Interactive influence of ENSO and IOD on contiguous heatwaves in Australia

To identify heatwaves, we employ the daily maximum and minimum temperature data from the Australian Water Availability Project (AWAP) observational gridded dataset (Jones et al 2009). The gridded dataset is developed using the hybrid gridding algorithm on high-quality station data across Australia (Jones et al 2009). The AWAP data is available at both 0.05° and 0.25° horizontal resolutions. We use the 0.25° spatial resolution data for the analysis because of computational limitations and also because it has been suggested that this decrease in grid resolution does not affect the spatial pattern and magnitude of heatwaves (Perkins and Alexander 2013). We use the observed Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST) monthly global sea surface temperature (SST) data (Rayner et al 2003) for the identification of climate variability modes (ENSO and IOD) during 1958–2020.

Due to the small observational sample size, we obtain daily maximum and minimum temperature and monthly SST data from the CNRM-CM6-1 SMILE (Voldoire et al 2019) of CMIP6 for the historical period 1958–2014 (November 1958–March 2014). Out of the 30 ensemble members (r(1–30)i1p1f2) of CNRM-CM6-1 SMILE, r2i1p1f2 and r25i1p1f2 members data is not considered for the analysis due to large errors in the calculation of contiguous heatwaves compared with the observational data (further details provided in supplementary material (available online at stacks.iop.org/ERL/17/014004/mmedia)). These 28 ensemble members of CNRM-CM6-1 SMILE provides a large sample size to robustly investigate the interactive influence of ENSO and IOD on contiguous heatwaves in Australia (Schaller et al 2018, Lehner et al 2020). We choose CNRM-CM6-1 SMILE because it performs well in simulating extreme temperatures across Australia (Deng et al 2021), and its horizontal resolution is finer (1.4008° × 1.40625°) compared to other SMILEs of CMIP6. A finer resolution is required to effectively capture the spatial variation of contiguous heatwaves across various climatic regions and coastlines of Australia, where a relatively coarser resolution would provide fewer grid cells, if any at all (Perkins et al 2014).

We defined heatwaves at each grid cell using the excess heat factor (EHF) (Nairn and Fawcett 2013), which has been used in many previous Australian heatwave studies (Perkins et al 2015, Reddy et al 2021a). The EHF is computed according to Reddy et al (2021a) and is expressed in the supplementary information (equations (S1)–(S4)). The EHF based heatwave definition identifies a heatwave as a period of three or more consecutive days with a positive EHF value (>0). However, recent research shows that heatwaves can continue after minor interruptions of cold days, and these heatwaves could have similar effects to those that persist through consecutive days (Baldwin et al 2019). To overcome this limitation and following Baldwin et al (2019) and Rastogi et al (2020), we allow for minor breaks in our heatwave definition. Therefore, a heatwave at a given grid cell is defined as a period of at least three consecutive days with an EHF value greater than zero, and the heatwave is considered to continue until at least two consecutive days with EHF value less than or equal to zero are recorded.

Contiguous heatwaves are identified using the three-dimensional connected component algorithm (Silversmith 2021). First, the gridded temperature data is converted into binary format, with a '1' assigned to grid cells that satisfy the heatwave criteria described above, and a '0' assigned to those grid cells that do not. Then, the three-dimensional connected component algorithm is applied to the gridded binary data by setting the connectivity as 26, which considers all the surrounding neighbours (9 in the previous day + 8 in current day + 9 in next day) by both adjacent and diagonal connectivity (Vogel et al 2020 see figure 1). This allows tracking the contiguous heatwave simultaneously, both in space and time dimensions. This algorithm also allows a contiguous heatwave event to break into parts and connect again providing they are contiguous either in space or in time. Figure 1 shows the tracking of contiguous heatwaves from 23 January 2009 to 09 February 2009, where each colour represents the different individual contiguous heatwave. The large and long event (represented in red, shown in figure 1) started as a small event on 23 January 2009 and through time grew bigger, moved in space, broke into parts, and connected again and vice versa. During the period 23 January–09 February 2009 several other contiguous heatwaves evolved in space and time, which are shown in different colours (see figure 1).

