The study site is a 1.37 ha catchment (named RBF) located on a dairy farm at Poowong East, in the Lang Lang River catchment, Victoria, Australia, 130 km south-east of Melbourne (Fig. 2). The study area has a humid climate and rainfall is uniformly distributed across the year, with an annual mean (1961–1990) of 1100 mm (Bureau of Meteorology, 2009). Annual areal potential evapotranspiration (1961–1990) is 1040 mm (Bureau of Meteorology, 2005).

A general description of the study catchment can be found in Adams et al. (2014). The study period was between September 2009 and December 2011. Elevation ranges from 160 to 210 mAHD (Australian Height Datum) and the slope varies from 2 to 50 %. Based on field observations, the topography, range of slopes and the groundwater behaviour, the catchment was divided into four different zones: (1) the riparian area located on the relatively flat convergent lower part of the catchment (outlined in red in Fig. 2) included sites 1, 2, 32 and 3; (2) the lower slope (low slope) area; (3) the mid-slope area with sites 4, 5, 6 and 7; and (4) the upper slope (upslope) area with sites 10, 11 and 15 (Fig. 2).

The catchment geology comprises sandstones and mudstones of the Cretaceous Strezlecki Group (VRO, 2013). Outcrops on the lower stream banks of the catchment (just downstream of the monitored hillslope) show weathered sandstone and mudstone bedrock. Hand augering revealed a soil depth of between 1 and 1.6 m, and the lower parts of the profile included mottled clay and weathered bedrock particles. The soils are acidic and mesotrophic brown dermosols (Isbell, 2002) that grade from a fine sandy clay loam to a medium clay with mottles and weathered bedrock. Soil profile depth decreases, moving downslope. These soils typically have a moderate hydraulic conductivity surface horizon (0–40 cm, Ks≈ 5×10-6 m s-1, about 20 mm h-1). The dominant land use is grazing by dairy cows.

Stream discharge was measured at the catchment outlet (Fig. 2) using an RBC flume (Clemens et al., 1984) and an Odyssey (Dataflow Systems inc. Christchurch, NZ) pressure transducer (PT) recorded stream water levels every 10 min, which were used to compute instantaneous discharge rates. After August 2011, the PT was replaced with an ISCO (Teledyne ISCO, Lincoln, NE, USA) model 730 bubbler. The following rating curve was used to calculate Q from water level: Q=0.001H2+0.0168H, where Q is stream flow discharge and H is water head.

Rainfall data were recorded using a tipping-bucket raingauge at an automatic weather station (AWS) which was installed in 2010 on the upper boundary of RBF. Weather variables (temperature, humidity, wind, rainfall, global radiation) were measured by the AWS. Areal potential evapotranspiration (APET) was also computed using the Morton (1983) wet environment method on a daily basis. APET was strongly seasonal, resulting in strongly seasonal soil moisture contents and intermittent streamflow at RBF. Soil moisture storage was calculated from the volumetric soil water content which was measured for the 0–30 and 30–60 cm layers and recorded hourly by the AWS logger using two vertically installed 30 cm long Campbell Scientific (CS625) soil moisture probes (Campbell Scientific, 2006). The 0–30 cm probe was inserted from the surface. To install the 30–60 cm probe, a 30 cm deep hole was dug, with the excavated soil set aside. The probe was inserted vertically and then the excavated soil was repacked so that the soil was replaced at a similar depth to that from which it had been removed.

The CS625 produces a pulse signal and the pulse period was temperature corrected using measured soil temperature, and the manufacturers recommended temperature correction, as follows (Campbell Scientific, 2006). τcorrectedTsoil=τuncorrected+20-Tsoil⋅0.526-0.052⋅τuncorrected+0.00136⋅τuncorrected2, where τuncorrected is the probe output period, Tsoil is the soil temperature and τcorrected is the corrected probe output. The VWC was then computed from τcorrected using (Campbell Scientific, 2006) VWC=-0.0663-0.0063⋅period+0.0007⋅period2. Soil water storage over the top 60 cm soil depth was computed by adding the VWC from the 0–30 and 30–60 cm layers and then multiplying by the 300 mm soil depth. We use this soil water storage at the start of each rainfall event as an index of antecedent soil water (ASI) and assume it represents the catchment wetness condition.

