echette et al. 1989;Gloor et al. 1994;Lou et al. 2000). In lakes experiencing hypoxia, such as Lake Erie (e.g., Rao et al. 2008), resuspension of organic biomass can increase the sediment oxygen demand by enhancing the surface area of decaying organic matter (Ackerman et al. 2001;Lorke and MacIntyre 2009). Resuspension by physical processes may originate from storm-driven currents (Lick et al. 1994;Beletsky et al. 2003;Churchill et al. 2004;Hawley and Eadie 2007;Marti and Imberger 2008), surface waves (Lou et al. 2000;Hawley et al. 2004), and/or shoaling and breaking of packets of high-frequency internal waves (HFIWs; Hawley 2004). HFIW-induced resuspension results from progressive nonlinear internal wave (e.g., solitary wave) shoaling in lab experiments and the coastal ocean (e.g., Quaresma et al. 2007;Stastna and Lamb 2008;Boegman and Ivey 2009); although there is also the potential for resuspension from shoaling of both obliquely propagating and convectively unstable linear modes (e.g., Kelvin-Helmholtz billows; e.g., Bouffard et al. 2012).
In the Laurentian Great Lakes, storm-induced mean currents, from winds with speeds of 15-20 m s 21 , frequently resuspend bottom sediments (Lick et al. 1994;Beletsky et al. 2003;Churchill et al. 2004;Hawley and Eadie 2007). In shallow systems, surface waves' orbital velocities have the potential to impinge on the bed leading to resuspension (Hawley 2000;Hawley et al. 2004). The simultaneous action of storminduced currents and surface waves lead to an increase in the *Correspondence: reza.valipour@canada.ca Additional Supporting Information may be found in the online version of this article.
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.
bottom shear stress, which may exceed the critical shear stress and cause resuspension (Hawley and Lesht 1992;Hawley 2000;Lou et al. 2000;Churchill et al. 2004;Hawley et al. 2004). In the nearshore zone of the Laurentian Great Lakes, and shallow lakes and ponds, it is well-established that surface waves have a more pronounced contribution to resuspend sediments than mean currents (Luettich et al. 1990;Hawley 2004;Mian and Yanful 2004;Chung et al. 2009;Reardon et al. 2014). Shoaling of HFIWs has been observed in lakes (MacIntyre et al. 1999;Boegman et al. 2003;Lorke et al. 2006;Dorostkar et al. in press); although their influence on resuspension remains speculative (Hawley 2004). Linear HFIW modes can generate nearbed patches of turbulence with the ability to resuspend sediment either in the form of a turbulent patch propagating toward the bed (Boegman 2009), or critical breaking of obliquely propagating HFIWs that are reflected off the bottom (Imberger 1998;Ivey et al. 2000).
The objective of the present study is to examine highresolution field data from central Lake Erie, to identify and quantify the occurrence of resuspension by storm-driven currents, surface waves, and HFIWs. We describe the mechanisms driving resuspension with particular emphasis on the enhanced nearbed turbulence by HFIWs. We parameterize resuspension events in terms of the observed instantaneous critical velocity (1 m above the bed), bottom stress and Shields parameter. Using in situ observations, an extended Shields diagram for silty bed materials is proposed, which can be generalized to model resuspension throughout Lake Erie and other aquatic systems with similar sediment composition.

Study area
Lake Erie ( Fig. 1; 388 km long and 92 km wide) is the shallowest of the Laurentian Great Lakes and consists of distinct western, central and eastern basins, which have maximum depths of 211 m, 225 m, and 264 m, respectively (herein depth is positive upward). The inertial period in the lake is 18 h (428 latitude), giving an inertial frequency f 0.97 310 24 rad s 21 . During seasonal stratified periods, there is a basin-scale near-inertial (Poincar e) waves cell in the central basin where the present moorings are located (Valipour et al. 2015b).

Identification of resuspension events
Increases in the beam amplitude of the ADV backscattering signal (at 1 mab) is used as indicator of resuspension (hereafter, ADV-amp, unit Count). This is qualitatively and statistically compared, and cross-correlated with turbidity, velocity timeseries and HR-ADCP acoustic backscatter. Fluorescence timeseries and satellite images (MERIS, ESA) are used to distinguish inorganic sediment resuspension events from algal bloom-type events in the turbidity timeseries.
We also relate elevated HR-ADCP acoustic backscatter to resuspension events (Quaresma et al. 2007). We correct HR-ADCP backscatter (ADCP-amp) for attenuation following Lohrmann (2001). The ADV-amp point measurements did not require a correction for attenuation.