Zoom In Zoom Out Reset image size

Contiguous heatwaves can be quantified by their total area, maximum area, median duration (Vogel et al 2020), and areal magnitude (Keellings and Moradkhani 2020). They are expressed as follows:

• Total area: The sum of the areal extent of a contiguous heatwave across all days.
• Maximum area: The maximum areal coverage of a contiguous heatwave in its duration.
• Median duration: The median value of the duration of all grid cells associated with the contiguous heatwave.
• Areal magnitude: Computed similarly as heat severity and convergence index defined by Keellings and Moradkhani (2020):

$$\begin{align}{\text{Areal magnitude}} = \sum\limits_{j = 1}^m {\left( {{{\left( {\frac{{\sum\nolimits_{i = 1}^n {{\text{EH}}{{\text{F}}_i}} }}{n}} \right)}_j} \times {a_j}} \right)} \end{align} \tag{ 1 }$$

where ${\text{EH}}{{\text{F}}_i}$ is the EHF value corresponding to the grid cell $i$, $n$ represents the total number of grid cells associated with the heatwave on a particular day $j$, ${a_j}$ represents the proportion of land area coverage of a heatwave event on a particular day $j$, $m$ is the heatwave duration measured in days.

We categorised contiguous heatwaves as large and extremely large events using their total area metric. A heatwave is classified as large or extremely large if its total area exceeds the climatological 80th or 90th percentile value of the selected base period (1961–90), respectively. We focus on analysing the changes in large or extremely large contiguous heatwaves because previous studies have indicated large contiguous heatwaves potentially cause adverse societal impacts such as increased electricity supply-demand, overwhelming pressure on emergency services, and food supply-demand rise in other regions (Lyon et al 2019, Vogel et al 2020), and which will likely have similar impacts for Australia.

The contiguous heatwave metrics of individual events were calculated using the AWAP gridded data across Australia in the extended heatwave season, i.e. from November to March during the period 1958–2020. We used the bootstrapped two-sample Kolmogorov-Smirnov (KS) test (Sekhon 2011) (further details provided in the supplementary material) to identify significant changes or shifts in large and extremely large contiguous heatwave metrics between the first half (1958/59–1988/89) and the second half (1989/90–2019/20) of the study period. We also calculated the seasonal contiguous heatwave metrics of all, large and extremely large heatwaves to see how they are changing through time. The seasonal contiguous heatwave metrics are calculated as follows:

• Accumulated areal magnitude: Seasonal sum of areal magnitude across all contiguous heatwaves (Keellings and Moradkhani 2020).
• Aggregate total area: Comprehensive sum of all individual contiguous heatwave's total area in a season.
• Maximum of maximum area: Maximum value of all individual contiguous heatwave's maximum area in a season.
• Mean of median duration: Average value of all individual contiguous heatwave's median duration in a season.

These seasonal metrics are used to visualise and capture the temporal changes of the respective metric, and in the rest of the study, we considered the individual contiguous heatwave metrics for the analysis.

The single realisation or the ensemble mean of the model will not be able to exactly reproduce the observations but can provide the better estimates of temporal variability (Wood et al 2021). Here, to evaluate the model's skill in representing the temporal variability of contiguous heatwaves, we considered the evaluation method followed by the Maher et al (2019), Suarez-Gutierrez et al (2020) and Wood et al (2021). By following this method, we compared the observed seasonal anomalies of contiguous heatwave metrics with the anomalies of the CNRM-CM6-1 SMILE members. Anomalies were calculated relative to the selected base period 1961–90. The model is considered to be overestimating the temporal variability if the observed anomalies cluster mostly (more than 85% of the time) in the 75th percentile range (i.e. 12.5th–87.5th percentile) of the ensemble and many (more than 15%) observations located outside the ensemble range (0–100th percentile) indicates the underestimation of variability. The model is considered to be in a good agreement with the observations in terms of representing the variability when only a few observed anomalies (less than 15%) fall outside the ensemble range and observations should not cluster most of the time (less than 85%) in the 75th percentile range of the ensemble.

