With water scarcity increasing globally, understanding the hydrology of rivers in arid and semi-arid regions has become especially important. In these regions, most streams and rivers are non-perennial, meaning surface flow ceases for some or most of the year. The understanding of the processes that lead to streamflow generation in non-perennial rivers is incomplete (Costigan et al., 2017; Gutierrez-Jurado et al., 2019; Shanafield et al., 2021). This is partly due to a lack of appropriate data; streamflow gauges are preferentially located on perennial rivers (Fekete and Vörösmarty, 2007; Poff et al., 2006). In addition, the particular challenges associated with characterising unsaturated flow and highly transient streamflow complicate the efforts to model non-perennial systems (Beven, 2002; Ye et al., 1997).

Among semi-arid regions, Mediterranean climate regions are relatively well represented in the literature because they are widespread globally, have experienced severe anthropogenic alteration, and are experiencing increasing anthropogenic water demands (Merheb et al., 2016). Intermittent rivers (i.e. those that flow seasonally) and ephemeral rivers (i.e. those that flow only after rain events) are already prevalent in these regions. Understanding streamflow generation mechanisms in these rivers is particularly needed because Mediterranean climate regions are sensitive to climate change (Cudennec et al., 2007) and are expected to experience significant drying due to shifting climate patterns, which will greatly impact streamflow regimes (Milly et al., 2005). Numerical models are used to explore the complex drivers that lead to streamflow production in a catchment. Several modelling studies of non-perennial river catchments have provided insight on the role of vegetation, soil cover, topography, antecedent wetness, and soil heterogeneity on runoff generation (Carr et al., 2014; Pierni et al., 2014; Ebel et al., 2016; Maxwell and Kollet, 2008; Vivoni et al., 2007) and on the evolution of saturated area patterns (Weill et al., 2013), as well as the importance of unsaturated storage dynamics as major controls on the processes of runoff generation (Vanderkwaak and Loague, 2001; Gutierrez-Jurado et al., 2019; Mirus and Loague, 2013; Table 1). Nevertheless, the required level of information to adequately parameterise boundary value problems has restricted the use of fully integrated surface–subsurface hydrologic models (ISSHMs) in non-perennial river catchments to mostly small-scale hillslope or headwater catchments (0.001–0.9 km2). Moreover, a majority of the numerical models for runoff and streamflow generation in non-perennial rivers investigated relatively short time periods ranging from 2 h to 330 d and reported the integrated system response to a set of scenarios (Carr et al., 2014; Di Giammarco et al., 1996; Heppner et al., 2007; Kollet et al., 2017; Mirus et al., 2009; Panday and Huyakorn, 2004). In contrast, there is still a knowledge gap regarding longer-term hydrological controls on the dry–wet transition.

The overarching, spatiotemporal processes that control key catchment-level dynamics in non-perennial rivers remains a knowledge gap. For instance, in small-scale systems, the hydrological processes occurring at a given time and place (i.e. active processes; Ambroise, 2004) might be the same as those contributing to flow generation at that same time. However, in large-scale systems the hydrologic response is influenced by different surface water–groundwater travel times, initial losses (e.g. evapotranspiration or infiltration), and the connectivity of the areas where hydrological processes are occurring. Consequently, the active processes occurring at a time and place do not necessarily contribute to the integrated catchment response at another given point at that or a later time. In contrast, a defining characteristic of the dry–wet transition in non-perennial rivers is that the dry initial conditions exacerbate initial runoff and streamflow losses due to high rates of streambed infiltration, causing the development of saturated areas and the generation of runoff and streamflow to occur discontinuously throughout the catchment.

Gutierrez-Jurado et al. (2019) used an ISSHM in an idealised concept development study to explore the processes leading to the transition from dry streambed to flowing stream. This theoretical study concluded that soil hydraulic properties and unsaturated storage dynamics exhibit strong control over streamflow generation and determine the spatiotemporal development of runoff generating areas and dominant flow generation mechanisms. It also highlighted the importance of understanding the development and progression of active areas (i.e. where processes are active) and their dominant flow generation mechanisms to understand the pathways and threshold of streamflow generation. But how applicable are the findings of this small-scale, idealised, and simplified model to real, complex, and larger-scale catchments?

