DROUGHT AND HOTTER TEMPERATURE IMPACTS ON SUICIDE: EVIDENCE FROM THE MURRAY–DARLING BASIN, AUSTRALIA

1. Introduction

Suicide is an important public health issue around the world, with more than 800,000 deaths globally due to suicide each year (World Health Organization, 2014). In addition, studies are increasingly focusing upon the link between higher average temperatures and suicide (Takahashi, 2017; Burke et al., 2018; Luan et al., 2019; Barve et al., 2021). In Australia, suicide has been identified as the leading cause of death for residents aged 18–44 and the second leading cause of death for those aged 45–54 (Handley et al., 2018). Suicide rates are much higher in rural than urban areas in Australia, along with many other countries (Hirsch, 2006; Miller and Burns, 2008; Gamm et al., 2010); with the suicide rate in remote Australia almost double that of major cities (Senate Community Affairs and References Committee, 2018). Furthermore, rural residents have been found to have less access to mental health services at the time of their death (Caldwell et al., 2004; Hunter, 2007). Within rural areas, farmers or members of farming families in particular have been identified as vulnerable groups (Page and Fragar, 2002; Fraser et al., 2005; Arnautovska et al., 2016; Kennedy et al., 2020).

In the context of global climate change, there has been growing concern of the impacts of extreme weather (e.g. drought, flooding) and ongoing incremental weather events (e.g. extreme temperatures) on mental health (Berry et al., 2018). These events induce post-traumatic stress disorder, anxiety and depression (Obradovich et al., 2018; Cooper et al., 2019). As the driest continent in the world, Australia regularly experiences rainfall deficiency, prolonged drought and hotter temperatures (Head et al., 2014), with climate change predicted to heighten these extremes (IPCC, 2021). Farmers and rural residents are vulnerable to extreme weather events through reduced agricultural production and increased financial hardship (Austin et al., 2020; Barve et al., 2021; Edwards et al., 2015; Quiggin et al., 2010; Sherval and Askew, 2012; Wheeler et al., 2013, 2020); degraded environmental conditions (Albrecht et al., 2007); and reduced employment (Austin et al., 2020). Moreover, prolonged drought can cause mental illness—known as “solastalgia” —through changed environmental conditions such as native tree damage, loss of topsoil, declining green space and dying wildlife. This negative effect of climate-associated environmental change is further exacerbated by a sense of powerlessness or lack of control over the unfolding change process (Albrecht et al., 2007). Berry et al. (2018) provide a causal process diagram linking climate change factors and mental illness.

Though climate change is considered a threat to emotional wellbeing and mental health, to date, there has been limited quantitative evidence examining the relationship between extreme weather events—such as drought—and rural suicide in Australia (Nicholls et al., 2006; Hanigan et al., 2012; Guiney, 2012). One New South Wales (NSW) study found that a decrease in annual precipitation was associated with an 8% increase in the total suicide rate from 1964–2001 (Nicholls et al., 2006); while another NSW study found this positive association between drought and suicide existed for rural males aged 30–49 years, but not rural females over 30 years from 1970–2007 (Hanigan et al., 2012). In Victoria, no evidence was found of increased farming suicides during the drought years of 2001–2007 (Guiney, 2012).

In Australia, the Murray–Darling Basin (MDB), covering four states and one territory (NSW, Victoria, Australian Capital Territory, Queensland, South Australia), accounts for 66% of Australia’s total irrigated area; 40% of its farm businesses; and 59% of its irrigation water. Over the last century, the MDB has incurred higher temperatures, with maximum temperature anomalies increasing in size and frequency (Fig. S1). While rainfall has also shown great volatility during this time-period, there is no clear reduction in trend (Fig. S2).