We used both the HadISST (Rayner et al 2003) and CNRM-CM6-1 SMILE (Voldoire et al 2019) detrended monthly SST data for the calculation of ENSO and IOD indices. The Niño3.4 SST index is used to quantify ENSO (Trenberth 1997). The anomalies were calculated relative to the base period 1961–90. Perkins et al (2015) found that the strength of ENSO is significantly correlated with heatwaves (not necessarily contiguous) over much of Australia. To accommodate the strength variations of ENSO in our analysis, we classified ENSO into two categories (moderate and strong events) according to Casselman et al (2021). Moderate events are defined as ±0.5–1 standard deviation ($\sigma$) of mean November–February Niño3.4 index, and strong events are defined as the exceedance of ±1$\sigma$ of mean November–February Niño3.4 index (see equation (2)):

$$\begin{align}{\text{ENSO phase}} = \left\{\!\! {\begin{array}{*{20}{l}} {{\text{strong}}\;{\text{El}}\;{{{\text{Ni}}\tilde{\text{n}}{\text{o}}}},\qquad\quad{{{\text{Ni}}\tilde{\text{n}}{\text{o}}}}3.4 > 1\sigma } \\ {{\text{moderate}}\;{\text{El}}\;{{{\text{Ni}}\tilde{\text{n}}{\text{o}}}},\qquad\!0.5\sigma < {{{\text{Ni}}\tilde{\text{n}}{\text{o}}}}3.4 \leqslant 1\sigma } \\ {{\text{ neutral}},\qquad\qquad\qquad - 0.5\sigma \leqslant {{{\text{Ni}}\tilde{\text{n}}{\text{o}}}}3.4 \leqslant 0.5\sigma } \\ {{\text{moderate}}\,{\text{La}}\,{{{\text{Ni}}\tilde{\text{n}}{\text{a}}}},\qquad - 0.5\sigma > {{{\text{Ni}}\tilde{\text{n}}{\text{o}}}}3.4 \geqslant - 1\sigma } \\ {{\text{strong}}\;{\text{La}}\;{{{\text{Ni}}\tilde{\text{n}}{\text{a}}}},\qquad\quad{{{\text{Ni}}\tilde{\text{n}}{\text{o}}}}3.4 < -1\sigma } \end{array}} \right..\end{align} \tag{ 2 }$$

IOD is quantified using the dipole mode index (DMI) (Saji et al 1999). Similar to ENSO, we classified IOD as moderate and strong events to incorporate the strength variations of IOD in our analysis. Moderate events are defined as ±0.5–1 standard deviation ($\sigma$) of mean September–November DMI, and strong events are defined as the exceedance of ±1σ of mean September–November DMI (see equation (3)):

$$\begin{align} {\text{IOD phase}} = \left\{\!\!\!\!\! {\begin{array}{*{20}{l}} {{\text{strong IOD positive}},\qquad\quad {\text{ DMI}} > 1\sigma { }} \\ {{\text{moderate IOD positive}},\qquad{ }0.5\sigma < {\text{DMI}} \leqslant 1\sigma } \\ {{\text{ neutral,}}\qquad\qquad\qquad\qquad\,\,\, - 0.5\sigma \leqslant {\text{DMI}} \leqslant 0.5\sigma } \\ {{\text{moderate IOD negative}},\qquad\! - 0.5\sigma > {\text{ DMI}} \geqslant - 1\sigma } \\ {{\text{strong IOD negative}},\qquad\quad{\text{ DMI}} < -1\sigma { }} \end{array}} \right..\end{align} \tag{ 3 }$$

We considered the co-occurring modes of ENSO and IOD to identify their relationship with contiguous heatwaves because these modes can interact with each other (Meyers et al 2007, Risbey et al 2009, Ummenhofer et al 2011). The above-mentioned ENSO and IOD classification resulted in 25 combinations (5 × 5) of phases of co-occurring modes (table 1). For example, the season is considered a strong El Niño active season (E_s) only when ENSO is in a strong El Niño phase and IOD in a neutral phase. If ENSO is in a strong El Niño phase and IOD in a strong positive phase, it is considered a co-occurring event strong El Niño + strong IOD positive (E_s-Ip_s). The season is considered neutral (NN) when both the ENSO and IOD are in their neutral phases, respectively. If ENSO is in a strong La Niña phase and IOD in a moderate negative phase, it is considered a co-occurring event strong La Niña + moderate IOD negative (L_s-In_m) (table 1).

