1 Introduction

Heatwaves are prolonged periods of extreme heat lasting at least three consecutive days (Perkins and Alexander 2013). They have the highest mortality rate of any natural disaster in Australia (Coates 1996) and cause substantial labour productivity losses (Zander et al. 2015). For example, in the most recent assessment of the economic burden of heat stress, Zander et al. (2015) showed existing risks associated with extreme heat represented an annual economic burden of around US$6.2 billion (0.33–0.47% of Australia’s GDP) for the Australian workforce. They can adversely impact agriculture by decreasing crop yield (Decker 1967; Farooq et al. 2016) and impact other sensitive ecosystems (e.g. Welbergen et al. 2008). Heatwaves affect human mortality and over 370 deaths were linked to the heatwave preceding the Black Saturday bushfires in 2009 (Victorian Department of Health 2009; Coates et al. 2014). Globally, heatwaves accounted for 148,000 of the 164,000 lives lost due to extreme temperatures and 92% of heatwave deaths occurred in high-income countries (CRED 2015). They have also caused infrastructure failures (McEvoy et al. 2012) and stress power supply (Colombo et al. 1999).

Since 1950, the intensity, frequency and duration of heatwaves have increased globally (Frich et al. 2002; Perkins et al. 2012) and over Australia (Perkins and Alexander 2013) at an accelerating rate in recent decades (Coumou and Rahmstorf 2012). Most recently, in 2016, Melbourne, Adelaide and Sydney all experienced record high day and night temperatures (Bureau of Meteorology 2016b). Sydney, for example, experienced 39 days in a row at or above 26 °C (Green et al. 2017).

The austral summer of 2015–2016 experienced one of the strongest El Niño events on record coincident with the warmest year on record globally (World Meteorological Organisation 2016). Australia also experienced above average temperatures (Bureau of Meteorology 2016a) and a record breaking heatwave in the southeast of the continent (Bureau of Meteorology 2016b). The El Niño–Southern Oscillation (ENSO), which consists of quasi periodic warm (El Niño) and cold (La Niña) phases of sea surface temperatures (SSTs) in the equatorial eastern Pacific, explains a large portion of the interannual variability of Australian heatwaves (Nicholls et al. 2005; Arblaster and Alexander 2012; Perkins et al. 2015; Loughran et al. 2017a). Understanding how ENSO influences the mechanisms responsible for heatwave development is important for predicting future seasonal heatwave behaviour, particularly in the context of a warming climate where hotter, longer and more frequent heatwaves are projected (Meehl and Tebaldi 2004; Cowan et al. 2014).

The three fundamental mechanisms of how heat accumulates in the atmosphere preceding and during Australian heatwave events are summarised by McBride et al. (2009). These mechanisms are: advection from lower latitudes; large-scale subsidence transporting higher potential temperature air from upper levels; and surface heating involving the development of the diurnal mixed layer, and replacement from below by the new mixed layer of the successive days.

The first mechanism refers to horizontal advection of heat from other regions, usually lower latitudes or hot and dry interiors of continental regions, driven by a slow moving anticyclone to the east of the heatwave affected area. The synoptic structure of anticyclones associated with southeast Australian heatwaves are well described as part of a baroclinic Rossby wave train that develops over the Indian Ocean and breaks near the Tasman Sea (Pezza et al. 2012). The sea level high pressure system is also associated with upper level anticyclonic potential vorticity that is advected into the region, and which also sustains the persistence of the anticyclone (Parker et al. 2014a). While the anticyclone is sometimes identified in these studies as a blocking high, Marshall et al. (2013) made the important distinction for the Australian case that they are in fact “persistent highs”. Persistent highs are located at a distinctly lower latitude than blocking highs, and do not feature a split in the polar jet stream. The difference in the synoptic patterns for each ENSO phase was explored for southeastern Australia by Parker et al. (2014b) who showed that despite the anticyclone being stronger during El Niño, Victorian heatwaves are more frequent during La Niña when the northerly air circulation and upper level potential vorticity anomalies are stronger. However the influence that ENSO has on the synoptic patterns for heatwaves in other regions of Australia has not been fully explored.

The second mechanism enabling heat to accumulate during a heatwave relates to vertical movement of air that is warmed adiabatically. For heatwaves in southeastern Australia, air tends to originate at high altitudes from the westerlies over the Southern Ocean, circulate around a high pressure system over the Tasman Sea, and descend to finally approach the heatwave affected region from the north (McBride et al. 2009; Boschat et al. 2014). As air descends it is warmed adiabatically. Over Europe, Bieli et al. (2015) used a Lagrangian approach to evaluate changes in potential temperature preceding hot extremes, and found that the adiabatic warming and diabatic heating contribution to temperature for air parcels that approach hot extremes can vary regionally. This approach has recently been applied to the state of Victoria in southeastern Australia by Quinting and Reeder (2017), who examined the origins of air masses and the sources of potential vorticity that sustains the anticyclone associated with heatwaves in the region. They identified a number of sources of potential heat and vorticity. For example, the temperature of air parcels approaching heatwave locations is raised equally strongly by adiabatic warming and diabatic heating processes. However, the exact role of the subsidence of potential heat from aloft during each ENSO phase is still unknown for any region of Australia.

