New Zealand has some well-known experimental catchments, which offer information on causes and effects of hydrological variability, focusing on the soil zone. In the Mahurangi catchment in Northland, Wilson et al. (2004) compared the variability of gridded soil moisture measurements in time vs. in space. They found that temporal variability was approximately 5 times greater than spatial variability. Temporal variability was highly predictable, and explained by seasonality, whereas spatial variability was less easily predictable and only partly explained by terrain indices. In the same catchment, Wilson et al. (2003) compared variability of soil moisture at 0–6 cm depth vs. 30 cm depth, and found differences in distribution and low correlations between the two depths. At Maimai catchment in Westland, nested arrays of tensiometers were used to estimate variability in the depth to water table. High variability was found within nests (plot scale) and between nests (hillslope scale) (McDonnell, 1990; Freer et al., 2004).

Some characteristics of the New Zealand climate and landscape may result in locally important controls on variability. Aspect is important in New Zealand hill country, due to high radiation and prevailing wind direction. Typically, Penman PET is 35–50 % greater on north-facing than south-facing slopes (Jackson, 1967; Bretherton et al., 2010), or more for sites exposed to the prevailing WNW wind (Lambert and Roberts, 1976). At one site, these differences translated into mean soil moisture differences of 10 % (Bretherton et al., 2010). In a similar environment to the catchment described in this paper (i.e. the eastern foothills of the Southern Alps, greywacke geology), aspect-induced microclimate differences were found to promote physical and chemical soil differences, with stronger leaching and weathering on south-facing slopes (Eger and Hewitt, 2008).

Controls on soil moisture are varied and may affect soil moisture mean (in either space or time), distribution (Teuling et al., 2005) and dynamics such as recession, stability or recharge rate (Kim et al., 2007). Examples from previous (international) studies are given in Table 1. Controls can also interact, such as soil type and topography (Crave and Gascuel-Odoux, 1997). Even though new technologies are available to measure soil moisture and its variation at larger scales, including remote microwave sensing (Njoku et al., 2002) and electrical resistivity tomography (Michot et al., 2003), there is still no accurate way of predicting soil moisture patterns, with studies based on topography typically predicting less than 50 % of the spatial variation (see the review by Wilson et al., 2004).

High variability in soil moisture has many implications for hydrological process understanding and modelling. There is a large body of work investigating causes of low vs. high variability, without attempting to predict exact spatial or temporal patterns, often using geostatistical methods to quantify the magnitude and the scales of variation (e.g. Western et al., 1998; Brocca et al., 2007). Causes of high variability have been found to be dry conditions (Brocca et al., 2007), mid-wetness conditions (Ryu and Famiglietti, 2005; Rosenbaum et al., 2012), wet or dry conditions conditional on climate, soil and vegetation types (Teuling and Troch, 2005; Teuling et al., 2007), increasing scale (Famiglietti et al., 2008; Entin et al., 2000), aspects of land use and topography (Qiu et al., 2001), groundwater influence, and contrasts between groundwater influenced/uninfluenced areas (Rosenbaum et al., 2012).

Studies of variability in groundwater dynamics are less common, reflecting the greater difficulty and expense in measuring groundwater levels, but a wide range of controls on groundwater levels have been identified. Detty and McGuire (2010a) considered surface topography controls, by dividing the landscape into landform units, e.g. footslopes, planar backslopes, or convex shoulders. They found statistical differences in metrics of water table hydrograph shape between different landform units. The water table response increased in duration and magnitude from shoulders to footslopes, but was most sustained on backslopes. The responses also differed between the growing and dormant seasons. Anderson and Burt (1978) showed that topography can control matric potential and downslope flow: at their field site, hillslope “hollows” had specific discharge an order of magnitude higher than hillslope spurs. Fujimoto et al. (2008) found that topography interacts with storm size to control subsurface processes. For small storms, a concave hillslope stored more water than a planar slope and produced less runoff, whereas for larger storms, transient groundwater in the concave slope caused greater expansion of the saturated area than in the planar slope, and correspondingly greater runoff. Bachmair et al. (2012) drilled 9 transects, each of 10 shallow wells (< 2 m deep) to study the effect of land use and landscape position on variability in groundwater dynamics. They found that patterns of groundwater response in winter reflected expansion of saturated areas at the base of the hillslope, whereas in summer, groundwater response was controlled by transient preferential flow networks and was highly spatially variable. The wells with the strongest response also varied between events. The relationship between topography and subsurface flow dynamics has been demonstrated theoretically (Harman and Sivapalan, 2009), although bedrock topography may be more important than surface topography (Freer et al., 2002; Graham et al., 2010; Tromp-van Meerveld and McDonnell, 2006a, b).

