Review of hydrological modelling in the Australian Alps: from rainfall-runoff to physically based models

ABSTRACT

In the context of global research in snow-affected regions, research in the Australian Alps has been steadily catching up to the more established research environments in other countries. One area that holds immense potential for growth is hydrological modelling. Future hydrological modelling could be used to support a range of management and planning issues, such as to better characterise the contribution of the Australian Alps to flows in the agriculturally important Murray-Darling Basin despite its seemingly small footprint. The lack of recent hydrological modelling work in the Australian Alps has catalysed this review, with the aim to summarise the current state and to provide future directions for hydrological modelling, based on advances in knowledge of the Australian Alps from adjacent disciplines and global developments in the field of hydrologic modelling. Future directions proffered here include moving beyond the previously applied conceptual models to more physically based models, supported by an increase in data collection in the region, and modelling efforts that consider non-stationarity of hydrological response, especially that resulting from climate change.

KEYWORDS:

• Review
• Australian Alps
• hydrological modelling
• snowmelt
• rainfall-runoff modelling

1. Introduction

Snow-affected mountainous areas globally are colloquially termed ‘water towers’, owing to their role in the storage of water in the form of snow during winter before its release as snowmelt come spring (Simpkins 2018; Viviroli and Weingartner 2004). On continental Australia, an area of approximately 5,000 km² lies above the snowline (Bilish, McGowan, and Callow 2018; Pepler, Trewin, and Ganter 2015), known as the Australian Alps, with 1,675 km² experiencing snow cover lasting 2 months or more (Sanecki et al. 2006). Only about 5% of the area is above the treeline and qualified as ‘alpine’, with the rest being ‘subalpine’ for being below the treeline (Bilish, McGowan, and Callow 2018; Costin et al. 2000). This unique environment is a water resource to the agriculturally important Murray-Darling Basin (Brown and Millner 1989; S. Y. U. Schreider, Whetton, et al. 1997; Worboys, Good, and Spate 2011) as well as hydro-power generation (Domicelj 1980; Hardman 1968), the mecca of snow sports and snow tourism in Australia (Bicknell and Mcmanus 2006; K. J. Hennessy et al. 2008; König 1999; Morrison and Pickering 2013; Pickering and Buckley 2010), and home to many distinctive flora and fauna (Costin et al. 2000; Happold 1998). It also holds great Indigenous cultural significance (Lennon 2003). For example, Indigenous groups from areas adjacent to the high country would move into the mountains in pursuit of Bogong moth (Agrotis infusa) during the moths migration, a tradition that has recently been revived (Searby 2008; Stephenson et al. 2020). Despite its wide-reaching importance, research in hydrological modelling of the Australian Alps lags that of countries and continents where snow is more prominent.

Runoff from the majority of the Australian Alps region flows into the Murray-Darling Basin (MDB) (Figure 1), providing water resources to Australia’s food bowl. It has been estimated that between 20% and 29% of water in the MDB is sourced from the Australian Alps (Brown and Millner 1989; Reinfelds et al. 2014; Worboys, Good, and Spate 2011), which makes up less than 1% of the total area of the MDB (1,059,000 km²). For example, the Kiewa Basin (Figure 1) in the Victorian portion of the Australian Alps contributes the highest runoff per unit area of all the Murray tributaries (mean annual discharge of 705 GL – equal to 223 mms over a year – from 1,985 km²), driven primarily by snowmelt from the basin’s headwaters between August and October (S. Y. U. Schreider, Whetton, et al. 1997). The percentage contribution increases in times of drought across the MDB (Brown and Millner 1989; S. Fiddes and Timbal 2016; Reinfelds et al. 2014). This is a result of these headwater catchments receiving larger volumes of precipitation (Grose et al. 2019; Landvogt, Bye, and Lane 2008; Sarmadi et al. 2019), that is more reliable than precipitation received in the adjacent lowlands (Dey et al. 2019; Murphy and Timbal 2008). Additionally, these catchments are commonly energy-limited on an annual scale, rather than precipitation-limited like areas downstream (Reinfelds et al. 2014), meaning that annual precipitation totals are larger than evaporative demand, resulting in a system where groundwater recharge and runoff is common. There is a seasonal dynamic where these energy-limited catchments can become water-limited (Bilish, Callow, and McGowan 2020).

Figure 1. The Murray-Darling Basin with the Australian Alps region highlighted. Major watercourses as defined by the Geoscience Australia are shown. The Murray and Murrumbidgee Rivers are highlighted, as these are the two major snow affected rivers in the basin.

Despite the importance of runoff from the Australian Alps to the MDB, there has been limited research investigating the contribution of snowmelt to water resources in the MDB. This science gap has been recognised for decades (see e.g. Brown and Millner 1989). Water sourced from snowmelt in the Australian Alps is used primarily for irrigation and hydropower generation as well as contributing to environmental flows in the Basin (Bilish, McGowan, and Callow 2018). The Snowy Hydro Scheme, consisting of nine power stations and 16 major dams, estimates on average 50% of inflows from snowmelt and rain during spring (Snowy Hydro Limited 2020).

In this review paper, we aim to summarise the current state of hydrological modelling that explicitly considers snowmelt in the Australian Alps and suggest future directions to guide the field. We juxtapose the state of hydrological modelling in the Australian Alps against other research disciplines (e.g. ecology, climatology) which conduct work in the region, and snow hydrology efforts in other parts of the world. To the best of the authors knowledge, this is the first such review.