Although changes in MDB weather pose a significant threat to farmers’ mental health (Wheeler et al., 2018; Daghagh et al., 2019, 2020), to date, there has been no study modeling suicide across the entire MDB. Furthermore, only a limited number of studies have examined the association between drought severity (as opposed to its presence) and suicide, given that suicide risk may increase with greater drought severity (Kolassa et al., 2010; Stain et al., 2011). However, recognizing the extreme weather influence on suicide, especially among vulnerable populations, is essential for policy purposes. Though some adaptation strategies are already used in the face of severe weather events such as drought in Australia (e.g. changing irrigated area, water trading, changing crop mix (Wheeler et al., 2013)), to the best of our knowledge, the link between suicide and the modifiers which could reduce the suicide impacts are unknown.

This study aims to narrow existing gaps and model entire population effects over a long time-period in several ways. First, we use a finer level of spatial and temporal dataset in Australia to identify the impacts of drought and hotter weather on suicide in the MDB. Second, heterogeneity analysis is applied to estimate the variation of relationships for a series of subgroups, including age, and gender groups. Third, we test for any association of potential modifiers (e.g. green space coverage, the proportion of health professionals per labor force, income, education) that could mitigate the suicide impacts of drought and hotter temperatures in the MDB.

In particular, we employ Poisson fixed-effect regressions to model longitudinal monthly suicide data from 2006–2016 by areas in the MDB to investigate the impacts of duration and severity of drought and hotter temperature on suicide rates (total, and by age and gender groups). The MDB provides an apt case study for this research, given the predominance of a variety of droughts during this time-period and its importance as a food producing region.

2. Data

This study uses consolidated data obtained from multiple sources. The main dataset, the National Cause of Death Unit Record File (COD URF), was obtained from Australian Coordinating Registry (ACR). The COD URF provides ongoing data starting from 2006, and the latest currently available finalized year is 2019. However, we only have access to the finalized datasets up to 2016. People who died from suicide are defined as death due to intentional self-harm and identified using the International Classification of Disease categories (International Classification of Disease, Tenth Revision, codes X60-X84 and Y87.0). COD URF data were recorded using various geographic units across the years. Specifically, the Statistical Local Area (SLA) under the Australian Standard Geographic Classification (ASGC) was used up to and including 2010, while the previous years version of the SLA under ASGC was used. The Australian Statistical Geography Standard (ASGS) was applied from 2010 onwards. From 2011 to 2015, ASGS 2011 Statistical Areas Level 2 (SA2) was used, and ASGS 2016 SA2 was used from 2016 onwards. Therefore all these data had to be concorded into 2016 SA2 boundaries using concordance files provided by the Australian Bureau of Statistics (ABS) to facilitate appending across years and merging with other datasets. To investigate the possible links between hot temperature, drought and suicide, suicide rate (%) was used as the dependent variable, which is calculated as the monthly number of suicides divided by the whole population in each SA2 in that year.

Other data include the specialized records from the Australian Population Census collected by ABS every five years. By the time of study, there were three waves (2006, 2011 and 2016) available at varying geographic unit levels. Population census data (2006) was collected at the ASGC 2006 level, and population census data (2011) was collected at ASGC 2011 level. ASGS 2016 SA2 was used for the 2016 population census data. All these variables in different waves were then concorded to the 2016 SA2 boundaries. The variables from the Population Census include proportion of indigenous population, proportion of unemployed population, average household annual income (in 2006 Australian dollars, measured in thousands), proportion of farmer population, proportion of health professionals and proportion of people with Bachelor’s degree or above. The five-year census data are transferred into monthly data from 2006–2016 using linear interpolation.

Weather data (i.e. monthly average maximum temperature, monthly total rainfall, long-term average annual rainfall (1976–2005) and long-term average annual maximum temperature (1976–2005)) were collected from the Australian Bureau of Meteorology (BOM). BOM provides gridded weather data, collected at 0.025^∘ (approximately 2.5km) for temperature, and 0.05^∘ (approximately 5km) for rainfall, which was then converted into SA2 regions. The monthly average maximum temperature and monthly total rainfall are calculated by rolling 12-month averages. The Palmer drought severity index (PDSI), which is a widely used agriculture drought index, was calculated using monthly total rainfall data, monthly average maximum temperature, monthly average minimum temperature, and available water capacity data from the Commonwealth Scientific and Industrial Research Organization (CSIRO). It was used to assess the dryness and different extents of drought. A PDSI lower than $-4$ indicates extreme drought, while a PDSI between $-2.0$ and $-2.9$ are moderate drought. To also capture the duration of different levels of drought and estimate their impacts, the number of months with drought that occurred during the past 12 months was used.