To capture the nature of hydrologic connectivity, runoff mechanisms and flow pathways, shallow (1.5–1.6 m) groundwater wells were installed at 12 sites across the RBF catchment using 40 mm PVC pipes and backfilled with sand, bentonite, the topsoil and grass. Figure 2 shows these sites, of which 1, 2, 3, 16 and 32 were in the riparian zone; 4, 5, 6 and 7 were on the mid-slope; and 10, 11 and 15 were on the upper slope. Sites 4, 5 and 6 were equipped with water level loggers from July 2010 until the end of study period in December 2011. Sites 3, 7, 16 and 32 were logged from winter 2011 until the end of the study period in December 2011. Water level loggers were not installed at sites 1 and 2 since they were nearly always saturated. At sites 1 and 2 groundwater levels were measured manually. Water levels were logged using Odyssey PT loggers.

A rainfall sampler (Kennedy et al., 1979) collected up to 10 sequential rainfall samples per event, each being equivalent to 6.6 mm of rainfall. The sampler was initially installed close to the AWS; however, due to instances of damage by animals, it was relocated near to the flume in August 2010 until the end of the study period (December 2011).

An auto sampler (Teledyne ISCO 6712) was installed at the flume and streamflow sampling was triggered based on the rising stage. Following triggering, the sampler was programmed to collect up to 24 samples of streamflow at hourly intervals. Samples were removed from the auto sampler within 48 h. To reduce the laboratory analysis workload, we plotted recorded water levels from the RBF catchment in the field prior to removing the event sample from the auto sampler and selected certain samples for analysis. All samples during the rising limb and the peak were selected and samples were typically selected at an interval of 4 h during the falling limb. Routine grab sampling was undertaken at weekly intervals during the main runoff season when water was flowing through the RBF flume. This was supplemented by additional grab sampling during visits to collect event samples from the auto sampler. Routine grab sampling was also undertaken at weekly intervals during the main runoff season when water was flowing through the RBF flume.

Sub-samples for isotopic analysis were taken of stream water from both manual and auto sampler samples, and from all full rainfall sample bottles; these were collected in glass bottles for isotope analysis. Bottles were completely filled. The samples were refrigerated (+4 ∘C) until analysis for δ18O and δ2H. Stable isotope ratios were measured either at Monash University using a Finnigan MAT 252 and ThermoFinnigan Delta Advantage Plus mass spectrometers (2010 samples) or at the University of Melbourne, where a Picarro L2120i cavity ring-down isotope analyser was used to determine isotope ratios (2011 samples). The instrumental precision was δ18O = ±0.15 ‰ and δ2H = ±1 ‰ for the isotope samples analysed at Monash University, and it was δ18O = 0.1 ‰ and δ2H = 0.4 ‰ for samples analysed at Melbourne University. The results from the isotope analysers were checked for systematic differences, by analysing duplicate samples in both laboratories. We developed and applied a correction between the two laboratories based on these samples. With the exception of determining pre-event end member uncertainty (described later), the analyses presented here only use samples analysed at the Monash laboratory, and hence this difference between laboratories only affects our hydrograph separation uncertainty estimates. For isotope analysis, in total we collected 115 samples of rainfall, 28 samples during low flows and multiple stream water samples of isotopes from five events.

For the measurement of major ions, sub-samples of grab and event streamflow samples were collected using 50 mL plastic bottles. Sub-samples were also taken from each water sample for selected major ion (Na+, K+, Ca2+, Mg2+ and Cl-) analyses, which were analysed in the NATA-certified, analytical chemistry laboratory of the Water Studies Centre at Monash University using standard methods. Major ions were determined in non-filtered samples as follows (Cm is ion concentration).

Mg2+ was determined by atomic absorption spectrometry (APHA, 2005). The issue of interference with Si, Al, and P was solved using a combination of lanthanum and caesium. The determined uncertainty for Mg2+ was Cm ⋅ 0.0596, with a 95 % confidence interval (CI).

In order to analyse event behaviour, it was necessary to identify rainfall and runoff events. Based on an examination of the time series of hourly rainfall in the catchment (in the study period which was between April 2010 and December 2011), rain events were defined as having > = 5 mm total rainfall, and peak hourly rainfall intensity, Ipeak > = 1.5 mm h-1. Distinct events were separated by > 12 h without rainfall.