Other factors may also control the variability in groundwater responses, such as variability in recharge. Gleeson et al. (2009) tracked snowmelt recharge to groundwater using 15 bedrock wells in a humid Canadian catchment with flat topography. In addition to widespread slow recharge, they found fast, localised recharge in areas with both thin soils and fractured bedrock. Riparian soils can form a fast conduit to groundwater, where a higher fraction of gravel leads to hydraulic conductivities an order of magnitude higher than the hillslope soils (Detty and McGuire, 2010b).

Characteristics of the groundwater aquifers are also important. Winter et al. (2008) and Tiedeman et al. (1998) monitored 31 bedrock wells and found water table gradients caused by different geological units within a catchment. Even in headwater catchments, variability in groundwater dynamics has been found due to multiple underlying aquifers (Kosugi et al., 2008, 2011). In Plynlimon catchment in Wales, Haria and Shand (2004) found that groundwater at 1.5, 10 and 30 m depth was not hydraulically connected, and was chemically stratified, with distinct pH, electrical conductivity and redox characteristics. Different groundwater pathways to the stream could therefore be identified, including discharge from fractured bedrock, and upwelling into the soil zone causing rapid lateral flow.

The division between stored water that is considered soil moisture or groundwater is not well defined. Soil moisture is typically measured as volumetric water content at a specific depth in the unsaturated zone, although soil moisture sensors can be subsumed by groundwater. Here, we use groundwater level synonymously with water table, referring to saturated subsurface layers, which may be above or below any soil–bedrock interface. Piezometers or shallow wells to measure groundwater level can be screened along their whole length (as in our study) or at specific depths if multiple perched or confined layers are suspected. Where the geology includes fractured rock or buried lenses of gravels, groundwater levels may be highly heterogeneous.

There are many processes by which soil moisture and groundwater interact. As soil water drains downwards, layers of low hydraulic conductivity may create perched water tables. Such layers include clay pans (Parlange et al., 1989) and the soil–bedrock interface (Tromp-van Meerveld and McDonnell, 2006a). Macropores provide a fast route for groundwater recharge (Beven and Germann, 2013). They may allow water to bypass confining layers or to flow quickly along them (e.g. lateral preferential flow along the bedrock interface found by Graham et al., 2010). If groundwater rises into upper soil layers, large increases in soil matrix porosity or macropores may “cap” water table levels, as additional water is quickly transported to the stream (Haught and Meerveld, 2011). Lana-Renault et al. (2014) found in a Mediterranean catchment that patterns of near-surface saturation and transient water tables were affected not only by topography, but also by soil properties and previous agricultural land use. The riparian zone facilitates mixing between soil water and groundwater, and tracers, temperature, electrical conductivity, flow gauging and head differences may all be used to quantify the interactions (Unland et al., 2013). Using modelling and tracer data, Binley et al. (2013) found that in a 200 m river reach the upper section was connected to regional groundwater, but lower section inflows were from local lateral and down-river flow paths.