We also collected a measure of green space, the Normalized Difference Vegetation Index (NDVI) from the BOM for the years 2006–2016, which includes both urban open space and natural reserves, and was concorded into 2016 SA2 boundaries.

Table S1 reports the descriptive statistics and variable definitions. The total suicide rate (namely total suicides divided by the relevant population within 100,000) varies across regions (namely SA2) in the MDB. Some SA2s in remote areas with very low populations have relatively high suicide rates (see Fig. 1). Across our time period, some very remote areas had a very small population (e.g. less than 10 people), and also incurred suicides, hence there were a few observations where there were very high suicide rates (above 10,000). These outliers were excluded from the database, although sensitivity testing indicated our results did not significantly change with them included.

Figure 2 provides evidence that suicide rates vary across time. Across the whole population and all gender-age groups, suicide rates show a noticeably upward trend (from 2006 to 2016) though with monthly fluctuations. Generally, men experience a higher suicide rate than women in the MDB (see Fig. 2(a)), confirming previous literature (Page and Fragar, 2002; Caldwell et al., 2004; Alston, 2012). Moreover, among all age groups in the MDB, the 30–49 age group have the highest suicide rate, followed by those 50+ years, while the 10–29 group have the lowest suicide rate. Figure 2(b) illustrates a higher suicide risk for the 30–49 age group, as compared to other cohorts, which is consistent with existing literature (Page and Fragar, 2002; Miller and Burns, 2008; Hanigan et al., 2012). Figure 2(c) further divides the population into gender-age groups.

3. Methods

As noted, suicide rates are modeled at the SA2 level in the MDB. Given the count data nature of suicide outcomes, Poisson fixed-effects regression model with robust standard errors was applied. There are several reasons to choose this regression model. First, the Poisson model avoids retransformation problems of linear regression (e.g. in the log-transformed model of linear regression, estimated results are biased when obtained by retransformation using the exponential or anti-log function); and the method can be used even when the dependent variable does not follow a Poisson distribution (Buntin and Zaslavsky, 2004; Cameron and Trivedi, 2013; Sparrow et al., 2014; Ogun, 2021). Second, the only requirement for the Poisson estimator to be consistent is that the conditional mean is correctly specified (Wooldridge, 2002) and therefore Poisson can also be applied to continuous non-negative data (Cameron and Trivedi, 2013). Third, a robust variance estimator is used in Poisson fixed-effects regression model to avoid the impact of possible issues of overdispersion (conditional variance is larger than conditional mean, which sometimes occurs in Poisson regression model) on standard errors. This is because, although estimation of the Poisson model is still consistent once the conditional mean is correctly specified, overdispersion leads to biased standard errors and a robust variance estimator is therefore needed to adjust it (Cameron and Trivedi, 2005). A number of other studies have used Poisson regression to model suicide rates (e.g. Isumi et al., 2020; Reisch et al., 2013; Wang et al., 2014).

Considering the possible correlation of unobserved regional characteristics with the regressors, the Poisson model with fixed instead of random-effects was used. Unobserved regional characteristics—including natural and environmental conditions—could bias the estimates, hence the need to incorporate regional fixed-effects into the model. According to the Hausman test that checks the possible existence of unobserved heterogeneity, the null hypothesis that the unobserved heterogeneity is uncorrelated with regressors was rejected ($p-\mathrm{\text{value}}=0.00$). Thus, Poisson fixed-effects model with robust standard errors to address potential heteroscedasticity was used, with random-effects Poisson models also estimated for robustness purposes.¹

Hence, let $y_{it}$ represent the suicide rate—calculated as the suicide number per 100,000 population—in region i at period t. The primary equation of the model is

where ${\lambda }_{it}$ is the rate parameter of the Poisson fixed-effects regression model. The formula for ${\lambda }_{it}$ is the loglinear model : where $d_{it}$ is the weather condition vector including drought variables based on PDSI, and monthly average maximum temperature in region i at period $t;x_{it}$ is a vector of time-variant covariates (evaluated with coefficient vector $\beta$); $\delta$ and $\beta$ are vectors of coefficients to be estimated; $r_t$ are time fixed-effects; and $a_i$ is the full set of 2016 SA2-level regional fixed-effects.

