Precipitation falling in the Snowy Mountains region of southeastern Australia provides fuel for hydroelectric power generation and environmental flows along major river systems, as well as critical water resources for agricultural irrigation. A synoptic climatology of daily precipitation that triggers a quantifiable increase in streamflow in the headwater catchments of the Snowy Mountains region is presented for the period 1958–2012. Here, previous synoptic-meteorological studies of the region are extended by using a longer-term, year-round precipitation and reanalysis dataset combined with a novel, automated synoptic-classification technique. A three-dimensional representation of synoptic circulation is developed by effectively combining meteorological variables through the depth of the troposphere. Eleven distinct synoptic types are identified, describing key circulation features and moisture pathways that deliver precipitation to the Snowy Mountains. Synoptic types with the highest precipitation totals are commonly associated with moisture pathways originating from the northeast and northwest of Australia. These systems generate the greatest precipitation totals across the westerly and high-elevation areas of the Snowy Mountains, but precipitation is reduced in the eastern-elevation areas in the lee of the mountain ranges. In eastern regions, synoptic types with onshore transport of humid air from the Tasman Sea are the major source of precipitation. Strong seasonality in synoptic types is evident, with frontal and cutoff-low types dominating in winter and inland heat troughs prevailing in summer. Interaction between tropical and extratropical systems is evident in all seasons.
Inflows generated from precipitation falling in the Snowy Mountains region provide vital water sources for irrigation and environmental flows in the agriculturally important and ecologically diverse Murray–Darling basin, as well as fuel for hydroelectric power generation. Located in southeastern Australia (SEA), the Snowy Mountains are one of only a few alpine regions in Australia. They form the highest part of the Great Dividing Range and include Australia’s highest peak—Mount Kosciusko—at 2228 m (Fig. 1). In contrast to younger mountain ranges such as the European Alps, the topography is not as steep, rugged, or high in elevation; instead, areas of undulating tablelands dominate the region.
In response to a series of severe droughts in the region, construction began on the Snowy Mountains Hydro-Electric Scheme (“Scheme” hereinafter) in 1949. The Scheme consists of a complex network of dams, hydroelectric power stations, tunnels, aqueducts, and pipelines that are able to divert eastward-flowing rivers under the mountain range inland to the Murray and Murrumbidgee Rivers. The Scheme provides on average 2360 Gl of water per year for irrigation, underwriting AUD $3 billion of agricultural production in the Murray–Darling basin. Furthermore, the water provides environmental flows along major rivers and offers a degree of flow regulation (Snowy Hydro Limited 2003; Ghassemi and White 2007; see also http://www.mdba.gov.au/about-basin/how-river-runs/murrumbidgee-catchment). The Scheme’s hydroelectric power generation meets the peak power demand for much of eastern Australia and currently provides 32% of all renewable annual energy production in Australia (http://www.snowyhydro.com.au/energy/hydro/snowy-mountains-scheme).
The Snowy Mountains are typically one of Australia’s wettest regions, but the precipitation is highly variable. Annual precipitation from days with ≥10 mm of precipitation in high-elevation regions of the Snowy Mountains varied between 2800 and 760 mm between 1958 and 2012 (Fig. 2). High precipitation variability also exists on intra-annual time scales. Precipitation in winter and spring is heavily influenced by the prevailing midlatitude westerly winds in conjunction with orographic enhancement, and these are commonly considered to be the wettest seasons (Snowy Hydro Limited 2003; Pook et al. 2006; Cai and Cowan 2008; Ummenhofer et al. 2009; Chubb et al. 2011). High-intensity and warm spring rains falling onto the snowpack are a major source of inflows (McGowan et al. 2009), generating as much as 50% of the total annual inflows to the hydroelectric catchments (Snowy Hydro Limited 2003). Although the warmer months are generally drier, heavy rainfall can result from trajectories of warm, moist air from lower latitudes and small-scale convective events that generate thunderstorms (Barry 1992; Basist et al. 1994; Snowy Hydro Limited 2003). Orographic enhancement of precipitation also occurs in summer as north and northwesterly winds flow perpendicular to the Snowy Mountains (Chubb et al. 2011).
The Snowy Mountains region lies toward the northern limits of the influence of the midlatitude westerly wind belt, where interaction between tropical and extratropical weather systems is an important factor in the generation of precipitation (Wright 1989). Accordingly, this region is sensitive to changes in the annual cycle of the subtropical ridge (STR) and shifts in the westerly storm tracks (Drosdowsky 2005; Murphy and Timbal 2008; Verdon-Kidd et al. 2013; Timbal and Drosdowsky 2013), as well as hydrological-cycle changes in a warming climate that can dramatically affect seasonal precipitation and runoff (Beniston 2003; Viviroli et al. 2011). In addition, the ascent mechanisms that are responsible for providing deep convective moisture and thus the potential for higher precipitation totals in the Snowy Mountains vary seasonally.
Historic snow-course data have shown significant declines in snow depth (Nicholls 2005; Hennessy et al. 2008) in association with rising alpine temperatures. Climate-change modeling of snow depth by Hennessy et al. (2003, 2008) and Hendrikx et al. (2013) shows a continuation of these declines. Regional-scale hydroclimatic modeling predicts rainfall and runoff to decrease on average and to increase in extremity over coming decades (Chiew et al. 2011). Meanwhile, demand for both water and energy are forecast to increase into the future (Christensen et al. 2007) resulting in increased vulnerability of water resources that are already under stress (Viviroli et al. 2011). Despite the importance of inflow-generating precipitation in the Snowy Mountains, there remains a knowledge gap regarding the long-term, historic climatological behavior of the synoptic weather systems that deliver precipitation to the region.
