1 Introduction

Long-term suspended sediment load parameters (e.g. mean annual yield) are useful metrics for prioritizing sediment management strategies among catchments (Sherriff et al. 2015, 2016); however, they provide limited information about the sources, in-stream storage and delivery of suspended sediment within the catchments. Therefore, river management must take the inherent variability and temporal dynamics of suspended sediment transport into account to design efficient sediment management measures and reduce fine sediment supply to downstream aquatic ecosystems (Fuller et al. 2014; Le Gall et al. 2017).

Analysing suspended sediment concentration (SSC) and flow (Q) hysteresis patterns during flood events improves our understanding of sediment transport pathways (Eder et al. 2010), the relative importance of eroding channel banks (Rinaldi et al. 2004; Lefrançois et al. 2007; Yu et al. 2017) and surface soils (Yeshaneh et al. 2014) as sediment sources and in-stream fine sediment deposition (Lawler 2005; Fang et al. 2008; Navratil et al. 2010). Such patterns are related to multiple factors, including variations in sediment generation from different sources within the catchment (Asselman 1999; Topping and Wright 2016), delay between runoff and geomorphic adjustment during floods (Topping et al. 2000), bank collapse during different stages of the hydrograph (Ashbridge 1995; Rinaldi et al. 2004; Lyu et al. 2020; Yu et al. 2017), proglacial sediment sources (Mao and Carrillo 2017; Tsyplenkov et al. 2020), distance of sediment sources from the monitoring sites (Seeger et al. 2004; Smith and Dragovich 2009), spatial and temporal distribution of runoff in small (Bača 2008) and large (Zhao et al. 2017) catchments, changes in particle size of suspended sediment (Topping et al. 2000; Kleinhans et al. 2007; Cheraghi et al. 2016), the effect of riparian vegetation on sediment retention and recycling during low and high flows (Pavanelli and Cavazza 2010), sediment supply changes during snowmelt periods (Kämäri et al. 2018) and due to climate variability through the year (Sun et al. 2016), and different runoff origins (e.g. overland flow versus subsurface flow—Terry et al. (2014)). Gellis (2013) and Vercruysse et al. (2017) have summarised further possible explanations for the occurrence of different hysteresis loops.

The influence of catchment characteristics, land use or land cover, and channel morphology on hysteresis patterns has been explored previously (Vaughan et al. 2017; Aguilera and Melack 2018; Misset et al. 2019). However, the majority of previous hysteresis studies have used datasets from a limited number of catchments, which has hindered understanding of the combined effects of catchment characteristics and hydrological drivers on suspended sediment transport and hysteresis patterns. This study aims to overcome this limitation by exploring Q-SSC relationships over storm runoff events from 17 catchments across New Zealand that vary in regard to land cover, erosion terrain, and catchment size—which determines the distance of sediment sources from the monitoring sites. In addition, attributes of flow and sediment-related variables were used to examine which of those control most of the variability in suspended sediment transport within storm events in each catchment.

Previous studies have also mainly focussed on the dynamics of suspended sediment transport in small (< 100 km2) (Gao et al. 2007; Deasy et al. 2009; Aich et al. 2014; Strauch et al. 2018; Vongvixay et al. 2018) and mid-size catchments (100–500 km2) (Gao and Josefson 2012; Yeshaneh et al. 2014), with less attention having been paid to larger catchments due to the difficulties of field measurements and long-term data collection (Neal et al. 2006; Oeurng et al. 2010; Zhao et al. 2017; Vale and Dymond 2020). In this study, we also evaluate the effect of catchment size on sediment concentration and flow hysteresis by including catchments with sizes varying from < 1 km2 to > 3000 km2 in our dataset.

Another aspect of suspended sediment dynamics that tends to have been mostly neglected in the literature is direct examination of the phase relationship between suspended sediment load and flow during runoff events. Arguably, this provides more direct information on the delivery and transfer of sediment and water in a stream than does the Q-SSC relationship (after all, SSC is simply sediment load divided, or diluted, by flow). Haddadchi and Hicks (2020) examined metrics of the within-event suspended sediment load using the same dataset as used in this paper. Their findings on the effect of catchment characteristics on suspended sediment load during different hydrograph stages should be borne in mind when interpreting the results of the current study of Q-SSC relations.

2 Methods

2.1 Study sites

Co-located flow and sediment concentration records from 17 catchments across New Zealand’s North and South Islands were used in this study (Fig. 1 and Table 1). Catchment sizes ranged widely, from very small (i.e. West Hoe with 0.5 km2 area, and Weiti with 1.65 km2 area) to very large (i.e. Oreti at Branxholme with 2121 km2 area and Mataura with 3595 km2 area).