2. Snow hydrology and data

Snow hydrology is the study of the water cycle as pertaining to snow, including its formation, accumulation, melting, and impact on water resources and the environment. With an improved understanding of snow hydrology, we can improve our streamflow forecast skills to optimise water reservoir and hydroelectric operations in light of other demands for water downstream (Asadzadeh et al. 2014; Zhao, Cai, and Li 2019). As precipitation that falls as snow is stored temporally in a snowpack, the processes that drive the hydrological cycle in snow-affected area differ in terms magnitude and timing compared to rain-dominated systems (Dozier 1987; Male and Granger 1981).

Energy fluxes are important drivers of snowmelt, particularly radiation (short- and long-wave) and turbulent (sensible and latent) energy fluxes (Male and Granger 1981). Latent heat flux involves the exchange of energy between the surface, here the snowpack, and the atmosphere where a phase change is involved (e.g. sublimation, condensation) (Ledley 2003), while sensible heat flux is the transfer of heat resulting from a difference in temperature between the surface and the atmosphere without a phase change (Schmid and Read 2022).

Data available to inform snow hydrology research are varied in type, being broadly classed as in-situ observations, modelling outputs and remotely sensed data (Dong 2018). In-situ data are collected in the environment being studied, as opposed to being collected remotely or through model simulations. In-situ data collection has a long history in snow hydrology, including measurements of snowfall amounts starting as early as the 1500s (Dong 2018). Although the collection of remotely sensed snow data does not have the long-term history of in-situ measurements, the breadth of data collection is possible at a much wider scale.

Measuring snowfall is notoriously difficult, with undercatch being a significant source of error (Groisman and Legates 1994; Rasmussen et al. 2012). Measurement error can account for upwards of 50% of solid precipitation, with windy conditions being a large contributor to undercatch (Groisman and Legates 1994; Rasmussen et al. 2012). Gauge types not suited for capturing solid precipitation also contribute to the issue (Rasmussen et al. 2012). Chubb et al. (2015) found undercatch in the Snowy Mountains in the New South Wales portion of the Australian Alps could be as great as 52% of precipitation when a precipitation gauge was not adequately shielded. Average seasonal undercatch in that region varied between 6% and 15% of precipitation (T. Chubb et al. 2015). As precipitation is the prime input variable to hydrological models, it follows that additional error would be introduced if poor quality precipitation data are used to force the model. Input errors can be reduced by using gauges that are double fenced as this configuration has been shown to minimise undercatch (Rasmussen et al. 2012), or by applying a transfer function to precipitation measurements (Kochendorfer et al. 2022; D. Yang et al. 2005). Gridded rainfall products are available for Australia (Jones, Wang, and Fawcett 2009), but large precipitation underestimates have been emphasised from such products over the Australian Alps (T. H. Chubb et al. 2016; Pepler, Trewin, and Ganter 2015). In past modelling efforts of snow-affected catchments in the Australian Alps, precipitation data errors were not explicitly addressed (G. Burns, Fowler, and Horne 2023; Duggan and Garr 2006; S. Y. Schreider and Jakeman 2000, 2001; S. Y. U. Schreider, Young, and Jakeman 1997a, 1997, 1997c, 2001; Tan et al. 2005; Trim et al. 2023), something also deficient in hydrological modelling in general.

In-situ measurements of snow on the ground can be undertaken manually or by automated instruments. Snow courses are sets of points at which manual measurements of snow depth and snow water equivalent (SWE) are made periodically and were first introduced in the United States in the early twentieth century. With the construction of the Snowy Mountains Scheme beginning in 1949, a similar measurement approach was adopted in New South Wales. More than 50 snow courses were operated during the initial investigations phase though these were consolidated over the following decades. In more recent years, many Australian studies have utilised data from three index snow courses (Spencers Creek, Deep Creek and Three Mile Dam) that are surveyed approximately weekly throughout the snow season and have become the primary long-term record of the Australian snowpack (e.g. Davis 2013; Duus 1992; Nicholls 2005; Pepler, Trewin, and Ganter 2015). Additionally, a limited number of the original snow courses have continued to be surveyed less frequently (K. J. Bormann, Evans, and McCabe 2014). In the present context, it is important to note that snow courses are typically chosen to inform forecasts of seasonal inflows and are not specifically ‘designed as measures of the absolute amount of SWE in a basin, or to yield SWE estimates across a basin’ (Dressler, Fassnacht, and Bales 2006, 706). This means that assumptions of representativeness must be carefully considered when using these data for hydrological modelling (Bilish et al. 2019).

The invention of computing dramatically enhanced the capacity for modelling of hydrological systems, allowing collected data to be applied to a larger area (Silberstein 2006). However, models are simplified representations of real-world systems (Sorooshian et al. 2008) and should be acknowledged as such. Model development should be driven by the research goals, which may include prediction, scenario analysis and diagnostic learning (Razavi et al. 2022). In this way, models have been classified based on resultant model properties, namely data-driven compared to process-based (Figure 2) (Beven 2005; Devi, Ganasri, and Dwarakish 2015; Silberstein 2006; Wheater, Jakeman, and Beven 1993).