The cool-season synoptic circulation over various regions of SEA has been well studied (e.g., Wright 1989; Pook et al. 2006; Landvogt et al. 2008; Risbey et al. 2009; Gallant et al. 2012), although few studies relate specifically to the Snowy Mountains area (Colquhoun 1978; Chubb et al. 2011; Fiddes et al. 2015). These studies commonly report cool-season precipitation declines. From these cool-season-focused studies, it is widely reported that cold fronts and closed lows, including cutoff lows, are responsible for the majority of wintertime precipitation. These types of weather systems can occur year-round (Landvogt et al. 2008; Wright 1989), however, and are also important for inflow generation outside the winter season, particularly when they interact with tropical systems. For instance, the highest 7-day accumulated rainfall total in the Snowy Mountains region fell between 27 February and 4 March 2012, causing widespread flooding (Bureau of Meteorology 2012). Strengthening of the East Australia Current and associated Tasman Sea warming due to climate change suggest increasing summer precipitation in SEA (Cai et al. 2005; Shi et al. 2008; Gallant et al. 2012).
Previous studies of synoptic circulation over SEA have predominantly used manual-classification schemes, which are, by their nature, subjective and limited in the number of meteorological variables on which they are based. The majority consider only the cool-season period. Manual approaches are considered to be time-consuming, and, with the exception of Pook et al. (2014), the majority of SEA studies have covered relatively short time periods of a few decades at most. Several SEA studies have focused particularly on the period of extended drought that persisted for much of the 1990s and 2000s in an attempt to understand the significant precipitation decline that occurred during this period (e.g., Risbey et al. 2013). Studies confined to shorter periods may not be fully representative of synoptic circulation over a multidecadal period. McGowan et al. (2009) identified the role that multidecadal ocean–atmosphere teleconnections may play in the hydroclimate of the Snowy Mountains region. Accordingly, there is a need to extend current understanding of synoptic circulation beyond the last few decades and to encompass year-round synoptic systems.
Previous manual classifications of precipitation systems worldwide and in Australia focus mainly on surface pressure fields with limited analyses of mid- and upper-level atmospheric properties. Between three and five synoptic types were identified for SEA (Wright 1989; Pook et al. 2006; Landvogt et al. 2008; Chubb et al. 2011; Risbey et al. 2013). Studies that have applied automated techniques have been based on a single atmospheric variable, usually mean sea level pressure (MSLP; Whetton 1988; Kidson 2000; Jiang 2011; Renwick 2011), because of difficulties encountered in combining multilevel data (Kidson 2000). Multivariable classifications either classify each variable separately (e.g., Bettolli et al. 2010) or use data-reduction techniques (e.g., Stahl et al. 2006; Moron et al. 2008). Self-organizing maps that are based on MSLP (Cassano and Cassano 2010) or 500-hPa geopotential height (Newton et al. 2014a,b) were employed in Canadian studies to assess catchment-scale hydroclimatic variability. These approaches, as noted by Stahl et al. (2006), conceal the complex three-dimensional nature and internal variability of synoptic types. Consequently, causes of variability in seasonal rainfall that depend not only on the annual cycle of surface pressure, but also on seasonal changes to the atmospheric circulation in the mid- and upper troposphere (Pook et al. 2006), may not be identified.
Here we present a seasonal, 55-yr (1958–2012) synoptic climatology of inflow-generating precipitation days for the Snowy Mountains region. A novel, objective method is developed that describes synoptic types on the basis of a suite of 21 meteorological variables throughout the depth of the troposphere. Use of reanalysis data and an automated approach allows a multidecadal time period to be investigated and removes much of the subjectivity of manual classifications (Yarnal 1993). Important is that this approach allows robust conclusions to be drawn with regard to patterns of synoptic circulation that are responsible for inflow-generating precipitation, enabling ongoing research to investigate trends in synoptic circulation variability and their significance for local to regional hydroclimate. Such knowledge is essential to better understand the drivers of variability in historical records of precipitation so as to make better-informed water-management decisions (Viviroli et al. 2011).
2. Data and methods
A network of private, tipping- and weighing-bucket precipitation gauges operated by Snowy Hydro Limited (SHL) and the Queensland government’s Scientific Information for Land Owners (SILO) “Patched Point Dataset” (https://www.longpaddock.qld.gov.au/silo/ppd/index.php?reset=reset, accessed January 2014; Jeffrey et al. 2001) from the Australian Bureau of Meteorology (BoM) recording stations provide the daily precipitation observations used in this study (Fig. 1b). Some records in the SHL dataset begin in the 1950s, but many of the early data records (before 1975) are discontinuous. Installation of an automatic weather station within the study area in 1996 and a substantial increase in precipitation gauge network size after 2004 vastly improved data quality. Newer, heated and weighing gauges at higher elevations, some with wind shields, more accurately record snowfall. Records from 56 SHL active gauges within the Snowy Mountains have been used in this study, which covers the period of 1958–2012. Data are recorded instantaneously and were aggregated to daily totals to 0900 local time (in line with the BoM convention for daily precipitation observations). SHL data from two inflow-recording stations, Yarrangobilly River at Ravine station and the Snowy River above Guthega station (Fig. 1b), were also aggregated to daily totals.
