Australia's Tinderbox Drought: An extreme natural event likely worsened by human-caused climate change

Abstract

We examine the characteristics and causes of southeast Australia's Tinderbox Drought (2017 to 2019) that preceded the Black Summer fire disaster. The Tinderbox Drought was characterized by cool season rainfall deficits of around -50% in three consecutive years, which was exceptionally unlikely in the context of natural variability alone. The precipitation deficits were initiated and sustained by an anomalous atmospheric circulation that diverted oceanic moisture away from the region, despite traditional indicators of drought risk in southeast Australia generally being in neutral states. Moisture deficits were intensified by unusually high temperatures, high vapor pressure deficits, and sustained reductions in terrestrial water availability. Anthropogenic forcing intensified the rainfall deficits of the Tinderbox Drought by around 18% with an interquartile range of 34.9 to -13.3% highlighting the considerable uncertainty in attributing droughts of this kind to human activity. Skillful predictability of this drought was possible by incorporating multiple remote and local predictors through machine learning, providing prospects for improving forecasting of droughts.

Fig. 1.. The drought focus region.

(A) The thick blue line shows the outline of the region in drought during 2017–2019. Basemap colors denote elevation. The map also shows agricultural areas, the Murray Darling Basin (MDB, thin aqua line), smaller river basins, locations of streamflow stations, and borewells. (B) The proportion of time in drought during April to September 2017–2019 based on standardized drought metrics. The thick black line denotes the drought area. The fraction of time spent in drought is calculated here as the mean proportion of time SPI-3/SPEI-3 ≤ −1 for data encompassing only the cool season months (April to September) of 2017–2019 based on three precipitation and two potential evapotranspiration (PET) datasets (Materials and Methods; fig. S1).

Fig. 2.. Historical context for the Tinderbox Drought.

(A) Total April to September rainfall over the drought focus region using the Australian Gridded Climate Data (AGCD) historical rainfall dataset. Shading indicates the decile over the full record length (1900–2022), where darkest brown indicates rainfall in the lowest 10% of all years and darkest green indicates rainfall in the highest 10% of all years. (B) Moving 3-year accumulated rainfall anomalies, relative to the 1961–1990 mean. Accumulated anomalies are based on all months and are plotted for the final month of each 3-year interval. Gray shading indicates the 2017–2019 Tinderbox Drought interval, and dashed horizontal line indicates the maximum 3-year accumulated rainfall anomaly at the end of the Tinderbox Drought.

Fig. 3.. Monthly anomalies in water cycle components in the drought focus region.

(A) Precipitation and evapotranspiration (ET) (%), (B) vapor pressure deficit (%), and T_max (°C) (C) soil moisture from the European Space Agency dataset (%) and terrestrial water storage from Gravity Recovery and Climate Experiment (GRACE) data (mm), with error bars indicating uncertainty estimates from the same datasets. (D) Monthly anomalies in water table depth (m) from borewell data. The water level anomalies in the Murrumbidgee and Upper Murray are shown on the right y axis. (E) Seasonal anomalies in streamflow (%). Anomalies are calculated with respect to a baseline (1980–2016), unless constrained by data availability. The figure shows the period covering 1 year before and after the drought (2016–2020), and vertical shading in (A) to (C) indicates the cool seasons of the Tinderbox Drought.

Fig. 4.. The evolution of the drought impacts on vegetation in southeast Australia.

(A) The 12-month rolling mean of vapor pressure deficit (VPD; 15:00 hours reading from AGCD) across the focal region is shown for 1981–2020. The shadings show the inner 50% and 90% range of the focal region’s VPD. The mean annual VPD and 10% deviation are overlaid. The Millennium and Tinderbox droughts are highlighted. (B) The relative VPD anomaly expressed as a percent deviation from the 2002–2016 mean annual value. (C) The relative anomaly of the vegetation optical depth (VOD) during September to November (SON) is plotted as a percent deviation from the 2002–2016 SON seasonal mean. (D) The relative anomaly of the Normalized Difference Vegetation Index (NDVI) is plotted as a percent deviation from the 2002–2016 seasonal (SON) mean. Regions that experienced burning during the 2019 Black Summer fires are denoted by orange points. (E) The annual mean of the daytime (13:30 overpass time) land skin temperature anomaly (LST; °C) as derived from the MODIS AQUA platform.

Fig. 5.. Probability that the 2017–2019 southeastern Australian meteorological drought occurred within the range of internal variability.

(A) The observed deficit in cool season (April to September, AMJJAS) rainfall of the Tinderbox Drought (2017–2019; red dashed line) relative to the first 60 years of the observational period (1900–1959). The likelihood of the observed 2017–2019 rainfall deficit is assessed relative to random resampling of the full historical period (1900–2019) 10,000 times and computing the precipitation anomaly of the last 3 years compared to the first 60 years of the resampled data (gray shaded distribution). The black dashed line indicates the 1% significance level based on the bootstrapping relative frequency distribution. (B) Probability of occurrence of the least severe annual (black), cool season (AMJJAS; blue), and summer (December to February, DJF; salmon) precipitation deficit observed during the 2017–2019 drought, for one, two, and three sequential years as estimated from the LIMs. The solid line shows the distribution constructed using SST data from COBE, and the dotted line shows the distribution constructed using SST from ERSSTv5. Dashed horizontal gray lines show 5% and 1% significance levels.