The third heat accumulation mechanism relates to the diabatic heating of air leading up to and during a heatwave via land surface heating and associated feedbacks. During conditions of high soil moisture, turbulent energy transferred to the atmosphere from the land surface is predominantly in the form of latent heat via evaporation. Higher latent heating necessarily leads to less sensible heating due to energy balance constraints. On the other hand, dry soil conditions results in the surface energy balance favouring sensible heat fluxes, leading to a warmer and commonly drier atmosphere (Alexander 2011). A warmer and drier atmosphere tends to further desiccate the soil leading to a soil moisture-temperature feedback (Seneviratne et al. 2010; Miralles et al. 2012). As the surface energy balance tends towards higher sensible heating and lower latent heat fluxes, a progressive accumulation of heat can develop within the atmospheric boundary layer, increasing the potential for extreme temperature and increasing heatwave duration (Hirschi et al. 2010; Lorenz et al. 2010; Miralles et al. 2014). For Australia, soil moisture plays an essential role in heatwave development, but is ultimately insufficient on its own (Herold et al. 2016). This is particularly relevant to the north and eastern regions where ENSO has a dominating influence on precipitation and hence soil moisture. Several other studies have examined the importance of ENSO on Australian rainfall variability in station observations, reanalysis and climate models (e.g. Nicholls et al. 1997; Risbey et al. 2009; Klingaman et al. 2013; Ashcroft et al. 2014; King et al. 2015). While the link between rainfall, ENSO and heatwaves is well-known (Perkins et al. 2015), it is by no means consistent. For example, the 1997–1998 El Niño was one of the strongest on record but had only weak impacts in Australia (Lim and Hendon 2015).

While some basic mechanisms have been identified, and the Australian heatwave–ENSO relationship has been described and quantified, an understanding of how these three mechanisms relate to ENSO spatially across Australia is lacking. The aim of this study is to identify the effects of ENSO on the three heat accumulation mechanisms outlined above, and thereby contribute to the explanation of how ENSO can modulate heatwave development.

2 Methods

2.1 Definition of heatwaves

We identify heatwaves following Nairn et al. (2009) using the Excess Heat Factor (EHF) index, which is a heat measure based on a relative threshold. The EHF has two components, one compares a three day average of daily mean temperature to the previous 30 days average, and the other compares the 3 days average to a relative threshold. The relative threshold used here is the day-of-year 90th percentile. A heatwave occurs if the EHF index is greater than 0 for at least 3 days. Since ENSO typically peaks in December, analysis was restricted to the extended austral summer period from November–March, which is when heatwaves are hottest and have their greatest impacts. We define the heatwave frequency as the sum of heatwave days, the duration as longest heatwave, and the amplitude as the hottest day of the hottest heatwave, all within the summer period as previously defined by Perkins and Alexander (2013).

2.2 Modelling approach

To investigate how ENSO controls Australian heatwave variability, a set of model simulations were performed using a climate model, the Australian Community Climate and Earth-System Simulator (ACCESS) v1.3. ACCESS has a horizontal grid resolution of 1.25° latitude by 1.875° longitude, and 38 vertical levels on height-based terrain-following co-ordinates. We used ACCESS in an atmosphere and land surface only mode with prescribed SSTs. ACCESS uses the UK Meteorological Office unified model (UM) (Davies et al. 2005) atmosphere coupled to the CSIRO Community Atmosphere Biosphere Land Exchange (CABLE) version 1.8 (Kowalczyk et al. 2013). CABLE describes land surface processes such as heat, water and carbon fluxes, for 13 surface types and 6 layers. CABLE has been extensively evaluated both independently (Abramowitz et al. 2008; Wang et al. 2011) and as part of major intercomparison projects (Zhang et al. 2013; Best et al. 2015). ACCESS was included in the CMIP5 experiments as part of Working Group 1 of the Intergovernmental Panel on Climate Change’s 5th Assessment Report (IPCC 2013), where it has been shown its simulation are close to other models and observations (Bi et al. 2013). It has also been shown to simulate the impact of El Niño on Australian heatwave frequency reasonably well (Loughran et al. 2017b).

To undertake our experiments, we spun-up ACCESS for 50 years with constant year-2000-equivalent CO2 concentrations, and an annually repeating seasonal climatology cycle (average of 1970–1999) derived from the Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) monthly SST data set (Rayner 2003). The last 30 years of this spin-up were used as the control simulation because it features neutral ENSO conditions and no interannual SST variability. We constructed a set of 2 years ensemble simulations to sample ENSO events, instead of a single simulation with successive events. Ensembles are more independent and would not be consistently biased by existing atmospheric and land surface conditions from consecutive years. We restarted the atmosphere initial conditions from each 1st of January from the last 30 years of the control simulation to generate up to 30 ensembles per experiment group, which is a common method to sample internal variability in models (e.g. Deser et al. 2012; Arblaster et al. 2014).

The two experiment groups were forced with globally defined SSTs, featuring a single El Niño or La Niña event. The El Niño or La Niña event was derived from the HadISST monthly SST data set from 1970 to 2014 by compositing 2-year periods where the Niño 3.4 index reaches greater than 1 standard deviation (for El Niño) or less than 1 standard deviation (for La Niña) from the mean. The December–February mean of these composites are shown in Fig. 1 and demonstrate the spatial patterns (see Fig. 1a, b) and their temporal evolution (see Fig. 1c). Each event develops and decays over a 2-year period and reaches its peak in the December of the first year.

Fig. 1
figure 1

DJF austral summer composite of sea surface temperature anomalies for El Niño years a where the Niño 3.4 index is greater than 1 standard deviation from the mean, and b La Niña years where Niño 3.4 index is less than 1 standard deviation. c Is the corresponding time series on the equator in the Pacific (93.5°W). d Contains the regions of interest for compositing and trajectory initiation. The regions are north (129°E–136.5°E, 12°S–19.5°S), northeast (139°E–146.5°E, 18°S–25.5°S), east (145.5°E–153°E, 24°S–31.5°S), southeast (141°E–148.5°E, 31°S–38.5°S)

Full size image

2.3 Heatwave composites

We anticipate heatwaves in various regions of Australia will be affected in different ways by ENSO and we therefore summarise results based on four regions of interest (Fig. 1d). These regions were selected based on an approximation of the observed influence of ENSO on heatwaves taken from Loughran et al. (2017a), where principal component analysis was used to identify the northern (129°E–136.5°E, 12°S–19.5°S), northeastern (139°E–146.5°E, 18°S–25.5°S), and eastern (145.5°E–153°E, 24°S–31.5°S) regions as being most heavily affected by ENSO. For comparison, we also included the southeast region (141°E–148.5°E, 31°S–38.5°S), where large amplitude heatwaves occur but interannual variation is either weakly or inversely related to ENSO (Boschat et al. 2014; Parker et al. 2014b).