Three drought measures—of varying duration and severity—were constructed to understand their association with suicide rates of MDB rural regions. Specifically, the PDSI was employed, which converts meteorological drought into agricultural drought. PDSI expresses drought severity within a period when the water supply in a region is continually lower than normal (Palmer, 1965). However, PDSI has been found to be inconsistent at various locations, making spatial comparisons of PDSI values difficult. Compared with the PDSI, self-calibrating Palmer Drought Severity Index (SC-PDSI) with different drought categories is more spatially comparable, and reports extreme dry conditions with frequencies that would be expected for rare conditions (Wells et al., 2004). Therefore, we applied SC-PDSI classification in this study to explore the impacts of different frequency and severity levels of drought. SC-PDSI was used to categorize drought events based on their severity (Wells et al., 2004). There are three mutually exclusive categories: (1) extreme drought (PDSI$\le -4.0$); (2) severe drought (PDSI $-3.0$ to $-3.9$); and (3) moderate drought (PDSI $-2.0$ to $-2.9$). Three levels of drought severity are created based on SC-PDSI classification: at least moderate (PDSI$\le -2.0$), at least severe (PDSI$\le -3.0$), and extreme (PDSI$\le -4.0$).² Drought duration was calculated by the number of months in each category during the last 12 months, and monthly average maximum temperature and rainfall were also calculated. Other variables (socioeconomic and demographic characteristics such as household annual income, indigenous population, unemployment, farmers per labor force, and health professionals) also potentially contribute to suicide rates (Caldwell et al., 2004; Qi et al., 2014; Fitzpatrick et al., 2021) and therefore were included as control factors.

The expected number of suicides per 100,000 per month is indicated as

The Poisson fixed-effects model is a nonlinear regression and is estimated by maximum likelihood techniques.

Considering the possibility of drought’s heterogeneous impact on different population groups, we also estimated the main regression specifications for a series of subgroups. To compare the associations of weather and socioeconomic conditions on suicide rates for different age and gender sub-groups, the sample was divided into different gender and age groups (10–29 years, 30–49 years, 50+ years). Due to no record of suicide in any SA2s for smaller age-gender groups, we combined the 10–29 and the 30–49 years into one group when they are modeled by gender. Therefore, there were four age–gender groups overall where impacts were estimated for each group: 10–49 male, 10–49 female, 50+ years male, 50+ years female). In this study, a large range of alternative models were used for robustness checks, including more-restrictive fixed-effects (SA2, year and seasonal fixed-effects) and less-restrictive fixed-effects (SA2 and year fixed-effects), inclusion and exclusion of various community socio-demographic variables.

After investigating the relationship between drought and suicide rate, we also investigated: (a) the modifiers which could reduce the suicide impact of drought and maximum temperature; and (b) who are the vulnerable groups that are more susceptible to higher drought or hotter temperature. To investigate (a) we tested whether various modifying factors (e.g. the proportion of health professionals per labor force, NDVI (green space coverage), income, education) could mitigate the impact of drought on the suicide rate. To investigate (b), we tested whether there were heterogeneous drought and maximum temperature impacts on areas with different numbers of farmers, indigenous populations and long-term climate conditions (measured by long-term average annual rainfall and annual maximum temperature). The formula for ${\lambda }_{it}$ in Eq. (2) is therefore

where ${\mathrm{\text{Mod}}}_{it}$ is the measure of modifiers or group factors in region i at period t. It is worth noting that the long-term climate variables are time-invariant. Hence, they cannot be included in FE regressions. Equation (5) will estimated instead, where ${\mathrm{\text{Mod}}}_i$ is the time-invariant long-term climate variables and only their interactions with drought variables are included.