Figure 2. Classification scheme of hydrological models, based on model structure and spatial representation of the area (i.e. catchment) to be modelled, and snowmelt models that are included in hydrological models.

Data-driven models, which relate input and output data mathematically without consideration of physical processes (Corzo et al. 2009; Elshorbagy et al. 2010; Wheater, Jakeman, and Beven 1993), are juxtaposed against process-based models (Beven 2005; Wheater, Jakeman, and Beven 1993). Process-based models can be further divided into conceptual, where effective parameters describe a conglomerate of processes, or physically-based, where parameters have physical meaning and can be measured in theory (Figure 2) (Devi, Ganasri, and Dwarakish 2015; Silberstein 2006). On a spatial scale, models can be further divided based on how the catchment to be modelled is represented. Lumped models treat the catchment area as a single homogenous unit, where a single runoff output at the catchment outlet is calculated (Figure 2) (Beven 2005; Devi, Ganasri, and Dwarakish 2015; Silberstein 2006). Semi-distributed and distributed models divide a catchment into smaller sub-catchments or via a grid to be summed later (Figure 2) (Beven 2005; Devi, Ganasri, and Dwarakish 2015; Silberstein 2006).

When a catchment is snow-affected, a hydrological model is modified to account for the storing of precipitation as a snowpack. Subsequent snowmelt can be modelled in two main ways – using a melt rate within a temperature-index model or by evaluating energy fluxes (Figure 2) (K. J. Bormann, Evans, and McCabe 2014; S. Y. U. Schreider, Whetton, et al. 1997). The melt rate is a calibratable (i.e. adjustable) parameter that relates snowmelt to air temperature, whereas physical models weigh up the energy inputs and loss to the snowpack with residual energy assumed to be used for melting (Beven 2005; K. J. Bormann, Evans, and McCabe 2014; S. Y. U. Schreider, Whetton, et al. 1997). These classifications are not prescriptive, with there being many models that straddle categories.

Remotely sensed data is collected at a distance, often using the sensors that collect information in the electromagnetic spectrum (Fussell, Rundquist, and Harrington 1986). Remotely sensed data has been highly useful in hydrological modelling (Ali et al. 2023; Kite and Pietroniro 1996; X. Xu, Li, and Tolson 2014) and snow studies (Andreadis and Lettenmaier 2006; Dong 2018). Remote-sensing imagery can provide information on several snow properties including snow covered area, snow depth and snow water equivalent (Andreadis and Lettenmaier 2006; Dong 2018; Kite and Pietroniro 1996).

Nevertheless, remotely sensed data collection can struggle in complex mountainous terrain, where clouds can cause confusion for optical sensors (Dong 2018). An example can be seen in Bormann et al. (2012), where there were only 15 days over an 11-year period where the Landsat satellite passing over the Australian Alps coincided with clear sky conditions during the winter months. Microwave remote-sensing and LiDAR offers support to overcome this, but is not without drawbacks (Dong 2018).

3. Hydrology of the Australian Alps

The hydrology of the Australian Alps received significant attention in 1950s and 60s primarily by prominent Australian ecologist, Alec Baillie Costin, and his colleagues (Costin and Wimbush 1961; Costin, Wimbush, and Cromer 1959, 1960, 1961, 1964). Their studies linked soils, vegetation and water over multiple landscape types in the Australian Alps, at both alpine and subalpine, and lower montane, elevations. Their research methods included field studies at many sites in different environments to represent the complexity observed in the region. Many future studies expanded the work of Costin and colleagues (Brookhouse, Farquhar, and Roderick 2013; Brown 1972; A. J. Schwartz 2021; A. J. Schwartz, McGowan, and Callow 2020).

Australia’s snowpack is prone to high variability, both between seasons and within a single season (Bilish et al. 2018, 2019; T. H. Chubb, Siems, and Manton 2011; Davis 2013; K. Hennessy et al. 2003; Nicholls 2005; Pepler, Trewin, and Ganter 2015). The winter snowpack that blankets the Australian Alps differs from snowpacks elsewhere, straddling the maritime and ephemeral classes of the popular Sturm et al. (1995) snow cover classification system. The maritime class is characterised by warm, deep snow cover, whereas the snow cover of the ephemeral class is thin and short-lasting, frequently melting completely in the time between snowfall events (Sturm, Holmgren, and Liston 1995). Higher-elevation snow cover in Australia most resembles the maritime class, but experiences higher winter air temperatures than those defined by Sturm et al. (1995) for the maritime class (Sanecki et al. 2006). Snow at lower elevations in the Australian Alps is more aligned with the ephemeral class (Sanecki et al. 2006).

The distinctive snow cover experienced by the Australian Alps and the spatial variability of the snowpack and its characteristics (Bilish et al. 2019; K. J. Bormann et al. 2013) make the Australian Alps an interesting study region. While only parts of the snowpack can be classed as ephemeral, the instability of phase (accumulation vs. ablation) throughout the Australian snow season means that melt is commonly generated mid-season and the conventional representation of a snow season as two distinct phases does not fit; the term ‘marginal snow environment’ has been proposed to describe this behaviour (Bilish et al. 2019). As the dynamics of snowpacks change globally with the warming climate (López-Moreno et al. 2020; Stewart 2009), the characteristic qualities of the Australian snowpack may become more widespread across the world in the next few decades, and therefore, lessons learned through more research in the Australian Alps could be leveraged to uncover possible future dynamics of snowpacks elsewhere.