We hypothesise that topography, groundwater level, and soil properties are also the dominant controls over streamflow generation mechanisms in mid-sized, Mediterranean climate coastal catchments such as those typical of South Australia. We first present a conceptual model of the potential flow mechanisms for a representative catchment, based on the theoretical findings of Gutierrez-Jurado et al. (2019), information of shallow soil profiles, and hydrological, meteorological, and geologic data. This conceptual model is then tested using an ISSHM, which is needed to fully capture the physical properties of interest. Given the inherent difficulties of modelling a large, unsaturated domain with contrasting soil layers during state changes (dry to wet), the goal of this study was not to reproduce the field observations, which in non-perennial rivers are strongly a function of antecedent moisture conditions, but to more broadly understand the interplay between shallow soil properties and groundwater levels found throughout the catchment. Finally, through simulations across the range of likely soil properties, and with varying degrees of model discretisation, we explore whether the general lack of detailed soil data at the catchment scale and computational difficulties in capturing a localised variation in stream geometry impact the characterisation of streamflow generation mechanisms.

The computational demands of modelling a large and variably saturated domain subject to sudden state changes from dry to wet conditions led to extremely slow model convergence. From the set of scenarios testing different values for incising the stream (Table 4), only scenario 8 (incision = 10 m) finished within 2 months of run-time. Scenarios 5–7 showed less than 10 % of progress after 20 d of computation time. Nevertheless, the comparable results between scenario 8 and 4, which shared the same soil hydraulic properties but had the two ends of the spectrum with respect to the river incision (the most vs. none) suggest that results from scenarios 5–7 would have likely shown similar results. Therefore, from here on we will only focus on the results from scenarios 1–4 (different soil hydraulic properties) and 8 (river incised).

The goals of this study were to provide insight into streamflow generation processes at the catchment scale for an intermittent Mediterranean climate catchment to better understand the importance of controlling characteristics identified during previous modelling efforts in non-perennial stream catchments (Table 1). This includes soil heterogeneity, as studies in both perennial and non-perennial catchments have shown that vertical soil heterogeneity can result in the development of perched saturated zones that contribute to flow generation (Hathaway et al., 2002; Maxwell and Kollet, 2008). Other studies indicated that horizontal heterogeneity contributed to the spatiotemporal variability in flow generation under different mechanisms which resulted in longer flow durations overall due to delays in runoff occurring from areas with high infiltration capacities (Ebel et al., 2016; Luce and Cundy, 1994; Smith and Hebbert, 1979).

In this study, we developed a conceptual framework of the hydrological processes identified in three distinct subregions of the Pedler Creek catchment. The steep hills are characterised by permeable shallow loam soils, steep slopes, and deep GW heads, which would result in sat-IF and SE-OF. The undulating hills are characterised by high permeable deep sandy soils, mild slopes, and deep GW heads, which would result in unsat-IF. The flat valley consisted of a mix of the previous soils, with the addition of a clay area, all located in low-gradient terrain and presenting the areas of shallow GW heads which would result in a mix of flow generation processes. We hypothesised that these distinct topographical conditions and soil types have a definite influence on streamflow generation mechanisms. However, although each catchment has its own set of conditions, most Mediterranean climate catchments would have similar topography to what was modelled here, with hills graduating to a coastal plain. Moreover, because the modelled catchment included a range of soil types, we were able to explore the variation in streamflow generation processes across several soil types. Moreover, it is likely that many other seasonally flowing catchments would have similar variation in soil, as the periodic and often flashy nature of streamflows carries fine material from the steeper headwaters and deposits it on the plains (Jaeger et al., 2017).