Composites are presented for heatwaves that occur within the four regions shown in Fig. 1d. Boschat et al. (2016) give caution regarding the formulation of hypotheses based on composites, and the implications of this will be explored later in the discussion section. Heatwave days are identified when at least half of the region’s area is affected by a heatwave which was done to remove single grid point exceedences of the EHF thresholds. Composites on these heatwave days were calculated for mean sea level pressure (MSLP) to ascertain changes in synoptic patterns and advection associated with ENSO. To ascertain the changes in the partitioning of net radiation associated with ENSO during heatwaves, we also composite the land surface turbulent surface sensible and latent heat fluxes (QH and QE respectively). A two-sided student’s t test was used to calculate statistical significance of the composites compared to the control climatology. Values are deemed significant when there is little difference between the experiment and control groups and the null hypothesis is rejected at the 5% level.

2.4 Observational data

To compare the simulations to observations we use the NOAA Twentieth Century Reanalysis v2 (20CR) dataset (Compo et al. 2011). This reanalysis uses the NCEP Global Forecast System model, assimilated with surface pressure observations from the International Surface Pressure Databank. This reanalysis was chosen because it allows a longer analysis period compared to other reanalyses, which is required when studying interannual variability. The analysis period used for these observations is 1901–2012 which contains 17 El Niño events and 17 La Niña events. Observed heatwave days are identified based on the EHF index (as above) using daily maximum and minimum temperatures at 2 m height, and the thresholds were calculated from the 1960–2012 base-period. As above, we calculated composites of summer heatwave days for MSLP, surface latent heat flux and sensible heat flux during La Niña or El Niño. El Niño and La Niña austral summer seasons are identified as in Sect. 2.2 using Niño 3.4 index derived from HadISST SSTs.

2.5 Backward trajectory analyses

Six hourly horizontal and vertical winds from the simulations were used for calculating 10 days backward air parcel trajectories to examine the vertical and horizontal origins of air approaching heatwaves and the evolution of potential heat. The trajectories were calculated using Traj3d, a three-dimensional spherical trajectory algorithm described in detail by Noone and Simmonds (1999) and Barras and Simmonds (2009). Sub-grid scale convection and circulations are not resolved in the ACCESS winds output data, therefore the effect of small scale winds on the trajectories cannot be accounted for. Temperature (T) and hydrostatic pressure (p) were tracked along the trajectories to calculate the potential temperature (θ) to reveal the diabatic heating that occurs as air approaches the heatwave region. The potential temperature along the trajectories was calculated as in Eq. (1), where p0 is the reference pressure of 1000 hPa and κ = R/cp = 0.286 (R is the gas constant of dry air and cp is the specific heat capacity).

$$\uptheta ={\text{ T}} {({{\text{p}}_0}/{\text{p}})}^\upkappa$$
(1)

The trajectories were initiated from the geographical centre of heatwave events such that there is one trajectory per heatwave day starting at 12 p.m. local time, and 950 hPa. This approach was used so that each trajectory is a reasonably independent sample of a heatwave day. We identified the co-ordinates for the centre of a heatwave event on any given day as the medians of the latitude and longitude of grid points that experienced an event. We also restricted the sample of heatwaves to large-scale spatial events (at least 9 grid points that occurred within each region in Fig. 1d). This eliminated single grid point events that can arise by chance and are not associated with an anticyclone to the east. Statistical significance in the median of the potential temperature and height of the El Niño and La Niña trajectories was calculated using a bootstrapping method. 10,000 sub-samples of the trajectories were taken with replacement to calculate a 5–95% confidence interval for the median. Significant differences in the median of the El Niño and La Niña experiments occur when their confidence intervals do not overlap. The robustness of the results was also tested against the entire distribution of trajectories (not just the median) using a Kolmogorov–Smirnov test, but similar significance patterns were found.

3 Results

Before analysing the effect of ENSO on the three mechanisms described in the introduction, we examine the ensemble means of heatwave frequency, duration and amplitude for El Niño and La Niña (Fig. 2). El Niño has approximately eight more heatwave days on average than the control simulation climatology, and La Niña can have about six fewer heatwave days on average (Fig. 2a, d). Statistically significant differences for frequency occur in the northwest, central inland and eastern areas of Australia for El Niño and most northern regions for La Niña. The duration of the longest heatwave (Fig. 2b, e) shows a similar pattern to frequency, with El Niño heatwaves being 2 days longer and La Niña being 2 days shorter than the control simulation. ENSO’s influence on heatwave amplitude (Fig. 2c) is less spatially uniform but heatwaves are generally warmer during El Niño in the southeastern region with a small area of relatively weaker heatwaves in the western side of southeast Australia. This pattern is approximately reversed during La Niña (Fig. 2f). However, the impacts of ENSO on heatwave amplitude are not statistically significant.

Fig. 2
figure 2

Ensemble means of heatwave aspects for a–c the El Niño experiment and d–f the La Niña experiment. Heatwave frequency is on the left (days), duration is the middle (days), and amplitude is on the right (C°2). Dotted contours indicate negative values. T-test statistical significance at the 5% level is indicated with cross hatching

Full size image

It is clear that the impacts of ENSO forcing on summer heatwave frequency and duration is prevalent for the north and eastern regions of Australia, typical of the observed impacts of ENSO from other studies (Perkins 2015; Loughran et al. 2017a). The spread of the ensemble members also varies across the north, northeast, east and southeast. Figure 3 demonstrates this by showing the area averaged number of heatwave days for each region in Fig. 1d. In the north and northeast regions, El Niño has a wider spread of heatwave days, while La Niña consistently has fewer. On the other hand, the eastern region has a broader spread during La Niña, and the southeastern region has a similar range of heatwave seasons between El Niño and La Niña. To examine how these impacts might arise, we will focus on each of the three mechanisms from the introduction in more detail.