Processing of current velocities
We calculate the burst-average current speed (from the ADV and HR-ADCP) at 1 mab, u mean 5 (û EW 2 1û NS 2 ) 0.5 wherê u EW andû NS are the mean velocity in the east-west and north-south directions, respectively. Raw ADV and HR-ADCP data were de-spiked (RC Filter method, Goring and Nikora 2002) but no difference was found between de-spiked and raw signals, except for two bursts on doy 256.8 and 258.7 in 2008, and two bursts on doy 146.2 and 273.8 in 2009, which were removed from the analysis.
In each burst, we define the 1 mab maximum velocity as u max (from the 4800 samples in each ADV burst; Table 1), and the 1% of velocities that exceed the 99 th percentile as u 99% . We assume a normal distribution where u 99% 5 u mean 12.7 r (Goring and Nikora 2002; Thomson and Emery 2014), where r is the standard deviation of the velocity per sampling interval. The calculation of u 99% enables us to obtain the maximum statistical estimate with 99% confidence in each ADV burst interval and compare it to the actual maximum value (Fig. 5).

Bottom stress from currents and waves
The bottom stress can be decomposed into a meancurrent bottom stress and surface wave-induced bottom stress, s cw 5 s current 1 s wave . The mean current bottom stress is s current 5 q o u * 2 where u * is the shear velocity and q o 5 1000 kg m 23 is the characteristic water density. In previous studies, we computed u * and bottom drag coefficient (C D ) by least-square fitting the burst averaged HR-ADCP velocity profiles to the law-of-the-wall and found C D 5 4.5 3 10 23 (method was explained in Valipour 2012; Valipour 2015a); which has been also used in recent Lake Erie field investigations and numerical modellings (Valipour et al. 2015b;2016). We follow Churchill et al. (2004), and band-pass filter the ADV velocity signals around 3-15 s to obtain the surface-wave induced orbital velocity (the median observed significant period of the surface waves is T s 5 s).

Critical shear stress
Resuspension of non-cohesive bed material is parameterized according to the Shields parameter (⍜), which is the ratio of the destabilizing forces (s b ) to the submerged particle weight (Van Rijn 1993): where q s is the grain density, with diameter of d s (typically d 50 ) and q 5 q o is the fluid density. For a Shields parameter greater than its critical value ⍜ cr , initiation of suspension, is modelled to occur. ⍜ cr is a function of d 50 and can be obtained from well-known diagrams ( Compared to purely non-cohesive sediments, establishing a s cr threshold for the initiation of suspension of mixed cohesive/non-cohesive material is more challenging because of the additional cohesive forces (Van Rijn 1993;van Ledden et al. 2004;Dalyander et al. 2012). As an alternative for the Shields diagram, s cr for mixed sedimentary materials can be obtained from an empirical relationship with q b ; s cr 5 0.015 3 (q b 21000) 0.73 (Mitchener and Torfs 1996).
To obtain s cr 5 q o C D u 2 cr from our field data, we estimate the observed instantaneous critical velocity u cr from timeseries plots of u max and turbidity. This observed u cr is validated against timeseries of ADV-amp, and contour plots of ADCP-amp. Our thus determined s cr are then compared with those predicted according to the Shields diagram for noncohesive sediments (e.g., following Dusini et al. 2009;Mier and Garcia 2011), and according to the Mitchener and Torfs (1996) empirical relationship for the mixture of sediments (e.g., following O'Callaghan et al. 2010).

Watercolumn turbulence parameters
The turbulent intensity is quantified from the turbulent kinetic energy (TKE) at 1 mab: where T int is the duration of each ADV burst (Table 1) we high-pass filter velocity data at the observed significant wave period T s . The dissipation of turbulent kinetic energy e is obtained from the ADV burst data by fitting to the theoretical wavenumber spectrum through the inertial subrange (Lorke and W€ uest 2005;Bluteau et al. 2011). Here, we follow standard practice (e.g., Lorke and W€ uest 2005) and use the mean velocity in each burst to convert the Reynolds decomposed velocity timeseries to wavenumber space (i.e., Taylor's hypothesis). We use ADV burst data to compute Reynolds stresses (s Reynolds ) at 1 mab as the covariance of u 0 v and u 0 h during each sampling interval: We also evaluate the potential for shear instability from the gradient Richardson number: where N(z) 2 is the square of the Brunt-V€ ais€ al€ a frequency profile: Here, @q/@z is the vertical density gradient calculated from the resampled 15-min temperature data using the UNESCO equation of state (Fofonoff and Millard 1983). Because velocity and temperature field data were recorded at different depths, we linearly interpolate the temperature data onto the 1-m upward looking ADCP bins in the above analysis.

Seasonal stratification
At Sta. 341, the seasonal temperature stratification formed in early June and the thermocline remained at 29 m depth, until deepening in late August intersect the lakebed  Table 1, (c) turbidity and Chl-a recorded at 1.5 mab using XR-620, (d) mean flow velocity (ADV-u mean ) and averaged beam amplitude from ADV (ADV-amp) at 1 mab. The gray rectangles show mooring refurbishment periods. Labels in (c) starting with T8 indicate the major turbidity events in 2008.