Interactions between soil moisture and groundwater provide possible explanations for relationships between the two. Results from three Nordic catchments showed a consistent negative correlation between soil moisture content and depth to water table, so that soil moisture distributions could be described as a function of depth to water table (Beldring et al., 1999). Kaplan and Munoz-Carpena (2011) studied soil moisture regimes in a coastal floodplain forest in Florida, and showed that groundwater and standing surface water elevations were successful predictors of soil moisture using dynamic factor analysis and regression models. Model-based studies demonstrate how capillary rise can lead to dependencies between groundwater level and soil moisture. Kim et al. (1999) used a hillslope model to show how gravity-driven downhill groundwater flow creates downslope zones with high water tables. In those areas, capillary rise keeps soil moisture content and evaporation rates high. Similarly, the model developed by Chen and Hu (2004) showed that soil moisture in the upper 1 m of soil was 21 % higher when exchange between soil moisture and groundwater was included; they inferred that groundwater variability may drive soil moisture variability.

A compact weather station was located centrally within the catchment (Fig. 1). It uses a Vaisala WXT520 weather transmitter, which measures wind speed and direction, air temperature, barometric pressure and relative humidity. A LiCOR LI200 pyranometer measures solar radiation. Rainfall was measured using an OTA OSK15180T 0.2 mm resolution tipping bucket gauge. All weather measurements were at 5 min intervals.

Streamflow was measured at three locations within the catchment (Fig. 1), all at 5 min intervals. Only data from the downstream gauge, a v-notch weir, were used in this paper. Periodical manual gaugings were used to confirm the theoretical weir flow rates.

Soil moisture and water table level were monitored by 16 instrument stations. The stations are divided into two groups: 10 on the north-facing slope, and 6 on the south-facing slope.

Our typical measurement site included an Acclima TDT soil moisture sensor at 30 cm (base of the root zone) and 60 cm, which were used with factory calibration as recommended by the manufacturer (Acclima, 2014). The sites also included a well drilled to a fixed depth of 1.5 m (except where a high fraction of stones prevented the full depth being reached) equipped with a Solinst levelogger to measure water level. The wells were sealed for the top 0.5 m to prevent ingress of surface water, with open screening below this. On each hillslope, we centred the sites around a shallow gully surface feature, with sites in the centre of the gully and on each bank. The sites were designed in two rows, at 10 m and 20 m from the stream centreline (Fig. 1). In this way, we aimed to sample across multiple variables of aspect, slope position and distance from stream. All sensors recorded at 5 min intervals, which were typically aggregated to 15 min before further analysis.

Each station aggregates sensor data and discards unneeded data. Each group is associated with a “master” station that polls the individual stations every 5 min for their sensor data. The master station comprises a Unidata Satellite NRT datalogger and a proprietary short-haul radio interface. The data received by the master station are stored temporarily in the logger until they can be relayed to a central database via satellite. Data in the central database are available to end users via internet and e-mail. To conserve power in the solar-recharged batteries, the sensors and radio system are only powered up to respond to data requests.

The data used in this paper were collected between March 2012 and July 2013 (Fig. 3). Climate and flow data are available for 14 months prior to this date. The largest rainfall event in the study period occurred in August 2012, which brought 80.6 mm of rainfall in 2 days, approximately a 1-in-2 year rainfall event when compared against the 62 year daily rainfall record from Melrose station, 2.0 km from the catchment. The 2012–2013 summer was unusually dry in many parts of New Zealand but, at Melrose, the summer months December/January/February recorded a rainfall total of 196 mm, only marginally below the long-term average of 210 mm.

Some data gaps occurred during the study period, with short outages due to sensor or battery failure. A long outage occurred in the aftermath of the storm event in August 2012, which caused water damage to the telemetry system on the north-facing slope.

To provide an overview of the soil moisture content and groundwater level for different time/space locations, a selection of summary statistics was used. To summarise the distribution of data, we calculated the median and 5th, 25th, 75th and 95th percentiles for each data series. This allowed us to compare absolute soil water content and groundwater level between sites. However, we also wanted to compare the extent to which each location is likely to contribute to runoff, especially as runoff generation is typically conceptualised as a threshold process (Ali et al., 2013). We therefore additionally used statistics that described the wet extremes of the data. For soil moisture, we calculated the percentage of time that the soil was saturated, as this represents the condition where the location would generate subsurface flow and, if saturation reached the soil surface, overland flow. Soil saturation points were defined individually for each sensor, using the co-located groundwater well record to determine times when the water table intersected the sensor, and taking the average soil moisture reading at those times. These values were confirmed (and in two cases adjusted) based on visual inspection of the soil moisture time series. For groundwater level, we calculated the percentage of time that the water table level was above the 75th percentile. This quantifies locations where groundwater is closer to the surface and would therefore have faster lateral velocity according to typical findings that hydraulic conductivity decreases rapidly with depth (Beven and Kirkby, 1979).