Fig. 1
figure 1

Erosion terrain maps of the study catchments and their locations across New Zealand’s North and South Islands. Sampling and flow and turbidity monitoring sites are shown with blue circles. Blue lines on inset maps indicate catchment boundaries. Note different scale for inset catchment maps

Full size image
Table 1 Catchment characteristics for suspended sediment monitoring sites. Percentages of areal coverage of catchment by different land cover and erosion terrain are shown in parenthesis.
Full size table

Erosion terrain and land cover maps for the study catchments (Figs. 1 and 2, respectively) were first extracted from the Manaaki Whenua—Landcare Research Land Resources Information portal (LRIS 2018), and then aggregated into functional groups, similar to Hicks et al. (2019), to make catchment characteristics more comparable. Erosion terrains describe susceptibility to erosion as determined from landform, rock type, slope and observed erosion activity (Dymond et al. 2010). As listed in Table 2, measurement periods were similar for sites from the same region but varied between regions (e.g. measurements between 2002 and 2009 were used for the Motueka catchments, whereas for the Hoteo catchments the data spanned the period 2012 to 2016). Similar to Haddadchi and Hicks (2020), land cover data concurrent with the period of flow and sediment records were used to generate land cover maps.

Fig. 2
figure 2

Land cover maps of the study catchments. Location of the catchments across New Zealand is shown in Fig. 1. Note different scales for inset catchment maps

Full size image
Table 2 Summary of measured flow and suspended sediment data for monitoring sites
Full size table

The dominant land cover and erosion terrain for the upstream catchments together with the features most proximal to the monitoring sites are listed in Table 1. Pasture is the dominant land cover in eight catchments (Hoteo, Waiteitei, Makarewa, Mataura, Oreti at Branxholme, Oreti at Lumsden, Pauatahanui and Whareama), whereas eight other catchments (Motueka River at Gorge and Woodmans, Motupiko, Wangapeka, Waiwhiu, Weiti, West Hoe and Horokiri) were mostly covered by forest and scrub (lumped as forest), with Porirua catchment equally covered by pasture, forest, and urban areas. More-erodible erosion terrains, including sand country and areas formed in tephra and loess and tertiary mudstone, sandstone and soft limestone, were the major terrains in eight catchments (Motupiko, Hoteo, Waiteitei, Waiwhiu, West Hoe, Makarewa and Whareama catchments), while seven catchments (Motueka at Gorge, Wangapeka, Mataura, Oreti at Lumsden, Porirua, Horokiri and Pauatahanui) were dominated by less erodible (hard) terrains formed in greywacke, argillite and hard limestone, schist and coarse crystalline plutonics and metamorphics. Two catchments of the Weiti and Oreti at Branxholme had approximately similar coverage of more-erodible and hard erosion terrains.

Continuous flow records were derived from rated stage records, while the matching continuous SSC records were based on turbidity monitoring ‘calibrated’ with relationships between SSC and turbidity for each monitoring site. The derived flow records, SSC-turbidity relationships and the technique used to select flood events for each study catchment are detailed in Haddadchi and Hicks (2020). The number of flood events with continuous records of flow and turbidity data used for the hysteresis analysis ranged between 12 events in the Motupiko catchment to 202 events in the Motueka at Gorge catchment. Mean annual precipitation in the study catchments ranged from 805 mm in the Oreti at Lumsden catchment to 1790 mm in the Motueka at Gorge catchment (Table 2).

The ranges of peak flows and event sediment loads for the events selected for each study catchment are listed in Table 2. The average of peak flows for all selected events ranged between 0.3 m3 s−1 on the West Hoe River and 168.3 m3 s−1 on the Mataura River. Event sediment load was highly variable among the study catchments, ranging between 0.5 t on the West Hoe River and 1236 t on the Oreti River at Branxholme.

2.2 Q-SSC hysteresis analysis

Six different patterns of Q-SSC relationship hysteresis were observed during flood events in the study catchments (Fig. 3). These include clockwise, anticlockwise, figure-eight shaped with anticlockwise peak (CAC), figure-eight shaped with clockwise peak (ACA), complex with multiple flow peaks and complex with a single flow peak.

Fig. 3
figure 3

Different types of Q-SSC hysteresis patterns. Arrows indicate direction through time. Clockwise hysteresis example was for an event on October 2016 event in the Oreti River at Branxholme site (Qmax = 145 m3 s−1, SSCmax = 475 mg l−1). Anticlockwise hysteresis example was for an event in Wangapeka River on March 2008 (Qmax = 229 m3 s−1, SSCmax = 244 mg l−1). Eight-shaped clockwise-anticlockwise-clockwise (CAC) example has been observed at Weiti River on Sep 2014 event (Qmax = 0.5 m3 s−1, SSCmax = 303 mg l−1). Eight-shaped anticlockwise-clockwise-anticlockwise (ACA) example was occurred at Motueka at Woodmans on Oct 2007 event (Qmax = 1350 m3 s−1, SSCmax = 1163 mg l−1). Complex with multiple flood example was for the Motueka River at Woodmans on May 2006 (Qmax = 123 m3 s−1, SSCmax = 19 mg l−1). Complex with single flood example was for the Hoteo River on April 2004 (Qmax = 58 m3 s−1, SSCmax = 422 mg l−1)

Full size image

A simple clockwise hysteresis loop appears when the peak flow occurs after the peak SSC. A common explanation for the clockwise pattern is that the bulk of the sediment carried during the event is sediment from either proximal surface soil erosion sources or channel storage that is readily available during the rising stage of the hydrograph but is depleted on the hydrograph recession (Rovira and Batalla 2006; Gao et al. 2007). However, alternative explanations include dilution by increased baseflow during the event recession (Bača 2008), the arrival of clean water from forested headwater areas (Oeurng et al. 2010) or higher rainfall close to the catchment outlet (Jansson 2002; López-Tarazón et al. 2009; Oeurng et al. 2010).