To omit the climate change signal and solely focus on investigating the interactive influence of ENSO and IOD on contiguous heatwaves, we detrended the contiguous heatwave metrics data by removing the linear trends. We used the non-parametric permutation test to assess the statistical significance of the difference in detrended mean contiguous heatwave metrics during the NN phase and various co-occurring phases of ENSO and IOD. This test does not assume any distribution and can be appropriately used to assess the significance of non-normally distributed contiguous heatwave metrics data. We set the permutations to 10 000 times and the significance level as 0.05 (Singh et al 2021) (for further details, refer to supplementary material). We also explored the underlying mechanisms associated with various co-occurring phases of ENSO and IOD that influence contiguous heatwaves in Australia during extended summers by analysing the composite maps of detrended mean sea level pressure (MSLP) and geopotential height at 500 hPa (z500). The composite maps are generated as the difference of detrended MSLP/z500 between the extended summers of various co-occurring phases of ENSO and IOD and NN phase.

Observed temporal changes of seasonal contiguous heatwave metrics of all heatwaves, large heatwaves, and extremely large heatwaves, respectively, during 1958–2020 in Australia are presented in figure 2. Results show that seasonal values of the maximum of maximum area metric are the same across all categories of contiguous heatwaves (all, large, and extremely large) (figure 2). This means that extremely large contiguous heatwaves with a massive total area also have the maximum areal coverage on a particular day in its duration, which is the maximum of all contiguous heatwaves in a season. Minor variations in seasonal values of accumulated areal magnitude and aggregate total area metrics are observed because the areal magnitude and total area of events other than large and extremely large contiguous heatwaves are minimal. However, seasonal metrics (accumulated areal magnitude, aggregate total area, maximum of the maximum area and mean of median duration) of the three size categories are increasing through time, with a more pronounced increase during the recent decade (figure 2). This substantial increase in recent decades is consistent with previous studies (Perkins-Kirkpatrick and Lewis 2020, Reddy et al 2021a), which highlighted a rapid rise in seasonal heatwave metrics over Australia.

Zoom In Zoom Out Reset image size

Figure 3 shows changes in the individual large and extremely large contiguous heatwaves during 1989/90–2019/20 compared to 1958/59–1988/89. We find that most contiguous heatwave metrics of extremely large heatwaves and the median duration of large heatwaves show a significant change in 1989/90–2019/20 relative to 1958/59–1988/89 (figure 3). This implies that extremely large heatwaves are growing larger and are lasting longer. Previous research on Australian heatwaves found that heatwaves in Australia are mainly driven by modes of natural climate variability, global warming, and synoptic systems (Pezza et al 2012, Perkins 2015). This suggests that the observed increase in extremely large heatwaves could be due to the combination of climate variability, global warming, and large and persistent synoptic systems. Further research is required to identify the causes behind the increase in extremely large contiguous heatwaves in recent decades compared to the past, which may consider investigating the influence of multidecadal climate variability (such as Interdecadal Pacific Oscillation (IPO)) and global warming on Australian contiguous heatwaves.