Nearbed high-turbidity events
As the stratification evolved in the watercolumn, there were changes in nearbed turbidity . We characterized high-turbidity events to be when the nearbed turbidity at Sta. 341 is > 10 NTU; consequently, there were four high-turbidity events in 2008 (T8-1 through T8-4, Fig.  2c) and twelve high-turbidity events in 2009 (T9-1 through T9-12, Fig. 3c). We have assumed a value of 10 NTU as a threshold to identify a high-turbidity event which is 3 times more than the average ambient near-bed turbidity 3.3 NTU (e.g., doy 120-210 in 2008), and is a reasonable value for nearbed aquatic ecosystem purposes (McCabe and O'Brien 1983;Newcombe and Macdonald 1991;Newcombe 1994;Bilotta and Brazier 2008).

Correlation between turbidity, velocity, and acoustic backscatter
To quantitatively compare timeseries from the turbidity sensors and backscatter amplitudes (ADV-amp and ADCP-  Table 1, (c) turbidity recorded at 1.5 mab using XR-620, and turbidity and Chl-a recorded at 5 mab using XR-420, (d) mean flow velocity (ADV-u mean ) and averaged beam amplitude from ADV (ADV-amp) at 1.0 mab. The gray rectangles are refurbishment periods. Labels in (c) starting with T9 indicate the major turbidity events in 2009. amp), correlations between data from these three instruments were examined. Timeseries of ADV-amp and ADCPamp were highly correlated when data from both instruments were available (doy 214-238 in 2008 and doy 135-196 in 2009;Fig. 5a,b). Timeseries of ADV-amp and turbidity were also highly correlated in 2008 (Fig. 5c), and weakly correlated in 2009. This difference is likely due to presence of algal biomass at Sta. 341 suggesting that the turbidity sensor recorded both sediment resuspension and algae, while elevated ADV-amps were only due to sediment resuspension. The turbidity sensor records particulates of varying diameter, including sediment and algal biomass, whereas ADV (and ADCP) backscatter signals scatter off of sediment but not algae (Lohrmann 2001). The correlation between ADV-amp and u mean is weaker than with u max (Figs. 5e,f, 6); suggesting that instantaneous burst-type currents are a better indicator of resuspension. This is consistent with recent studies showing that instantaneous burst currents, as opposed to mean currents, are better predictors of resuspensions (Gloor et al. 1994;Mathis et al. 2014;Aghsaee and Boegman 2015). In the Great Lakes, surface waves induce orbital velocities with 3-15 s periods (Churchill et al. 2004;Hawley and Eadie 2007) that will be best captured with u max as the averaging filters out the wave-induced component of velocity. Thus, we conclude that u max is a better parameter to characterize resuspension. In many field deployments instantaneous velocity data are not available (e.g., 15 min to 1 h time-averaged ADCP data is more common than instantaneous ADV data). Therefore, it is useful to relate u mean to u max using quantilequantile analysis (Fig. 7a,c; also see Supporting Information 1). This shows that u mean and u max are normally correlated with a ratio of u max u 21 mean 2.5 for u mean < 0.1 m s 21 , above which the ratio increased to u max u 21 mean 10 (i.e., for u mean > 0.1 m s 21 , we expect u max 10 u mean ).