Fig. 3
figure 3

Box and whisker plots depicting the ensemble spread of the number of heatwave days in each experiment, area averaged over each region in Fig. 1d. Red indicates the El Niño experiment and blue are the La Niña experiment. The black dot marks the control simulation climatology averaged for the northeastern region

Full size image

3.1 Surface energy balance and atmospheric heating

Figure 4a, b show the QH and QE November–March climatology for the control simulation, which demonstrates where in Australia sensible or latent heat fluxes are largest. QH ranges from 105 W m−2 in the southwest to 60 W m−2 in the northeast (Fig. 4a). The QE has greater variation across the continent: it is smallest in the southwest regions (15 W m−2) and increases gradually in a northeasterly direction, reaching a maximum of 100–120 W m−2 along the north and east coast of Australia (Fig. 4b). The greater soil moisture availability is driven by the nearby tropical conditions and the warm East Australian Current and this leads to higher precipitation and QE (eg. Chambers et al. 2014; Pepler et al. 2016).

Fig. 4
figure 4

November–March means from the control simulation for a sensible heat flux (contouring 20 W m−2 intervals), b latent heat flux (contouring every 20 W m−2 intervals), and c, d the corresponding differences between the ensemble means of the El Niño and La Niña experiments expressed as anomalies from the control simulation climatology. Dashed contours indicate negative values where the El Niño ensemble mean is less than the La Niña ensemble mean

Full size image

The difference between the ensemble means of the El Niño and La Niña experiments (Fig. 4c, d), shows how QH and QE vary with the ENSO phase. The El Niño ensembles tend to have a higher QH across the eastern half of Australia (12 W m−2 in the southeast and 24 W m−2 in the northeast, see Fig. 4c). This corresponds to a decrease in QE (− 8 W m−2 in the southeast and − 20 W m−2 in the northeast, see Fig. 4d). The changes to QE and QH in the western half of Australia are opposite, but much smaller in comparison to those in the east. In short, El Niño re-partitions the surface energy balance towards higher QH and less QE over the eastern parts of Australia, increasing the rate the surface warms the atmosphere which in turn tends to lead to warmer summers and the potential for more heatwaves.

Compared to the reanalysis, the variability of QH and QE in the model experiments is similar despite the existence of some climatological bias in the model. Figure 5a shows that the 20CR QH November–March climatological mean is approximately 100 W m−2 in the north and northeast, which is much higher than ACCESS. There is also slightly less QE in the reanalysis (Fig. 5b), although the spatial pattern is the same. Nevertheless, the difference between El Niño and La Niña years in the reanalysis (Fig. 5c, d) is similar to the model in spatial pattern and magnitude.

Fig. 5
figure 5

Reanalysis November–March 1960–2012 means for a sensible heat flux (contouring 20 W m−2 intervals), b latent heat flux (contouring every 20 W m−2 intervals), and c, d the corresponding differences between the means of the El Niño and La Niña years expressed as anomalies from the climatology. Dashed contours indicate negative values where the El Niño mean is less than the La Niña mean

Full size image

To examine these effects on heatwaves we will also explore the changes in surface heat fluxes that occur specifically on heat wave days. Figures 6 and 9 show the QH in ACCESS and 20CR composited for heatwaves that occur in the four regions shown in Fig. 1d, while Figs. 7 and 10 show the same for QE. The change in heat fluxes on heatwave days is characterised by significant and highly localised increases in QH around the heatwave region and moderate decreases elsewhere. In the north of Australia, this change is much larger in La Niña phase than El Niño (compare 30–50 W m−2 in Fig. 6a, b). As we look at regions in the northeast and moving south the magnitude of QH increase become more similar between El Niño and La Niña (Fig. 6c, e, g and d, f, h).

Fig. 6
figure 6

Modelled composites of sensible heat flux (both shading and contours at 10 W m−2 intervals) for heatwave days in a, b the north, c, d northeast, e, f east, and g, h southeast (see the rectangles in Fig. 1d). Values are expressed as anomalies from the control simulation climatology. The El Niño experiment is on the left and the La Niña experiment is on the right. The ‘n’ numbers in the top of each plot are the number of heatwave days used in each composite

Full size image
Fig. 7
figure 7

Modelled composites of latent heat flux (both shading and contours at 10 W m−2 intervals) for heatwave days in a, b the north, c, d northeast, e, f east, and g, h southeast (see the rectangles in Fig. 1d). Values are expressed as anomalies from the control simulation climatology. The El Niño experiment is on the left and the La Niña experiment is on the right. The ‘n’ numbers in the top of each plot are the number of heatwave days used in each composite

Full size image

The above increases in QH correspond to decreases in QE. The ACCESS QE composites on heatwaves days in Fig. 7 show significant and localised decreases in QE which mirror the anomalies in Fig. 6. For example, the anomalies for heatwaves occurring in the north are approximately − 30 W m−2 in El Niño and − 50 W m−2 in La Niña (Fig. 7a, b), which is proportional to the increases in QH. This means that on heatwave days the energy balance favours sensible heating which contributes to higher air temperatures. The larger heat flux anomalies in the La Niña experiments suggests that during La Niña strong surface heating is an important component in order to generate the extreme temperatures in otherwise adverse conditions as demonstrated in Fig. 2.