The runoff hydrograph was also divided into events. Runoff events began when the stream discharge hydrograph started to rise from its initial low flow value or moved above a threshold of 0.05 mm h-1 following the commencement of a rainfall event. Events ended either when (1) the discharge returned to its initial value, (2) a new rainfall event started, or (3) 96 h after the end of the rainfall event in unusually wet situations where elevated flow continued. For each event, a number of characteristics were determined as shown in Tables 1 and 2.

The Antecedent Soil moisture Index (ASI) was represented as the amount of the soil water storage in the top 60 cm of the profile at the AWS at the start of each rainfall event. The topography of the catchment was surveyed using a differential Global Positioning System (dGPS) and a 1 m horizontal resolution DEM (digital elevation model) was developed by interpolation methods. Instrument and well locations were also determined using dGPS. Then these data were used to produce maps of the study area, including sampling sites using Arc GIS. The lateral boundary of the saturated area was topographically constrained and field inspections suggested it was stable over time, while the upstream boundary moved up and down the riparian zone. The saturated area was estimated by locating the upper boundary through field inspection and then measuring the distance from either well 2 or 3 (the saturated area extended towards site 3 during very wet conditions). Figure 2 demonstrates the approximate maximum boundary location of the saturated area at site 3 and the main stream located close to the outlet of the catchment. The saturated area was then estimated from this information combined with the mapping of the riparian zone boundary in ArcGIS. These measurements were made between events. The proportion of saturated area was estimated using these data and then used to estimate saturation excess runoff generation for the different events. Consistent with our definitions in Fig. 1, here we conceptually separate return flow and flow resulting from direct precipitation on the saturated area and use saturation excess runoff to refer to the latter. The event runoff depth (mm) and event runoff coefficient (RC %) were calculated by separating the event hydrograph using the method of Hewlett and Hibbert (1967), which has been widely applied (Buttle et al., 2004; Fujimoto et al., 2008; McGuire and McDonnell, 2010). The method assumes that baseflow increases at the rate of 0.55 L s-1 km-2 h-1 (0.002 mm h-1) from the start of the rising limb. While this method, like other hydrograph based baseflow separation methods, is arbitrary, the results are insensitive to the assumed rate of rise.

Isotope samples are used for hydrograph separation into pre-event and event contributions. We applied the well-known one-tracer, two-component model of hydrograph separation approach (Pinder and Jones, 1969; Sklash and Farvolden, 1979). We determined uncertainties following Genereux (1998). We undertook the analysis using both the 2H and 18O data. We used the mean of the rainfall samples during the event to estimate the event water end member and low flow samples from the few days to the event for the pre-event end member. In the uncertainty analysis, the standard deviations of the event rainfall samples and of all the low flow samples across the study period were used. Half the analytic uncertainty was used to represent the standard deviation of the streamflow sample.

Figure 3 shows time series of weekly rainfall, APET, soil water storage and runoff. The rainfall, although variable from week to week, exhibited seasonality, while there was strong seasonality in PET. This drove a strong seasonality in soil water storage. An examination of the weekly runoff data shows that there was generally no flow from about October to May due to the seasonal nature of this catchment; however, an exception was that persistent low flow occurred from 26 November 2011 to the end of the event on 10 December 2011. During this period the ASI was often relatively low, but there was frequent and substantial rainfall (> 200 mm in 30 days). While a strong link between runoff and soil water storage is evident at the seasonal scale in Fig. 3, there are exceptions at the event scale. For example, in February 2011, there was runoff response despite the catchment being near to the lowest soil water storage for the study period.

Moving to the event timescale, Table 1 summarises 38 rainfall–runoff events and Table 2 shows a summary of 22 rainfall events that did not produce a runoff response. A further 16 rainfall events occurred over the study period which are not included in the analysis due to missing stream discharge data. For the 38 runoff events, total event rainfall varied from 7 to 72 mm, Ipeak ranged from 2 to 31 mm h-1, the ASI ranged from 130 to 286 mm and total event runoff varied between 0.23 and 41 mm. For the no-flow events (Table 2), total rainfall varied from 5 to 28 mm, Ipeak ranged from 2 to 10 mm h-1 and the ASI ranged from 146 to 238 mm. Figure 4 shows rainfall–runoff responses for selected events at RBF. These graphs are ordered from lowest (27 November 2010) to highest ASI (7 June 2011) for the selected events. All events (except events on 26 November 2011 and 10 December 2011; Fig. 4c) presented in Fig. 4 had zero or very low initial flow. For the events on 26 November 2011 and 10 December 2011, the initial discharge was 0.07 and 0.13 mm h-1, respectively.