Fig. 6.. Sea surface temperature and large-scale influences on rainfall in the Tinderbox Drought region.

(A) Spatial correlation of April to September rainfall anomalies in our study region (hatching) with SST anomalies, showing only correlations significant at P < 0.1. Colored boxes show the southeast tropical Indian Ocean (purple, 0 to 10°S and 90° to 110°E), northern Australia (orange, 5° to 20°S and 100° to 160°E), Niño4 (blue, 5°N to 5°S and 160°E to 150°W), and Niño3.4 (red, 5°N to 5°S and 120° to 170°W) regions explored further in (B) to (E). (B to E) Relationship between rainfall anomalies in our study region with SST averaged over regions indicated in (A), for April to September anomalies between 1982 and 2020 (circles), with the 2017–2019 Tinderbox Drought years indicated by diamonds. Data in this figure use the OISST v2 0.25° × 0.25° SST product and the ACGD rainfall product (Materials and Methods). All data in (A) to (E) are linearly detrended to isolate interannual variability and the Niño3.4 relative index (D) is calculated by first removing the tropical ocean mean (83). Anomalies in (B) to (E) are relative to 1982–2016 climatology (Methods). (F) Time series of the Niño3.4 (red), Niño4 (blue), DMI (purple), and SAM (orange) indices between 2016 and 2020. Months that exceed 1 SD of the respective index (computed over 1980–2016) are indicated with markers. Gray shading denotes the cool seasons of the Tinderbox Drought.

Fig. 7.. Sources of moisture during the Tinderbox Drought.

Anomalies of (A to C) oceanic moisture sink (mm/day), (D to F) terrestrial moisture sink (mm/day), and (G to I) 850 hPa winds (m/s, vectors) and wind speed (m/s, shading). Anomalies are calculated relative to April to July 1980–2016 climatology from April to July for [(A), (D), and (G)] 2017, [(B), (E), and (H)] 2018, and [(C), (F), and (I)] 2019 relative to April to July 1980–2016 climatology. Note that the analysis uses a shorter cool season (April to July) due to ERA-Interim data availability (stops in August 2019). April to September moisture source and sink anomalies for 2017 and 2018 can be seen in fig. S9.

Fig. 8.. Rainfall anomalies associated with heavy rainfall days and weather systems during the Tinderbox Drought.

(A) The distribution of the anomalous proportion (%) of seasonal rainfall stemming from heavy rain days from DJF 2015/16 to SON 2020, computed for each grid box in the Tinderbox Drought domain. The anomalous proportion is defined as the proportion of the seasonal rainfall total that falls on days exceeding the climatological 90th percentile of rain days (>0.01 mm/day). Whiskers show the 5th/95th percentiles, the box shows the interquartile range, and the median and mean are denoted by the horizontal line and dot, respectively. For example, if the climatological mean contribution of heavy rain days to a seasonal rainfall total is 70% and during a given year of the drought it was 20%, the value shown is −50%. (B to E) The attribution of weather object frequency and intensity change to the daily rainfall anomalies (in mm/day) averaged over the Tinderbox Drought domain are shown for winter (JJA) of (B) 2016, (C) 2017, (D) 2018, and (E) 2019. Darker colored bars represent rainfall changes related to changes in object frequency, and the lighter shading to the intensity of rainfall associated with each object. The numbers within each bar are the rankings of the frequency and intensity anomalies compared to the full 40 years of data from 1980 to 2019, with 1 being the largest negative anomaly and 40 being the largest positive anomaly. The daily rainfall anomalies are calculated with respect to all winter days in the period 1980–2016. The numbers indicated in the bottom of (B) to (E) indicate the area mean rainfall anomaly for each JJA in mm/day equivalent.

Fig. 9.. Contribution of anthropogenic forcing.

Percentage changes in area-averaged cool season rainfall of (A) the Tinderbox Drought (2017–2019) period, and (B) the driest 3-year period between 2014 and 2023 relative to the 1900–1959 period average in CMIP6 models. Box plots show the spread of change in rainfall based on historical simulations (to 2014) extended to year 2024 with 33 models under SSP5.85, 28 models under SSP3.70, and 31 models under SSP2.45 and SSP1.26 scenarios. We group all SSPs together for this analysis owing to the similarity of forcing in 2017–2019 across all scenarios. The vertical line in the box indicates the median, the box represents the interquartile range, and the whiskers indicate the 5th and 95th percentiles. (C) The range of a possible 3-year change due to internal variability alone based on CMIP6 models under preindustrial conditions. One and two SDs of the distribution due to internal variability alone are shown as vertical dashed lines in black and orange colors. The blue probability distribution is the same as the shaded curve, except that it is shifted left by the median value (i.e., our estimate of the externally forced response) of the box plot shown in (A). The observed % change is indicated using the thick vertical red dashed line in all panels.

Fig. 10.. Characteristics, drivers, and impacts of the Tinderbox Drought.

(A) Map showing the area most affected by the drought, highlighting regions of warm and cool SST anomalies likely to have influenced the evolution of the drought. Aqua arrows show the path of moisture deflected away from the drought region, resulting in precipitation deficits. Graphics in the drought area show some of the major characteristics of the drought. (B) Timeline of key events and amplifiers of the drought, showing the magnitude of anomalies in relevant metrics (blue and warm colors), and the strength of remote climate drivers (browns). The intensity of shading indicates the strength of the respective drivers and anomalies.