Compared to the north and northeastern regions, the east and southeast regions show some displacement in heat flux anomalies from the region where the heatwaves occur. The eastern region heat flux anomalies are situated slightly north from the respective compositing box (e.g. Figs. 6e, f, 7e, f). The southeast regions anomalies are the weakest of all the regions and are situated to the northeast of its respective compositing box (Figs. 6g, h, 7g, h). This may indicate a change in the dominant processes responsible for the development of the heatwaves. That is, the heatwaves in the east and southeast accumulate their heat from sensible heating in the north or northeast which is then advected towards the heatwave region driven by the Tasman Sea anticyclone. This is consistent with the findings of Quinting and Reeder (2017) who used similar trajectory analysis to identify that air masses advecting into heatwaves in southeastern Australia accumulate their heat from remote regions.

For all but the southeast region, there are more than double the number of heatwave days in the El Niño experiment compared to the La Niña experiment. For example, the eastern region has 255 heatwave days for the El Niño experiment, while the La Niña experiment had only 95. On the other hand, the southern region has 204 heatwave days for El Niño, and the La Niña experiment has not much fewer with 196 days. This agrees with observational studies that ENSO has relatively little influence on the number of heatwave days in the southeastern region (Boschat et al. 2014; Parker et al. 2014b).

While results from the La Niña experiments fairly consistently yield very few heatwave days, an El Niño event does not necessarily guarantee a summer season with more than average heatwave days (Fig. 3). The reason a given El Niño event has fewer than expected heatwaves may be because the soil is moist and the surface energy balance is dominated by QE rather than QH. Figure 8 shows the QH, QE and soil moisture of the differences between the summer ensemble means of 11 El Niño ensembles that had fewer heatwave days than the climatology in the northeast and the remaining 19 that had more (see black dot in Fig. 3). This demonstrates that for El Niño summers that have fewer than average heatwaves, there is a decrease in the QH (Fig. 8a) and an equivalent increase in QE (Fig. 8b), which is associated with an increase in the available surface soil moisture (Fig. 8c) and precipitation (Fig. 8d) (Seneviratne et al. 2010). Clearly, the concurrent and antecedent soil moisture conditions are important for predicting the association between ENSO and extreme heat in a given season (Kala et al. 2015).

Fig. 8
figure 8

November–March mean difference of the below average northeastern heatwave frequency El Niño ensemble members from the above average El Niño members, for a sensible heat flux (contouring interval 5 W m−2), b latent heat flux (contouring interval 5 W m−2), c surface level soil moisture (contouring interval 0.2 kg m−2) and d precipitation (contouring interval 0.5 mm s−1)

Full size image

In order to compare the effect of ENSO on QH and QE in ACCESS and observations, the observed heat fluxes from 20CR on heatwave days are shown in Figs. 9 and 10. They show general agreement in sign with the model simulations, but sometimes disagree with the simulations in the precise spatial distribution. For example, Fig. 9 shows that the northern region has weak local changes to QH in El Niño and strong increases in the northeast in La Niña (Fig. 9a, b), however the robustness of this is poor as it consists of only six heatwave days from two heatwave events. The northeastern region has a similar pattern, but weaker anomalies due to a slightly larger sample of heatwave days (Fig. 9c, d). The east and southeast regions contrast with the simulations especially during La Niña. Generally, QH during El Niño on heatwave days is higher than normal and less than normal during La Niña (Fig. 9e, h).

Fig. 9
figure 9

Observed, composites of sensible heat flux (both shading and contours at 10 W m−2 intervals) for heatwave days in a, b the north, c, d northeast, e, f east, and g, h southeast (see the rectangles in Fig. 1d). Values are expressed as anomalies from the November–March, 1901–2012 climatology. The El Niño years are on the left and the La Niña years are on the right. The ‘n’ numbers in the top right of each plot are the number of heatwave days used in each composite

Full size image
Fig. 10
figure 10

Observed composites of latent heat flux (both shading and contours at 10 W m−2 intervals) for heatwave days in a, b the north, c, d northeast, e, f east, and g, h southeast (see the rectangles in Fig. 1d). Values are expressed as anomalies from the November–March, 1901–2012 climatology. The El Niño years are on the left and the La Niña years are on the right. The ‘n’ numbers in the top right of each plot are the number of heatwave days used in each composite

Full size image

The observed QE on heatwave days is more consistent with ACCESS but still contrast somewhat with the simulations. Figure 10a–d shows significant decreases in QE for both El Niño and La Niña for heatwaves occurring the north and northeast. However, for heatwaves occurring in the southeast and eastern regions, the QE is less during El Niño and generally not significant during La Niña (Fig. 10e–h). The number of heatwave samples in the observations is much less than the simulations per ENSO phase. Additionally, there is greater spread (not shown) of QH and QE in the observations which would interfere with identifying a difference between El Niño and La Niña. Another factor is a variety of strong and relatively weak ENSO events in observations, while in the ACCESS experiments the ENSO events are all the same.

To summarise, the El Niño experiment exhibited higher climatological atmospheric warming than La Niña linked with the re-partitioning of the surface energy balance to increased QH and decreased QE, which led to more heating of the lower atmosphere and therefore a tendency to more extreme temperatures. On heatwave days specifically, the northern region had much larger QH and lower QE during La Niña. This was likely due to the importance of soil moisture in the region, which tends to suppress heatwave development even during El Niño when other conditions might be conducive to heatwave development. This effect could be seen in the observations to some extent, however there were several points of disagreement in the southern regions of Australia, and the lack of samples in observations would make conclusions tenuous.

