The bed-level dynamics at each site were monitored using recently developed SED sensors (Hu et al., 2015; see Fig. 2). These sensors are stand-alone instruments with all parts for measuring and data logging and include batteries enclosed in a transparent tube. The measuring part is an array of light-sensitive cells that measure light intensity. When in use, a sensor is inserted vertically into the bed, leaving about half of the measuring array above the bed. The cells above and below the bed receive different amount of the daylight, which will lead to different voltage outputs in the array of cells. By using an autonomous script, the noise in the raw signal is reduced, and the bed level is determined as where the large transition from high to low voltage occurred (Fig. 2d; see Willemsen et al., 2018). When bed accretion or erosion occur, the transition point moves up or down in the measuring array. Thus, by recording the changes of the transition point, we can measure the bed-level changes. In some cases, scouring holes occurred around some of the deployed SED sensors, with the maximum depth of 5 cm. They typically result in two transition points in the array, corresponding to the bottom and the top of the scouring holes. In such cases, the bed level was determined as the vertical position at the top of the scouring holes. Details of SED sensor data processing are included in Willemsen et al. (2018). We note that the SED technique does not include effects of deep subsoil subsidence on bed-level changes. For a typical deployment period of the SED sensors (10–15 months), subsidence in the study areas is mainly related to glacial isostatic adjustment after the last ice age, with values in the order of less than 1 or a few millimetres over the considered time periods (Vink et al., 2007), and therefore these values are mostly much less than those of vertical bed-level changes recorded by the SED sensors.

As the sensor is dependent on the presence of daylight, the measuring window is daytime during low tide. Data acquired while the sensors were submerged or during night were excluded from the analysis. For most of the time, SED sensors provided at least one measurement every day, i.e. daily temporal resolution. To avoid recording bed-level data when sensors were submerged, an effective measuring window was set as 2 h around low tide. The tidal fluctuation of water level was recorded by pressure sensors deployed close to SED sensors. In such a window, we used the averaged readings as a bed-level observation point.

The accuracy of the sensors has been compared to a precise manual method (i.e. sedimentation–erosion bar) (Hu et al., 2015). The manual measurements were conducted weekly from 13 June to 17 July 2014 at the second most seaward measuring station of site 1 (Zuidgors A). These observations serve as an independent quality control of our automatic SED sensor measurements. Good agreement (R2=0.89) has been obtained between these two methods (detailed in Hu et al., 2015). The estimated operational accuracy of the SED sensors is 5.0 mm with a 3.9 mm standard deviation. Additionally, good agreement between the SED sensors and sedimentation–erosion bar measurements has been obtained at site 8 (Galgeschoor) over an 18-month parallel measurement (Belliard et al., 2019).

Bed-level changes in the intertidal environment are closely related to the local hydrodynamic forcing. We measured hydrodynamic parameters of water level, wave height, and tidal current velocity simultaneously with the bed-level measurement at some of our observation sites (Table 1). To measure the water level and wave height, we deployed pressure sensors 0.05–0.10 m above the bed in the vicinity of the SED sensors at some of the sites (see Table 1). At sites 1, 3, 4, and 6, OSSI-010-003C pressure sensors (Ocean Sensor Systems, Inc.) were used to measure pressure at a frequency of 5 Hz over a period of 7 min, with a 15 min interval. The mean water level is determined by the mean pressure in an interval. The significant wave height (Hs) and peak wave period (Tp) were derived from the dynamic wave pressure signals. The attenuation of pressure signals with water depth was corrected using the standard calculation methods as described in Tucker and Pitt (2001). The attenuation correction was only applied over the frequency range 0.05–0.4 Hz, and the maximum correction factor was set as 5 to avoid over-amplification of high-frequency signals (i.e. noise). A detailed description and the source of the data-processing routines can be found at http://neumeier.perso.ch/matlab/waves.html (last access: 28 January 2021). At sites 2, 5, 9, and 10, pressure sensors (series PDCR 1830, Druck Ltd.) were used. Pressure was recorded at 4 Hz for 4096 readings (∼ 17 min) around high-tide slack water, as determined by an onboard algorithm on the data logger (Möller et al., 1999). This typically results in one set of wave parameters per tide. For site 8, both OSSI-010-003C and PDCR 1830 pressure sensors were used. The measuring frequency was 16 Hz for the PDCR sensors and 20 Hz for the OSSI-010-003C sensors. More details on the sensor deployments at site 8 are included in Belliard et al. (2019).

At sites without pressure sensor measurements, the water-level data were obtained by nearby tidal gauge stations operated by Rijkswaterstaat (the Directorate-General for Public Works and Water Management, the Netherlands) or the British Oceanographic Data Centre (BODC). These data were obtained from Terneuzen (for site 2 at Zuidgors B and site 5 at Paulina) and Eemshaven (for site 7 at Uithuizen) with 10 min interval. For site 9 (Tillingham) and site 10 (Donna Nook), water-level data were obtained at the stations at Sheerness and Immingham with a 15 min interval. Tidal current velocity was measured by acoustic Doppler current profilers (ADCPs, Nortek Aquadopp) with a 5 or 10 min interval at sites 1, 3, and 8. Additionally, near-bed 3D current velocities were measured at site 8 using two acoustic Doppler velocimeters (ADVs, Nortek Vector). All hydrodynamic data are included in the current dataset.