To understand how total water storage in the catchment changes through the year, we estimated the water stored in the soil moisture and groundwater components. For soil moisture, we divided the catchment by soil type, according to the classification described in Sect. 2. For each type, we estimated total soil depth as the deepest functional soil horizon described in the S-Map database (Lilburne et al., 2004). The fraction soil moisture for soil from 0 to 45 cm depth was taken from the 30 cm sensor, and soil moisture for 45 cm depth to the base of the soil column was taken from the 60 cm sensor. For each time step, we derived the total soil moisture volume as Total soil moisturem3=∑SoilType∑AspectAream2⋅Soil depth[m]⋅fraction soil moisture. Dividing by total catchment area then gave average depth of soil water.

For groundwater, we do not know the total aquifer depth, and therefore use instead groundwater depth above minimum recorded. For each time step, we derived the variable groundwater storage above minimum as Total groundwaterm3=∑AspectAream2⋅∑WellsGW level[m]-Min.GW level[m]/Number of wells. Dividing by total catchment area then gave average depth of groundwater above minimum.

We recognise that this calculation involves a significant and uncertain extrapolation from the 32 soil moisture time series to the remainder of the 0.7 km2 catchment. However, given that the sensor locations were installed across aspect, distance from stream, and landscape position and depth, we anticipate that the estimated storage dynamics are a reasonable guide to true behaviour. We also note that, in the riparian zone, some water will be double counted where the groundwater rises into the soil column. However, given that this occurred only in near-stream locations, i.e. within 10 m of the stream centreline, this volume would be negligible compared to the total catchment storage.

To compare rainfall and runoff depths for individual storm events, and to identify seasonal changes in the rainfall–runoff relationship, the data were pre-processed to define storm and inter-storm periods, based on the method of McMillan et al. (2014). The start of a storm was defined by a minimum rainfall intensity: either 2 mm h-1 or 10 mm day-1 was required. The end of the storm was defined when 12 h without rainfall occurred. Runoff for a maximum of 5 days after rainfall ended, or until a new storm started, was deemed to be associated with the storm event. No baseflow separation was used.

To compare the frequency and strength of soil moisture responses to rainfall for different locations, we used the concept of a “wetting event”. A wetting event was defined as a period of rainfall during which soil moisture rose by at least 3 %. We calculated events on a per-site basis, and then averaged across sites, either for northern/southern aspects or near-stream/far-stream sites. The average % soil moisture rise was used as an indication of the strength of response.

Both soil moisture and groundwater level show strong variations over event and seasonal timescales. Figure 3 shows soil moisture, and depth to groundwater for the study period; for clarity we average the 32 soil moisture sensors and 14 water level sensors by location (aspect, depth and distance from stream).

In Fig. 4, we show the summary measures for each season. The summary statistics show that both the mean and extremes of catchment water storage vary seasonally. The yearly cycle of soil moisture (Fig. 3) shows an extended wet season from April/May to November, followed by a slow drying until February, when the catchment reaches its summer state. The return to wet conditions occurred over a very short time period during a May storm event. Water table dynamics also display a yearly cycle (Fig. 4), although the range during any season is large compared to seasonal changes. As shown in Fig. 4a, soil moisture quantiles are typically lowest in summer, and water tables are lowest in summer and autumn. The driest conditions in terms of extremes (Fig. 4b) occurred in late summer for both soil moisture and water table, and remain low into autumn, particularly for the water table, suggesting that the lowest potential for runoff generation occurs at that time. Note that the autumn season values represent an average between the wetter conditions of the 2012 autumn and the drier conditions of the 2013 autumn; for example, mean autumn (March–May) soil moisture at 0–30 cm for the upper rows of sensors was 17.9 % for 2012, 15.2 % for 2013.