3.2 Synoptic anticyclones and horizontal advection

We now examine the synoptic pressure systems that occur during heatwaves in each ENSO phase and how they might affect the synoptic patterns advecting heat towards each heatwave region. Figure 11 shows the synoptic MSLP anomaly patterns composited for heatwave days in each ENSO phase and region. The synoptic pattern associated with heatwaves depends on the region. During El Niño, heatwaves in the north, northeast and east (Fig. 11a, c, e) are characterised by anomalously high pressure to the northeast, and low pressure to the south. The La Niña phase does not feature high pressure to the north, but tends to maintain the low over the Southern Ocean (Fig. 11b, d, f). The lack of a consistent synoptic high anomaly in these regions for the La Niña experiment is due to the dominance of the Southern Oscillation on the pressure field. The composite for northern Australian heatwaves (Fig. 11b) also has a large high pressure system southeast of New Zealand, though it should be noted that this composite contains only 50 heatwave days. There are broad regions of high pressure in the Indo-Pacific region during El Niño and relatively lower pressure in the central equatorial Pacific, driven by the SST forcing. This pattern is reversed in the La Niña experiments. This might suggest that winds would flow from the west–northwest and the interior of the continent, but low latitude winds seldom follow the geostrophic balance. The corresponding wind barbs that are overlaid in Fig. 11 indicate that air flow is generally from the east, usually associated with a mid-latitude anticyclone.

Fig. 11
figure 11

Modelled mean sea level pressure anomaly (Pa, contours at 100 Pa intervals) composites for heatwave days in a, b the north, c, d northeast, e, f east, and g, h southeast. Values are expressed as anomalies from the control simulation November–March climatology. Contouring interval is 200 hPa. The El Niño experiment is on the left and the La Niña experiment is on the right. The numbers in the top right of each plot are the number of heatwave days used in each composite

Full size image

On the other hand, the southern region is dominated by an anticyclone situated over the Tasman Sea, geostrophically advecting air from the north and northeast (Fig. 11g, h). The high anomaly tends to be weaker in the La Niña experiment, helping to account for the fewer heatwave days compared to El Niño. For this region, the anomalies are generally only statistically significant in the subtropics and extratropics, which suggests that Southern Oscillation related subsidence and convection is unrelated to heatwaves in the southeast.

The MSLP patterns derived from 20CR are inconsistent with the simulations, likely because of the poor sample sizes. The MSLP associated with heatwaves in the northern region are weak and mostly not significant (Fig. 12a, b). The northeast and eastern region composite shows the increased pressure surrounding the northern parts of Australia associated with the El Niño phase of the Southern Oscillation, but the La Niña phase does not (Fig. 12c–f). The eastern and southeastern region La Niña phase does display the Tasman anticyclone (Fig. 12f, h), but not in the El Niño phase.

Fig. 12
figure 12

Observed composites of MSLP (Pa, contours at 100 Pa intervals) for heatwave days in a, b the north, c, d northeast, e, f east, and g, h southeast (see the rectangles in Fig. 1d). Values are expressed as anomalies from the November–March, 1901–2012 climatology. The El Niño years are on the left and the La Niña years are on the right. The numbers in the top right of each plot are the number of heatwave days used in each composite

Full size image

To summarise, the effect of ENSO on winds and pressure systems related to heatwaves, the phase of ENSO can weakly affect the pressure field near Australia through the action of the Southern Oscillation. But this would only affect the northern region, and does not much affect the positioning of the synoptic patterns that would advect warm air towards the heatwave. The Southern Oscillation and extratropical Rossby wave structures could not be consistently identified based on 20 CR MSLP due to small heatwave sample sizes and inconsistent ENSO variations, hindering the observation of changes to these structures for each phase. Therefore, it does not appear that ENSO could influence heatwave development by geostrophically advecting air from different regions.

3.3 Analysis of subsiding heat using back trajectories

While ENSO may have little effect on the synoptic structures, it may still impact the vertical movement of air which can contribute to a large amount of warming to the heatwave adiabatically. Figure 13 shows the tracks of the air parcels up to 10 days before the heatwave events for each ENSO phase and region, demonstrating where air parcels come from and where they descend. There are fewer trajectories than the heatwave days sampled in the composites because some heatwave events did not meet the requirement of at least nine grid points in size, or because the trajectory scheme advected the parcel below the surface of the earth and terminated it. This mostly affected trajectories for heatwaves in the northern region. Heatwaves in northern Australia (Fig. 13a, b) have air masses approach the heatwave region from the south to southeast, descending as they move north. Prior to that, the origins of air for northern heatwaves can be from the Pacific Ocean to the east or from far to the south. There does not seem to be a consistent path taken by air approaching northern heatwaves, but future studies may benefit from a trajectory clustering analysis if more samples are available (e.g. Harpaz et al. 2014).

Fig. 13
figure 13

10 day backward air parcel tracks for heatwave events in the EL Niño (left) and La Niña (right) experiments. Tracks are plotted for each region in Fig. 1d, a, b north, c, d northeast, e, f east, and g, h southeast. Line shading indicates the hydrostatic height of the trajectory in hPa

Full size image

For heatwaves in the northeast, eastern and southeastern regions (Fig. 13c–h) air approaches the heatwave from the north or northeast. These air masses originate from the Tasman Sea and rotate and descend around an anticyclone before arriving at the heatwave location. Most of the trajectories originate from the westerly winds anywhere between 800 and 400 hPa over the Southern Ocean. Many of them can circulate beyond New Zealand and over the south Pacific before approaching the heatwave in Australia. There appears to be no ENSO related difference in the horizontal origins of heatwave air for North and Northeastern heatwaves. However, there does appear to be a difference for the East Australian region (Fig. 13e, f). Many more trajectories originate from the Southern Ocean in El Niño, while more air originates from the South Pacific region during La Niña.