In Fig. 4 most events showed rapid response to rainfall, except for the event on 8 December 2010 (Fig. 4b), which did not produce any significant runoff, and the event on 7 June 2011. The events on 27 November 2010 and 10 December 2011 in particular showed a very flashy response. These events had the highest peak hourly rainfall intensity (30.4 and 31 mm h-1, respectively) during the study period, and they occurred at the end of the flow season with low ASI (ASI was 161 and 192 mm, respectively). The highest peak runoff rates for the study period were for the events on 27 November 2010 and 10 December 2011, which were 2.4 and 5.6 mm h-1, respectively. In contrast to most events, the runoff response for the event on 27 November 2010 was transient with very rapid recession. For the event on 10 December 2011, a second peak of moderate rainfall intensity (about 10 mm h-1) produced a second runoff peak, and there was a more significant recession flow following the rainfall bursts. This was also true for the other events (8 and 26 November 2011, 11 May 2011, 1 August 2010, 7 June 2011) shown in Fig. 4, which were typical of responses to lower intensity rainfall during wetter (in terms of soil water) periods.

For events with Ipeak < 10 mm h-1 there was a general increase in response as the ASI increased. The event on 12 November 2010 had 184 mm ASI and total rainfall was 28 mm, and it did not produce any runoff. This was a typical example of no-flow events. Coming into the runoff season, as the ASI increased (e.g. 220 mm on 11 May 2011), RBF started to respond gradually, producing small amounts of runoff (e.g. for events on 11 and 14 May 2011). When the ASI was > 250 mm for the event on 7 June 2011, it can be clearly seen that RBF responded to this low intensity, small size rainfall event with a delayed and smooth discharge hydrograph with continued flow following the event. This also occurred for the next event on 1 August 2010.

In Fig. 1 we set out a number of thresholds that are important in runoff production mechanisms. We now explore the event data from the perspective of thresholds, concentrating on two key ones: catchment wetness and rainfall intensity. Figure 5 builds on approaches by Detty and McGuire (2010), who considered thresholds in ASI and ASI + Rain (i.e. Fig. 5c and d), and Janzen and McDonnell (2015), who considered the impact of event rainfall and rainfall intensity on event runoff (i.e. Fig. 5a). As we move from Fig. 5a to d, the various influences on rainfall–runoff behaviour and thresholds become clearer. Figure 5a shows event runoff as a function of event rainfall, with the highest hourly rainfall intensity (Ipeak) indicated in colour. This illustrates that there is little relationship between event rainfall and runoff or between rainfall intensity and runoff. For some quite large events (up to ∼ 25mm), zero runoff can occur. There was also a wide variation in runoff coefficients (indicated by the scatter). It is also clear that the events with high peak hourly intensity also had relatively large total rainfall accumulations. Overall, Fig. 5a shows that event rainfall and intensity do not effectively differentiate rainfall–runoff behaviour when considered by themselves.

Figure 5b begins to separate out different effects by showing the impact of five factors together. The cumulative curve shows the distribution of soil water storage as observed throughout the study period. Specific events are shown with the ASI identified (left-hand end of the grey lines) and the ASI + Rain depth (filled markers at the right end of the horizontal grey lines). The length of the lines is the rainfall depth for the event. The colour of each bubble shows the peak hourly rainfall intensity (Ipeak) and the size of the bubbles shows the event runoff coefficient. Squares indicate events that did not produce any runoff or where the peak runoff rate was less than 0.05 mm h-1. The vertical grey dashed line shows the ASI (left-hand end of the horizontal grey line) or ASI + Rain = 250 mm (right-hand end of the horizontal grey line).

There are several trends that can be discerned from Fig. 5b. First the rainfall events analysed here occurred across the full range of catchment wetness and were relatively evenly spread, showing that the rainfall events occur over a representative range of catchment antecedent conditions. The larger rainfall events generally occurred in summer when ASI < 250 mm and a mix of low and high intensity events occurred for these conditions, also in summer. All the events on a wet catchment (ASI > 250 mm) had low Ipeak (≤ 6.2 mm h-1). Events where the ASI + Rain was less than 250 mm usually did not generate any runoff, although there were some high intensity rainfall events that were exceptions and a small number of events with very low runoff coefficients (1–4 %) where the ASI + Rain was generally between 230 and 250 mm. These low runoff coefficient events were at the end of the runoff season.

