The different datasets of cyclones, fronts, thunderstorms, and anticyclones described above are then combined into a single dataset of weather types. First, the two cyclone and two front methods are each combined to make more robust datasets of cyclones and fronts that are common to two independent methods, which allows us better to identify those associated with significant rainfall. These are then combined with the thunderstorm datasets to identify seven different weather types that consider the interaction between co-occurring events, as per DC17. Finally, the full weather type dataset is combined with the rainfall datasets to identify the relationship of each weather type with southern Australian rainfall.
The combined cyclone dataset
We first combine the two distinct cyclone datasets to identify those cyclones that are found using both methods. To do so, each unique cyclone area from WS06 is assessed against the UM cyclone centre dataset to check if a cyclone centre is contained within the cyclone area. These cyclones are considered “Confirmed cyclones”, as they were identified by both methods, while areas that do not contain a UM cyclone centre are considered “Unconfirmed events”. As there is no perfect cyclone method, the terms confirmed and unconfirmed are not intended to indicate whether a cyclone is “real” or not, merely the level of agreement between the two methods. Areas where a UM cyclone centre is located within a 5° radius but there is no corresponding WS06 cyclone area are also considered “Unconfirmed events”.
Figure 1 shows the percentage of 6-hourly observations influenced by a cyclone area that is confirmed by both methods (left), in comparison to areas where only one method identifies a cyclone (middle and right). Both methods were developed for application outside the tropics, and there is generally strong agreement between the two cyclone identification methods south of about 25° S in both seasons, with both methods detecting the main storm track to the south of Australia, as well as the area of high cyclone frequency in the Tasman Sea.
Both cyclone methods identify relatively few cyclones in northern Australia during the cool season, and a large frequency of cyclones in northwestern Australia during the warm season, consistent with the high frequency of cyclones and tropical depressions (Lavender and Abbs 2013). However, there is also considerable uncertainty between cyclone methods in this region, with large numbers of observations where a cyclone is detected by only one method (Fig. 1e, f). Each cyclone identification method could also be detecting a variety of other weather systems that may not produce substantial rainfall, including the semi-stationary West Coast Trough (Kepert and Smith 1992), the monsoon trough, and the Pilbara heat low (Sturman and Tapper 1996).
To associate cyclones with rainfall, we then add an additional 5° area of influence beyond the definite (or unconfirmed) cyclone region. This is slightly larger than the 3° radius used in DC17 for examining the more intense rainfall events, but better matches the full region influenced by rainfall in cyclone composites for southeast Australia (e.g. Figure 7a in Pepler and Dowdy 2020). In the case of unconfirmed UM cyclones, this 5° region is added to the 5° cyclone area used for Fig. 1, for a 10° total radius of influence. This is similar to the radius of 10–12° from the cyclone centre used for attributing rain to cyclones by Hawcroft et al. (2012) and the ~ 10° region with rain from cyclones in (Pepler et al. 2018).
It is important to note that we have used a constant region in degrees of latitude/longitude to identify the area of influence of each weather system, for simplicity and consistency with previous studies e.g. DC17. This means that a given weather system will be influencing a smaller spatial region in the south of the domain than it does in the north. However, the frequency of undefined events is no higher in Tasmania than it is elsewhere in Australia (Sect. 3.4), suggesting our regions of influence are sufficiently broad to identify the majority of rainfall associated with a given weather system at all our latitudes of interest. This effect becomes more significant for polar regions, which are not included in this study.
Figure 2 shows the average rainfall anomaly on cyclone days compared to the mean rainfall across all days. Days with a confirmed cyclone are more likely to be associated with rainfall than cyclones identified using only one method, with the average rainfall on Confirmed cyclone days double the average daily rainfall for all days. Averaged across the country, 28% of confirmed cyclone days have rainfall of at least 1 mm, double the climatological likelihood of rainfall across all days (14%) and substantially higher than unconfirmed WS06 days (18%) or unconfirmed UM days (17%). Averaged across southern Australia (south of 25° S), for a given location 20% of days are influenced by a confirmed cyclone, but these days contribute on average 46% of the annual rainfall total (Table 1). Unconfirmed cyclones make a smaller contribution to rainfall, noting that unconfirmed cyclones may co-occur with other weather types.
Table 1 Annual average proportion of days with a cyclone present, and proportion of rainfall record on days with a cyclone present, for northern (north of 25° S) and southern (south of 25° S) While confirmed cyclones make a similar contribution to rainfall in northern Australia, there is also a large number of unconfirmed WS06 cyclones in this region, which make a lower contribution to total rainfall than confirmed systems (Table 1). While these results suggest the confirmed cyclone dataset improves on each individual cyclone dataset for detecting rainbearing lows in the tropics, there are larger uncertainties in northern Australia than southern Australia including the potential for both methods to miss small-sized systems such as some tropical cyclones and tropical depressions as both methods were developed to identify extratropical systems. Consequently, the primary focus of this paper is on southern Australia and the weather systems that are associated with rainfall in this region.
The confirmed cold front dataset
We similarly combine the two cold front datasets to identify cold fronts where a wind shift is combined with a wet bulb temperature change, as significant fronts are expected to satisfy both criteria (Hope et al. 2014). First, the WND dataset and the cold fronts identified using TFP are each expanded to a grid of frontal area, with a region considered to be influenced by a front if it is within a 5° radius of the front. This is a larger region than used by DC17 for their focus on more intense rainfall amounts, but better accounts for the full area of potential pre- and post-frontal rainfall as well as the movement of the front location throughout the 6 h period, as this study aims to consider all rainfall intensities. As with cyclones, regions influenced by both front datasets are considered “Confirmed cold fronts”, while areas influenced by only one method are considered “Unconfirmed cold fronts”.