There are some differences in the vertical origins of air between ENSO phases. Figure 14 shows the medians of pressure along the trajectories for each region. The northern region does not show any significant differences in height (Fig. 14a). In the last 5–6 days before the heatwave, the air pressure level shows few significant differences between the ENSO phase. During days 10–6, heatwaves in the northeast tend to have air originate from lower in the atmosphere during El Niño (Fig. 14b). On the other hand, in the east and southeastern heatwaves region, air tends to originate from lower in the atmosphere during La Niña (Fig. 14c, d). This could be due to differences in the origins of air being advected into the Tasman Sea anticyclone, which may account for the differences in height. That is, during El Niño air originates from high over the Southern Ocean and descends in the Tasman anticyclone. On the other hand, during La Niña westerly flowing air enters the anticyclone from the subtropical ridge over the South Pacific at slightly lower latitudes and altitudes, where the climatological South Pacific anticyclone is stronger during La Niña, or weaker during El Niño (Trenberth 1976; Rasmusson and Carpenter 1982).

Fig. 14
figure 14

Median parcel height (hPa) for each ENSO phase along 10 day back trajectories starting at heatwaves in a the north, b northeast, c east, and d southeast. Shading indicates the 5–95% confidence interval in the median calculated using the bootstrapping technique. Significant differences in the median between ENSO phase occurs when the confidence intervals do not overlap. The ‘n’ number in the legend indicates the number of trajectories for each experiment group

Full size image

The adiabatic warming and diabatic heating that occurs to the air along the trajectories can be seen in the potential temperature in Fig. 15. Potential temperature at days 10–4 before the heatwaves undergoes minimal change for all regions. This is the temperature the air would have if it were reduced to 1000 hPa and therefore is the adiabatic component of the heat content of the heatwave. In the last 4 days, the potential temperature increases as it is heated diabatically within the boundary layer, as can be seen by the diurnal oscillations over this period.

Fig. 15
figure 15

Median potential temperature (K) for each ENSO phase along 10 day back trajectories starting at heatwaves in the a north, b northeast, c east, and d southeast. Shading indicates the 5–95% confidence interval in the median calculated using the bootstrapping technique. Significant differences in the median between ENSO phase occurs when the confidence intervals do not overlap. The n number in the legend indicates the number of trajectories for each experiment group

Full size image

There is no significant difference in the median potential temperature between ENSO phases for any region. It is possible that the distribution of potential temperature for one of the experiments could be bimodal. To account for this, a Kolmogorov–Smirnov test was also used to test the entire distribution of heights and potential temperatures, but the non-significance of the results remains, indicating that the distributions between each experiment are not significantly different. If ENSO affected heatwave development by warming the atmosphere globally and hence increasing the amount of potential temperature subsided from aloft, we might expect to see greater baseline potential heat in the El Niño experiments, but we do not see this in our results.

4 Discussion and conclusion

We have conducted a 30 member ensemble of ACCESS 1.3 experiments, forced with SSTs containing El Niño and La Niña events, and analysed their effects on three heatwave development mechanisms. The simulations displayed increases in heatwave frequency and duration for northern and eastern Australia during El Niño and decreases for La Niña, similar to the observed impacts of ENSO in Perkins et al. (2015). The El Niño experiments are characterised by summertime decreases in QE, compared to the La Niña experiment, which drives more frequent extreme temperatures through sensible surface heating. Meanwhile, maximum heatwave temperatures in the northeast are not significantly different between ENSO phases. On heatwave days, the northern regions of Australia require much larger changes in surface fluxes for a heatwave to occur during the La Niña experiment than El Niño, likely due to the dominating effect of soil moisture in the region and the tendency for wet conditions during La Niña. In fact, existing soil moisture conditions can potentially suppress heatwaves in the northeast, regardless of the occurrence of an El Niño. Hence, surface moisture and energy fluxes are an important mechanism through which ENSO modulates the frequency and duration of north and northeast Australian heatwaves. The MSLP patterns and horizontal advection along trajectories showed little difference between each ENSO phase and the direction from which hot air approaches a heatwave depended primarily on which region that was being affected. This was also true for the trajectory analysis on the subsidence of potential temperature, which did not show any significant differences between ENSO phase for the north and northeastern regions. On the other hand, the trajectories from heatwaves in the eastern and southeastern regions showed stronger subsidence during El Niño, but equivalent potential temperature along the trajectories. Because the El Niño trajectories tend to begin from higher altitude there must therefore be more potential heat aloft that is being subsided through the anticyclone. The difference in the subsidence during each ENSO phase is also evident in the horizontal origins in heatwave air parcel trajectories, whereby air parcels originating from midlatitudes experience stronger adiabatic warming due to the poleward tilt of the isentropes, compared to those originating from the subtropics (Fig. 13e, f).

We recognise that caution needs to be applied in interpreting our finding that the La Niña phase must have stronger diabatic surface heating in order for heatwaves to occur. Boschat et al. (2016) argues against the sole use of composite analysis for the formulation of hypotheses, and shows that the spatial pattern of a field produced by composites can often occur without the occurrence of heatwaves. In the context of our study, composite analysis alone cannot infer that the decrease in QE anomalies causes the heatwave days, and that the occurrence of the QE anomalies does not necessarily mean that a heatwave will occur. It is plausible that a heatwave may not occur despite the existence of dry land surface conditions and strong surface heating of the atmosphere, especially if the conditions do not persist for more than 3 days. As a result, we do not conclude that the decreased QE necessarily causes heatwaves, or that the lower QE in La Niña causes more heatwaves (clearly the opposite is true, as demonstrated by Fig. 4). Instead, the composites suggest that the decrease in QE and enhanced diabatic heating on heatwave days tends to be stronger during La Niña, but there are multiple factors that contribute to the occurrence of heatwaves. The timing and intensity of rainfall is therefore crucial for the suppression or development of heatwaves during each ENSO phase. Nevertheless, there may be many other adverse conditions for heatwaves that account for the decreased occurrence during La Niña, such as a generally cooler atmosphere, or weaker subsidence in the northern region.

