Methods

2.1 Modification of the PASS sampler to operate in ephemeral waterways.

The PASS sampler was originally designed for use in perennial waterways (e.g., small permanently flowing streams and rivers) (Figure 1) (Doriean et al., 2019). However, we propose that reconfiguration of the sampler to operate in ephemerally flowing systems (e.g., gullies) will provide an affordable alternative or complimentary monitoring method to those currently used for the measurement of suspended sediment concentration and particle size in ephemerally flowing systems, such as gullies.
The following modifications have been made to allow the use of the PASS sampler in gully systems (Figure 1):

2.2 Laboratory evaluation of gully monitoring methods

Water quality conditions typical of a gully flow event, based on previous observations at relevant field sites, were simulated in a laboratory setting to evaluate the modifications made to the PASS sampler design and to compare its performance with the other established methods. A falling suspended sediment concentration trend (high (~10,500 mg L-1) to low (4,500 mg L-1)) over a 6-hour period (typical of flow events in the active gully used for the field evaluation of this study, determined by preliminary field study data) was simulated using sediment sourced from a gully at the field study site (median particle size of 29 µm). An agitation vessel, similar in design to a churn splitter (20 L cylindrical polypropylene container with four baffles (vertical strips of aluminium, 0.5 cm thick and 3 cm wide, placed perpendicular to the side wall, from the bottom to the top, of the vessel) (Ward et al., 1990)) was used to create a turbulent flow of water during the experiment (Figure 2).
A triplicate set of PASS sampler intakes were placed approximately 0.15 m above the bottom of the vessel. Sample inlets for discrete sample collection, identical to the outlet of a churn splitter (Ward et al., 1990), (6 mm ID polypropylene tube tapped through agitation vessel wall) and the automatic sampler (Sigma® 900) inlet were placed at the same level as the PASS inlets to collect discrete samples. The discrete automatic sampler was elevated (2 m) above its intake point to simulate the configuration that would typically be used in the field. A turbidity logger (Observator, NEP495 (measurement range 40-4000 nephelometric turbidity units (NTU)) was placed at the same level as the sampler inlets and programmed to record a turbidity measurement every ten-minutes.
The RS sampler did not fit inside the agitation vessel, thus, a substitute dataset using the discrete manual sample data (collected from the isokinetic outlet) was generated to simulate the RS sampler data and allow comparison with the other techniques. Laboratory test samples collected using the discrete collection method and an RS sampler were compared and found to be similar in suspended sediment concentration and particle size distribution (< 2% ± 1% for both), which provided confidence to rely on the discrete sample dataset to simulate RS sample data. Flow event data gathered during a preliminary study, from the gullies monitored at the field-test site show that there is little hysteresis between suspended sediment concentration and water level. Thus, the RS sample data was constructed based on time after initial flow, estimating the peak stage to occur relatively early during the simulated event (i.e., the peak water height of the simulated event occurred 75-minutes into the 6-hour event).
To simulate the flow event, dry gully soil was weighed, suspended in a small volume of rain water (collected from the laboratory roof) to aid dispersion, and then diluted in rain water to a predetermined suspended sediment concentration with a final volume of 15 L. The sediment was kept in suspension using an overhead stirrer (OS40-S paddle stirrer) operating at 500 rpm. The concentration was changed by exchanging the water and sediment solution in the agitation vessel at 30-minute intervals. Triplicate PASS samplers continuously sampled water from the agitation vessel during the simulated flow event and repeat discrete samples (three samples per method) were collected from the same vessel using flow-proportional discrete sampling, simulated RS sampling and discrete autosampling methods every 30 minutes (15-minutes after each change in concentration).