Model results overall supported our conceptual understanding that distinctive topographic and soil characteristics explain flow generating processes in Pedler Creek. Results from the active areas showed distinct mechanisms developing in the three major areas (Fig. 5), supporting the idea that there is a spatial and temporal variation in flow generation processes in Pedler Creek. In the model, flow developed first in the steep hills areas (fast flow), and the dominant mechanism was SE-OF, with a few areas showing unsat-IF as hypothesised in our conceptual model (Figs. 3b and 5). An unexpected development was the contribution of pre-event GW during flow recessions. In this area, the pre-event GW was likely to be pre-event soil water since an evaluation of the model showed that the groundwater level did not rise to intersect land surface in this area. The flows generated in the valley near the outlet were similarly simulated via the conceptualised mechanisms (Figs. 3d and 5). We saw small areas with flow originating from IE-OF from the clay areas and a combination of unsat-IF and pre-event GW for the rest of the active areas in this region. The GW did rise above the surface elevation in this area, supporting the GW contribution to flow in this area. Finally, in the few small areas close to the sandy undulating hills region, flow was simulated through the unsat-IF mechanism, as predicted in our conceptual framework (Figs. 3c and 5). These results support the findings by Gutierrez-Jurado et al. (2019), who suggested that soil properties largely dictate the dominant flow generation mechanisms and that unsaturated storage dynamics control the thresholds and pathways of flow. For real catchments such as Pedler Creek, soil properties and topography evolve in tandem, and it is impossible to fully disentangle their relative influence on streamflow generation. Such undulating and variable topography is not captured by theoretical models. The development and extent of active areas and the dominant flow generation mechanisms estimated by the HMC method and the water balance results were almost identical for scenarios 1–4, indicating that knowledge of the exact hydraulic conductivity value of a given soil type is less importance than capturing the general vertical and longitudinal soil heterogeneity across the catchment. The small differences simulated between scenarios 2 and 3 and 1 and 4 show that the models were more sensitive to variations in the hydraulic properties of the loam soil, as the scenarios with identical responses (2 and 3 and 1 and 4) shared the same loam but different sand hydraulic properties. The loam's smaller hydraulic conductivity for scenarios 2 and 3 (0.0624 m3 d-1) limited infiltration, which translated to more OLF. At the same time, the higher water holding capacity in the loam areas might have resulted in slower subsurface flows to either exfiltrate to the surface or to contribute to the fluid transfer (FT) through the subsurface boundary. In contrast, both tested values for the sand resulted in a low water holding capacity that allowed the incoming precipitation to drain past the root zone and move in the subsurface contributing to the FT. This is supported by the higher FT values shown for scenarios 1 and 4 (Fig. 6).

The greatest differences for both the water balance and HMC results among the tested scenarios were caused by differences in stream definition between scenarios 1–4 (no incised stream) and scenario 8 (incised stream). The effects of the incised stream in scenario 8 resulted in a larger OLF flux, a smaller ET flux, and in an overall smaller extent of active areas (Fig. 5 and Supplement). The roles of a good channel representation in ISSHMs is extensively discussed by Käser et al. (2014). While their discussion of channel representation revolves on the ability of ISSHMs to quantify GW–stream interactions (which was not a major component for this model), the hydrological principles are relevant and transferable to explain the importance of channel representation to capture streamflow generation processes. Without a defined channel, the connection to the Pedler Creek floodplain was lost, and the shallow sheet flow and most of the precipitation infiltrated and stayed within the porous media. The importance of the connection of the floodplain to the channel is also discussed by Käser et al. (2014), as they argued that not only is the channel topography important but also its connection to the floodplain, given that riverbank geometry is key for bank storage and overbank flooding. Although overbank flooding is not considered important for this study (flows in Pedler only rarely will experience overbank flooding), the stream–floodplain connectivity and bank storage were key aspects under our model conceptualisation. That is, the predicted dominant mechanisms relied upon the saturation to build up along the riverine zone in the loamy areas, which would lead to saturation excess overland flow. We expected a perched groundwater to develop on the sandy hillslopes which would, after intersecting the stream, contribute with interflow into the stream. While we observed as these processes developed, they only occurred briefly as very shallow runoff.