Anticlockwise Q-SSC hysteresis appears when the peak flow occurs before the SSC peak. Delayed sediment delivery from distant tributaries, variable rainfall patterns in larger catchments (López-Tarazón et al. 2009; Mano et al. 2009) and bank collapse on hydrograph recessions (Rinaldi et al. 2004) are some possible explanations for anticlockwise hysteresis. It can also occur in large catchments when the flow wave travels faster than, and ‘outruns’, the sediment (Brasington and Richards 2000).

Figure-eight hysteresis with a clockwise loop for low flow and anticlockwise loop for high flow (CAC) occurs when entrainment of more local and readily available sediment sources prevails early in the event until exhaustion of these sediment sources occurs, but this is then followed by sediment arriving at the monitoring site from more distant sources (Seeger et al. 2004). Conversely, the figure-eight anticlockwise for low flow and clockwise for high flow (ACA) hysteresis can occur, for example, when proximal sources can be activated late on the flood recession (e.g. associated with bank collapse).

Complex hysteresis patterns (with more than two linked hysteresis loops) can appear during multi-peak floods (Lecce et al. 2006; Sadeghi et al. 2008a; Zhao et al. 2017) (complex-multiple flood) or single peak floods (Asselman 1999) (complex-single flood). Complex hysteresis with multiple flood peaks can occur due to the heterogenous spatial and temporal distribution of rainfall events which results in several flow peaks or a broad peak flow period with similar sediment responses to flood generation. In this situation, the overall Q-SSC relationship shows less variance between rising and falling stages since the mixing of sediment from multiple sources tends to dampen the amplitude of loops (see complex hysteresis with multiple peak example in Fig. 3 from Motueka at Woodmans catchment). Re-entrainment of fine sediment drapes on channel beds due to a sequence of flow waves may also result in complex hysteresis patterns. Complex hysteresis patterns with a single peak flow can appear when uniform rain intensity over a catchment activates several erodible sources with different transport distances. Transport of suspended sediment with a range of particle sizes and therefore different travel and residence times is another probable cause of complex hysteresis over events with either single or multiple flow peaks.

As with previous studies (Langlois et al. 2005; Lawler et al. 2006; Aich et al. 2014; Zuecco et al. 2016), we have used multiple indices to quantify hysteresis patterns. The hysteresis index (HI) is defined as the ratio of the flow-weighted sediment concentration, which equals the sediment mass load divided by the water volume, before the flow peak (SSCR) to that after the flow peak (SSCF):

$$ \mathrm{HI}=\mathrm{Log}\ \left({\mathrm{SSC}}_{\mathrm{R}}/{\mathrm{SSC}}_{\mathrm{F}}\right) $$
(1)

This essentially indicates whether sediment concentrations are higher or lower overall between rising and falling stages. It identifies the synchroneity of the flow and SSC hydrographs, and also indexes the vertical (i.e. SSC axis) amplitude of the hysteresis loop. Hysteresis indices close to zero indicate near-synchronous SSC and flow hydrographs, with consequently little difference in the Q-SSC relation between rising and falling stages, while higher absolute values of HI indicate more nonsynchronous behaviour between flow and SSC with a wider hysteresis loop.

The hysteresis index therefore provides insight into the changes of flow and SSC during the event, but to better understand how sediment concentration varies around the flow peak, especially when patterns are complex, we defined a dimensionless measure of the time difference between the peaks of flow (TQ-max) and SSC (TSSC-max), termed the peak phase difference. This was defined as:

$$ \mathrm{Peak}\ \mathrm{phase}\ \mathrm{difference}=\left({T}_{\mathrm{Q}-\max }-{T}_{\mathrm{SSC}-\max}\right)/{T}_{\mathrm{flood}} $$
(2)

where Tflood is the flood duration, which is the time between the onset of the flood and the return to baseflow. Peak phase differences close to zero indicate negligible phase differences between flow and sediment concentration peaks. Positive peak phase differences show clockwise hysteresis under the flood peaks, while negative peak phase differences indicate anticlockwise hysteresis under the flood peaks.

2.3 Deriving storm event flow and sediment explanatory variables

In addition to the catchment characteristics, several attributes of flow and sediment during storm events were used to determine hysteresis differences in sediment concentration between the rising and falling limb at a given flow (Table 3). The hydrological variables include flood duration and time between the onset of flow and the peak flow (rising duration), time after peak flow until end of event (falling duration), peak flow, total runoff volume during the event, runoff during the rising and falling limbs of the hydrograph, the ratio of peak flow between the current and previous event, the ratio of runoff between the current and previous event, and the time since the previous flood event. The sediment explanatory variables were maximum SSC during the flood event, flow-weighted sediment concentration for the entire event, flood event suspended sediment load and maximum instantaneous suspended sediment load. Calculations to extract the variables were carried out in MATLAB (version R2017B).