Data from 10 gauges within the Snowy Mountains were selected from the SILO dataset (Fig. 1b) and used to ensure a continuous daily precipitation record. The SILO dataset consists of observations from BoM gauges, filled in with interpolated values where observations are unavailable. Rainfall is interpolated using the geostatistical method of ordinary kriging (Jeffrey et al. 2001). An interpolated surface (0.05° × 0.05° grid resolution) is produced for each day, from which missing data values for point locations are extracted from the nearest grid cell. The provision of daily data for individual stations rather than a gridded product allows direct insertion of data (Pook et al. 2010) into discontinuities in the SHL record. For any point location, the nearest 30 stations and all within 100 km are used for the interpolation, whichever is greater. It has been shown that normalized precipitation removes the component of precipitation variability that is due to topographic influences and can be reliably interpolated at time scales from hourly to monthly (Jeffrey et al. 2001). Topographic effects are subsequently accounted for by denormalizing the interpolated surface to derive the final rainfall surface. The SILO dataset is available for the period from 1889 to present. Precipitation data from all seasons were considered. The standard climatological seasons have been used throughout this study: December–February (DJF), March–May (MAM), June–August (JJA), and September–November (SON).
For the purpose of producing a synoptic climatology, the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim and ERA-40 reanalysis products (Dee et al. 2011) were used as input to a clustering algorithm and to construct composite maps of the synoptic atmospheric circulation associated with precipitation days. ERA-Interim spans the period from 1979 to near–real time and is available at a default 0.75° latitude × 0.75° longitude grid resolution. ERA-40 spans the period 1958–2002 and has a default resolution of 2.5° × 2.5°, but a 0.75° × 0.75° resolution can also be obtained and is used here. Reanalysis products at coarser scales have been shown in previous studies to be unsuitable for regional climate assessments (Eichler and Gottschalck 2013) and to have difficulty accurately detecting features such as surface pressure fronts and troughs. As a result, past synoptic studies have needed to supplement coarse-resolution reanalysis data with satellite imagery and manual-analysis charts (e.g., Pook et al. 2006). Furthermore, ECMWF reanalyses have improved representation of Southern Hemisphere high-latitude atmospheric circulation when compared with other reanalysis products (Marshall 2003).
ERA-Interim represents the latest reanalysis product (at the time of writing) from ECMWF and has addressed several data-assimilation issues that were encountered in ERA-40 (Dee et al. 2011). Comparison of clustering results from the two products for an overlapping 22-yr period (1979–2001) demonstrated no significant difference in the output (not shown). This result is in agreement with Hoskins and Hodges (2005) who evaluated the impact of changes to observing systems in their analysis of Southern Hemisphere storm tracks. They concluded that their climatological description remained robust between different reanalyses and the pre- and postsatellite eras.
Daily mean values of the following variables were used in this study: MSLP; 500-hPa geopotential height; wind vectors at the surface, 850, 700, 500, and 250 hPa; relative humidity and temperature at 850, 700, 500, and 250 hPa; and (the computed) 1000–500-hPa atmospheric thickness. The subtropical jet stream (STJ), the strength and position of which over Australia is known to influence steering and development of synoptic systems, was calculated as the magnitude of the wind vector at 250 hPa (Risbey et al. 2009). For our purposes, temperature and humidity profiles are standard airmass indicators of atmospheric stability and available moisture, respectively (Davis and Kalkstein 1990). Wind vectors provide critical information on direction of moisture transport and steering of synoptic systems (Pook et al. 2006). Mean sea level pressure provides an indicator of the large-scale current atmospheric state (Eder et al. 1994) while thickness provides information on advection and frontal position (e.g., Pook et al. 2006). Variables were obtained for the synoptic analysis area bounded by latitudes 20°–46°S and longitudes 120°–160°E (Fig. 1a). This region is considered to be extensive enough to capture all synoptic weather systems affecting the Snowy Mountains, including those originating in, and interacting with, tropical latitudes.
1) Quality control
For the purposes of this study, all data were subject to quality control and any data flagged as unsuitable for climatological purposes were automatically removed. Data were checked on a gauge-by-gauge basis for anomalous values by calculating the maximum, minimum, and range. Furthermore, any data with a quality flag indicating potentially bad data were subject to additional quality control and in a few cases were removed because of anomalously high half-hourly values. In addition, SHL data coded as “good” but with a zero amount where the corresponding SILO dataset showed a value greater than zero were disregarded when calculating the daily mean values.
2) Precipitation threshold
This study investigates days on which precipitation generates inflow to reservoirs within the Snowy Mountains water catchment (Fig. 1b). To define a threshold amount for a precipitation day, a Lyne and Hollick filter (Lyne and Hollick 1979) was applied to separate “quickflow” (surface runoff) from base flow in the inflow dataset (Nathan and McMahon 1990) at the Snowy River and Yarrangobilly River stations. As alpine and subalpine sites, respectively, these two stations provide indicative conditions across the region. Precipitation from the most relevant gauges to the inflow stations was then correlated with quickflow at a lag time of +1 day (based on prior knowledge of the behavior of precipitation and runoff in these catchments). The application of the Lyne and Hollick filter to inflow data resulted in a threshold precipitation amount of 10 mm, above which quickflow and therefore inflow was enhanced (not shown).
Only precipitation during the period from December to April was considered for the determination of this threshold. During the cool season, precipitation can be stored as snowpack or can enhance inflow during spring snowmelt, and, particularly during this period, a static precipitation–runoff threshold does not apply. It is acknowledged that such a precipitation–runoff threshold is typically dynamic and depends upon numerous catchment conditions but is necessary for the purpose of synoptic classification of precipitation days in this study.