To determine the median bed sediment grain size (D50), surface sediment samples (upper 2–3 cm) were collected at most of the sites (see Table 1). D50 of these samples was measured by a Malvern laser particle sizer. The chlorophyll a level in the sediment is an indicator for diatom biomass. Diatoms act as bio-stabilizers on tidal flats by producing extracellular polymeric substances (EPSs) and as such can affect sediment bed-level dynamics (Underwood and Paterson 1993; Austen et al., 1999; Andersen et al., 2005). At sites 2, 5, 9, and 10, chlorophyll a samples were collected from the upper 1 cm of the sediment using a small cut-off syringe. The processing procedures that were used to determine chlorophyll a are described in Willemsen et al. (2018). Additionally, at site 8, organic matter content was determined for the upper 2 cm of surface sediment samples by loss on ignition.

At our study sites, daily bed-level observations were conducted for 9–20 months, which includes conditions with various hydrodynamic forcing. As an example, we show the daily bed-level change and the accompanying wave height at site 4 (Zimmerman) and site 6 (Hellegat) from February 2015 to May 2016 (Fig. 3). Waves in front (5 m) of the marsh cliffs at site 4 (Zimmerman) were generally smaller than at site 6 (Hellegat) (Fig. 3b vs. 3d). Additionally, at both sites, there was a strong reduction in wave height from the bare tidal flats into the marshes (Fig. 3a vs. 3b and 3c vs. 3d). We observed that the bed-level fluctuation was more apparent on the bare tidal flats than in the marshes. Over the whole observation period, the bed-level fluctuation on the bare tidal flat was in the order of 5 cm at both sites, whereas bed level in the marshes stayed stable (station 1 of site 6) or experienced mild accretion (station 1 of site 4).

Notably, a number of storm events with high incident waves were captured during our measurements. During the two storm events in November 2015, Hs (significant wave height) exceeded 0.6 m on the bare flat stations at both site 4 and site 6 (Fig. 3g and h), whereas the mean Hs at the most seaward pressure sensors over the whole observation period was 0.104 and 0.113 m at these two stations, respectively. During the two storm events, sudden erosion of 2–3 cm occurred on the two bare flat stations. However, bed-level changes at the two marsh stations remained small (0.025 and 0.05 m at sites 4 and 6, respectively). Across the 10 sites, the most severe short-term erosion was observed at site 1 (Zuidgors A) on 27 and 28 October 2013 during the St. Jude storm (Hu et al., 2015). In that event, severe bed erosion of 10.5 cm depth was captured by our SED sensor on one of the bare flat stations at site 1 (data not shown).

Our observations at most sites were longer than 12 months. Thus, seasonal bed-level changes were captured in our dataset. Examples of seasonal bed-level changes at site 2 (Zuidgors B), site 4 (Zimmerman), and site 6 (Hellegat) show complex spatiotemporal variations (Fig. 4). Our data show that all stations at these three sites have alternating erosion and accretion seasons. There is no consistent seasonal erosion–accretion pattern for all stations. Winter is a typical season of bed erosion for stations on the bare flat but not for the stations in marshes.

Spatially, bed-level variations were generally smaller at the landward stations in the marshes and increased towards the seaward stations at all three sites. We further observed that the most seaward station at site 4 (Zimmerman) experienced net erosion over an annual timescale, whereas stations at the other two sites were in equilibrium; i.e. the degree of erosion was comparable to accretion. Profile elevation data show that marsh cliffs were distinct at site 2 (Zuidgors B) and site 6 (Hellegat), with the cliff height being 0.88 and 0.35 m, respectively, whereas a cliff was absent at site 4 (Zimmerman) (Fig. 4). Notably, the magnitude of bed-level changes reduced from bare flat stations to the stations on the marsh plateaus at sites with marsh cliffs (sites 2 and 6), whereas there was no clear difference between the bare flat station and the neighbouring marsh station at the site without cliffs (site 4).

Out of six sites with surface sediment grain size measurements, two sites (sites 4 and 6) in the Westerschelde had the largest median sediment grain size (Fig. 5). At these two sites, D50 of the surface sediment was in the range of 66.7–131.8 µm, which was significantly coarser than the rest of the shown sites (p=0<0.05). Within each site, there was no apparent difference in D50 between the marsh and bare flat stations around the marsh edge (50 m seaward and landward to the marsh edge). However, there was a gentle trend of coarsening from the landward to the seaward stations on bare flats.

Chlorophyll a levels in surface sediment, a proxy for the diatom biomass and their bio-stabilization effect, were also obtained at some of our observation sites (Table 1). The chlorophyll a levels at site 2 (Zuidgors B) showed great temporal variability (Fig. 6). For all stations, the chlorophyll a levels were generally low in winter (January) but reached their maximum at the end of the spring (May). However, there was no clear spatial pattern in the chlorophyll a levels across different stations, as the marsh stations had similar levels compared to the bare flat stations.