Table 3 The hydrological and sediment explanatory variables used in the analysis of suspended sediment dynamics.
Full size table

2.4 Principal component and classification analysis

Principal component and classification analysis (PCCA) was performed to determine which of the above explanatory variables caused significantly different hysteresis patterns within each catchment. By reducing the dimensions of the original variable space, the PCCA can classify the cases (here as hysteresis type) and highlight the relations among variables and cases. To do this, the variables and cases were plotted in the space generated by the principal component axes. The input variables used in the PCCA are listed in Table 3.

Principal component analysis (PCA) has been previously used to identify the effect of hydrological variables on sediment transport (Sherriff et al. 2016; Vongvixay et al. 2018). PCCA has a crucial difference from PCA in that the individual samples must be assigned to different categories before the analysis. PCCA produces weightings that enable identification of the variables causing the highest differences between the observed hysteresis patterns. The principal components with eigenvalues higher than one were retained and subjected to a varimax rotation to identify hydrological and sediment-related variables that have high loadings on each PC and therefore higher impact on deriving different hysteresis patterns. All statistical analyses were performed using STATISTICA-V.13.4 (2018).

3 Results

3.1 Variability of hysteresis patterns in study catchments

The types of hysteresis patterns observed varied across the 17 study catchments. Figure 4 shows the proportion of events categorised by different hysteresis patterns (top plot, Fig. 4a) and the contribution of each hysteresis type to the total loads during the monitoring period (bottom plot, Fig. 4b).

Fig. 4
figure 4

Percentage of flood events categorised by hysteresis pattern (top), and contribution of each hysteresis pattern to total event load (bottom) for each of the 17 study catchments. Catchments ordered by region

Full size image

Seven catchments (i.e. Motueka at Gorge, Wangapeka, Motueka at Woodmans, Hoteo, Waiteitei and Weiti) each showed all six patterns of hysteresis listed in Fig. 3. This inter-event variety in hysteresis patterns in these catchments indicates variability in potential sediment sources with different distance from the monitoring sites. In contrast, only clockwise and complex hysteresis patterns were observed at the two monitoring sites on the Oreti River, and only anticlockwise and complex patterns were observed in the Waiwhiu catchment.

From Fig. 4a, clockwise hysteresis was the most frequent pattern in catchments mostly covered by pasture—including Whareama (79% of 14 events), Pauatahanui (85% of 62 events), Horokiri (71% of 31 events), Oreti at Lumsden (92% of 65 events), Oreti at Branxholme (98% of 44 events), Mataura (51% of 94 events), Makarewa (56% of 54 events), and Waiteitei (75% of 89 events). In contrast, anticlockwise was the main hysteresis pattern for flood events in catchments with forest, scrub, and urban areas as the dominant land covers—including Porirua (78% of 187 events), Waiwhiu (78% of 65 events), Motueka River at Woodmans (35% of 150 events) and at Gorge (40% of 202 events), Wangapeka (64% of 231 events) and Motupiko (75% of 12 events).

Anticlockwise hysteresis was observed in a low proportion of events in the Whareama (20%), Pauatahanui (5%) and Horokiri (13%) catchments, but those events contributed significantly to the total load with 75%, 60% and 70%, respectively (Fig. 4b). This is probably because distant sediment sources remain active and their sediment continues to be transported long distances downstream during larger flood events, only reaching the monitoring sites after the flow peak and generating higher sediment concentrations during the flow recession.

3.2 Influence of flood size and catchment characteristics on hysteresis pattern

To understand how sediment sources respond to flood size and catchment characteristics (i.e. land cover, erosion terrain), and to investigate geomorphic processes during suspended sediment transport, flood event suspended sediment load (SSL) was compared against event peak flow for different hysteresis patterns at each catchment (Fig. 5).

Fig. 5
figure 5

Flood event suspended sediment load against peak flow for the 17 study catchments showing different hysteresis patterns

Full size image

Clockwise was the dominant hysteresis type across all sizes of flood event in the Oreti and Pautahanui catchments (Fig. 5a, d, j). We interpret that this reflects catchment characteristics. For example, the Oreti catchment is dominantly in pasture land cover, with both monitoring sites in reaches with adjacent floodplains and alluvial terraces and long upstream braided reaches; thus, we expect suspended sediment to be relatively rapidly mobilised from the riverbed and proximal sources in a first flush on rising stages, and we expect waning loads on recessions due to exhausted sources and deposition in upstream braided reaches.

In the Horokiri catchment (Fig. 5b), while the dominant hysteresis type was clockwise, anticlockwise hysteresis occurred in events with higher peak flows. This suggests that in smaller flood events, proximal sources from pasture land cover on floodplains made a high contribution to the suspended load, while in larger floods (Qmax > 25 m3 s−1) suspended sediment from the catchment headwaters with pasture land cover made a higher contribution to event loads. Similar results were found in the Whareama catchment (Fig. 5q), with clockwise hysteresis dominant during events with relatively low peak flows and anticlockwise in events with higher peak flows. In contrast, in the pasture-dominated Hoteo catchment (Fig. 5c), most smaller floods (flow peaks smaller than ~ 30 m3 s−1) showed anticlockwise hysteresis while larger floods had clockwise hysteresis, suggesting exhaustion of sediment sources only during large floods. This clockwise hysteresis only during large flood events was also observed in the Waiteitei River (Fig. 5 l) and the Mataura River (Fig. 5f), both of which are also pastoral catchments.