3) Grouping data
Different precipitation regimes are experienced across the Snowy Mountains region, in part because of topographic interaction with the prevailing flow (Chubb et al. 2011; Fiddes et al. 2015). Following Chubb et al. (2011), the 56 SHL gauges were divided into western-, high-, and eastern-elevation groups (Fig. 1b) on the basis of the similarity of the statistics shown in Table 1. Although the high-elevation group contains a greater number of gauges, most of these began operating after 2004. Prior to 2004, the number of gauges was more similar across all groups (Table 1). Such regionalization of data has been used successfully in previous studies in a variety of locations worldwide, helping to reduce noise and spatial autocorrelation in the data (Whetton 1988; Widmann and Schär 1997; Kidson 2000; Chubb et al. 2011; Plavcova et al. 2014).
A continuous daily precipitation record back to 1958 was not possible using only the SHL data; therefore, comparisons between the SHL and SILO datasets were carried out to determine the suitability, and associated uncertainty, in combining the two datasets. Correlations, contingency tables, and their associated skill-score statistics [bias, probability of detection (POD), and mean absolute error (MAE)] were calculated (Wilks 2006; Beesley et al. 2009; Tozer et al. 2012). Data from all 10 SILO gauges within the watershed boundary provided the best comparison with the most recent SHL data (which include high-resolution, heated, and fenced gauges), suggesting that using all gauges in the SILO dataset back to 1958 provides the most robust estimate of precipitation.
Contingency-table statistics were better overall for gauges within the high-elevation group, with POD scores indicating that at least 78%–84% of precipitation days over 10 mm are accurately detected by SILO. The MAE between the SILO and SHL datasets varies between 0.03 mm (high elevations) and 3.66 mm (eastern elevations), in good agreement with Jeffrey et al. (2001) for the study region. SILO estimates of precipitation do not demonstrate a forecast bias for the high elevations, but there is a tendency to underforecast (overforecast) precipitation days after (before) 2006 in the western and eastern elevations. Lower skill scores in the western and eastern elevations are likely due to fewer gauges in these areas. Particularly in the east, the only two gauges within the SILO dataset are located in the southern part of the catchment. Lack of representation in the northern catchment may explain the lower skill there. Despite the lower skill scores for the eastern and western elevations, only 4.5%–4.9% of days are detected solely in these groups.
A daily precipitation amount for each group was calculated as the mean daily precipitation from each gauge within the group (Chubb et al. 2011; Dai et al. 2014; Fiddes et al. 2015). Following WMO guidelines and Chubb et al. (2011), a minimum of four data values were used to calculate a mean (WMO 2011). When this was unachievable using only SHL data, the SILO data were included in the calculation. The change in the number of SHL gauges, particularly within the high-elevation group, was shown to have minimal effect on the calculation of the mean, with an average difference of 0.11 mm between mean values. Prior to the commencement of data recording in the western (eastern) elevations by the Snowy Hydro Scheme in 1961 (1960), daily totals were calculated solely from the SILO dataset.
Each day with a precipitation total greater than the threshold amount of 10 mm was extracted from the daily record. The corresponding ECMWF reanalysis variables for each of those days were collated into a data matrix and standardized on a monthly basis (over the full study period), using z scores, to account for the inconsistent units (Yarnal 1993; Hart et al. 2006; Wilson et al. 2013; Gao et al. 2014). Standardizing climate data has been demonstrated to have the additional benefit of removing seasonal variations in intensity of synoptic systems, allowing comparison between data with different variabilities and means (Gao et al. 2014; Jiang 2011; Wilks 2006; Kidson 2000; Yarnal 1993). The strong influence of the seasonal cycle of the STR on weather patterns in SEA means clustering applied to standardized data better reflects the daily changes in atmospheric circulation (Yarnal 1993). For example, Pook et al. (2006, their Fig. 3) show the variability in monthly mean MSLP across the cool season. In our study, the monthly standardized MSLP data reveal that precipitation days are always associated with negative anomalies, which become stronger during the cooler months and are at a maximum in May (not shown). Similar month-to-month variations are apparent in the other variables used in this study (not shown for brevity). The standardized data matrix was used as input into the cluster analysis.
4) Cluster analysis
To generate synoptic types for the period 1958–2012, cluster analysis was applied to the standardized meteorological variables for each daily precipitation total greater than the threshold amount of 10 mm. The cluster analysis was performed across all seasons simultaneously, given the prior removal of the seasonal cycle by data standardization. Equal weighting was given to each variable, given the importance of each in influencing the synoptic circulation and precipitation received, as outlined in section 2a. Hierarchical average-linkage clustering gave an initial indication of the number of groups contained in the data (Wilks 2006; Hart et al. 2006; Trauth 2007). The k-means clustering method of Wilson et al. (2013), with a city-block distance measure, was then applied to the standardized variables to assign each precipitation day to a synoptic type (Hart et al. 2006; Wilson et al. 2013). The iterative nature of the k-means technique refines the clusters by reclassifying days until the smallest within-cluster solution is found and days with similar meteorological characteristics are classified in the same cluster (Hart et al. 2006). The algorithm was tested for a range of clusters between 2 and 20. Examination of a plot of the distance measure against number of clusters for the point at which the line flattens out, and after which distance increases again, commonly gives an indication of the optimum number of clusters (Wilks 2006; Tan et al. 2006; Wilson et al. 2013). This was used in conjunction with physical interpretation of composite maps (generated from the mean value of all days assigned to each cluster; Kalkstein et al. 1987). The k-means technique was initialized several times using random subsets of the data as cluster seed values. The iteration for which the sum of distances was smallest was then used as the cluster seeds for the full dataset.