4. Findings

4.1. Effects of influences of drought and temperature on suicide

Results indicated that drought and higher monthly average maximum temperatures were associated with an increased suicide rate across MDB regional areas (Table S2). The effects were heterogeneous across all gender and age groups (Figs. 3 and 4; Tables S3 and S4). The relationship between the number of months in extreme drought during the previous 12-month period and the total suicide rate was statistically significantly positive (at the 5% level); which indicates the more frequent the extreme drought in the previous year, the higher the suicide rate in an SA2 (see Fig. 3). A marginally statistically significant relationship (at the 10% level) was also found between the number of months of moderate drought and suicide rate—albeit at a far lower impact. Specifically, one more month of moderate drought was associated with the total suicide rate increasing by 2% (approximately 0.03 within a 100,000 population)—but the suicide rate increased by 32% (approximately 0.45 within a 100,000 population) with one more month of extreme drought (Table S2). However, it is important to note that the average number of months in the previous year, during which a SA2 in our study time-period was exposed to moderate drought, was 1.73 months (52 days); while it was only 0.01 month (0.3 days) for extreme drought. The difference between moderate and extreme drought impacts has been found previously, in that individuals with greater and repeated exposure to traumatic events over a longer period have poorer mental health outcomes (Kolassa et al., 2010). The results for moderate and extreme drought are presented together for comparison purposes; while results for severe drought are included in Table S6.

The findings of a significant temperature impact on suicide rates are consistent with previous studies elsewhere (e.g. the U.S (Burke et al., 2018), Finland (Hiltunen et al., 2014), Japan (Sim et al., 2020), India (Carleton, 2017), Korea (Kim et al., 2011) and Mexico (Burke et al., 2018)). This implies the possible existence of impairment of thermoregulation amongst suicide victims (Hiltunen et al., 2014; Burke et al., 2018) (see Fig. 4). Moreover, the coefficient in Table S2 indicates that the total MDB suicide rate increases by 8% (approximately 0.11 within a 100,000 population) with a 1^∘C increase in average monthly maximum temperature in the previous 12 months. Compared with studies in the U.S., Mexico, India and New Zealand, examining the temperature–suicide relationship (Williams et al., 2015; Carleton, 2017; Burke et al., 2018), our results show that the MDB—as Australia’s most important agricultural production area which suffers considerably from hotter and more variable weather—experiences much higher suicide impacts from temperature. For example, a 1^∘C increase in average monthly temperature increases the monthly suicide rate by 0.68% in the U.S, 1.8% and 2.1% in Mexico (Burke et al., 2018), both of which are much lower than Australia. This result is probably because the MDB consists primarily of rural areas, while the U.S. and Mexico studies focused on both urban and rural areas, and excluded suicide data for counties with less than 100,000 people. Combined, this may have underestimated the average suicide rate in regional areas and thereby reduced the impact of temperature. In Australia, SA2s generally have a population range of 3,000–25,000 and, for MDB SA2s, the average population size in our time-period was 6,301. Furthermore, we estimate the impact of average monthly maximum temperature on suicide rates, while the U.S. and Mexican studies mainly focused on average monthly temperature. In this study, the temperature impact is even greater for suicide rates of residents aged 30–49, as much as 21% (approximately 1.42 within a 100,000 population) from a 1^∘C increase in average monthly maximum temperature.