Rainfall events are superimposed on the seasonal cycle. In winter, the large events cause saturation at many of the soil moisture sensors, and water tables rise in many of the wells, including some in the upper row where the water table was previously lower than the well. In early summer, rainfall can return soil moisture and water tables to winter levels, but only briefly. In summer, the catchment response to rainfall is highly subdued.

The strong seasonality of catchment conditions is due to seasonality in PET. Although rainfall depths are similar throughout the year, in summer the combination of higher temperatures, high solar radiation and frequent hot, strong winds from the northwest contributes to seasonal drying of the catchment. The effects are illustrated by storm runoff depths in winter vs. summer (Fig. 5a). In summer, even large rainfall events produced almost no streamflow response. To demonstrate the effect of antecedent wetness on storm runoff depths, we plotted runoff depth against the sum of antecedent soil moisture storage (ASM) and storm precipitation (Fig. 5b), following Detty and McGuire (2010b; their Fig. 4a). Antecedent soil moisture storage was taken as the total soil moisture value from Eq. (1). The results show a threshold relationship between ASM + precipitation and runoff depth, although it is not linear, as was found by Detty and McGuire (2010b).

Figure 3 shows distinct differences between the water storage dynamics on north-facing and south-facing slopes, and between the far-stream and near-stream rows of soil moisture sensors. The near-stream sensors on the south-facing slopes showed more frequent and pronounced wetting events, as defined in Sect. 3.7 (Table 4). South-facing slopes at 60 cm depth had 33 % more wetting events that were on average 22 % larger than north-facing slopes at 60 cm depth.

Spatial controls act differently on different water stores. These differences are illustrated in Fig. 6, using the same summary statistics as in the previous section, but grouping sites by aspect and distance from stream. We did not include water table statistics for the far-stream rows as water tables only rarely rose into the wells, and therefore distribution estimates would not be accurate. Figure 6a shows that when comparing north-facing vs. south-facing slopes, soil water content at 30 cm has similar distributions, but the underlying groundwater level is on average 20 cm closer to the ground surface for the south-facing slopes, and has a smaller range. Spatial controls also act differently on average vs. extreme conditions; e.g. average soil moisture on the south-facing slope is similar at 30 and 60 cm depths (Fig. 6a), but the fraction of time that the soil was saturated is 11 % at 60 cm against 0.5 % at 30 cm (Fig. 6b). Note that the statistics describing the extremes of the data are highly variable between locations (e.g. some locations are saturated much of the time; others, almost never); however, we show averages by location to assist interpretation of the spatial control.

To quantify the relative importance of different water storage components of the catchment, we calculated the average depth of water stored as soil moisture and groundwater using the method described in Sect. 3.5 (Fig. 7a). The groundwater component dominates, with an average depth of 0.27 m against 0.15 m for soil moisture. The difference may be further enhanced given that the part of the soil moisture volume below wilting point is not likely to be mobilised. The difference is most pronounced in the wettest conditions, with groundwater storage peaking at approximately 4 times that of soil moisture. During the driest summer conditions, groundwater and soil moisture storage are similar.

To visualise the changes in variability for each store over time, we plotted the time series of spatial standard deviation in soil moisture and groundwater, separated by aspect and sensor depth (Fig. 7b and c). All stores have the highest standard deviation in winter, and the lowest in summer, as the range in values tends to be compressed as the catchment dries out. Previous studies have shown that the relationship between soil moisture and soil moisture standard deviation varies by catchment (Sect. 1.1). Soil moisture at 60 cm maintains a high standard deviation even during summer, as both slopes have one sensor that retains high soil moisture and therefore has a strong influence on the standard deviation.