Zoom In Zoom Out Reset image size

The larger interannual variability of all considered seasonal contiguous heatwave metrics is seen during 1958–2020 and is more pronounced during the recent decade (figure 2). Previous studies have linked variability in heatwave metrics to climate variability (Perkins-Kirkpatrick et al 2017). Hence, here we investigated the influence of natural climate variability modes (ENSO and IOD) on large and extremely large contiguous heatwaves. As previous studies suggested that ENSO and IOD interact with each other (Meyers et al 2007, Risbey et al 2009, Ummenhofer et al 2011), we consider co-occurring phases rather than individual phases for further analysis (see section 2.4). The CNRM-CM6-1 SMILE data provides a larger sample of co-occurring modes (table 1) relative to observations. We evaluated the performance of the CNRM-CM6-1 SMILE in simulating temporal variability of the seasonal contiguous heatwave metrics against the observations. Results suggested that the CNRM-CM6-1 SMILE is able to skilfully represent the temporal variability of the aggregate total area (figure 4), accumulated areal magnitude (figure S1), maximum of the maximum area (figure S2) and mean of median duration (figure S3) of contiguous heatwaves in Australia.

Zoom In Zoom Out Reset image size

Figures 5 and 6 show the variations of detrended contiguous heatwave metrics of large and extremely large events of co-occurring variability modes. The results obtained without detrending are presented in the supplementary figures S4 and S5. However, detrending the contiguous heatwave metrics had minimal effect on the results. Results suggest that areal magnitude, maximum area, total area, and median duration of both the large and extremely large contiguous heatwaves are significantly higher during the strong El Niño (E_s), strong El Niño co-occurring with strong IOD positive (E_s-Ip_s), and with moderate IOD positive (E_s-Ip_m) seasons relative to the neutral seasons (where both ENSO and IOD are in neutral phase; NN) (figures 5 and 6). This means that E_s, E_s-Ip_s, and E_s-Ip_m conditions significantly influence the increase in size (spatial extent) and duration (temporal length) of these events. These results are in line with those from previous studies, which found that El Niño seasons are associated with dry conditions, above-normal temperatures, and more heatwave days over much of Australia (Arblaster and Alexander 2012, Perkins et al 2015). In addition to El Niño seasons, the drier conditions are also observed during co-occurring El Niño and IOD positive seasons over many parts of Australia (Meyers et al 2007, Ummenhofer et al 2011). During these phases, the abnormally dry conditions caused by both of these phases likely favour heatwave conditions and thus result in significant increases in extremely large contiguous heatwaves.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The co-occurring strong La Niña and strong IOD negative (L_s-In_s) seasons are associated with the most significant decrease in most of the considered metrics of the large and extremely large contiguous heatwaves compared to the NN seasons (figures 5 and 6). Both the La Niña and IOD negative modes are associated with below normal temperatures and enhanced rainfall conditions over many parts of Australia (Pepler et al 2014, Perkins et al 2015). Their interactions (La Niña and IOD negative) likely exacerbate wet conditions (Meyers et al 2007, Ummenhofer et al 2011), which are not favourable for heatwaves. Co-occurring modes such as strong El Niño with strong IOD negative (E_s-In_s) and moderate IOD negative (E_s-In_m), moderate El Niño with strong IOD negative (E_m-In_s) and moderate IOD negative (E_m-In_m) exhibit a significant association with some of the considered metrics of large and extremely large contiguous heatwaves compared to the NN conditions. However, there is a limited sample of data during these co-occurring modes (E_s-In_s, E_s-In_m, E_m-In_s, and E_m-In_m), and as such, these results might not be statistically robust. The observations and CNRM-CM6-1 SMILE data suggest that the combinations of El Niño co-occurring with IOD negative (E_s-In_s, E_s-In_m, E_m-In_s, and E_m-In_m) and La Niña with IOD positive (L_s-Ip_s and L_s-Ip_m) occur very rarely, consistent with previous studies (Meyers et al 2007, Ummenhofer et al 2011).

Larger variability of heatwave metrics is seen across various co-occurring modes of ENSO and IOD, even when their mean values are significantly higher compared to the neutral conditions (figures 5 and 6). The high probability of occurrence of large and extremely large contiguous heatwaves, which cover massive areas, particularly during the E_s, E_s-Ip_s, and E_s-Ip_m seasons (table S1), results in a larger composite mean, even though relatively moderate events are also possible. Although the mean values of large and extremely large contiguous heatwave metrics are significantly lower during the L_s-In_s seasons compared to the neutral conditions, there is a (low) probability for the occurrence of those events with a massive area (table S1). This significant decrease in their composite mean during the L_s-In_s seasons is likely due to the low (high) probability of occurrence of large and extremely large contiguous heatwaves which cover massive (moderate) areas (table S1).