Gender and age have been shown to be critical predictors of suicide (Nicholls et al., 2006; Alston, 2012; Kunde et al., 2017), and therefore modelling was conducted by various age and gender groups (see Figs. 3 and 4; Tables S3 and S4). Results suggested that only the overall male suicide rate was statistically significantly associated with both moderate and extreme drought duration and higher monthly temperatures in the MDB, but with no such findings for the overall female suicide rate. When the models are disaggregated by gender and age groups, the suicide rates of both males and females aged 10–49 were both statistically significantly positively associated with extreme drought. However, higher monthly temperatures impacted only male suicide rates. For males and females aged 50+, no statistically significant association of drought or temperature was found. The age–group models in Table S3 illustrated that extreme drought duration was associated with higher impact in younger residents (aged 10–29), followed by those aged 30–49. All these findings suggest that in the MDB younger people are more vulnerable to drought—and especially extreme drought—while older people seem more resilient (Hanigan et al., 2012). It is known that younger men’s mental health maybe be more likely to deteriorate in a drought because of finances, debt, a lack of accessed services, family stress, and increased social isolation (Dean and Stain, 2010; Carnie et al., 2011). Although previous studies have suggested that female farmers are more resilient than men in the face of drought and recover quicker (Stain et al., 2011); our results suggest this is only true for rural resident females living through moderate drought in the MDB, and older women (50+). In times of extreme drought, women aged 10–49 were still significantly exposed to suicide risk.

4.2. Socioeconomic and demographic influences

Beyond drought and temperature influences, suicide rates are also jointly affected by other demographics. Generally, an increase in farmers’ proportion of the area’s labor force increased the total suicide rate, which suggested that farmers were at a higher risk compared to other occupations, as found elsewhere (Page and Fragar, 2002; Arnautovska et al., 2014). For instance, it has been highlighted that the suicide rate of farmers in Australia is one and a half to two times higher than the national average (Page and Fragar, 2002). This has been related to risk factors for suicide such as farming masculinity issues (Bryant and Garnham, 2015); natural disaster impacts (Hanigan et al., 2012); financial stressors (Fraser et al., 2005; Edwards et al., 2015) and other socioeconomic distress (Hirsch, 2006).

An increase in the proportion of people with a bachelor’s degree or above in an area was negatively associated with all suicide rates, which suggests that a greater educated workforce may improve wellbeing through strengthening various forms of human, social and cultural capital and increasing access to mental health services (Phillips and Hempstead, 2017). Moreover, an increase in the proportion of the indigenous population within an MDB area was positively significantly associated with higher suicide rates, which suggests that indigenous people in rural and remote areas may be at particular risk given their multifaceted disadvantages and limited access to social and health services (Hunter, 2007).

4.3. Robustness checks

Several alternative specifications were run to check the robustness of results. This included employing various year and season fixed-effects (FE) and using a variety of dependent and independent variables. First, models with less-restrictive fixed-effects (SA2 and year fixed-effects) were used for robustness checks. Our results showed that changes in model specifications using less-restrictive fixed effects did not alter the significance and magnitude of our key variables of maximum temperature and extreme drought on suicide rates, for all population and subgroups (see Fig. S3). Second, models with only weather variables, excluding socio-demographic variables, were also estimated, with no significant change in extreme drought impact observed (see Fig. S4). Third, a variety of other socio-demographic, agricultural and environmental variables were tested, with no significant change in key results. Fourth, three mutually exclusive drought variables were combined under one model to compare drought impact of different severity. Results again support the significant impact of extreme drought and temperature on suicide rate (see Table S5) and that the magnitude and significance of drought increase with its severity. Finally, we modeled our dependent variable for suicide as a count variable (rather than as a suicide rate) and controlled for the SA2 population as an independent variable. Again, the results were similar to our overall findings. Overall, robustness test results indicate that the impacts of maximum temperature and extreme drought on MDB suicide rates are stable.

4.4. The impact of modifiers on the weather-suicide relationship

Finally, we considered variables that may modify the relationship between weather conditions (i.e. drought,³ maximum temperature) and suicide rates. Influences that increase (or decrease) the identified impact of drought variables are used to identify vulnerable groups where further intervention might be useful (Mullins and White, 2019; Austin et al., 2020).