Comparison of clustering results between the ERA-Interim and ERA-40 reanalyses for an overlapping 22-yr period produced no significant differences in the resulting synoptic types, with a hit rate of >80% in the assignment of individual days to the same synoptic type. This is in agreement with Hoskins and Hodges (2005) whose comprehensive comparison of synoptic climatologies remained robust between different reanalyses products and pre- and postsatellite eras.
The automated clustering procedure was validated by comparison with a manual classification for a 5-yr period (2008–12) and nonparametric hypothesis testing of the precipitation assigned to each cluster. Surface and 500-hPa height charts, readily available from BoM (http://www.bom.gov.au/australia/charts/archive/), and NOAA satellite imagery (http://www.ncdc.noaa.gov/gibbs/year) were used to classify each day on the basis of identification of key surface and upper-air features (Davis and Kalkstein 1990; Yarnal 1993), and the presence of cloud bands.
Classification on the basis of these variables revealed several days that could have been placed into more than one cluster. Further inspection of additional reanalysis-generated variables (temperature, humidity, and wind vectors) showed that most days could belong to only one particular cluster. Overall, only 8% of days were moved into a different cluster on the basis of the manual analysis, from that generated by the k-means algorithm.
3. Results and analysis
a. Synoptic classification of precipitation-bearing systems
The application of the threshold precipitation amount to daily precipitation for the period of 1958–2012 resulted in 3443 days being identified with inflow-generating precipitation and requiring synoptic classification. A day was classified if at least 10 mm of precipitation was recorded in the western, high, or eastern group. These specific days account for almost 40% of all precipitation days, that is, those for which precipitation ≥ 1 mm is recorded. It is acknowledged that this definition of a precipitation day does not account for multiday precipitation events, in which a series of synoptic types may traverse the region as a precipitation-bearing weather system evolves through time. Instead, each individual day is assigned to a specific synoptic type, following Pook et al. (2006) and Risbey et al. (2009). Figure 2 shows the annual precipitation across the different elevations (Fig. 2a) and the annual number of precipitation days of at least 10 mm (Fig. 2b) between 1958 and 2012, demonstrating the high degree of interannual variability in the precipitation of the Snowy Mountains region. A statistically significant trend in precipitation of +38 mm (10 yr)−1 in the eastern elevations is apparent, but western and high elevations and annual precipitation days exhibit nonsignificant decreases.
A two-sided Wilcoxon–Mann–Whitney test demonstrated that the median rainfall amount in corresponding clusters of the manual and automated classification was equal at a 95% confidence level (p < 0.05). Furthermore, the manual classification confirmed that the automated scheme was capable of detecting expected synoptic patterns. The types were not as clearly defined as in previous manual-classification studies (e.g., Pook et al. 2006; Chubb et al. 2011), however, likely because of the greater number of variables being considered and thus the greater possible combinations of variables as well as the multitype nature of some synoptic systems.
The k-means clustering method, applied to the daily, standardized reanalysis data for a range of cluster numbers k, suggested the optimum number of clusters to be 10 or 11 (Fig. 3). Comparison of composite maps for each of these solutions further suggested that 11 synoptic types better represented known synoptic systems and reinforced initial hierarchical clustering results. A Wilcoxon–Mann–Whitney test determined that median precipitation between types was significantly different, rejecting the null hypothesis that all medians were equal (p < 0.05), for 80% of all possible combinations (considering precipitation across all areas of the catchment). Given that only those days experiencing precipitation ≥ 10 mm have been classified, rather than rain versus no-rain days, and the natural variability in precipitation amount from individual occurrences of the same type, it is more likely that the null hypothesis will not be rejected in all cases. This result, along with physical interpretation of composite maps for each reanalysis parameter (Kalkstein et al. 1987; Wilson et al. 2013), demonstrates that each of the 11 clusters represents specific synoptic types.
Composite charts showing average meteorological conditions for a number of key parameters for each of the 11 synoptic types are presented in Figs. 4–9. In addition to the clustering parameters, an analysis of columnar precipitable water (PW) and relative vorticity for each synoptic type was conducted, and it is shown in Fig. 10. These additional parameters give further information on available moisture and system development, respectively. Some between-type similarities exist in individual variables for a given level (e.g., MSLP), but when all variables are considered together each type has distinct characteristics and distinguishing features. Table 2 summarizes the three-dimensional characteristics, ascent mechanisms, and moisture pathways for each synoptic type. The frequency of occurrence of each synoptic type and the resulting precipitation contributions in each elevation group are presented in Table 3.