The above-presented results show that large and extremely large contiguous heatwaves in Australia are significantly increasing (decreasing) during the E_s, E_s-Ip_s, and E_s-Ip_m (L_s-In_s) conditions. To examine the large-scale underlying physical mechanisms associated with these phases, we calculated the composite differences of detrended MSLP and geopotential height at 500 hPa (z500) between these phases and the neutral phase. This composite analysis provides more insights into how the E_s, E_s-Ip_s, E_s-Ip_m, and L_s-In_s phases influence contiguous heatwaves in Australia.

The composite maps of MSLP and z500 during the E_s, E_s-Ip_s, E_s-Ip_m, and L_s-In_s are shown in figure 7. The composite difference patterns of MSLP are similar with z500 during the E_s, E_s-Ip_s, and E_s-Ip_m, and L_s-In_s phases. The E_s, E_s-Ip_s, and E_s-Ip_m phases show positive changes in MSLP and z500 along the Southeast of Australia over the Tasman Sea and negative changes along the northwest of Australia over the Indian Ocean (figure 7). These positive and negative changes of MSLP and z500 are significant, particularly during the E_s-Ip_s, and E_s-Ip_m phases. No significant changes of MSLP and z500 are observed during the L_s-In_s phase. The zonal pressure dipole-like circulation features such as low-pressure cyclonic circulation over the northwest and high-pressure anticyclonic circulation over the southeast of Australia observed during the E_s, E_s-Ip_s, and E_s-Ip_m phases favour the persistence of blocking highs over the Tasman Sea (Pezza et al 2012, Boschat et al 2015). The blocking of high-pressure systems over the Tasman Sea favours the development and amplification of heatwaves over Australia (Pezza et al 2012, Perkins et al 2015). Hence, the increased geopotential height and the pressure dipole-like circulation features near Australia associated with the E_s, E_s-Ip_s, and E_s-Ip_m phases influence the contiguous heatwaves in Australia.

Zoom In Zoom Out Reset image size

The results presented here have a number of important implications for Australian industry and society. The increase in extremely large contiguous heatwaves coupled with prolonged duration (figure 3) results in simultaneous exposure of crops to heatwaves over large connected regions. The prolonged exposure of crops to heatwaves can decrease their productivity (Asseng et al 2011). The intensification of the areal magnitude of extremely large contiguous heatwaves (figure 3) will also expose a greater proportion of the Australian population to intense heatwaves in neighbourhood regions at the same time. Severe heatwaves seriously affect human health (Hanna et al 2011) and have a negative impact on the emergency health services (Lindstrom et al 2013) and electricity supply systems (McEvoy et al 2012). Many previous studies investigated the effects of heatwaves (not necessarily contiguous) on Australian society (Hanna et al 2011, Lindstrom et al 2013), but there is a lack of research on the impacts of large contiguous heatwaves on society. Further research is needed to consider and quantify the impacts of simultaneous exposure of regions to contiguous heatwaves.

There are a few caveats and limitations to this study that need to be highlighted. In this study, the spatial connectivity of heatwaves is only allowed for those grid cells with adjacent land area, which omits Tasmania from contiguous events on the mainland. Moreover, we have only considered ENSO and IOD and their interactions but not the SAM, which also influences heatwave characteristics over southern parts of Australia (Perkins et al 2015). Nevertheless, this study is the first to systematically quantify observed changes in spatial characteristics of contiguous heatwaves in Australia during the historical period and comprehensively investigate their association with the interactions of ENSO and IOD. Our results will help understand how the contiguous heatwaves in Australia respond to ENSO and IOD and their interactions. An enhanced understanding of the interactive influence of ENSO and IOD on contiguous heatwaves can help improve the seasonal prediction skill of contiguous heatwaves.