At the Motueka at Gorge catchment (Fig. 5g), at events with similar peak flow, the anticlockwise events had typically higher sediment loads than the clockwise events. This suggests sediment sources from further upstream (mainly grassland) prevailing during the higher-load events but more sediment from proximal sources (e.g. riverbed, adjacent native forest) during lower-load events. Conversely, in the Motueka at Woodmans (Fig. 5h) and the Wangapeka (Fig. 5n), both within the greater Motueka catchment, clockwise hysteresis events tended to have higher sediment loads (for a given flood size) compared with anticlockwise events. This suggests that in these two rivers, areas covered with pasture and erodible terrains proximal to the monitoring sites generated more sediment than did distal forested areas covered by less erodible terrains. In addition, in these three monitoring sites in the Motueka catchment, complex hysteresis with single flow peaks was dominant in small floods. This suggests uninterrupted sediment supplies from multiple catchment sources even during small flood events. Overall, these Motueka sites demonstrate how spatially variable rain and erosion can produce at-a-site variability in event hysteresis pattern.

In the Porirua (Fig. 5k) and Waiwhiu (Fig. 5m) catchments, sediment loads (at similar peak flows) were higher in events with complex hysteresis with multiple flow peaks than in events with anticlockwise hysteresis. This is because the multiple peak events have longer duration and higher runoff and are therefore capable of transporting a higher event sediment load (e.g. multiple waves of flow can re-entrain fine sediment deposited within the river bed after the initial peak, and therefore increase the overall event load).

Figure-eight-shaped CAC was the main hysteresis pattern observed during large events (Qmax > 0.6 m3 s−1) in the small forest-dominated Weiti catchment (Fig. 5o). This might be explained by initial sediment contribution from accumulated streambed deposits at the first stage of the hydrograph (generating the upper segment of the low-flow loop), followed by a delayed contribution of sediment from distant pasture-dominated areas during high flow (producing a high flow anticlockwise loop), then eventual exhaustion of sediment sources during the flow recession (producing the lower segment of the clockwise pattern).

In the West Hoe catchment (Fig. 5p), figure-eight-shaped ACA hysteresis was the dominant pattern across all sizes of flood event. No clear relationships between hysteresis patterns, peak flow and event load were observed in the Makarewa (Fig. 5e) and Motupiko (Fig. 5i) catchments.

3.3 Hysteresis and peak phase difference indices

Variability of sediment concentration during the rising (SSCR) and falling (SSCF) stages of the flow hydrographs were compared in all study catchments using the hysteresis index (Fig. 6). This index is aimed primarily at establishing whether a river gets cleaner or muddier on the event recession.

Fig. 6
figure 6

Boxplots of the hysteresis index (HI) for all monitored events in each study catchment

Full size image

Comparing the three sites in the Hoteo River catchment, all with similar erosion terrain but different land cover, indicates that in the forest-dominated Waiwhiu catchment the HI ranged between − 2.34 and 0.37 and averaged − 0.71, indicating generally muddier event recessions. In contrast, the two pastoral catchments of Hoteo (− 1.09 to 0.84, average of − 0.17) and Waiteitei (− 1.37 to 0.72, average of − 0.01) had smaller HI values with averages close to zero, indicating a more even balance of SSC between rising and falling stages. A similar land cover controlled pattern was found in the catchments draining into Porirua Harbour. The average HI was highest (algebraically) and near zero in the pasture-dominated Pauatahanui catchment (58% pasture, average hysteresis index of 0.03) while the forest/scrub-dominated Horokiri had a more negative HI (60% forest/scrub, average HI of − 0.14). The Porirua (33% urban, 32% pasture and 35% forest, average HI of − 0.53) had the most negative HI, likely by virtue of its significant areas of urban land cover.

The three forest-dominated Motueka River catchments were compared to assess the effect of catchment size on HI. All had negative average HI values, indicating muddier falling stages overall, with the average HI becoming increasingly more negative as the catchment area decreased (− 0.33 in the very large Motueka at Woodmans, − 0.66 in the mid-size Wangapeka, and − 0.98 in the small Motupiko). This indicates that at least in forest-dominated catchments, falling stages tend to become increasingly more muddy than rising stages as catchment size decreases.

Figure 7 lists the peak phase difference for all study catchments. Comparing two small catchments (Waiwhiu and Waiteitei) in the greater Hoteo River catchment, with similar erosion terrain but different land cover, indicates that in the pasture-dominated Waiteitei catchment, flow peaks occur after SSC peaks with a lag averaging 10% of flood duration, whereas in the forest-dominated Waiwhiu catchment, flow peaks led the SSC peaks by 11% of flood duration on average. In the larger Hoteo catchment, with mixed land cover (56% pasture, 44% forest), the phase difference is smaller with an average flow peak lag just 3% of flood duration.