The synoptic classification reveals that 8 of the 11 types (all except T1, T5, and T9) represent atmospheric circulation patterns with a connection to tropical latitudes, in particular where a north or northwesterly airflow (specifically between 700 and 500 hPa; Figs. 7–9) advects a conveyor of warm, moist air originating from the warm oceans surrounding tropical Australia toward the Snowy Mountains (Figs. 6–8). These tropical-connected systems deliver over 70% of total precipitation greater than 10 mm across the whole catchment (Table 3). Three of the tropical-connected synoptic types—T8, T11, and T4—display synoptic circulation conducive to northwest cloud bands (NWCBs). For T4, however, NWCBs are detected on only ~10% of occurrences, when conditions match those in Wright (1989, their Fig. 3). NWCBs form over the warm surface waters to the northwest of Australia, where deep convection feeds moisture from the tropical Indian Ocean along the cloud band to southeastern Australia (Tapp and Barrell 1984; Sturman and Tapper 2006). This circulation, evident in Figs. 4 and 5 for T8 (and to a lesser extent T4 and T11), features moisture aligned with the core of the STJ and its region of maximum intensity [as shown in Tapp and Barrell (1984)]. Together these three synoptic types account for 26% of all days over 10 mm. In a similar way, and as confirmed by the manual classification, T10-classified synoptic types have circulation that is conducive to cloud bands extending northward along the east coast, commonly seen as an easterly dip and cloud-band pattern (Gallant et al. 2012; Fiddes et al. 2015). Downstream anticyclones and ridging—apparent in T3, T4, T6, T7, T8, T10, and T11—contribute to tropical moisture transport and enhancement of warm air advection (WAA; Fig. 6), which, combined with an anticlockwise rotation of winds with height (“backing”), signifies forced ascent (Figs. 7–9). With the exception of T11, all tropical-connected systems demonstrate upper-level divergence in the exit quadrant of the STJ (Fig. 5). All synoptic types exhibit airflow directions that are conducive to orographic enhancement of precipitation.
Synoptic types T1, T2, T5, T6, and T8 can be grouped as cold-cored frontal-type days [including contributions from closed and cutoff lows (T2 and T6) and prefrontal troughs (T5 and T8)], and together account for 53% of all classified precipitation days (Table 3). Cutoff lows here follow the definition in the SEA studies of Pook et al. (2006), Risbey et al. (2009), and Chubb et al. (2011), among others, in which closed circulation can be apparent at the surface or midlevels with a trough above or below, respectively. The manual analysis showed that T2 fulfils the traditional criteria of cutoff lows on approximately 60% (50%) of occurrences and T6 fulfils them on 72% (44%) of occurrences when considering MSLP (500-hPa geopotential height). In addition, closed-low types have associated fronts on 75% (T2) and 45% (T6) of occurrences. Cold-frontal types have in common differential cyclonic vorticity advection (CVA) at 500 hPa as an ascent mechanism (Fig. 10), but closed-low types demonstrate stronger cyclonic vorticity maxima than do embedded cold fronts and prefrontal types. The types T4 and T7 show the signature of heat lows and troughs interacting with cold-cored extratropical fronts to the south of Australia (Sturman and Tapper 2006; Gallant et al. 2012). Localized acceleration of the STJ core along the enhanced thickness gradient, jet-stream divergence in the poleward exit quadrant, and strong WAA (associated with strong downstream anticyclones) are consistent with enhanced system development and higher precipitation totals (Risbey et al. 2009). Cold fronts (T1) occur at a frequency that is similar to those of other frontal types, although they deliver smaller amounts of precipitation across all elevations, consistent with lower humidity and precipitable water. Strong cold-air advection (CAA), clockwise rotation of winds with height, and jet-streak divergence downstream of the front indicate ascent is provided primarily by frontal lift. Development of closed and cutoff lows is associated with a localization and concentration of the STJ, often forming on the poleward side of the jet. This relationship is represented in Fig. 5, which shows a strong jet core located to the north of the closed low for types 2 and 6. Similarly, for classifications that include the passage of a cold front (T1 and, to some extent, T5), meridional excursions of the polar jet that cause it to merge with the subtropical jet near the location of the front are a known feature (Sturman and Tapper 2006; Risbey et al. 2009). Furthermore, Risbey et al. (2009) associate cyclonic curvature of the jet-stream core, the exit region of which is divergent, with the largest amounts of precipitation (>5 mm and, in particular, for synoptic systems that produce ≥ 15 mm of precipitation) from frontal systems—evident here in Fig. 6 for frontal-type days over 10 mm.
Synoptic type T3 is representative of inland heat troughs extending from a low pressure center in northern Australia—known locally as the Cloncurry low (Sturman and Tapper 2006; Gallant et al. 2012). Strong WAA, into the divergent region of the jet stream, is apparent in the vicinity of the trough (Fig. 6).
Topographic interaction of the prevailing airflow in each synoptic type creates differences in precipitation contributions between elevation areas. Synoptic types generating the greatest precipitation totals across westerly elevations are the result of approaching cold fronts or troughs, closed lows and troughs that extend toward northwest Western Australia (T2, T4, T6, T7, and T8; Table 3). This is similar for the high elevations, where NWCBs associated with days that are classified as type 8 and with closed lows (T6) bring the highest percentage of precipitation totals per day. A common feature in each of these types (T2, T4, T6, T7, and T8) is a downstream anticyclone, with WAA and relatively high PW to the northeast and northwest of the study region (Figs. 6 and 10), along with divergence in the poleward exit quadrant of the jet stream (Fig. 5) and orographic enhancement. As a result, thermal wind is enhanced and directed southeastward along the thickness gradient, causing acceleration of the STJ (Fig. 5)—apparent here for those synoptic systems classified as T2, T6, and T8. Classifications that include closed lows alone account for approximately 20% of days and contribute 26%, 23%, and 16% of total precipitation to the western, high, and eastern groups, respectively.