Figure 5c and d examine the impact of catchment wetness, quantified as ASI and ASI + Rain, respectively, at the start (Fig. 5c) and end (Fig. 5d) of the event, on event runoff response, combined with the impact of rainfall intensity, Ipeak. Catchment wetness is plotted on the x axis and rainfall intensity on the y axis. Values ASI = 250 mm (Fig. 5c), ASI + Rain = 250 mm (Fig. 5d) and Ipeak = 15 mm h-1 are shown by grey dashed lines. The bubble size shows the event runoff coefficient, as before, and crosses indicate rainfall events that did not generate any runoff. Colour indicates the runoff volume. The runoff coefficient behaviour is separated into groups more clearly in Fig. 5d than in Fig. 5c. In Fig. 5d, three different groups of events can be identified. Looking along the x axis, there is a threshold at ASI + Rain = 250 mm that separates most events with a significant runoff response from those without. Looking along the y axis direction, it can be seen that the ASI + Rain threshold is not successful at distinguishing runoff when Ipeak is high and there is also a threshold in runoff response at Ipeak = 15 mm h-1. Thus the three groups are (1) events without runoff where ASI + Rain < 250 mm and Ipeak < 10 mm h-1, (2) events that produce runoff when ASI + Rain > 250 mm and (3) events with ASI + Rain < 250 mm and Ipeak > 15 mm h-1 that did produce runoff (Tables 1 and 2). Where ASI + Rain exceeded 250 mm (group 2), some runoff was always produced.

Both the first and third groups had ASI + Rain less than 250 mm, but they behaved differently in that some produced runoff and others did not. In the first group low intensity rainfalls mostly happened in drier periods when ASI varied between 146 and 227 mm. It is assumed that rainfall completely infiltrated into the soil and these events did not produce runoff (see Table 2 for event characteristics).

The third group (Table 1) occurred during dry periods at the end of the flow season when the ASI was < 200 mm. The runoff coefficients for the four events with peak hourly intensity of 15 mm h-1 and higher are 3, 12, 20 and 68 % for peak hourly intensities of 16, 30, 15 and 31 mm h-1 and ASI + Rain of 202, 215, 257 and 245 mm, respectively. Note that one of these events exceeds both the wetness and intensity thresholds. These events were distinguished by having maximum hourly rainfall intensities above ∼ 15 mm h-1, and they did produce runoff. In particular, two of these events on 27 November 2010 and 10 December 2011 had the highest rainfall intensities observed (Ipeak > 30 mm h-1), and they produced the highest peak runoff rates (8.1 and 9.1 mm h-1) and hourly runoff totals (2.4 and 5.6 mm) observed during the study period (Fig. 3). These runoff peaks happened at the same time as the highest recorded rainfall intensities. Antecedent stream discharge for the events on 27 November 2010 and 4 February 2011 was zero and the hydrograph rose and recessed quickly. For the event on 10 December 2011, the ASI was 192 mm, and the total rainfall was 53 mm, with an initial discharge of 0.13 mm h-1 (Table 1). It produced the highest observed peak hourly runoff of 5.6 mm, the runoff duration was 32 h and total runoff was 41 mm. The highest intensity was observed in the first 2 h of the event and the runoff coefficient calculated for the first 2 h of the rainfall event was 18 %. The RC calculated for the duration this event was 68 %. These are large compared with the maximum surface saturated extent throughout the study period of about 6 % of the catchment area, indicating that processes other than saturation excess runoff are important. The event on 10 December 2011 marked the end of a particularly rainy period, with more than 200 mm over 30 days. Note that due to flow measurement equipment being removed after this event, it was the last recorded at RBF.

Table 3 provides a summary of the average conditions for each group of events. In terms of rainfall characteristics, the events that produced runoff (groups 2 and 3) tended to have higher total rainfall, with the highest intensity events also having the largest total rainfall. Of course the grouping criteria mean larger groups are more likely to fall into group 2 compared with group 1. Peak rainfall intensities were both low and almost identical for groups 1 and 2. The runoff behaviour is quite different, with almost no flow and very low runoff coefficients on average for group 1 and average runoff coefficients of 17.9 and 25.8 for groups 2 and 3, respectively. These runoff coefficients are clearly much higher than the observed maximum saturated area proportion (6 %).