Critical velocity and shear stress
Individual resuspension events were examined to determine u cr from the u max observed to resuspend sediment. For example, in Fig. 6 there was stronger backscatter and elevated turbidity when u max > 0.25 m s 21 . The duration of suspension in the water column ranged from 0.03 d to 5 d at 1 mab (Table 2; defined as the time until turbidity/backscatter returned to background levels). There were also instances when the HR-ADCP or ADV recorded short duration velocity impulses without an increase in the backscatter signal (e.g., doy 151.7, 152.2 in Fig. 6a-c). These are suggested to occur when the turbulence was not energetic enough to resuspended material to 1 mab. For example, on doy 151.7, 152.2 ( Fig. 6a-c), the velocity impulses are consistent with the elevated HR-ADCP backscatter signals below 1 mab.
The probability distribution of u max against turbidity using normal quantile-quantile plots (  T9-01  T9-02  T9-03 T9-04 T9-05   T9-06  T9-07  T9-08 T9-09  T9-10  T9-11 T9-12 velocities will result in random scattering of sediments into the overlying watercolumn and a breakdown of the normal relationship. For velocities stronger than u max > u cr 5 0.25 m s 21 (Fig. 7b,d), the turbidity is > 10 NTU; which is also in agreement with the only reported in situ observations of 0.08-0.26 m s 21 (i.e., s cr 5 0.03-0.3 N m 22 ) for a nearby site (Hawley and Eadie 2007). We argue that the error range results from differences in the amount of local consolidation and bioturbation between resuspension events (Sanford 1992;Hawley and Eadie 2007). amplitudes (ADV-amp) at 1 mab, (c, d) turbidity recorded at 1.5 mab using XR-620 vs. ADV-amp at 1 mab, (e, f) ADV-amp vs. flow velocity (ADV-u) including mean (u mean ), u 99% and maximum (u max ) currents using ADV at 1 mab -legend for (e) and (f) is the same. On the vertical axis, values close to one indicate a higher correlation. On the horizontal axes the lag time between the two timeseries, negative denotes the delay between the first and second signals. The period of cross-correlation analysis is shown in each panel.
To calculate the critical bottom stress and Shields parameter, we followed Gloor et al. (1994) and Mathis et al. (2014) and used the instantaneous observed u cr 5 0.25 m s 21 to obtain s cr 5 q 0 C D u 2 cr 5 0.28 N m 22 and (from Eq. 1) ⍜ cr 5 2.48. In order to compare these observed values with theoretical ones, we followed Mitchener and Torfs (1996) and obtained s cr 5 0.41 N m 22 for q b 51093 kg m 23 . Note that d 50 5 10 lm, which is the key parameter to calculate ⍜ (Eq. 1), is below the minimum particle diameter of the existing Shields diagram (Van Rijn 1993; his Fig. 4.1.4).

Observations of algae blooms and high-turbidity events
Our fluorescence measurements showed the occurrence of spring and summer algae blooms, which may be causing high-turbidity events, not caused by sediment resuspension. For example, at Sta. 341 in 2009 (Figs. 3c, 8b) on doy 140-147 the elevated turbidity (T9-01) was not associated with a change in the acoustic backscatter signals, while fluorescence timeseries spiked (Fig. 3d). Similar trends (i.e., elevated turbidity but not acoustic backscatter) were observed on doy 234-240 (T9-07) in the absence of fluorescence data (Fig.  8b). During these periods, satellite images indicated patches of elevated Chl-a (Fig. 9b,c). Likewise for T9-06, T9-08 and T9-09 (Table 2). Of the high turbidity events, there was no evidence of algae on doy 226 in 2009 (T9-05; Fig. 9a) or in 2008 (Fig. 2). Table 2 and Fig. 8b summarize algal peaks associated with the high-turbidity events, showing the phytoplankton contribute to turbidity 12.57 d out of 20 d (63.7%).
Zooplankton have particle sizes comparable to sediment sizes that may appear as spikes in the acoustic backscatter and/or a turbidity signal (Lohrmann 2001;Walsh et al. 2012). Therefore, it is necessary to eliminate zooplankton from being spuriously interpreted as resuspension. During the resuspension events (Table 2), spikes in Chl-a (as required for zooplankton grazing; Blukacz et al. 2010) and/or diurnal migration patterns were not observed. High backscatter measurements during resuspension events are not due to zooplankton, because the turbulence generated velocities are stronger than the typical zooplankton swimming speed, preventing the development of coherent patches of zooplankton (Stich and Lampert 1981;Blukacz et al. 2010;Walsh et al. 2012). Thus we conclude the resuspension events listed in Table 2 are all due to bottom sediment resuspension (see also Churchill et al. 2004;Hawley and Eadie 2007).

Observations of resuspension by surface waves and storms
Surface waves were observed to intensify u wave and the u cw to values greater than u cr (Fig. 8a,c). Specifically, waves with H s 1.5 m and T s 5 s increased u wave > u cr 5 0.25 m s 21 and resuspended sediment (Fig. 8, see doy 240 in 2008 as an example).
Despite the strong correlation between wind forcing and storm-driven currents at this location in Lake Erie (Valipour et al. 2015a,b), we did not observe many direct correlations between wind speed and resuspension from storms (Figs. 2c,d,3c,d,8c). The sole exception is T9-04 with u mean > 0.1 m s 21 , which had a significant contribution of mean currents to the bottom stress (Table 2). Overall, surface waves and wind-induced currents cause resuspension during 3.49 d (17.7%) and 0.97 d (4.9%), respectively, out of 20 d ( Table 2).