The contribution of precipitation from frontal systems and the midlatitude westerly airflow is reduced in eastern elevations, in the lee of the mountain range. Instead, onshore easterlies in the subtropics associated with downstream anticyclones or ridging that advect warm, humid air from a moisture corridor along the east coast via inland, meridional troughs (T3, T7, and T10), and offshore lows (T9), are the major sources of precipitation (Table 3). WAA is a common ascent mechanism for these types. Differences in the spatial distribution of precipitation between synoptic types shown here are consistent with investigations by Chubb et al. (2011) and Fiddes et al. (2015).
b. Seasonality of synoptic types
Clear seasonality in the frequency of synoptic types is evident in Fig. 11, which reflects seasonal movement of the STR. The mean contribution of each type to seasonal precipitation accumulations (Fig. 12) demonstrates the high degree of intra-annual variability. The greatest between-type variability, in terms of both frequency and precipitation, occurs in winter and summer. Orographic enhancement of precipitation for all types is evident, with highest elevations consistently receiving the largest precipitation totals in all seasons (Fig. 12).
The majority of types can produce seasonal precipitation accumulations beyond the 95th percentile (>2 standard deviations) and could be considered as extreme (Pook et al. 2012). Notable large falls of precipitation, exceeding 300 mm in the western and high elevations, have resulted from the dominant types T4 and T7 in summer (not shown). As noted in section 3a, these types display properties that are consistent with enhanced system development, strong WAA, and higher precipitation totals.
Table 4 summarizes the mean precipitation from all days per season and reinforces previous studies that this region is dominated by cool-season precipitation (nearly 60%), often associated with frontal and closed-low systems in the lower-to-midtroposphere (surface–500 hPa; Figs. 11 and 12). However, this study highlights that a significant proportion of inflow-generating precipitation days of ≥10 mm (approximately 20%) are recorded during summer months, often related to the occurrence of inland heat troughs (i.e., T4 and T7) and increased convection. A further 20% of days occurred in the transitional autumn season. Mean daily precipitation from all synoptic types is similar in all seasons, but the fewer number of precipitation days occurring in summer and autumn and the higher PW indicate a higher intensity for warm-season precipitation days (Table 4).
In austral summer (DJF), precipitation is dominated by warm-cored heat-trough types T3, T4, and T7. Summer types are associated with weaker midlevel troughs and weaker cyclonic vorticity, often displaced upstream of the Snowy Mountains. Instead, stronger ridges, higher moisture, enhanced WAA, and low-level (upper level) convergence (divergence) dominate and act as ascent mechanisms. Moist air, with high PW, is entrained from tropical latitudes toward the Snowy Mountains region (Figs. 6–8, 10). These systems deliver relatively consistent falls across all elevations (Fig. 12).
The occurrence of each synoptic type and their associated precipitation are more consistent in autumn (MAM), although T1, T3, T4, and T5 dominate slightly, representing a mix of warm- and cold-core synoptic types. This is a transition season with precipitation still generated from a moist airflow and WAA from tropical latitudes (T3 and T4). However, the northward movement of the STR and midlatitude westerly wind belt is apparent as prefrontal troughs and cold fronts associated with embedded lows in the Southern Ocean (T1 and T5) begin to cross the region more frequently.
Cold-cored frontal types T1, T2, T5, T6, and T8 and offshore low pressure centers (T9) dominate in winter (JJA). All demonstrate CVA with a maximum either over SEA (T1, T2, and T9) or upstream (T5, T6, and T8). Winter types generally display higher maximum cyclonic vorticity than do summer types. Only one-half of these types have a tropical moisture corridor at 700 hPa, and the influence of the midlatitude westerly wind belt is clear, with a moisture source in the lower atmosphere (850 hPa) over the Southern Ocean or Tasman Sea. Low-level convergence, a northwest moisture corridor, and WAA are common features for T6 and T8, which deliver the highest winter precipitation totals. In eastern elevations the influence of orography is apparent, with much lower precipitation resulting from the predominantly westerly flow. Instead, low pressure centers situated off the east coast (T9) are the dominant source of precipitation.
Similar to autumn, spring (SON) observes an even spread in the percentage of total seasonal precipitation days, although cold-core types T2, T6, and T8 along with T4 occur slightly more frequently. Prefrontal and closed-low systems (T2, T6, and T8) make the highest contributions to precipitation totals, but, given that this is a transitional season, systems connected to tropical latitudes are also common (e.g., T4 and T7). Occurrence of these warm systems, on top of a late-season isothermic (“ripe”) snowpack, is considered to generate the greatest snowmelt and inflow in the Snowy Mountains.
4. Discussion and conclusions
Presented here is a 55-yr (1958–2012) synoptic climatology of daily synoptic circulation systems that deliver greater than 10 mm of precipitation for the Snowy Mountains region of southeastern Australia. This is the first study to link synoptic circulation throughout the tropospheric column to precipitation at the surface in the Snowy Mountains. The use of a suite of variables throughout the depth of the troposphere, applied to a large gridded analysis area, expands on previous studies (e.g., Wilson et al. 2013) and results in 11 synoptic types. The clustering method reveals subtle differences in, for example, the position and orientation of surface troughs and the location of moisture corridors, which directly affect the amount of precipitation received at different sites in the Snowy Mountains region. Some types display similar attributes at certain levels, but each represents a particular synoptic atmospheric circulation for this region. The method used in this study has demonstrated that difficulties in combining variables from different atmospheric levels (Kidson 2000) can be overcome, providing a vertical profile of atmospheric conditions during specific precipitation days. It offers an automated approach to the traditional map classification (Yarnal 1993) and generates synoptic types with no a priori forcing of the clustering algorithm, minimizing subjectivity. The importance of moisture source regions and ascent mechanisms in delivering precipitation to the region of interest is demonstrated.