Snowpack dynamics in the Australian Alps have continued to be studied further (e.g. Bormann et al. (2013), Bormann et al. (2014), Bilish et al. (2018), Bilish et al. (2019)), highlighting the variability in snowpack characteristics (i.e. SWE, snow depth, snow density) over small spatial scales (Bilish et al. 2019) and between years (K. J. Bormann et al. 2013). For example, Bormann et al. (2012) characterised snow cover in Australia spatially and temporally using satellite data, Bormann et al. (2013) improved understanding of snow density in Australia, and Bormann et al. (2014) improved the applicability of Temperature-Index Models (TIMS) in an Australian setting. Bilish et al. (2018) first characterised the energy fluxes controlling snowmelt in an Australian setting and Bilish et al. (2020) investigated the seasonality, interannual variability, and contribution of the snowpack to streamflow in a small subalpine catchment. However, improved runoff modelling of the Australian Alps that incorporates these advances is yet to be implemented.

Snow accumulation dynamics in an Australian setting are influenced by vegetation, with deep snow found in small clearings as a result of snow drifting (Costin et al. 1961). Less snow accumulation is observed under canopies, with shallowest snow depths observed in open areas (Costin et al. 1961; A. J. Schwartz 2021; A. J. Schwartz, McGowan, and Callow 2020). Canopies also play an important role during periods of snowmelt (Costin et al. 1961), providing for a reduction in shortwave radiation to the snowpack, an increase in longwave radiation, and a varying impact on the turbulent fluxes (Bilish et al. 2019; A. J. Schwartz, McGowan, and Callow 2020). Fire disturbances impact snow accumulation dynamics, enhancing ablation (i.e. loss) rates and reducing snowpack longevity post winter (A. J. Schwartz 2021; A. J. Schwartz, McGowan, and Callow 2020).

Advances have been made in modelling snow-affected areas, but these advances tend to be geographically limited, focused predominantly on the Northern Hemisphere (de Jong, Lawler, and Essery 2009; Knoben et al. 2020; Parajka, Merz, and Blöschl 2007). Alpine research in Australia has focused on mountain climatology (T. Chubb et al. 2011, 2015, 2016; Davis 2013; Duus 1992; Pepler, Trewin, and Ganter 2015; Ruddell et al. 1990; Slatyer, Cochrance, and Galloway 1984) and ecology (e.g. response of alpine vegetation to disturbance, impact of introduced species) (Green and Osborne 1981; Hughes 2003; Sanecki et al. 2006; Scherrer and Pickering 2005). Precipitation-focused research across the Australian Alps has featured predominantly, including the analysis of precipitation historical frequencies and trends (T. H. Chubb, Siems, and Manton 2011), evaluation of precipitation gauge undercatch (T. Chubb et al. 2015), comparison of precipitation datasets (T. H. Chubb et al. 2016), and the development of long-term precipitation and related datasets (Davis 2013).

Investigations into snow season length and snow cover have occurred, with Slatyer et al. (1984) modelling snow season length, Ruddell et al. (1990) determining relationships between meteorological drivers and snow cover, Duus (1992) modelling snow cover, and Pepler et al. (2015) assessing the impact of climate drivers on snow cover. Recently the soils of the Australian Alps have garnered research interest (Wilson et al. 2022). High elevation soils in the Australian Alps are of ecological and hydrological importance, however, the understanding of how these soils form and function is limited (Wilson et al. 2022). Threats to alpine soils in Australia are varied (e.g. climate change, fire, feral animals) even though the long-term impacts of these threats have been identified as research priorities (Wilson et al. 2022).

Climate change has understandably also been an important focus area for research in the Australian Alps (K. Hennessy et al. 2003; McGowan et al. 2018; Nicholls 2005; Ruddell et al. 1990; Slatyer, Cochrance, and Galloway 1984), with climate change projected to significantly impact areas with average winter air temperatures close to 0°C, as occurs in the Australian Alpine region (Adam, Hamlet, and Lettenmaier 2009; Di Luca, Evans, and Ji 2018; McGowan et al. 2021; Nolin and Daly 2006). Within alpine Australia, climate change has manifested in decadal decreasing trends in maximum snow depth, snow season duration, and snow-covered area (K. J. Bormann, McCabe, and Evans 2012; Davis 2013; S. L. Fiddes, Pezza, and Barras 2015; Green and Pickering 2009; K. Hennessy et al. 2003; Sánchez-Bayo and Green 2013; Whetton, Haylock, and Galloway 1996), with spring snow depth particularly affected (Nicholls 2005; Pepler, Trewin, and Ganter 2015). Warming, rather than reductions in precipitation, are the suggested cause of declining snow depths come spring (Nicholls 2005; Pepler, Trewin, and Ganter 2015). However, a reduction in autumn and spring precipitation over the period 1958–2012 has also been reported (Theobald, McGowan, and Speirs 2016).