Observations of HFIW breaking and resuspension
Shoaling of progressive nonlinear internal solitary waves has long been associated with sediment resuspension in the coastal ocean (e.g., Klymak and Moum 2003;Quaresma et al. 2007;Boegman and Ivey 2009;Aghsaee and Boegman 2015). These waves are commonly observed in lakes (e.g., Finger lakes Farmer 1978; Dorostkar et al. in press), but have not been recorded in the Great Lakes (indeed the present set of moorings was specifically designed to measure resuspension from internal solitary waves). Despite this lack of observation Hawley (2004) provides indirect evidence that near-inertial wave-induced and/or HFIW-induced resuspension may be occurring in large lakes. Similar speculations on near-inertial waves induced resuspension in large lakes were also reported by Austin (2013). Given the lack of observations of internal solitary waves, here we examine the ability of the linear HFIWs regularly observed on the thermocline of Lake Erie (Bouffard et al. 2012), to resuspend sediments. To our knowledge, this process has not heretofore been observed nor investigated in detail in a large lake. An example of these HFIWs can be seen at the trough of the Poincar e waves

Valipour et al. Sediment resuspension mechanisms and their contributions
between doy 226.5 and 229 (Fig. 10), when Chl-a was low (Figs. 8, 9a), and winds, surface waves (Fig. 10a) and mean near-bed currents (Fig. 10g) were very weak. The flow field associated with these waves shows that after the first wave packet (near doy 227; Fig. 10b,e), the metalimnion expanded and deepened towards the bed. At the same time, we observed periods when the hypolimnion re-stratified and isotherm detachment from the metalimnion region allowed HFIWs to propagate toward the bed. At 3.5 mab, Ri < 0.25 below the thermocline (Fig. 10c) suggesting the possibility of shear instability and associated billowing (Boegman 2009 Fig. 10f). High-pass filtered temperature timeseries data (2 nd order Butterworth < 1 h where the HFIWs are 5-45 min as in Bouffard et al. 2012; Fig. 10e) at 1.05 mab showed the elevated dissipation was preceded by packets of HFIWs. Similar temperature fluctuations were recorded at 1 mab, but at lower frequency ( Fig. 10d; Table  1), which confirmed again the presence of high-frequency temperature oscillations and revealed weak nearbed stratification. In the presence of HFIWs the temperature timeseries showed a peak at 17 min (Figs. 10d, 11). In the absence of HFIWs, the ADV frequently energized at the predominant period of 17 min (Figs. 10f, 11a-c,e-g). This 17 min are the predominant peak of HFIWs as previously was expected (Bouffard et al. 2012;Valipour et al. 2015b, Fig. 5 therein).
During and after the passage of the HFIWs packets, the ADV timeseries recorded u max > u cr (doy 227.3-227.4 and 228.2-228.5 in 2009, Fig. 10g), along with an increase in TKE (Fig. 10f) and resuspension (elevated turbidity and ADVamp on doy 227.6 and 228.5 in 2009, Fig. 10h); with the HFIWs being strongest during the weakly stratified periods (i.e., waveguide present) and instantaneous velocities/dissipation strongest between wave events, when the nonstratified watercolumn was more amenable to overturns and turbulence production. Resuspension occurred during these turbulence events, after the passage of the wave packets, for a total of 0.87 d out of the 20 d with high near-bed turbidity (4.4% of the total; Table 2).

Other processes observed to cause nearbed turbidity
Of the 16 high-turbidity events (Figs. 2c, 3c, 8), the processes leading to high-turbidity during T8-03 and T9-03 (Table 2) remained unknown because no significant wind, waves, elevated bottom velocities or Chl-a were observed. Because our hydrodynamic data were limited to one field station and local spatial and temporal variability of near- * The averaged ratio of u 2 mean (at 1 mab) over the u 2 cw during the turbidity event is used to calculate the contribution of the wind-induced currents; the remainder is associated with surface waves. † The percentages in parenthesis denote the percent of the total days. turbidity has been observed (Fig. 4), we cannot rule out the possibility that the turbidity was advected to Sta. 341. However, it is unlikely that this is movement of resuspended sediment, which we showed to primarily result from spatially and temporally coherent surface wave-induced resuspension patterns ("Nearbed high-turbidity events" section). Satellite images suggest the variability in turbidity results from advection of algae from the south or south-east (Fig. 9c), potentially from the nutrient rich Maumee or Sandusky River plumes. Lateral advection of suspended sediment has been observed in the Great Lakes (e.g., in Lake Erie by Hawley and Eadie 2007;or in Lake Michigan by Cardenas et al. 2005). These two unknown events collectively contributed 9.3% of the 20 d of high-turbidity events (Table 2).

Discussion
We have presented high-resolution field observations of surface waves, storm-induced mean currents and enhanced nearbed turbulence induced by HFIWs leading to resuspension. We have also quantified the contribution of these processes in generating near-bed high turbidity events. To assess the generality of these observations, we compare our results to published observations from Lake Erie and elsewhere and develop mechanistic resuspension models.