The use of a daily precipitation threshold of ≥10 mm differs from previous southeastern Australia studies in which all precipitation days have been considered. Important is that it allows classification of synoptic circulation associated with precipitation days that trigger a quantifiable increase in streamflow in the headwater catchments of Australia’s iconic and economically important river system: the Murray–Darling basin.
Seasonal variations in the frequency of occurrence of each synoptic type highlight the variability of atmospheric circulation affecting the Snowy Mountains region. Although certain synoptic systems are predominant during the cool or warm seasons, they may occur at any time (Wright 1989), and a clear seasonal signal in their incidence is apparent (Fig. 11).
The seasonal movement of the STR is critical to the synoptic systems experienced in SEA. In winter the STR is in its most northerly position over central Australia, allowing the passage of frontal systems associated with the midlatitude westerly wind belt to traverse southern Australia. Accordingly, winter synoptic types frequently relate to the passage of cold fronts and closed lows and are typically associated with CVA. Areas of descent coincide with CAA while CVA and ascending motion are enhanced downwind of the front or trough (Risbey et al. 2009). Jet streaks display greater intensity in winter. Divergence in their poleward exit quadrant, along with CVA in midlevels, results in rising motion (Sturman and Tapper 2006) and is a common ascent mechanism in winter. In addition, the highest mean winter precipitation totals result from types demonstrating convergence at 850 hPa and trajectories of warm, moist air from tropical latitudes (T6 and T8).
In summer, the southward movement of the STR sees southern Australia under the influence of a band of high pressure, associated with the descending branch of the Hadley cell. Accordingly, frontal systems associated with the midlatitude westerly wind belt are pushed south of Australia. Instead, downstream anticyclones and ridging enhance WAA and entrainment of tropical moisture. The warmer seasons are indicative of higher precipitable water and weaker ascent (Milrad et al. 2014). Areas of low-level convergence that are present in the dominant summer types indicate positive vertical velocity and, in addition to strong WAA, provide an ascent mechanism, in the absence of the CVA present in winter (Risbey et al. 2009).
In all seasons, interaction between tropical and extratropical systems (Wright 1989), tropical moisture pathways, and vertical ascent profiles with low-level (upper level) convergence (divergence) (Gyakum 2008) are highlighted as important factors generating precipitation of ≥10 mm, confirming trajectory analyses conducted by McIntosh et al. (2007) and Chubb et al. (2011). In a similar way, event precipitation isotope analyses conducted by Callow et al. (2014) identified significant synoptic-type variability in the isotopic signature of precipitation in the Snowy Mountains. The results of our study further support the dominance of moisture-source pathways in controlling isotopic variability in this region.
It is acknowledged that using daily variables may mask some of the contributions of subdaily systems such as pre- and postfrontal flow to precipitation over the region (Gallant et al. 2012) and excludes multiday and multitype events. Accordingly, slow-moving synoptic systems may be overrepresented. However, when considering synoptic-scale circulation, which generally operates on time scales of a few days, the use of daily variables is considered to be appropriate (Barry and Carleton 2001; Sturman and Tapper 2006; Pook et al. 2006). In addition, the nature of the composite plots may result in some features in the synoptic-scale circulation becoming masked (Milrad et al. 2014). Highly variable precipitation, complex terrain, and a relatively low density gauge network in the SILO dataset may all contribute to interpolation errors. Extensive testing and comparison of the SHL and SILO datasets have been used to quantify the uncertainty in using these interpolated precipitation data. This has permitted construction of a continuous daily precipitation record for the Snowy Mountains region; such a record is considered to be essential for compiling a long-term synoptic climatology (Chappell and Agnew 2001).
In summary, this novel method proposes that synoptic typing can be successfully based on atmospheric variables throughout the depth of the troposphere. The higher number of types in this study compares well to other automated typing schemes, which have generally produced more types than manual methods, with more subtle differences between types (e.g., Newton et al. 2014a,b; Plavcova et al. 2014; Kidson 2000). Although the k-means technique has been widely used in synoptic classifications, this is the first time, to our knowledge, that it has been applied to multilevel and multiparameter gridded meteorological data. It has revealed the complex three-dimensional nature of synoptic-scale circulation, including the importance of the influence of moisture pathways from tropical latitudes in the generation of high precipitation totals. It provides a method for linking regional-scale precipitation data to synoptic-scale atmospheric circulation. The spatial distribution of precipitation associated with synoptic types has implications for water resources management (Frei and Schär 1998; Newton et al. 2014a,b) in this region.
The synoptic-typing method that was applied here allows long-term and climatologically significant periods to be examined, enabling a robust investigation of the impacts of synoptic circulation on the hydroclimate of the Snowy Mountains region. Future research will investigate temporal variability of the synoptic types in relation to interannual drivers of climate variability. Increased understanding gained from this synoptic climatology has implications for water resource management in regional areas, and this method could be readily applied to other hydroclimate studies and to other regions worldwide.
We thank Snowy Hydro Limited scientific staff for helpful and constructive discussions, and we thank three anonymous reviewers whose comments have greatly improved this manuscript. We also thank Joshua Soderholm for initial assistance with programming. Alison Theobald was supported by an Australian Postgraduate Award and by Snowy Hydro Limited. We thank Snowy Hydro Limited and the Queensland Government Department of Science, Information Technology, Innovation and the Arts for providing precipitation data. ERA-Interim and ERA-40 data were provided by ECMWF (Reading, United Kingdom; http://apps.ecmwf.int/datasets/). Satellite imagery was provided by NOAA (http://www.ncdc.noaa.gov/gibbs/year).