Large amounts of long-term hydrologic data have been collected and continue to be collected in the Australian Alps at select sites, mostly to inform hydropower generation (Brown and Millner 1989). Research in the Australian Alps has largely relied on a few long-term measurement sites (e.g. Davis 2013; Duus 1992; Nicholls 2005) or small headwater catchments (e.g. Bilish, Callow, and McGowan 2018, 2019, 2020) with findings extrapolated to larger areas (K. Hennessy et al. 2003). The scarcity of ground-based data has been overcome to some extent with the use of remotely sensed data (K. J. Bormann, Evans, and McCabe 2014; Rasouli et al. 2022). But remotely sensed data are yet to be incorporated into hydrological modelling in the Australian Alps as has happened elsewhere (Andreadis and Lettenmaier 2006; Dong 2018; Kite and Pietroniro 1996; Molotch and Margulis 2008). Additionally, no study has yet estimated the runoff from the entirety of Australian Alps (Brown and Millner 1989; S. Y. Schreider 1996), except in general terms (Worboys, Good, and Spate 2011), or tried to quantify the contribution of snowmelt to runoff in the snow-affected catchments, with the exception of estimation by Bormann (2013, 224–225). However, it has been suggested that the existing marginality of the snowpack in the Australian Alps reduces the direct control it has on runoff at present, and limits the impact on runoff of future shifts towards proportionally less snowfall, when compared to regions with more persistent snow cover (Bilish, Callow, and McGowan 2020).

4. Current state of hydrological modelling in the Australian Alps

Only a handful of local studies considered snow when modelling runoff from snow-affected catchments on mainland Australia (G. Burns, Fowler, and Horne 2023; DeWalle and Rango 2008, 308; Duggan and Garr 2006; S. Y. Schreider and Jakeman 2000; 2001; S. Y. Schreider et al. 1997a, 1997, 1997c, 1996; Tan et al. 2005; Trim et al. 2023)¹ (Supplementary Material). These studies use one of the six different conceptual rainfall-runoff models (Table 1) – IHACRES (Identification of unit Hydrographs And Component flows from Rainfall, Evaporation and Streamflow; Jakeman, Littlewood, and Whitehead 1990), AWBM (Australian Water Balance Model; Boughton 1993), SIMHYD (Simple Hydrology; Chiew, Peel, and Western 2002; Li et al. 2010), SRM (Snowmelt-Runoff Model; Martinec 1975), Wapaba (water partition and balance; Q. J. Wang et al. 2011) and GR4J (Perrin, Michel, and Andréassian 2003).

Table 1. Past hydrological modelling of snow-affected catchments on mainland Australia (excluding Tasmania).

Source	Model	Model timestep	State	Snow-affected catchments	Modelling summary
Schreider et al. (1997a)	IHACRES with addition of snow-melt-accumulation module based on modified degree-day approach	Daily	NSW/Vic	Mitta-Mitta, Big River, Tooma River, Snowy Creek, Gibbo River, Upper Murray at Biggara	Degree-day approach is easy to apply, requiring only (daily) temperature and precipitation data. Performance of model with added snow module in snow-affected catchments similar to model without snow module in snow free catchments.
Schreider et al. (1997)	IHACRES with addition of snow-melt-accumulation module based on modified degree-day approach	Daily	Vic	Mitta-Mitta, Kiewa River	No comparison between models.
Schreider et al. (1997c)	IHACRES with addition of snow-melt-accumulation module based on modified degree-day approach	Daily	Vic	Mitta-Mitta, Kiewa River	During both calibration and simulation, the model performance is similar in snow-affected catchments as in snow-free catchments.
Schreider and Jakeman (2000)	IHACRES with addition of snow-melt-accumulation module based on modified degree-day approach	Daily	NSW	Snowy River above Island Bend Pondage	No comparison between models or catchments. Calibration model efficiency (R^2) was 0.78. Validation model efficiency (R^2) was 0.61.
Schreider and Jakeman (2001)	IHACRES with addition of snow-melt-accumulation module based on modified degree-day approach	Four-hourly	NSW/Vic	Mitta-Mitta, Big River, Tooma River, Snowy Creek, Upper Murray at Biggara	Model accuracy constant when moving from daily to sub-daily timestep. However, the model simulated diurnal fluctuations in flow, caused by temperature cycles, that were not observed in measured flow data.
Schreider et al. (2001)	IHACRES with addition of snow-melt-accumulation module based on modified degree-day approach	Four-hourly	Vic	Snowy Creek	Model efficiency slightly higher when the snow melt/accumulation module applied.
Tan et al. (2005)	SIMHYD with an added degree-day snow component and updated routing scheme	Daily	Vic	Upper Mitchell	Model additions (i.e. snow module and routing scheme) improved daily flow modelling.
Duggan and Garr (2006)	AWMB with addition of snow-melt-accumulation module based on modified degree-day approach (from Schreider et al. (1997c))	Daily	NSW	Tumut River above Happy Jacks Pondage	Correlation between observed and model simulated streamflow improved when the snow-melt-accumulation module was included.
Dewalle and Rango (2008, 308)	SRM	Daily	NSW	Geehi River	Six winter seasons were modelled with an average model efficiency (R^2) of 0.7 (out of 1) and percentage difference between measured and simulated runoff of 6.6%.
Burns et al. (2023)	Wapaba model with the addition of a degree-day approach (C.-Y. Xu, Seibert, and Halldin 1996)	Monthly	NSW	Snowy Mountains	Snow module was able to replicate snow dynamics well throughout snow-affected areas of south-east Australia.
Trim et al. (2023)	Modified GR4J	Monthly	NSW	Snowy Mountains	Explicit discussion of snow modelling results not included.