Figures S5 and S6 plot the marginal effects of weather conditions (e.g. extreme drought and maximum temperature) on suicide rates for different levels of the modifiers. Figure S5 focuses on the modifiers which mute the impacts of drought and hotter temperature while Fig. S6 indicates vulnerable population groups in the face of extreme drought and hotter temperature. For modifiers that might moderate the impact of extreme drought and hotter temperature, we consider: NDVI (green space coverage); access to health care (represented by the proportion of health professionals per labor force in SA2); average household annual income in SA2 and access to education (represented by the proportion of people with Bachelor’s degrees or above in SA2). Figure S5 illustrates that green space coverage moderates both the relationship between extreme drought and suicide rate, and the relationship between hotter temperature and suicide rate. Such findings confirm other existing studies which emphasize the benefits of green space area on mental health (Maas et al., 2006; Nutsford et al., 2013; Alcock et al., 2014), possibly through providing outdoor activity opportunities, fresh air and scenery (Xu et al., 2018). However, our results show no statistically significant evidence that modifiers such as access to health care, access to education, or household annual income at the community level moderate the effect of extreme drought on suicide. Regarding the relationship between maximum temperature and suicide, we only find a weak moderating effect of household annual income, but no moderating effect of access to health care or access to education. There are similar findings between our study and several other studies. Barreca et al. (2016) found that access to health care (represented by doctors per capita) was not statistically related to reductions in heat-related mortality. Mullins and White (2019) found no evidence of modifying effects of mental health professionals or increases of county-level income on suicide effects of higher temperatures.

For vulnerable groups, we considered two variants that potentially modify the relationship between extreme drought and suicide, namely: (1) long-term average annual rainfall⁴; the proportion of farmers per labor force; and the proportion of the Indigenous population; and (2) long-term average annual maximum temperature⁵; the proportion of farmers per labor force; and the proportion of the Indigenous population (Fig. S6). The marginal effect of extreme drought had a greater and more significant effect in areas with lower long-term average rainfall. Specifically, extreme drought significantly affects suicide rates in areas with long-term average annual rainfall lower than 387.4mm, and the effect diminishes as an area becomes wetter. It indicates that in the MDB, areas accustomed to a dry climate have a higher drought effect on suicide rate and therefore are more susceptible to drought. It possibly suggests that the cumulative drought effects in these areas are more significant than the adaptation effect. Conversely, our results show that hotter areas in MDB adapt better (with a smaller marginal effect of maximum temperature on suicide) to higher temperatures, while colder areas (with a long-term average annual maximum temperature lower than 22.5^∘C) are significantly affected by higher maximum temperatures.

Though there are few existing studies on modifiers of the relationship between drought and suicide, some studies (Barreca et al., 2016; Mullins and White, 2019; Heutel et al., 2021) use the same approach to explore the modifiers of the temperature impacts. Barreca et al. (2016) use the state-year-month level data from 1900–2004 in the United States and investigate the relationship between mortality rates of all causes and high temperatures and the modifiers that change that relationship. They find that the areas more accustomed to temperature extremes have adapted better because a more muted temperature-mortality relationship is observed. Heutel et al. (2021) use United States data from 1992–2013 and find that hot days are less deadly in warm places. However, Mullins and White (2019) do not find a significant modifying effect regarding the impact of ambient temperatures on a broad set of mental health outcomes including mental illness, suicides and self-reported days of poor mental health.

Our results also show that the marginal effects of extreme drought on suicide increase as the proportion of farmers or indigenous populations increases. It indicates that besides dry areas, local areas that have more farmers and indigenous populations represent hot spots that will benefit more from targeted extreme drought interventions. In contrast, the marginal effect of maximum temperature on suicide increases as indigenous populations increase, but decreases as the proportion of farmers increases. It indicates that areas with more Indigenous people are vulnerable to hotter temperatures and extreme drought, while areas with a higher proportion of farmers are affected more by extreme drought.

One limitation of this study is that it focuses on the short-term impact of drought and maximum temperature occurring during the past 12 months of the study years, while the long-term effects are unknown due to the limited access to the data. Also, the paths of suicide impacts of weather conditions are still unknown and need to be explored by future studies.