We also undertook a bootstrap analysis to test for differences between the group means. This showed that groups 2 and 3 are statistically different from group 1 at the 1 % significant level for total event runoff, quick flow and the quickflow runoff coefficient. Groups 1 and 2 (which are distinguished by ASI + Rain) are statistically similar in terms of Ipeak, suggesting rainfall intensity does not explain the runoff response differences between these two groups. Groups 1 and 3 (which are distinguished by Ipeak) are statistically similar in terms of ASI + Rain, suggesting that catchment wetness does not explain the runoff response differences between these two groups.

Overall these results demonstrate a range of different rainfall–runoff responses. The responses depended on both the catchment wetness as quantified by ASI + Rain and on the peak hourly rainfall intensity, Ipeak, with thresholds of 250 mm and 15 mm h-1 being identified. Runoff was produced whenever thresholds in either of these were exceeded.

The above presentation of results from Fig. 5d identifies a threshold catchment wetness expressed as antecedent soil water storage plus event rainfall depth of 250 mm above which runoff always occurred and another threshold of hourly rainfall depth exceeding 15 mm which also led to runoff production. It should be noted that there is some uncertainty in the Ipeak threshold as the most intense event (for ASI + Rain < 250 mm) that failed to produce runoff was 10 mm h-1, while the least intense event that did produce runoff was 14.8 mm h-1. Looking at events in the lower right quarter of Fig. 5d also shows that the event runoff coefficient tends to increase as either catchment wetness or peak hourly intensity increases. In fact, runoff and non-runoff producing events are very well separated by the relationship Ipeak=3/11⋅(ASI-260). These results suggest that there are both wetness dependent and intensity dependent runoff production mechanisms operating. This section examines the evidence for different runoff mechanisms contributing to event runoff.

The hydrometric data enable us to identify the important runoff mechanisms under different circumstances. The isotope and major ion geochemistry data provide further supporting evidence. The rainfall plus antecedent soil water threshold of 250 mm that needs to be exceeded for runoff in most circumstances shows that wetness dependent runoff processes are important, that is, either saturation excess or subsurface stormflow. Shallow groundwater data combined with field mapping of surface saturated areas show that complete profile saturation is limited to about 5 % of the catchment area and that this saturation is persistent, with only a small variation (see Table 1) over the winter–spring season but reducing over the dry summer–autumn period. The extent of the saturated area at the riparian zone varied seasonally and between events. Field observations of surface saturation extending to the flume and well hydrograph measurements of the water table at the surface showed that the saturated area was highly connected to the catchment outlet and suggest that it would be expected to produce saturation excess runoff. This surface connection disappeared during the dry summer season when the water level in all wells fell below the surface (see Fig. 6), so that the saturated area in the riparian zone disappeared. The isotope results for 12 August 2010 enabled the event runoff to be separated into event water and pre-event water contributions. Four to five percent of the rainfall volume on the catchment appeared in the event runoff, which corresponds well to the proportion of surface saturated area in the catchment (5.5 %, Table 1), supporting the identification of significant saturation excess runoff from this part of the catchment, as observed elsewhere (McGlynn and McDonnell, 2003; Penna et al., 2016).

While saturation excess runoff undoubtedly occurs, many of the event runoff coefficients were well in excess of 5 %, and they approach 100 % under very wet conditions (Fig. 5). The event on 12 August showed a substantial pre-event water contribution; logged shallow wells show that the water table did not reach the surface in the steeper areas of the catchment, even within the convergent drainage lines under very wet conditions (e.g. sites 5 and 7). The recession behaviour of wells in the catchment suggests subsurface flow moves down the catchment under wet conditions and the recession constant analysis shows that this connection becomes stronger as the catchment wets beyond 250 mm of stored water. This is all consistent with a substantial contribution of subsurface flow to event runoff once the catchment is sufficiently wet to establish subsurface connection with the riparian area, as has been inferred in other studies (Buttle et al., 2004; Detty and McGuire, 2010; Hewlett and Hibbert, 1967; Jencso et al., 2009; Penna et al., 2011).