NSW = New South Wales; Vic = Victoria, IHACRES = Identification of unit Hydrographs and component flows from rainfall, evaporation and streamflow data; SIMHYD = simple hydrology; AWBM = Australian Water Balance Model, SRM = Snowmelt-Runoff Model, Wapaba = water partition and balance.

IHACRES, AWBM, SIMHYD and Wapaba were modified to account for snowmelt with the addition of an empirical, degree-day approach within a temperature-index model (TIM) (Figure 2). It was noted that the GR4J model used in Trim et al. (2023) was also modified, presumably to GR4JSG (Nepal et al. 2017) where a degree-day approach is used. SRM was developed specifically to predict snowmelt runoff and also includes a degree-day approach (DeWalle and Rango 2008, chap. 11). Both of the two most recent studies (G. Burns, Fowler, and Horne 2023; Trim et al. 2023) run at a monthly timestep, differing from the daily or sub-daily timestep of earlier modelling efforts (DeWalle and Rango 2008 −308;, pp. 306; Duggan and Garr 2006; S. Y. Schreider and Jakeman 2000, 2001; S. Y. U. Schreider, Young, and Jakeman 2001, 1997a, 1997, 1997c; Tan et al. 2005).

TIMs are frequently used to model runoff from snowpacks, including ‘marginal’ snow environments characterised by winter air temperatures close to freezing, short snow seasons, and high intra- and inter-season variability (Hock 2003). Their popularity owes much to their limited data requirements of only daily temperature and precipitation (S. Y. U. Schreider, Jakeman, Dyer, et al. 1997). TIMs are most effective when modelling over longer periods (e.g. months) (DeWalle and Rango 2008). Under marginal snow conditions, the effectiveness of TIMs has been questioned (K. J. Bormann, Evans, and McCabe 2014). This is because TIMs assume the contribution from energy balance components is constant throughout the season, which is an oversimplification in the Australian Alps and elsewhere (Bilish, McGowan, and Callow 2018).

In addition to journal articles, conference papers and textbooks, two theses that modelled runoff from the Australian Alps were found (K. Bormann 2013; S. Y. Schreider 1996) (see Supplementary Material). Modelling of snow-affected catchments in Schreider (1996) was published in Schreider et al. (1997a), Schreider et al. (1997), and Schreider et al. (1997c). Bormann (2013) applied multiple-snowmelt schemes, including that of Schreider et al. (1997c), within a TIM to estimate snowmelt contributions from the snowpack to the Murray-Darling Basin. Mean annual snowmelt contributions were reported as 7,514 ML/day (K. Bormann 2013, 225). In Bormann (2013), only snowmelt was assessed independent of direct rainfall-runoff processes. The peer-reviewed published work to come out of Bormann (2013) (i.e. K. J. Bormann, Evans, and McCabe 2014, 2012, 2013) did not report snowmelt beyond the point-scale (K. J. Bormann, Evans, and McCabe 2014).

Beyond published research on hydrological modelling in the Australian Alps, Snowy Hydro Ltd. uses two models to provide hydrological guidance for the operation of the Snowy Mountains Scheme: a semi-distributed daily forecast model with a temperature-index snowmelt component and a tuned artificial neural network model that provides insight on sub-daily variability. Snowy Hydro Ltd. also produces estimates of water volumes stored in the snowpack and probabilistic monthly forecasts of winter-spring inflows using regression models based on measurements at the index snow courses.

On an Australian national scale, several models operate to simulate runoff. The Bureau of Meteorology’s and CSIRO Australian Water Resources Assessment’s AWRA-L and AWRA-R models (Dutta et al. 2017; Frost and Wright 2018) are widely used (e.g. Wallace et al. (2013), Kunnath-Poovakka et al. (2016), Khan et al. (2020)), but do not consider snow (Dutta et al. 2017; Frost and Shokri 2021). eWater’s Source model (Dutta et al. 2013; Welsh et al. 2013) is the Australian National Hydrological Modelling Platform (Carr and Podger 2012) and has been tested and used extensively (e.g. Dutta et al. (2012), Rassam et al. (2013), Yang et al. (2017), Zafari et al. (2022)). As Source is a modelling platform with the potential to embed different modelling modules within it, it has the ability to handle snow processes (Dolk, Penton, and Ahmad 2020, 202; Iqbal et al. 2018; Nepal et al. 2017; Rawat, Lahari, and Gosain 2018). Although the Source model has been applied to snow-affected catchments in Australia, the necessary modifications were not made to account for snow (Welsh et al. 2013).