Fig. 7
figure 7

The peak phase difference for different catchments. Black large circles indicate average peak phase difference for all events, and grey small circles show the peak phase difference for individual events

Full size image

A similar land cover influence was observed among the catchments draining into the Porirua Harbour, each with similar erosion terrain but different land covers. In the pasture-dominated Pauatahanui catchment, flow peaks arrived well after the SSC peaks, with an average phase lag equal to 19% of flood duration, whereas in the Porirua catchment (with a greater proportion under urban land cover) SSC peaks occurred after the flow peaks (average peak phase difference of − 5%). Again, similar patterns were observed among the large catchments from the South Island, with positive peak phase differences in the pastoral catchments (Oreti at Branxholme and Lumsden, Makarewa and Mataura) and a near-zero average phase difference in the more forested catchment (Motueka at Woodmans).

3.4 Effect of hydrological and sediment-related variables on hysteresis patterns

Twelve of the 17 study catchments (i.e. Motueka at Gorge and at Woodmans, Motupiko, Wangapeka, Hoteo, Waiteitei, Weiti, West Hoe, Makarewa, Mataura, Horokiri and Whareama catchments) were analysed by PCCA to investigate which hydrological and sediment-related parameters controlled the variation in hysteresis patterns within each catchment. These twelve all showed a variety of hysteresis patterns. The two Oreti sites, Waiwhiu, Porirua and Pauatahanui were excluded because their events were almost exclusively of one type.

The PCCA used the independent variables listed in Table 3 to separate the hysteresis patterns. Figure 8 shows the projections of these variables for the first and second principal components using circle biplots. The first three principal components (PCs) of the PCCA, with eigenvalues higher than one, accounted for between 81% (West Hoe) and 96% (Motupiko) of the range of variability shown by the input parameters for hysteresis patterns across the different catchments. These high percentages confirm the suitability of the PCCA test for selecting the variables responsible for variation in hysteresis patterns.

Fig. 8
figure 8

Projection of the variables on the PC-plan circle biplot using the PCCA test. The numbered variables are defined in Table 3

Full size image

In all catchments except the Makarewa, total runoff volume during the flood event, with loadings higher than 0.84 for PC1, was the main driver differentiating hysteresis pattern. Flood duration was the next best variable at defining the hysteresis groups, with loadings higher than 0.64 for PC2. Thus, flood size defined in terms of runoff volume rather than peak flow and flood duration appear to be significantly affecting the phasing difference between the flow and sediment peaks and the transport of sediment from sources to monitoring sites. In the Makarewa catchment, flow-weighted sediment concentration over the flood event and maximum sediment load, with PCCA loadings of higher than 0.9 for PC1, were the main drivers discriminating the hysteresis patterns.

Rising stage duration was another main driver to distinguish hysteresis patterns in catchments where flood events with anticlockwise hysteresis patterns around the peak flow contributed most to the total sediment load (i.e. Motueka at Gorge, Motupiko, Weiti and Whareama catchments). In these catchments, the PC2 loading ranged between 0.6 for the Weiti catchment and 0.94 for the Whareama catchment. Conversely, recession duration was an important factor discriminating hysteresis in catchments where flood events with clockwise patterns made the greatest contribution to the total sediment loads (i.e. Hoteo, Waiteitei, Makarewa, Mataura catchments). In these cases, the loadings for PC2 ranged between 0.67 for the Mataura catchment and 0.77 for the Makarewa catchment.

4 Discussion

4.1 Hysteresis patterns in different catchments

Anticlockwise hysteresis was the most common hysteresis pattern, being observed in 603 out of 1553 events (39%) over all 17 study catchments.

Three monitoring sites (Waiwhiu and two Oreti River sites) showed a similar hysteresis pattern between the flow and SSC in different flood sizes. In the Oreti River at both upstream (Lumsden) and further downstream (Branxholme) monitoring sites, sediment concentration invariably peaked before the flow peaked. Conversely, on the forested Waiwhiu catchment, the peak flow occurred prior to the peak SSC in all events except those with multiple flow peaks.

In the other 14 catchments, different directions for sediment rating curves have been observed with changes in flood size. Comparing flood peak against event load for different hysteresis patterns (Fig. 5) suggests that the spatial distribution of land cover and erosion terrain within the catchments and their distance from the monitoring site together with flood size were the main controls on phasing differences between sediment concentration and flow which results in different directions and shapes of the hysteresis patterns.

Previous studies have shown that in small catchments, sediment transport processes are commonly dominated by clockwise hysteresis patterns (Seeger et al. 2004; Langlois et al. 2005; Lefrançois et al. 2007; Sadeghi et al. 2008b; Smith and Dragovich 2009). This was not the case in this study, for example, in the 8.5 km2 Waiwhiu catchment 80% of events had anticlockwise hysteresis, while in the nearby West Hoe catchment (1.65 km2 area), figure-eight-shaped ACA hysteresis patterns were most frequent. This shows that relating hysteresis patterns only to catchment size insofar as that impacts sediment transport distance is not an accurate hypothesis. Other catchment characteristics such as land cover and erodibility of the sediment sources (Vaughan et al. 2017; Misset et al. 2019), within-event hydrological variables including rainfall variability (Lawler et al. 2006; Yang and Lee 2018), and catchment soil moisture (Seeger et al. 2004) can all play important roles in the flow and sediment phasing. Moreover, a limitation of this study and, as also noted by Misset et al. (2019), other relevant research on Q-SSC hysteresis patterns is the lack of consideration given to the influence of channel properties such as slope and morphology on hysteresis patterns. For example, steeper streams, with more rapid runoff, might subdue phase differences between water and sediment runoff, while multi-thread, braided morphologies might promote more fine sediment recycling from the channel bed compared with single thread streams, thereby developing figure-eight patterns. These other drivers could be tested in future work.