5. Conclusion

Though climate change has been considered the biggest threat to global mental health in this century, the mental health—especially suicide impacts—of climate change have received less research attention. Using comprehensive monthly data covering the main period of the Millennium drought (from 2001/02–2009/10) from 2006–2016 in the MDB, this paper makes four primary discoveries about suicide in light of climate change.

First, drought, as well as higher temperatures, increased total MDB suicide rates. The higher the duration and severity of the drought (i.e. extreme droughts), the higher the total suicide rate. Second, the effects of extreme drought and temperature on suicide were heterogeneous across gender and age groups. Specifically, males and younger people aged under 50 are more affected by either hotter temperatures or extreme drought compared with other groups. People aged 10–29 in the MDB are the most vulnerable in the face of extreme drought, while people aged 30–49 suffer more from increases in maximum temperature. Third, areas with higher percentages of Indigenous populations are more susceptible to extreme drought and hotter temperatures, while areas with a higher percentage of farmers suffer more from extreme drought than hotter temperatures. Therefore, areas with higher percentages of farmers and Indigenous populations are hot spots, and therefore would benefit more from appropriate interventions. Fourth, the empirical results point to increased green space coverage as an important determinant in moderating the effect of both extreme drought and maximum temperature on suicide in the MDB. Some evidence was also found that average household annual income partially moderates the relationship between maximum temperature and suicide.

Given the projection of higher temperatures and longer (and more extreme) droughts in Australia and the subsequent impact on agricultural production within the MDB, it is clear that rural areas’ mental health will increasingly deteriorate if no effective interventions are developed. However, current interventions may not be sufficient to assist vulnerable areas in adapting to extreme drought and hotter temperatures, especially in hot spot areas. Future research is warranted in regards to how to implement policies to address/improve drought preparedness, mental health, farm financial viability and farm exit, natural resource management and natural capital—as well as economic and social development policies in reducing the negative consequences of climate change.

Acknowledgments

We are grateful for the research assistance provided by Juliane Haensch and Natthanij Soonsawad, and for comments received by reviewers that improved this manuscript. Australian Research Council grants FT140100773 and DP200101191 funded this study.

Supplemental Materials

The Supplemental Materials are available at: https://www.worldscientific.com/doi/suppl/10.1142/S2010007823500240.

Notes

¹ Poisson Conditional fixed-effects model (PCFE) with spatial variance estimator was also tested. The spatial variance estimator generated by Bertanha and Moser (2016) is consistent as we fail to reject the null that spatial dependence is time-invariant ($P-\mathrm{\text{value}}=0.99$). Given results for drought were overall consistent between the models, and the desire to assess the modifying effects, we only report the Poisson fixed-effects models.

² At least moderate and at least severe levels are referred to as moderate and severe levels, respectively, in most tables and figures, except Table S5, in which three drought variables in one model need to be mutually exclusive.

³ The relationship between drought variables, including extreme drought, severe drought and moderate drought and suicide are estimated.

⁴ We test for the modifying effects of long-term average annual rainfall for subgroups and found the same results as shown in baseline impacts. For example, females in general are more resilient than males as only the impact of extreme drought in the areas with long-term average annual rainfall lower than 187.4mm significantly impact female suicide, but extreme drought in the areas with long-term average annual rainfall lower than 387.4mm significantly affected male suicide. Regarding impacts on different age groups, younger people are more affected. Specifically, extreme drought has significant impact on the 10–29 and 30–49 age groups in drier areas, but not in wetter areas (long-term average annual rainfall higher than 387.4mm for 10–29 age group and long-term average annual rainfall lower than 437.4mm for 30–49 age group). However, people who are 50+ do not seem affected by extreme drought.

⁵ The modifying effects of long-term average annual maximum temperature are also tested for subgroups, and the results are the same as shown in baseline impacts. More specifically, males in colder areas (with a long-term average annual maximum temperature lower than 23.5^∘C) are more affected, compared with males in hotter areas. In contrast, females are not significantly affected by maximum temperature in any areas. Younger people especially who are 30–49 years old are more affected by maximum temperature, but older people who are 50+ are not significantly affected.