Perhaps more surprisingly there was a group of events that produced runoff under conditions of relatively low soil water storage (ASI + Rain < 250 mm) but high rainfall intensity. This suggests that an intensity dependent runoff process is being triggered when rainfall exceeds some threshold for sufficient time, in this case about 15 mm of rainfall in an hour. It is worth noting that the 15 mm h-1 threshold will be dependent on the time step used in calculating the intensity. The choice of time step needs to consider the travel time from hillslope to stream as run-on infiltration processes occurring on the catchment surface will impact the connectivity to the stream. Thus the most appropriate time to use for averaging intensity would be the average travel time to the stream because it is this time period which is available to infiltrate rainfall that has become runoff at the point scale. Here the choice of hourly rainfall was pragmatic, as that was the recording time step of the AWS, but it is also likely that it reasonably approximates the flow times.

It is tempting to assume that this evidence suggested surface runoff occurred due to infiltration excess runoff, but it is also possible that the high rainfall intensities are efficiently activating macropore networks (Beven and Germann, 1982) and that the flow could be following subsurface pathways. For these events there were also rapid responses in water levels in wells on the hillslope and it is not clear exactly how the water moved rapidly into the wells in these cases, but it could be due to preferential flow through macropores. Furthermore, for the high intensity event on 8 November 2011, the flow after the main peak had a surprisingly low concentration of chloride (the cluster of low concentration red points in Fig. 10a), which may suggest that the higher intensity activated either overland or preferential flow paths, limiting soil contact time and leading to this low concentration.

The hydrograph from the event on 8 November 11 (the event that exceeded both thresholds, Fig. 10) shows both a rapid and a delayed runoff response. The concentration of chloride was unusually low during the delayed runoff component compared with all other events (with major ion data available). This may suggest limited contact with the catchment soils which could occur if macropore flow was important, but this explanation is not definitive. Of the two events with peak hourly intensities around 15 mm h-1, one also exceeded the wetness (ASI + Rain) threshold of 250 mm and the other had a very low runoff coefficient (only 3 %), and hence these two events are somewhat equivocal in terms of the importance of intensity. However, the two events with peak hourly rainfall intensities around 30 mm h-1 both produced rapid runoff responses without a significant delayed component (Fig. 4) and had runoff coefficients (12 and 68 %) well in excess of the surface saturated area (5 %) in the catchment, showing clear evidence of the role of rainfall intensity of 30 mm h-1. Unfortunately isotope and major ion data were not available for those events to attempt to determine whether surface or subsurface pathways are important. Overall there is clear evidence for intensity dependent runoff mechanisms, especially for the largest 30 mm h-1 events.

In summary, the process evidence relating runoff behaviour to catchment wetness thresholds, together with the data from shallow wells, suggests a catchment where subsurface flow leads to a seasonally saturated riparian area that produces saturation excess runoff in immediate response to rainfall. This saturation excess is augmented by subsurface stormflow when the catchment wetness (ASI + Rain) exceeds a 250 mm threshold. This subsurface flow exfiltrates in the riparian area. Volume considerations provide evidence that, for many of the monitored events, water must be coming from outside the riparian area. This area is only about 5 % of the catchment, but quickflow runoff coefficients very regularly exceed this, often (about 1/3 of runoff events) exceeded 20 %, and on occasion exceeded 50 %. The role of both saturation excess runoff and subsurface stormflow is corroborated to some extend by isotopic data from one event where the hydrograph separation shows a volume of event water similar to the volume of rainfall on the saturated area and by the highly damped isotopic signal in the stream in general. When hourly rainfalls exceed a threshold of 15–30 mm h-1, an intensity dependent runoff process is activated that also contributes flow from the hillslope area outside the riparian zone. It is not clear whether this is a purely surface runoff process or not. One of the key contributions of this work is clear hydrometric evidence of the interplay of both wetness and intensity dependent runoff processes in the one catchment. These responses include two wetness dependent processes, saturation excess runoff and subsurface stormflow, and an intensity dependent process for which we cannot distinguish between surface and subsurface flow pathways with our data.