Another potential modelling framework for snow hydrology in Australia is based on land-surface models (LSMs). LSMs simulate and predict exchanges of energy, water and carbon between the surface and the atmosphere (Blyth et al. 2021). In this way, LSMs are viewed as physically based (Figure 3) (Overgaard, Rosbjerg, and Butts 2006). LSMs are influenced by, and link, several scientific disciplines including atmospheric sciences and hydrology (Overgaard, Rosbjerg, and Butts 2006). Simulation of the water cycle (e.g. precipitation, evaporation, transpiration, infiltration, and runoff) is common in LSMs to predict streamflow, estimate water resources, and assess land-use impacts (Blyth et al. 2021). Remotely sensed data can be incorporated into LSMs (Overgaard, Rosbjerg, and Butts 2006; X. Xu, Li, and Tolson 2014). Examples of LSMs that include snow dynamics are the Community Land Model (CLM) (Lawrence et al. 2019), Joint UK Land Environment Simulator (JULES) (Best et al. 2011), and the Noah-Multiparameterization Land Surface Model (Noah-MP) (Niu et al. 2011; Z.-L. Yang et al. 2011). All these models have been applied within an Australian setting (Dumedah and Walker 2014; Hostache et al. 2020; Kala, Valmassoi, and Hirsch 2023; Kumar et al. 2021; X. Yang et al. 2020), but the Australian Alps have not been the focus region. LSMs applied to snow-affected parts of Australia have intentionally neglected the impact of snow (Lievens et al. 2015). The rationale for this omission was an assumption that snow has a limited influence on streamflow in the entirety of the MDB due to the limited spatial and temporal extent (Lievens et al. 2015). However, as detailed earlier (Section 1), this has been shown not to be the case.

5. Future of snow-runoff modelling in Australia

There are several research directions to extend the current emerging state of hydrological modelling of snow-affected catchments in Australia, leveraging both recent research in hydrology and related fields in the region and practices employed in the hydrological modelling community generally.

5.1. Model intercomparison and uncertainty analysis

It is common practice to compare model performance of multiple models on a single catchment (or set of catchments) using a common performance metric or metrics (Chiew, Stewardson, and McMahon 1993; Coron et al. 2012; Donnelly-Makowecki and Moore 1999; Franchini and Pacciani 1991; Knoben et al. 2020) – a practice called model intercomparison (Clark et al. 2008). To date, model comparisons have not been undertaken for catchments in the Australian Alps. This is unsurprising given that only six models have been applied to the region (Table 1). It would be beneficial to directly compare the models already applied as well as additional hydrological models that include snow dynamics to catchments in the Australian Alps to assess performance differences between models. Identifying the cause of differences in appropriate performance metrics between models could inform future model selection and development for runoff modelling from snow-affected catchments in marginal snow environments.

Although model intercomparison studies are useful by providing a range of possible simulation results typically for the same input data, the reasons behind model performance differences may be harder to understand (Clark et al. 2008; Pappas et al. 2013). Sensitivity analysis is another type of model diagnostics that is able to give further model insights, including identifying the critical sources of output uncertainty (Do and Razavi 2020; Koo et al. 2020; Razavi et al. 2021; Shin et al. 2013). As there are multiple ways uncertainty can be introduced to a modelling exercise (e.g. observed data, model structure, model assumptions, parameter estimation, calibration or a combination of these (Sitterson et al. 2018)), it is important to choose a sensitivity analysis method that is holistic (e.g. Koo et al. (2020)). Although a useful form of model assessment, sensitivity analysis is not often undertaken as part of the modelling process (Saltelli et al. 2020), and if it is, the analysis may not account for all or most uncertainty sources (Koo et al. 2020). Future hydrological modelling in the Australian Alps, and globally, should endeavour to analyse and report on uncertainties (Jakeman, Letcher, and Norton 2006; Saltelli et al. 2020).

5.2. Non-stationarity and climate change assessment

When the hydrological response of an area changes, it is characterised as hydrologic non-stationary (Ajami et al. 2017; Chiew et al. 2014; Razavi et al. 2015). Hydrologic non-stationarity has already been detected (Ajami et al. 2017; Chiew et al. 2014; Saft et al. 2015). Climate change is likely to alter hydrologic processes, introducing non-stationarity (Chiew et al. 2014). Physically based models have been suggested to cope well with non-stationarity, as a result of being comprised of equations that describe theoretically observable state and flux variables that adhere to the laws of physics (e.g. conservation of mass, energy, and momentum) (Fatichi et al. 2016). As the parameters of conceptual models often require calibration, the results of which depend on the period calibrated to, and there is potential for non-stationarity to render the current generation of models unable to capture future responses (Duethmann, Blöschl, and Parajka 2020). However, Vaze et al. (2010), Chiew et al. (2014) and Fowler et al. (2016) suggest that this can be overcome with well-thoughout calibration periods and techniques (i.e. searching for periods in the historic record that are extreme or similar to climate projections and critically assessing common calibration strategies).

Climate change has impacted the Australian Alps and will continue to do so (e.g. K. Hennessy et al. 2003; McGowan et al. 2018; Nicholls 2005; Whetton, Haylock, and Galloway 1996). It is common to use data created under climate scenarios to force models and assess climate change impacts (e.g. Chiew et al. (2009)). Climate change projections have been incorporated into hydrological models in the past (e.g. Coron et al. (2012), Rasmussen et al. (2011), Livneh and Badger (2020), Jung et al. (2012)). Past modelling work that compared uncertainty in climate change impacts, as assessed using a hydrological model, between rain-dominated and snow-dominated catchments found increased uncertainty in the latter catchment (Jung, Moradkhani, and Chang 2012). The increased uncertainty in future response to climate change impacts was attributed to increased difficulty in modelling snow processes compared to rain processes (Jung, Moradkhani, and Chang 2012). This sentiment is not unique; snow-affected catchments are consistently more difficult to model generally (Knoben et al. 2020). In an Australian setting, Schreider et al. (1997), Burns et al. (2023) and Trim et al. (2023) have tested in hydrological implications of climate change on catchments in the Australian Alps using models. Under dry scenarios streamflow from selected Australian alpine catchments was modelled to reduce by over 50% by 2070 (S. Y. U. Schreider, Jakeman, Whetton, et al. 1997), with low streamflow years projected to occur more commonly into the future (G. Burns, Fowler, and Horne 2023). There is still scope to improve our understanding about the possible impacts of climate change on runoff from the Australian Alps using multiple models of different types forced by climate projections.