Resuspension criteria
By observation (Figs. 2,3,8), we found an instantaneous critical velocity u cr 5 0.25 m s 21 , shear stress s cr 5 0.28 N m 22 and Shields parameter ⍜ cr 5 2.48 (in 8-258C water) for sediment resuspension in central Lake Erie. This is comparable to ⍜ cr 5 2.3 proposed by Van Rijn (1993) for similar sediments in 158C water. Our observed s cr 5 0.28 N m 22 is also within the observed range of 0.03-0.3 N m 22 in central Lake Erie (Hawley and Eadie 2007), and greater than 0.1 N m 22 , 0.2 N m 22 , and 0.18 N m 22 by Lick et al. (1994), Fukuda and Lick (1980) and Dusini et al. (2009), respectively, for Lake Erie. It is also greater than s cr 5 0.25 N m 22 in Lake St. Clair (Tsai and Lick 1986), and s cr 5 0.13 N m 22 in Lake Michigan (Lou et al. 2000). Our observed s cr is based on in situ observations of resuspension at 1 mab, and includes silty bed material with significant cohesive forces. The main composition of the bed material is silt ( 75%) which is noncohesive. However, van Ledden et al. (2004) and Dalyander et al. (2012) reported that for a percentage of clay > 7.5%, cohesive forces can influence resuspension, which is relevant for our study site with 20% clay (see "Sediment size,  T9-03  T9-04  T9-07  T9-08  T9-09  T9-10  T9-11  T9-12  T9-06  T9- density and type" section). Cohesiveness increases the required shear stress for resuspension with the same d 50 . The experimental s cr 5 0.41 N m 22 from Mitchener and Torfs (1996) is greater than our observed s cr 5 0.28 N m 22 , which is in agreement with the statement by Mitchener and Torfs (1996) that their laboratory experiments overestimate the critical shear stress for field purposes.
Our observations show instantaneous velocities, and in particular u max , are better predictors (Figs. 6,8,10) for resuspension events. Our nearbed mean currents of 0.15-0.2 m s 21 give much lower s cr 5 0.11-0.20 N m 22 and ⍜ cr 5 0.99-1.77; which are comparable to s cr 5 0.1 N m 22 obtained following Parker (2004). Our observations are also consistent with observations in Lake Alpnach (Gloor et al. 1994), which show instantaneous currents of 0.07 m s 21 resuspended bed material where the mean currents were 0.02 m s 21 ; coastal ocean data (Mathis et al. 2014), which demonstrated the role of instantaneous shear stress in resuspension; and laboratory experiments (Boegman and Ivey 2009; Aghsaee and Boegman 2015), which parameterize resuspension beneath internal waves as a function of the maximum vertical velocity.
Further comparisons between our observed u cr , s cr and ⍜ cr in Lake Erie with other locations are listed in Table 3, including rivers (e.g., St. Clair River), estuaries (e.g., Westerschelde Estuary), Coastal Oceans (e.g., Portuguese mid-shelf), and laboratory experiments.

Mechanisms for storm and surface wave induced resuspension
Our observations (Fig. 8) show that surface waves in Lake Erie with significant period T s 5 s and significant height H s 1.5 m are able to resuspend bed material. These waves are transitional/shallow water waves in a depth of h 5 17.5 m (h > 0.5L is the deep water limit, USACE 2002) and using the dispersion relationship (e.g., Dean and Dalrymple 1984) with wavelength, L5 35 m (h/0.5) we can estimate the theoretical T s 5 4.8 s < 5 s, above which surface . Timeseries at Sta. 341, in 2009, associated with T9-05 (Figs. 3, 8, 9). (a) wind speed and surface wave heights, (b) isotherm contour plot (numbers denote temperature in 8C) using 10-s temperature data from sensors moored at depths indicated by triangles, (c) Richardson number in logarithmic scale at 3.5 mab, (d) near bed temperature timeseries from the TR-1060 at 1.05 mab logged at 10 s, XR-620 at 1 mab logged at 3 min, (e) high-pass filtered < 1 h temperature timeseries from the TR-1060 at 1.05 mab, (f) dissipation and turbulent kinetic energy, TKE from the ADV (using Eq. 2) at 1 mab, (g) mean (u mean ) and maximum (u max ) currents per sampling interval from the ADV, (h) turbidity (XR-620) at 1.5 mab and amplitude (ADV-amp) from the ADV at 1 mab. wave induced velocities theoretically reach the bed (see Supporting Information 2). The orbital velocities for waves T s 5 s, therefore, always can reach the bottom, increasing bottom stress and TKE, and may resuspend bottom material. In central Lake Erie, waves with T s 5 s usually have H s 1.5 m (NDBC data) and we conclude that the significant wave period for lake Erie is a critical parameter for resuspension. Our results are consistent with a numerical investigation of surface waves in Lake Erie (Dusini et al. 2009), which found waves with periods of T s 5.9 s cause sediment mobilization to a depth of 217.5 m. Our results are also in agreement with Hawley and Eadie (2007), Lick et al. (1994), Churchill et al. (2004), andBeletsky et al. (2003) who find wind speeds of 10 m s 21 , 10-20 m s 21 and 20 m s 21 have the potential to resuspend sediment in Lake Erie at 224 m depth, Lake Superior at 290 m depth, and Lake Michigan at 260 m depth, respectively. These suggest that resuspension can also be expected throughout Lake Erie (see Supporting Information 2).