There remain some issues with the inconsistency of the observed and modelled heat fluxes, whereby the model produces larger composite anomalies that are more concentrated in a specific region, when compared to the observations. Firstly, the combined effect of having fewer observed samples, larger variance and inconsistent ENSO events as highlighted in the result section, makes comparing the observed heat fluxes to the models somewhat difficult. Alternatively, the model itself may be overestimating the magnitude and underestimating the spatial extent of the heat flux anomalies. Similar problems would also apply to the composites for MSLP. Therefore, future studies for observed impacts of ENSO on heatwaves should choose an approach (e.g., case studies) that is not affected by averaging of a small number of samples, as is the case for compositing.

An extraneous variable that might influence the occurrence of heatwaves in Australia is the Southern Annular Mode (SAM), which relates to the strengthening and southerly movement of the circumpolar westerlies during the positive phase. This brings cooler, wetter conditions to eastern Australia (Hendon et al. 2007; Risbey et al. 2009; Marshall et al. 2012). Multiple studies have noted a relationship between the ENSO and SAM, whereby the westerlies move south (positive SAM) during La Niña, or north during El Niño (Karoly 1989; Seager et al. 2003; L’Heureux and Thompson 2006). Therefore, the SAM conditions within ACCESS may co-vary with ENSO and also influence the occurrence of heatwaves. Table 1 lists the December–February mean of the SAM using the Antarctic Oscillation index (AOI) outlined in Gong and Wang (1999). This index consists of the difference in normalised zonal mean sea level pressure at 40°S and 65°S. The ensemble mean AOI in the El Niño experiment is − 0.1, and − 0.04 in the La Niña experiment. A strong positive SAM would be indicated by an AOI of 1, but both experiments ensemble mean SAM conditions are weakly negative. Therefore there is no preference for a particular phase of SAM in either experiment. Neither is there a preferred phase of SAM for the 11 El Niño (the first 11 in Table 1) ensemble members that had fewer than the expected number of heatwave days.

Table 1 December–February southern annular mode index for each El Niño and La Niña ensemble member
Full size table

This study focused on using a climate model to conduct experiments, and so comparison to other observational studies is needed in future studies to evaluate the reliability of the simulations. A key method of this study examines the role of surface heat fluxes in heatwave variability, but there are no surface flux observations for Australia that cover a long enough period to allow an analysis of ENSO conditions. The reanalysis used simulates rainfall and land surface processes, but the comparison being made is really one model with another. Nevertheless, the importance of rainfall and soil moisture in the development or suppression of heatwaves and extreme temperatures is apparent in observations (Herold et al. 2016) and in a regional climate model study (Kala et al. 2015). These studies found that the southeast regions that have low summer precipitation and soil moisture (as contrasted to the northern regions) are more strongly influenced by the synoptic climatology and climate variability modes other than ENSO. This generally agrees with our results whereby the advection of heat from lower latitudes towards southeast Australia contrasts with other northerly regions where ENSO dominates heatwave variability. The path of trajectories presented in this study is similar to previous observational studies that have examined different research questions but used similar methods. Parker et al. (2013), Boschat et al. (2014) and Quinting and Reeder (2017) demonstrate that air parcels ending at heatwaves in southeastern Australia originate over the Southern Ocean and circulate anticyclonically around the Tasman Sea. Our results demonstrate that this is true for the other eastern regions of Australia, but also that the air flows from a southeasterly direction for heatwaves in the north. It is also encouraging that other studies have found that the magnitude of the observed median diabatic heating of air approaching hot extremes is about 10 K. This is true for southeastern Australia (Fig. 10b of Quinting and Reeder 2017) and the similar latitude of central Europe (Fig. 6e of Bieli et al. 2015), both of which correspond to the 11 K in Fig. 15c.

While there is generally good agreement of the ACCESS model to observational studies, the Traj3d tracking scheme suffers from some of the common limitations of Lagrangian tracking schemes. For example, because of the discrete wind field and advection method, the calculated trajectory would have some error compared to a “true” path advected continuously in a real wind field. This would also make the advection non-reversible, such that back trajectories would not be identical to the forward trajectories. Furthermore, sub-grid scale process are not explicitly resolved and might have a detectable effect on heat accumulation along the trajectories. Using a high resolution regional climate model may address these problems and provide further insight into the mechanisms of heat accumulation in the atmosphere.

This study may also provide a framework for investigating how other climate variability modes might affect heatwaves. The Indian Ocean Dipole (IOD) is another mode of variability based on SST and air sea interaction that may influence heatwaves in Australia and an investigation of it may benefit from the methods used here. Other climate modes, such as the SAM and Madden–Julian Oscillation, may also interact with ENSO to produce heatwave seasons that are inconsistent with the usual impact of ENSO and should be studied further. There is also some suggestion that ENSO variants can have unique impacts on rainfall in Australia, and this may lead to differences in temperature extremes; however, Loughran et al. (2017b) have recently shown that this is unlikely for heatwaves.

In conclusion, this study has used the ACCESS climate model to simulate the heatwave response of the atmosphere to ENSO forcing, and examined how ENSO affect three heatwave development mechanisms. While the three mechanisms are all necessary for the development of a heatwave event, we have highlighted the relevance of each for different regions of Australia in the development of heatwave variability during ENSO phases. Our findings suggest that different regions of Australia are affected by ENSO in different ways, with the north and northeast being dominated by variability of land surface processes, while the east and southeast are more strongly affected by advection and subsidence of potential heat. Also, despite prevailing El Niño conditions, high soil moisture conditions may mitigate the severity of heatwaves in the north and northestern regions. This knowledge may provide useful insights to seasonal range predictions for heatwaves, and possibly shorter duration extreme heat events. Future research might focus on the role of other climate modes in heatwave development as these may affect heatwaves in similar or contrasting ways.