5.3. Data collection and modelling

TIMs type models are the only snow representation that has been applied to snow-affected areas of Australia (Table 1). Due to the drawbacks of TIMs, physically based models have been proposed as potentially offering more accurate representations of snowpacks and resulting runoff over smaller timescales (Bilish, McGowan, and Callow 2018). Models that simulate snowmelt and subsequent runoff based on energy balance components have a long history, proving useful in other snow-dominated areas (Gao et al. 2015; Lawrence et al. 2019; Niu et al. 2011; Pomeroy et al. 2007; Z. Wang, Zeng, and Decker 2010). Physically based models allow for detailed examination of how different factors influence snowmelt and do not require calibration (however, may still benefit from calibration (Essery et al. 2013; Mendoza et al. 2015)). As of yet, such models have not been applied in the Australian Alps due to data availability. Only Bilish et al. (2018), Schwartz et al. (2020b) and Schwartz et al. (2020a) have measured energy fluxes to the snowpack in the Australian Alps, and only for relatively short periods (one winter season, two winter seasons, and two winter seasons, respectively). Only Schwartz et al. (2020a) collected flux data at multiple sites, with the other studies focusing on a single site. The single-site flux data collected by Schwartz et al. (2020b) are available for download (McGowan, Schwartz, and Callow 2019). The flux data collected were leveraged by several follow-on studies: Schwartz et al. (2020b), Schwartz et al. (2021a), Schwartz et al. (2021b). In other snow-affected environments, there is a long history of flux measurements, for example, at Niwot Ridge and Forest in Colorado (Blanken et al. 2009; S. P. Burns et al. 2016; Knowles et al. 2015; Turnipseed et al. 2002). Fortunately, the newly established Australian Mountain Research Facility will instal long-term flux towers in the Australian Alps (Mountain Research Facility 2023). The development of physically based models requires detailed additional data (S. Y. U. Schreider, Whetton, et al. 1997). Combining a TIM with physical elements is a practical step to improve model performance using available data and recent advancements in the understanding of snow processes (K. J. Bormann, Evans, and McCabe 2014) while detailed flux data acquisition is beginning in the Australian Alps. Once these datasets are available, careful consideration will need to be given to their application at the catchment scale given the variability and frequent patchiness of the snow cover (Bilish, McGowan, and Callow 2018).

Although efforts have begun to increase instrumentation throughout the Australian Alps, there is scope for models to inform where and what type of in-situ data would be most valuable in reducing model uncertainty. This could be undertaken in two steps: 1) sensitivity analysis to identify the dominate sources of uncertainty (e.g. Abdelhamed et al. (2023), Sheikholeslami et al., (2017), Haghnegahdar et al. (2017) and Koo et al. (2020)) and 2) targeted data collection to reduce the identified uncertainties. Using models to inform optimal design of monitoring networks has proved useful, for example, in groundwater systems (e.g. Fienen et al (2011, 2010) and Janardhanan et al. (2020).

Acknowledgements

Natasha would like to acknowledge Dr Joseph Guillaume for assisting in the setup of the literature review. The authors would like to acknowledge the Traditional Owners of the land on which this research was completed (Canberra and the Snowy Mountains), Ngunawal, Ngunnawal, Ngambri, Ngarigu and Walgal Country, and pay our respects to Elders past, present and emerging. Many thanks to the two anonymous reviewers for their constructive feedback on the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/13241583.2024.2343453

Additional information

Funding

Natasha is supported by an Australian Government Research Training Program (RTP) Scholarship https://www.education.gov.au/research-block-grants/research-training-program/research-training-program-frequently-asked-questions-students.

Notes on contributors

Natasha Harvey

Natasha Harvey is a PhD student at The Fenner School of Environment & Society, ANU, having previously obtained a Bachelor of Environmental Systems (Hons) from The University of Sydney and a Master’s degree from The University of Colorado Boulder. Natasha’s research interests include hydrology and water modelling.

Saman Razavi

Dr Saman Razavi is an Associate Professor at the University of Saskatchewan. He holds a PhD in Civil Engineering with a focus on Hydrology and Water Resources from the University of Waterloo, and a Master’s degree in Hydraulic Engineering from Amirkabir University of Technology. His Bachelor’s degree in Civil Engineering is from Iran University of Science and Technology. His research interests include hydrology, water management, and methods to address uncertainty and sensitivity in environmental modeling.

Shane Bilish

Dr Shane Bilish is a Hydrologist at Snowy Hydro Limited. He completed a Bachelor’s and Master’s in Geophysics at The University of Auckland, followed by a PhD in Physical Geography and Hydrology at The University of Western Australia, focusing on snowpack dynamics in the Australian Alps.

Notes

1. Schreider (1996) can be considered a preprint of Schreider et al. (1997c) (Sergei Schreider, pers comms., 6 July 2023).