Mechanisms for HFIWs breaking
HFIWs, and their associated resuspension during shoaling, are expected but have not been previously observed in a large lake (Hawley 2004). The presently observed HFIWs form during linear instability (i.e., convective-shear instability implied by a subcritical Richardson number) of the thermocline during the passage of Poincar e waves (Figs. 10,11). The resuspension, by HFIWs, results either directly from turbulence created during the gravitational degeneration of the Kelvin-Helmholtz billows near the thermocline, or by breaking of obliquely propagating high-frequency internal waves on the lakebed. The latter mechanism is similar to the speculated resuspension by direct shoaling of shorter periods waves in Lake Michigan (Hawley 2004). These two mechanisms are investigated below.
The first mechanism is that the instabilities grow and eventually billow; causing overturns and nearbed turbulence that resuspend sediments. This occurs at the billowing timescale T b 20 DU g 021 (Turner 1973) where DU is the velocity difference over the layer of interest (above and below the 128C isotherm, which is between 211 m and 215 m depth; Fig. 10), g' 5 (q w 2 q) g q w 21 is the reduced gravity,  (the time from the formation of shear-induced high frequency waves until the billow starts growing; Fig. 10e). For the first resuspension event there is a 7 h lag between the elevated u max and the induced resuspension (doy 227.6 Fig.  10g,h); but there is no such lag during the second resuspension event (doy 228.5) resulting from elevated u max (doy 228.2-228.5). A lag could result from the need for eddies to mix the weak stratification near the bed prior to the first, but not the second resuspension event. The first resuspension event is consistent with the time scale of decay of turbulence associated with a turbulent event over the period Dt 5 226.7-227.2 doy (12 h), <TKE> 5 3.2 3 10 24 m 2 s 22 and <e> 52.7 3 10 210 m 2 s 23 where < > denotes averaging over Dt (Fig. 10f). The ratio of TKE to dissipation <TKE> <1.2e Dt > 21 23 (1.2 is for the effect of the buoyancy flux in the TKE budget) suggests that 1/23 4% of the turbulent energy has been dissipated, and the remnant energy has the potential to reach the bed (after overcoming the weak stratification) and resuspend bed material. The second resuspension event on doy 228 reached the bed with zero lag between the gravitational degeneration of overturns, created in the hypolimnion (Fig.  10b,e), and caused resuspension on doy 228.5 because there was no stratification preventing the downward propagation of eddies (Dt 5 227.3-228.6 doy, <TKE> 5 2.0 3 10 23 m 2 s 22 and <e> 52.1 3 10 210 m 2 s 23 and the ratio <TKE> <1.2e Dt > 21 74; the rest of TKE is dissipated below 1 mab). The alternative mechanism driving HFIWs induced resuspension is that the shear instabilities generate linear waves that propagate through the weakly stratified hypolimnion and break on the bed (Ivey et al. 2000); thus resuspending sediments (Fig. 10f). The waves will radiate toward the bed at a fixed angle b to the horizontal with a period (x 5 2p T HFIW

21
) according to N where x 5 N sin b (Ivey et al. 2000;Boegman 2009), which in our case T HFIW 5 17 min . In the hypolimnion (below thermocline, Fig. 10b) we obtained N 2 5 0.5-1 3 10 23 s 22 (Valipour 2012;Valipour et al. 2015a). On doy 226.9 and 228, we estimate b 5 1182168 for T HFIW 5 17 min and N 2 5 0.5-1 3 10 23 s 2 . Alternatively, we can estimate b by multiplying the time vector by a background 0.3 m s 21 near-inertial wave speed (Valipour et al. 2015b) to transform to a spatial coordinate under the frozen turbulence hypothesis. We then measured an observed angle b 158 for the 128C isotherm (Fig. 10b). The radiating linear waves will shoal on the lakebed and potentially resuspend sediment (Fig. 10b,h). The bed slope at Sta. 341 a 0.1% (Fig. 1) and so we do not expect critical breaking (a 5 b) with associated near-bed turbulence (Ivey and Nokes 1989;MacIntyre et al. 1999;Hawley 2004). However, for our case sin b ð Þ= sin a ð Þ1 and so near-bed mixing from linear wave reflection can be more intense, with larger overturn scales, than the critical case (Ivey et al. 2000). Both HFIW resuspension mechanisms are theoretically possible, and may be simultaneously occurring, strongly suggesting that HFIW are driving occasional resuspension events in central Lake Erie. Similar results may be expected in other aquatic systems.