Our observed dominance of clockwise hysteresis during smaller peak-flow events and anticlockwise hysteresis during larger peak-flow events in the Horokiri and Whareama catchments is similar to the findings of Goodwin et al. (2003) and Oeurng et al. (2010), who observed anticlockwise hysteresis when flood magnitudes had sufficient capacity to transport sediment from distant erodible areas up-catchment to the outlet.

Similar to the results of Sherriff et al. (2016), the contributions of events classified by hysteresis pattern to the total event sediment load were not generally correlated with the frequencies of the hysteresis classified events (Fig. 4b). However, events with the sediment concentration peak occurring after the flow peak (i.e. anticlockwise, eight-shaped CAC) did generate more sediment overall at most sites. For instance, while flood events in the Motueka at Gorge showed all six hysteresis patterns, events with anticlockwise hysteresis (40% of the total of 202 events) delivered most of the sediment load (93% of 688000 tons total sediment load) transported during the monitored flood events. In the Weiti catchment, a figure-eight-shaped pattern with anticlockwise peak (CAC) was the least common hysteresis pattern (19% of a total of 105 events) but made the highest contribution (57%) to the total load (128 tons).

Comparing the hysteresis index among catchments (Fig. 6), we found that the average HI was positive in only two catchments (Oreti at Lumsden and Pauatahanui), both dominated by pasture land cover and less erodible terrains. Three other catchments (Oreti at Branxholme, Mataura, Waiteitei) with negative but near-zero average HI (average HI > − 0.05) were mostly covered by pasture. Conversely, the lowest average HI values (average HI < − 0.45) were found in four forested catchments (Motueka at Gorge, Motupiko, Waiwhiu, Wangapeka). This HI dependence on land cover could, therefore, be used to incorporate hysteresis into sediment routing models that generate SSC as a function of Q. Also, the HI could serve as a metric for assessing the performance of sediment routing models sophisticated enough to develop sediment transport hysteresis (Sander et al. 2011; Zhong 2013). This can be done by comparing observed HI with HI derived from modelled SSC and flow records.

To understand the distribution of the hysteresis patterns by size of flood (here defined by peak flow) and event sediment load, the proportion of the hysteresis patterns associated with a range of event-specific sediment loads (event load divided by catchment area) and normalised event peak flows were determined, combining the results from all 17 catchments (Fig. 9). The peak flows were normalised by catchment area elevated to the power of 0.81 following Hicks et al. (2000).

Fig. 9
figure 9

Frequency of hysteresis patterns in all catchments for different ranges of event-specific sediment load and normalised peak flows (normalised by catchment area0.81)

Full size image

The results indicate that the hysteresis patterns in which the flood peak occurs before the SSC peak (i.e. anticlockwise and figure-eight-shaped CAC) were the main patterns irrespective of event peak flow and load. The proportion of complex hysteresis patterns with a single flood peak was highest in events with lower sediment load (25% in events with specific sediment load less than 0.01 t km2), and this proportion reduced as event peak flow and load increased. Conversely, the proportion of complex patterns with multiple flood peaks was larger in events with higher event-specific sediment load, probably because of higher sediment generation and evacuation from the catchment due to multiple waves of flow. Previously, complex hysteresis patterns have been associated with higher sediment yield (Yeshaneh et al. 2014) and longer flood durations (Zhao et al. 2017), which implicitly indicates higher runoff.

4.2 Hysteresis and peak phase different indices

By comparing the hysteresis indices in different catchments, we observed that in catchments from the same region but with different land cover (e.g. catchments within the greater Hoteo River or catchments draining into the Pauatahanui Harbour), the rising to falling flow-weighted sediment concentrations were higher on average in the pasture-dominated catchments compared with the forested and urban catchments. This might suggest that larger volumes of suspended sediment were transported during the first stage of the hydrograph in pasture catchments, which is probably due to the higher erodibility of surface soils during the rising stage of the hydrographs and exhaustion of those sources after the flood peaks. However, the analysis of event load splits over rising and falling stages from the same dataset by Haddadchi and Hicks (2020) showed that there was a reasonably even balance between rising and falling stage loads in these pasture-dominated catchments—which attests to the importance of the longer duration of the recessions, even if their concentrations are relatively dilute.

The negative hysteresis indices observed in the forest-dominated catchments in the current study align with the Haddadchi and Hicks (2020) finding that forested catchments delivered most of their sediment loads on event recessions. A likely explanation for smaller negative hysteresis indices in larger forested catchments compared with medium and small forested catchments (as observed in the Motueka River catchments) is that in large catchments, the fine sediment from distant sources may deposit on the riverbed upstream from the monitoring site on flow recessions, to be re-entrained early in the next event. This means that the river bed can act as a buffering reservoir of fine sediment sourced from far upstream (Misset et al. 2019), and the effect of this should be more apparent in larger catchments. Such fine sediment deposition can be significant in gravel bed rivers (Petticrew et al. 2007), especially in low shear stress zones where suspended sediment settles out and accumulates as fine patches after overpassing gravel armour where shear stress remains locally high (Wiele et al. 1996; Wilcock et al. 1996; Lisle and Hilton 1999). Commonly observed low shear stress zones in gravel bed rivers include the lee of cobbles, boulders and logs, voids of the bed framework of coarser gravel, and pools, backwaters and embayments. Of course, where it is the dominant control, this streambed buffering process can induce clockwise hysteresis.