2.3 Field evaluation of gully monitoring methods

The gullies monitored in this study were located at Crocodile Station in North Queensland (15°40’08.4”S, 144°35’38.4”E), Australia, and drain directly into the Laura River, which is connected to the coastal waters of the northern Great Barrier Reef via the Normanby River (Olley et al., 2013) (SI-1). The gullies are identified as alluvial gullies because they are located in alluvial soils of the Laura River floodplain (Brooks et al., 2013). Two gullies were studied to evaluate the accuracy and limitations of the monitoring techniques for measuring suspended sediment dynamics of gullies at different stages of erosion: an actively eroding gully (gully-1) with high suspended sediment output consisting of fine sediment (<63 µm) and some suspended sand (63-2000 µm); and a gully remediated in 2016 (gully-2) with relatively low suspended sediment output dominated by fine sediment (<63 µm) (SI-2). The suspended sediment particle size data used to describe the suspended sediment characteristics of the test gullies was gathered in a pilot study conducted at the study site.
The evaluated monitoring techniques consisted of a Sigma® 900 autosampler, a modified PASS sampler, a RS sampler array (six stages), and an Observator® NEP495 turbidity logger). Instruments were deployed in close proximity to each other in a straight section of channel approximately 50 m and 110 m downstream from the head of gully 1 and 2 respectively. The autosampler was placed on the bank beside the channel (elevated approximately 2 m above its intake) with the intake positioned in the middle of the channel cross section, 0.2 m above the channel bed with the inlet facing downstream (SI-3). A float switch, placed at the intake, was used to initiate and halt sampling. A PASS sampler was also placed at the midpoint of the channel affixed to a steel fencing post, driven into the channel bed; the intake and float switch were placed approximately 0.2 m above the channel bed. RS samplers were placed in a line along the channel centre at various heights above the channel bed, ranging from 0.2 to 0.45 m at 0.05 m intervals. The turbidity logger was placed alongside the autosampler inlet. A level logger (In-situ® R100) was placed at the midpoint of the channel directly on top of the bed, fixed to the steel support for the rising stage samplers. A barometric pressure logger (In-situ® baroTROLL) was placed nearby above maximum flood elevation, to allow accurate calibration of the level logger. A rain gauge (Hydrological Services tipping bucket rain gauges - 0.2 mm/tip with Hobo data logger) was also placed within the catchment area of the gullies.
Once activated, the autosampler collected a sample of approximately 800 mL every ten minutes, whilst the PASS sampler continuously sampled until the ambient water level dropped below the float switch. The turbidity logger was programmed to record a measurement every 10-minutes whilst deployed. The RS samplers collected a sample when the water level covered the intake and caused a pressure difference in the sampler, resulting in rapid filling of the sampler (Braatz, 1961){Braatz, 1961 #362}. Manual flow-proportional samples were collected using a DH-48 sampler using the equal discharge method when flow velocity and depth were sufficient (>0.3 m sec-1 and >0.17 m, respectively), or taken directly from the stream with a sample bottle when flow velocity and depth was too low for accurate use of the DH-48 sampler (Edwards et al., 1999).

2.4 Sample analysis and statistics

Samples collected from the laboratory and field evaluations were analysed for suspended sediment concentration using ASTM standard method D 3977-97 and particle size distribution using laser diffraction spectroscopy (Malvern Mastersizer 3000, Malvern Instruments). Discrete samples from the autosampler were analysed as received, whilst the PASS samples (composites of main settling column and intake sediment trap), were placed in cold storage (4°C) for five days to settle, after which they were decanted to 1 L and analysed. The supernatant was filtered through a pre-weighed glass fibre filter (Whatman GF/F (0.7 µm)), to account for the mass of any sediment that may have remained in suspension. The time-weighted average (non-continuous) suspended sediment concentration was determined by averaging the concentration of multiple discrete samples, weighted by the time span between two sequential samples. The PASS sampler continuously samples whilst in operation, thus, the time-weighted average suspended sediment concentration is calculated by weighting the total mass of suspended sediment collected by time as a function of volume (Doriean et al., 2019). Turbidity measurements were calibrated using the discrete samples from the autosampler. A linear regression between turbidity and suspended sediment concentration was used to convert NTU measurements into suspended sediment concentration, when appropriate (Rasmussen et al., 2009). Statistical analysis was conducted using GraphPad-Prism®. The sample data did not share similar standard deviations, thus, the unpaired nonparametric Mann-Whitney t-test method was used to compare differences between methods (p = 0.05).

2.5 Data quality and uncertainty

The uncertainty of each measurement method must be considered when evaluating their relative performance. The uncertainty assigned to a particular technique was determined based on laboratory evaluations conducted during this study and the scientific literature. If the difference in suspended sediment concentration between two sampling methods was equal to or less than the cumulative error associated with those methods, the individual results were not considered significantly different (Horowitz 2017). For example, manual sampling uncertainty is typically ~10% of the sample suspended sediment concentration (Sauer et al., 1992), whereas, the PASS sampler was previously demonstrated to exhibit ~6 to 17% uncertainty (Doriean et al., 2019). Cumulatively this suggests a sample concentration difference range in the order of 16 to 27%. Thus, suspended sediment concentrations of samples collected by these methods that differed within this uncertainty range were not likely to be statistically different (Horowitz, 2017).