Extended Shields diagram
Our particle diameter (d 50 5 10 lm) is not within the useable range of the existing Shields diagram. To extend the curve, we plot our observed ⍜ cr 5 2.48 against particle number, D * 5 (d 50 (S21)gm 22 ) 1/3 where S 5 q s q 21 , along with published data ( Fig. 12a; Table 3), on the Shields diagram for non-cohesive resuspension (Van Rijn 1993). With the exception of the Gloor et al. (1994) observations in Lake Alpnach, our results are consistent with published data for d 50 10-200 3 10 26 m showing an extended trend (Fig. 12a), outside the already well defined range for D * > 1(Van Rijn 1993). This suggested ⍜ cr 5 0.2 is the threshold for bed mobilization (not resuspension) in the laminar flow regime Re * < 1 (where Re * 5 u * d 50 m 21 is the Reynolds number) such as Lake Erie with Re * 5 0.2 (Van Rijn 1993). The data propose an inverse relationship between ⍜ cr and d 50 for D * < 1 (Fig. 12a). Mier and Garcia (2011) presented a similar diagram showing that for cohesive fine sediments, ⍜ cr significantly increases as d 50 decreases. The data include unconsolidated fine sediments (the samples were not consolidated as in the field) and noncohesive Pliolite particles with d 50 100 lm (Boegman and Ivey 2009). The lack of consolidation and presence of cohesive material between the non-cohesive material may explain the proposed relationship in Fig. 12a (e.g., Ara ujo et al. 2008;Mier and Garcia 2011), because s cr required to lift the sediments has to be sufficiently strong (increasing ⍜ cr ) to overcome the cohesive forces.
The Shields diagram shows s cr is inversely proportional to water temperature for d 50 < 600 3 10 26 m (Fig. 12b). Again, an extended trend is proposed between the data by Van Rijn (1993) and our observed and collected published data, except for Gloor et al. (1994) and Mier and Garcia (2011). The inconsistency with the Gloor et al. (1994) and Mier and Garcia (2011) d 50 50 3 10 26 m data was not unexpected, as it is also inconsistent with Fig. 12a, suggesting different resuspension dynamics (e.g., a laminarizing pressure gradient; Aghsaee and Boegman 2015) or significantly different sediments. Our extended Shields diagram captures the Mier and Garcia (2011) s cr for d 50 3 3 10 26 m, which is well aligned with the other observational trends (Lou et al. 2000;Boegman and Ivey 2009;Aghsaee and Boegman 2015) in the Shields diagram (Fig. 12a). We did not have sufficient data to further investigate the Mier and Garcia (2011) observations and so we cannot conclude if the s cr is valid for d 50 < 10 3 10 26 m (Fig. 12b).

Conclusions
We presented and analyzed a unique set of highresolution physical and biogeochemical data in Lake Erie during two ice-free seasons to understand the physical process driving nearbed sediment resuspension. Over the 334 d of measurement, spanning both years, 20 cumulative days were observed with significant nearbed turbidity (> 10 NTU). Algae, surface waves, storm-induced currents and turbulence generated by HFIW reflection accounted for 63.7%, 17.7%, 4.9%, and 4.4% of this time, respectively. Therefore, the majority of near-bed turbidity, in central Lake Erie, is suspended algae. We present the first evidence of resuspension by HFIWs in a large lake. Further laboratory or nonhydrostatic numerical investigations on HFIW-sediment interaction are recommended to clarify the details of the resuspension dynamics. Our results also suggest an extended Shields diagram for silty bed materials, which can be used to predict the resuspension processes in other aquatic systems.  Aghsaee and Boegman, (2015); d =107 μm 50 * Hawley and Eadie (2007); d =10 μm 50 Fig. 12. (a) Critical Shields parameter (⍜ cr ) for initiation of suspension of silt (with 20% cohesive sediment) on an extended Shields diagram. Particle parameter D * 5 (d 50 (S21)gm 22 ) 1/3 where S 5 q s q 21 , against ⍜ cr , both dimensionless, values of ⍜ cr on lines for D * > 1 are from Van Rijn (1993) Shields diagram of non-cohesive material. (b) Critical shear stress (s cr ) for erosion of silt (with 20% cohesive sediment) on an extended critical shear stress diagram, values of s cr on lines for d 50 >100 lm are from the Van Rijn (1993) diagram of non-cohesive materials. In (a, b) the dashed magenta line is a guide from the end of the Van Rijn (1993) data to our results. See Table 3 for details.