Comparing the flow and SSC peak times indicates that the concentration peaks tend to coincide with or lag the peak flows in pasture catchments. Longer peak phase differences in pasture-dominated catchments than in forested catchments from the same region with similar erosion terrain suggest that the sources of sediment were eroded faster in the pasture catchments due to entrainment of readily available proximal sediment sources during the rising stage of the hydrographs and/or re-entrainment of fine sediment embedded in the river bed. In addition to the erodibility of different land covers, the impact of the land cover on the timing of the flood peaks and therefore peak phase differences between flow and sediment concentration also needs to be considered.

4.3 Statistical test results

The PCCA was carried out to investigate the importance of hydrological and sediment-related variables on the hysteresis between SSC and flow. Flood event total runoff and flood duration were the two main hydrological factors that distinguish the hysteresis patterns. This aligns with previous studies that have found that the flood duration (Seeger et al. 2004) and runoff volume (Gellis 2013) influence the shape of hysteresis patterns. These two parameters implicitly reflect the areas of the catchments that contribute as sediment sources.

The PCCA results also indicate that the durations of the rising and falling hydrograph limbs are important factors causing different hysteresis patterns, with longer rising limbs associated with catchments where anticlockwise events make dominant contributions to the total load over time, and likewise for longer falling limbs and clockwise events.

4.4 Applications and future work

The purpose of this study was to understand the effect of hydrological and sediment-related variables and catchment characteristics on different hysteresis patterns in terms of their shape and direction. One application of the findings is around event-scale sediment routing models which predict SSC off flow via Q-SSC relations which could be parameterised by indices such as we have used herein (e.g. HI, peak phase difference, etc.). Alternatively, model predictive performance can be evaluated by comparing model-generated indices with those observed from field datasets such as presented herein. More broadly, the study findings also provide useful information for river management monitoring programmes aimed at mitigating suspended sediment generation and assisting development and implementation of policy to ensure land use sustains downstream receiving environments.

However, as noted in this paper and previous studies, a given Q-SSC hysteresis pattern can be explained by various erosion and physical processes; therefore, sole reliance on hysteresis information to infer the major processes acting in a catchment can be a speculative pursuit (Esteves et al. 2019; Misset et al. 2019). Further research to mitigate this ambiguity, and better understand the simultaneous effect of catchment and hydrological drivers on hysteresis patterns, could involve catchment characteristics being coupled with the hysteresis patterns as dependent variables in a statistical selection procedure similar to the PCCA approach employed herein but using larger datasets from further catchments. By having larger datasets, the results from the PCCA test can be compared with machine learning variable selection procedures such as support vector machines (Boser et al. 1992) and the random forest methodology (Breiman 2001) to evaluate the accuracy of selected variables.

Another way to overcome the interpretive ambiguity issue is to combine hysteresis analysis with additional sediment source characterization techniques. For example, high spatial and temporal resolution sediment tracing techniques can inform on the phasing of flow and sediment load during events and thereby help to verify the cause of hysteresis patterns (Haddadchi et al. 2019; Vale et al. 2020).

5 Conclusions

  1. 1.

    Hysteresis patterns in the relationship between flow and SSC across our 17 study catchments showed a consistent dependence on land cover, with clockwise hysteresis being the most frequent pattern in pasture catchments while anticlockwise patterns prevailed in forest-dominated catchments. Moreover, in pasture-dominated catchments, higher rising to falling sediment concentration ratios and longer time differences between the peaks of flow and SSC lead to relatively high volumes of suspended sediment being transported before the flood peaks, which means that the rising and falling stages transport near equal proportions of the total event load. In contrast, in forest-dominated catchments, more of the total event load is transported on falling stages.

  2. 2.

    No consistent relation between catchment size and hysteresis pattern was observed in this study, although within nested catchments, catchment size (manifest as distance from source) effects did appear to be a significant factor promoting anticlockwise and complex hysteresis patterns, particularly when considering spatial variability in land cover and erosion terrain within catchments.

  3. 3.

    Our principal component and classification analysis of hydrological and sediment-related variables showed that the main variables controlling the hysteresis patterns within our study catchments were flood event total runoff and flood duration.

  4. 4.

    While the findings of this study should be of immediate value in developing and testing event-scale catchment-based suspended sediment routing models, the complexity of geomorphic processes during suspended sediment transport combined with variability of sediment and runoff sources leaves residual ambiguity when linking at-a-site Q-SSC relationships with causes upstream. Thus, we recommend that further work could repeat the statistical analyses used herein with larger datasets spanning more catchments and by using other information streams such as high spatio-temporal resolution sediment tracing to reduce ambiguity in interpreting the hysteresis driving factors.