All of the soil moisture standard deviations rise sharply during rainfall events, especially in winter, which is due to saturation of some sensors, while others remained unsaturated. Accordingly, soil moisture at 30 cm on the north-facing slope has smaller rises in the spatial standard deviation, as none of those sensors showed saturation. Groundwater standard deviation has a different behaviour by aspect: on the north-facing slope, rainfall events cause the standard deviation to rise; on the south-facing slope, rainfall events cause the standard deviation to fall. This finding reflects that, on the south-facing slope, all wells react to rainfall events, albeit at different rates, but on the north-facing slope, behaviour is more variable, with one well often showing no response (i.e. water table lower than 1.5 m), and other wells split between a weak or strong response.

As was apparent from the time series of streamflow, soil moisture and water table depth presented in Sect. 3.1, there is significant spatial variability between different parts of the catchment as represented by the range of sensor locations, but this variability is not constant. In this section, we investigate the specific types of variability that occur, and seek to attribute them to different catchment conditions.

We found that an overarching driver of variability is the wetness condition of the catchment. As shown in Fig. 5, there is a strong seasonal differentiation in runoff coefficients. This seasonal cycle determines which of the catchment water stores are active, and where the greatest scope for variability exists. To assist our description of the seasonal changes in variability, we selected one event that illustrates each variability type. We selected the following events: dry period: 17–27 March 2013, 15.9 mm rainfall; wet period: 5–25 October 2012, 164.9 mm rainfall; winter wet-up: 15–30 April 2013, 80.0 mm rainfall; recession period: 7 September–5 October 2012.

Temporal variability is characterised by a strong seasonal cycle in catchment wetness; the mean and extremes of the soil moisture and water table distributions are higher in winter than summer. The cycle is driven by PET rather than rainfall depth, and causes significantly higher runoff coefficients in winter. The seasonal cycle in soil moisture shows a long, high winter plateau compared to water table levels that respond mainly to individual events. The catchment wets up quickly in autumn, but takes longer to dry out in spring, and spring rainfall can briefly return soil moisture and water table levels to their winter state. The volume of stored water in the catchment also has a seasonal cycle, mostly due to increased groundwater in winter, especially during the largest storms.

Spatial variability is controlled most strongly by aspect and distance from stream. South-facing slopes have similar mean soil moisture to north-facing slopes, but more events lead to a soil moisture response, and experience soil saturation more often. Water table levels are higher in south-facing slopes and more consistent between locations within the south-facing slope. Near-stream locations have higher soil moisture for both mean and extremes, and experience more wetting events. Near-stream locations frequently record saturation in winter, whereas far-stream locations have water tables below the soil moisture sensors and the 1.5 m wells for almost the whole study period. We found a strong interaction between groundwater level and soil moisture distribution. Sites where water tables peaked above the 30 cm sensor had a significantly higher soil moisture distribution compared to sites where water table remained below 30 cm for the whole study period. The finding that soil moisture distribution is dependent on water table depth agrees with measurements in Nordic catchments by Beldring et al. (1999).

Our conclusion that aspect is an important control on soil moisture echoes the results of previous studies in NZ hill country (e.g. Bretherton et al., 2010; Lambert and Roberts, 1976). The mechanisms linking aspect with soil moisture are varied. For example, Lambert and Roberts (1976) found complex interactions between air temperature, soil temperature and ET, driven by wind direction and aspect-induced radiation differences. They note that the specific heat capacity of soil drops as it dries, leading to a positive feedback cycle. In the Langs Gully catchment, the south-facing slopes are also steeper than the north-facing slopes. This is not obviously due to geological bedding – the main trend of syncline–anticline pairs in the wider Waipara catchment is northwest–southeast (transverse to catchment slopes), and in the immediate area of Langs Gully, known dip directions are highly variable. However, feedbacks are likely to exist between slope angle, vegetation (denser shrub cover on south-facing slopes), soil depth (thinner on south-facing slopes) and downslope sediment transport. Shading by denser vegetation and increased lateral flow are possible causes of the increased number of wetting events on the south-facing slope. Typical hydrological models do not account for aspect, but our results suggest that this is an important factor to consider in hillslope runoff generation.