Another important consideration for the channel representation in streamflow generation studies for non-perennial streams is the relationship between flow and the wetted area. The larger the channel (both vertically and horizontally), the larger the area of exchange to the unsaturated zones during a flow event (Doble et al., 2012), which would be exacerbated under low flows (Käser et al., 2014). This is particularly significant when evaluating streamflow generation for non-perennial streams where high streambed infiltration and transmission losses are common (Gutierrez-Jurado et al., 2019; Levick et al., 2008; Shanafield and Cook, 2014; Snelder et al., 2013) and often prevent flows from even reaching the catchment outlet (Keppel and Renard, 1962; Aldridge, 1970). In this study, we observed that, without a defined stream to channel the water, the little overland flow that was simulated in scenarios 1–4 (no incised stream) spread over a larger area than in scenario 8 where the stream was incised (Fig. 5). The same was true for the patterns of increased saturation of the porous media across the catchment (Fig. 6). Results from the water balance reflected the effects of having both flows and porous media saturation spread over larger areas by exacerbating ET and decreasing the overall amount of overland flow for scenarios 1–4 (see the Supplement). This is consistent with the remarks by Käser et al. (2014) regarding the likely impacts to the water flow budget by the spatiotemporal aspects linked to channel representation due to spatial exchange patterns.

Finally, this study highlighted both the need for further studies examining streamflow generation processes in additional non-perennial catchments. For instance, our results underlined the importance of channel representation, and future studies should investigate the effects of channel morphology in streamflow generation in non-perennial catchments. Moreover, while for Pedler Creek the GW–stream interactions were conceptualised to occur only near the catchment outlet (and likely only during certain wet years), flow intermittence in many rivers can be attributed to a water table fluctuation relative to the stream channel elevation (Snelder et al., 2013). Future work on flow intermittence as a result of GW–stream interactions would be valuable. Similarly, the inherent challenges associated with capturing unsaturated zone dynamics at the catchment scale were underpinned in this study. Indeed, modelling this non-perennial river system confirmed the inherent difficulties of using ISSHMs in medium-sized non-perennial river catchments and reiterated why so few studies have been done on this topic. Extensive work was needed to set up this model, and the large computational time to run the simulations was a major constraint to both establishing initial conditions and exploring scenarios. For example, when draining the model, simulations running for over 10 d only progressed to day 100 of the simulation. Relaxing the mesh allowed us to develop reasonable initial conditions after testing over 37 scenarios. For the scenarios running for the full 4-year simulation, simulation convergence consistently slowed down when the simulation encountered a precipitation input, particularly during prolonged precipitation events (i.e. consecutive precipitation inputs), which is expected given the highly nonlinear equations for unsaturated flow in the unsaturated surface domain. Despite the conceptual advantages of using a fully physically based model to explicitly capture all surface and groundwater processes, future studies may try to identify a suitable simplified surrogate model to speed up simulations and focus on specific areas where particular streamflow generation processes are thought to be dominant.

There are hundreds of similar non-perennial river systems in the semi-arid coastal Mediterranean climate regions of South Australia, Western Australia, California, South Africa, and around the Mediterranean itself (Davies et al., 1993; Tzoraki and Nikolaidis 2007; Skoulikidis et al., 2017). This study provides an initial step towards understanding non-perennial streamflow generation processes at the catchment scale and provides a template for using ISSHMs for process understanding in these stream systems. The development of a conceptual model of the most important factors impacting flow generation processes within Pedler catchment presented a hypothesis that combined our understanding of field data with lessons learnt from previous studies. This conceptualisation informed the model setup and captured the dynamics of streamflow generation in this non-perennial stream system. Difficulties in setting up and running the model reaffirmed the numerical difficulties experienced in large-scale unsaturated models, such as accurately reproducing the topography and observed initial conditions, was a challenge. Model runtimes prohibited extensive exploration of multiple scenarios. In particular, the importance of preserving channel representation to model streamflow generation on non-perennial systems became apparent in the scenarios. Yet, overall, the model results confirmed our conceptual understanding that soil type, unsaturated storage dynamics, and topography are major controls for streamflow generation processes in non-perennial streams. The similarity in the results from scenarios comparing soil hydraulic properties across the literature range for each soil type showed that exact knowledge of these values for a given soil type is not critical for identifying streamflow generation processes if the conceptual model is accurate and the vertical and longitudinal soil heterogeneity is captured. Given that soil properties are often highly heterogeneous within a catchment and rarely well known, this result is will be important for future modelling studies.