Figure 3 compares the frequency of Confirmed cold fronts to those identified using a single method. While confirmed fronts have a clear decrease in frequency from a maximum in the storm track to the south of Australia to a minimum in the tropics, those fronts identified by a single method show very different spatial structures. Fronts only identified by the TFP are most common along the coastline, particularly during the warm season and in the afternoon (not shown), potentially reflecting stationary temperature gradients between the land and ocean in these areas. In comparison, the WND method identifies a large frequency of fronts in northwestern Australia, particularly overnight, which may be detecting diurnal changes in wind direction associated with the sea breeze.
Figure 4 shows the average rainfall anomaly on front days compared to the mean rainfall across all days. In southern Australia (south of 25° S), where cold fronts are more common, days where a cold front is identified using two different methods are more likely to produce rainfall than where a front is identified using only the TFP or only the WND method. While days with a Confirmed cold front tend to be wetter than average across the country, for most of southern Australia days where a cold front is identified by only one method are drier than the average across all days, and TFP-only fronts explain a smaller proportion of rainfall than expected from their frequency (Table 2). TFP fronts are particularly dry in parts of southwestern Australia, while WND-only fronts are particularly dry along the east coast.
Table 2 Annual average proportion of days with a front present, and proportion of rainfall record on days with a cyclone present, for northern (north of 25° S) and southern (south of 25° S) Australia In northern Australia cold fronts are typically not shown on Australian synoptic charts, and cold fronts are both less common and less important for total rainfall (Table 2). Interestingly, there are parts of the tropics where WND-only fronts have relatively high likelihoods of producing at least 1 mm of rainfall (Fig. 4c). This suggests that, while the WND method was designed for the extratropics, it may be able to detect some sort of squall lines or other systems of relevance to tropical rainfall, noting that midlatitude troughs and fronts have been identified as contributing to the occurrence of monsoon bursts (Narsey et al. 2017). However, the low rainfall rates shown in Fig. 4a and the large differences between methods shown in Fig. 3 for the tropics highlight the fact that neither front method gives a useful indication of rainbearing systems in the tropics. Due to the lower skill of both the front and the cyclone methods in northern Australia, the remainder of the paper will focus on southern Australia (south of 25° S) where cyclones and fronts can be more skilfully identified and where these weather types explain a large proportion of annual rainfall.
Weather types
The approach used for identifying combined types is similar to that used in DC17. However, the use of different cyclone, front and thunderstorm datasets, and particularly the exclusion of warm fronts from the combined results, will be expected to produce somewhat different frequencies of each weather type.
The confirmed cyclone, confirmed front and thunderstorm environments are first used to classify each location and time into one of seven combined weather types based on which of the cyclone, front or thunderstorm datasets are observed at that location and time. As well as observations influenced by a single weather system (Cyclone Only (CO), Front Only (FO), and Thunderstorm Only (TO)) there are four compound weather types: Cyclone + Front (CF), Cyclone + Thunderstorm (CT), Front + Thunderstorm (FT), and Cyclone + Front + Thunderstorm (CFT), called “Triple storm” in DC17. Note that a “Cyclone + Front” type requires a point to be simultaneously impacted by both a cyclone and front; this does not include fronts that may be connected to a distant low pressure system in the Southern Ocean.
The “Other” observations that are not classified as any of these seven rain-related weather types are then further subcategorised into.
-
1.
High: observations where a high pressure centre was located within a 10° radius, consistent with the region of dry conditions associated with Australian highs in Pepler et al. (2019a, b).
-
2.
Warm front (WF): observations with a warm front present using the TFP dataset.
-
3.
Unconfirmed events (Unconf): observations where a cyclone and/or cold front was present using one of the detection methods but not both.
-
4.
Undefined events (Undef): remaining weather types.
In addition to the 6-hourly dataset, to compare the weather types dataset against daily gridded rainfall data we create a daily version of the dataset. For this version, each individual weather type is first aggregated across each day, so that a day which had a front at any of the 4 observations that day (0000, 0600, 1200 or 1800UTC) is considered a front day. The classification process described above is then applied to the new daily dataset. This means that a day could be considered a "Cyclone + Front" day for a region if, for example, there was a cyclone detected at 0000UTC and a front detected at 1200UTC. Consequently, the frequency of combined event types is generally higher and the frequency of other and undefined events lower in the daily dataset (Fig. 5).
Compared to southern Australia, the weather types explain a smaller proportion of total observations in northern Australia, where cold fronts and anticyclones are uncommon and the majority of the year is dominated by prevailing easterly or westerly flow. In addition, Sects. 3.1 and 3.2 demonstrated that the cyclone and front methods, designed for the midlatitudes, are less consistent in northern Australia. Consequently, results will focus on the region south of 25° S where the dataset is expected to be more applicable.
Weather type rainfall
To calculate the rainfall associated with each weather type, we use the daily version of the synoptic type database, which aggregates the weather types at the four observations 0000-1800UTC. The 0.75° data is then converted to a 0/1 flag for each weather type and bilinearly interpolated to the 0.05° AWAP resolution, with values of 0.5 or higher used to indicate the presence of the type. Finally, the rainfall recorded at 9am local time on the subsequent day (equivalent to 2200-2300UTC in eastern Australia and 0100-0200 UTC in Western Australia) at each gridpoint in the AWAP analysis is attributed to the weather type present. Rain days are defined as experiencing ≥ 1 mm of rainfall, with additional thresholds of 10 mm and 25 mm used for moderate and heavy rain days.
For the 6-hourly rainfall from BARRA and station observations, rainfall is accumulated into four 6-hourly time periods each day (0000-0600, 0600-1200, 1200-1800 and 1800-0000UTC), with each rainfall observation compared to the weather types identified at both the initial and final observation, similar to the process used for daily rainfall data.