We identified two important thresholds in the catchment response. The first is a wetness threshold of ASI + Rain exceeding 250 mm. Under these conditions the water table approaches the surface in the riparian area and water tables rise on the hillslope into what is inferred from relatively rapid hillslope water table recessions to be a more transmissive part of the soil profile (within ∼ 120–140 cm of the surface). Based on the volume of event runoff (Table 1), the lack of surface saturation on the hillslope and the behaviour of shallow wells (Fig. 6), we infer that the hillslope becomes connected to the riparian zone under these conditions. The existence of a large volume of pre-event water in the event on 12 August 2010 (Fig. 8) adds further corroboration to this. Similar catchment wetness thresholds for connectivity and runoff generation have been reported elsewhere (Detty and McGuire, 2010; Penna et al., 2011; Tromp-van Meerveld and McDonnell, 2006b). These have been expressed either in terms of rainfall depth, antecedent soil water storage conditions, or a combination of these. Similar to our results, in a study of a forested subcatchment with highly permeable soils and a small riparian area, Detty and McGuire (2010) found a relationship between a threshold of 316 mm for ASI + Rain and the start of the event streamflow response. They showed that the ASI + Rain threshold corresponded to a water table height threshold. Based on this, they suggested that subsurface flow, transmissivity feedback and preferential flow from hillslope to stream could be used to explain runoff mechanisms in their catchment. They did not observe either Hortonian overland flow or saturation overland flow (SOF) even during the largest events.

A second threshold to initiate event streamflow response associated with high rainfall intensities was also evident. A similar role of intensity has also been observed by Janzen and McDonnell (2015), who found that the Panola hillslope can produce significant runoff from dry antecedent conditions when high intensity rainfall occurs. In general, event runoff from Panola is controlled by catchment wetness, similar to our catchment. Mixed wetness and intensity dependent runoff processes have also been reported in Mediterranean catchments (e.g. Lana-Renault et al., 2014). Nevertheless, there are only a few studies we know of that have reported intensity thresholds in catchments where runoff is normally dominated by wetness thresholds. This may be a consequence of such events being relatively rare in any given catchment (roughly 10 % of runoff producing events in our case).

We now consider the three runoff processes occurring in the catchment in relation to the framework proposed in Fig. 1 and the various flux and timescale thresholds identified therein. Essentially Fig. 1 is posing a series of questions that allow us to systematically think through the runoff processes. Above we have identified three groups of rainfall events: those that do not produce runoff; those that produce runoff by saturation excess and subsurface stormflow from a wet catchment; and those that produce runoff from higher intensity events.

The rainfall events that do not produce runoff are not exceeding infiltration capacity for a sufficient time for runoff to flow from the catchment (box 1, “No”). They may or may not produce significant percolation (box 2), but if any of these events does produce percolation to a perched water table, this only results in an ephemeral water table that dissipates before lateral flow can move water down the hillslope (box 3, “No”), and they do not saturate the full profile (box 4, “No”). As a consequence no event runoff is produced. These conditions correspond to the local control state of Grayson et al. (1997).

Some high intensity rainfall events on a dryer catchment do exceed the infiltration capacity for sufficiently long periods of time (box 1, “Yes”; hourly intensity of 15–30 mm h-1) and these events produce runoff. The increase in runoff coefficient as intensity increases for a given ASI + Rain in Fig. 5 suggests that infiltration thresholds may also be playing a role for wetter conditions (i.e. box 1 “Yes”) but that infiltrated water (the dashed link in Fig. 1) also contributes through other mechanisms.

The final group of events are those that produce runoff from a wet catchment at low intensity rainfall. These follow the path box 1 “No”, box 2 “Yes” and box 3 “Yes” in Fig. 1. On the upper hillslope segments of the catchment, the subsurface flow capacity is sufficient so that the water table does not reach the surface and water drains either during or shortly after the storm (box 6, “Yes”). Subsurface connectivity develops during the event and subsurface flow dominates. The shallow well data (Fig. 6, wells 1 and 2, to some extent 32) show that the riparian area in the lower part of the catchment drains very slowly. There is a substantial reduction in slopes (and probably lower hydraulic conductivity associated with poorly structured, poorly drained soils) that suggests a substantial reduction in lateral subsurface flow capacity. The surface topography also suggests subsurface flow would converge in this area. From the perspective of Fig. 1, this leads to a situation where water takes longer than the typical time between events in the wet season to drain (box 7, “Yes”), resulting in persistent surface saturation and saturation excess runoff generation from the lower catchment. Hence under wet conditions this catchment produces a mix of saturation excess and subsurface storm flow, but from geographically distinct parts of the catchment.

The above illustrates how the framework in Fig. 1 can be used to understand the role of different thresholds regarding fluxes and timescales in determining runoff mechanisms. Such a framework is likely to be particularly valuable where there is a mix of runoff mechanisms operating for different events or in different parts of the catchment. Our study catchment nicely illustrates such a mixture.