We suggest that spatial variability can be classified as being in “summer mode” or “winter mode”. These modes are illustrated as a schematic diagram in Fig. 13. In “summer mode”, variability is controlled by shallow processes, e.g. interaction of water with soils and vegetation. Water does not typically penetrate to deeper soil moisture or groundwater. Summer variability is therefore disconnected from the channel, and will not directly affect the flow response. However, summer variability affects land surface processes such as evapotranspiration, and may have a lagged effect on the autumn/winter wetting-up process. An example of the disconnect is that the 30 cm soil moisture sites that reacted most strongly to the selected summer rainfall event did not correspond to the “saturating” sites identified in Sect. 4.4.2 as having consistently wetter soil moisture and shallow water tables.

In “winter mode”, variability is controlled by deeper processes, e.g. groundwater movement and bypass flow. The change from shallow vertical flow in dry conditions to vertical bypass flow and lateral flows from upslope in wet conditions is very similar to that found by Detty and McGuire (2010a). However, the summer and winter modes in Langs Gully differ from those found by Bachmair et al. (2012). In their catchment, intense summer storms onto dry soil caused preferential flow and fast, strong, spatially variable water table responses throughout the hillslope. In contrast, their winter storms led to slower water table responses that were strongest at near-stream locations.

In the shoulder seasons, there is a spatially variable shift between the summer and winter modes. Sensors in near-stream locations, particularly those with responsive water tables, stay longer in winter mode. As locations switch between summer and winter modes at varying rates, spatial variability is increased. This effect is particularly evident on the north-facing slope, where soil moisture standard deviation at 30 and 60 cm has a sustained rise during the spring drying period. Rosenbaum et al. (2012) similarly found that seasonal differences between groundwater-influenced and groundwater-distant locations had a strong effect on soil moisture standard deviation. This effect provides one explanation for why high spatial and temporal variabilities tend to co-occur, as has been found in previous work in New Zealand (McMillan et al., 2014).

It is common for some parts of the Langs Gully catchment to wet up or become saturated, and hence potentially contribute to a runoff response, while other parts of the catchment remain dry. Near-stream and south-facing locations have higher water tables and experience more wetting events. We were able to classify the near-stream sensors into “saturating zones” and “non-saturating zones”. The saturating zones had higher water table and wetter soil moisture distributions. These zones remained distinct throughout the year.

The saturating zones are likely to be dominant areas for runoff generation, as wetter soils facilitate vertical drainage and high water tables increase lateral transmissivity. For example, Jencso et al. (2010) found that connectivity between hillslopes and riparian zones led to fast turnover times of riparian groundwater. However, the saturating/non-saturating zones did not correspond to the pattern of sensors wetted by infiltration during a summer storm event. The different patterns imply that shallow soil moisture storm responses may not provide a good guide to winter runoff generation pathways, as also found by Tromp-van Meerveld and McDonnell (2005). Rainfall–runoff model structures that delineate catchment landscape components according to dominant processes (e.g. Gharari et al., 2011) may need to use different spatial disaggregations for shallow soil water and groundwater.

Understanding catchment variability has further implications for predictions of catchment behaviour. Variability controls which parts of the catchment are generating runoff and controlling water partitioning: it therefore controls uncertainty in flow predictions, depending on our knowledge or lack of knowledge about those water stores or fluxes. Similarly, variability controls how quickly water flows through a catchment, as the different response modes direct water into flow paths with different transit times (Heidbuechel et al., 2013). Variability also provides clues into unmeasured fluxes which are important for catchment response; for example, areas with more rapid water table movement suggest locations of preferential flow paths, either vertical or horizontal. Signatures of the catchment variability are seen in the flow response, such as a double or prolonged peak caused by slower groundwater pathways (also found by Bachmair et al., 2012), and seasonally variable changes in contributions between different hillslopes. These features suggest that understanding catchment-scale variability in hydrological processes is essential for predicting the hydrograph.