The 2019/20 Black Summer bushfire disaster in southeast Australia was unprecedented: the extensive area of forest burnt, the radiative power of the fires, and the extraordinary number of fires that developed into extreme pyroconvective events were all unmatched in the historical record. Australia’s hottest and driest year on record, 2019, was characterised by exceptionally dry fuel loads that primed the landscape to burn when exposed to dangerous fire weather and ignition. The combination of climate variability and long-term climate trends generated the climate extremes experienced in 2019, and the compounding effects of two or more modes of climate variability in their fire-promoting phases (as occurred in 2019) has historically increased the chances of large forest fires occurring in southeast Australia. Palaeoclimate evidence also demonstrates that fire-promoting phases of tropical Pacific and Indian ocean variability are now unusually frequent compared with natural variability in pre-industrial times. Indicators of forest fire danger in southeast Australia have already emerged outside of the range of historical experience, suggesting that projections made more than a decade ago that increases in climate-driven fire risk would be detectable by 2020, have indeed eventuated. The multiple climate change contributors to fire risk in southeast Australia, as well as the observed non-linear escalation of fire extent and intensity, raise the likelihood that fire events may continue to rapidly intensify in the future. Improving local and national adaptation measures while also pursuing ambitious global climate change mitigation efforts would provide the best strategy for limiting further increases in fire risk in southeast Australia.

This Supplementary Information provides further details on the datasets and data processing (Section 1) and the calculation of CO 2 sensitivity coefficients (Section 2). Description of datasets Study basinsThe 190 study basins were selected from the Zhang et al. (2013) 1 streamflow collection based on the completeness of streamflow records and land use within the basins. Basins meeting the following criteria were included in the analysis:• Intensive land use or irrigated land constituted less than 5% of the basin area• Daily streamflow time series during 1982-2010 was >95% complete• Fewer than 12 consecutive months had missing daily values present

Quantifying how climate change drives drought is a priority to inform policy and adaptation planning. We show that the latest Coupled Model Intercomparison Project (CMIP6) simulations project coherent regional patterns in meteorological drought for two emissions scenarios to 2100. We find robust projected changes in seasonal drought duration and frequency (robust over >45% of the global land area), despite a lack of agreement across models in projected changes in mean precipitation (24% of the land area). Future drought changes are larger and more consistent in CMIP6 compared to CMIP5. We find regionalized increases and decreases in drought duration and frequency that are driven by changes in both precipitation mean and variability. Conversely, drought intensity increases over most regions but is not simulated well historically by the climate models. The more robust projections of meteorological drought compared to mean precipitation in CMIP6 provides significant new opportunities for water resource planning.

Compound events (CEs) are weather and climate events that result from multiple hazards or drivers with the potential to cause severe socio-economic impacts. Compared with isolated hazards, the multiple hazards/drivers associated with CEs can lead to higher economic losses and death tolls. Here, we provide the first analysis of multiple multivariate CEs potentially causing high-impact floods, droughts, and fires. Using observations and reanalysis data during 1980–2014, we analyse 27 hazard pairs and provide the first spatial estimates of their occurrences on the global scale. We identify hotspots of multivariate CEs including many socio-economically important regions such as North America, Russia and western Europe. We analyse the relative importance of different multivariate CEs in six continental regions to highlight CEs posing the highest risk. Our results provide initial guidance to assess the regional risk of CE events and an observationally-based dataset to aid evaluation of climate models for simulating multivariate CEs.

Knowledge of how water stress impacts the carbon and water cycles is a key uncertainty in terrestrial biosphere models.We tested a new profit maximization model, where photosynthetic uptake of CO 2 is optimally traded against plant hydraulic function, as an alternative to the empirical functions commonly used in models to regulate gas exchange during periods of water stress. We conducted a multi-site evaluation of this model at the ecosystem scale, before and during major droughts in Europe. Additionally, we asked whether the maximum hydraulic conductance in the soilplant continuum k max (a key model parameter which is not commonly measured) could be predicted from long-term site climate.Compared with a control model with an empirical soil moisture function, the profit maximization model improved the simulation of evapotranspiration during the growing season, reducing the normalized mean square error by c. 63%, across mesic and xeric sites. We also showed that k max could be estimated from long-term climate, with improvements in the simulation of evapotranspiration at eight out of the 10 forest sites during drought.Although the generalization of this approach is contingent upon determining k max , it presents a mechanistic trait-based alternative to regulate canopy gas exchange in global models.

Aridity is a complex concept that ideally requires a comprehensive assessment of hydroclimatological and hydroecological variables to fully understand anticipated changes. A widely used (offline) impact model to assess projected changes in aridity is the aridity index (AI) (defined as the ratio of potential evaporation to precipitation), summarizing the aridity concept into a single number. Based on the AI, it was shown that aridity will generally increase under conditions of increased CO 2 and associated global warming. However, assessing the same climate model output directly suggests a more nuanced response of aridity to global warming, raising the question if the AI provides a good representation of the complex nature of anticipated aridity changes. By systematically comparing projections of the AI against projections for various hydroclimatological and ecohydrological variables, we show that the AI generally provides a rather poor proxy for projected aridity conditions. Direct climate model output is shown to contradict signals of increasing aridity obtained from the AI in at least half of the global land area with robust change. We further show that part of this discrepancy can be related to the parameterization of potential evaporation. Especially the most commonly used potential evaporation model likely leads to an overestimation of future aridity due to incorrect assumptions under increasing atmospheric CO 2 . Our results show that AI-based approaches do not correctly communicate changes projected by the fully coupled climate models. The solution is to directly analyse the model outputs rather than use a separate offline impact model. We thus urge for a direct and joint assessment of climate model output when assessing future aridity changes rather than using simple index-based impact models that use climate model output as input and are potentially subject to significant biases.

South‐East Australia has recently been subjected to two of the worst droughts in the historical record (Millennium Drought, 2000–2009 and Big Dry, 2017–2019). Unfortunately, a lack of forest monitoring has made it difficult to determine whether widespread tree mortality has resulted from these droughts. Anecdotal observations suggest the Big Dry may have led to more significant tree mortality than the Millennium drought. Critically, to be able to robustly project future expected climate change effects on Australian vegetation, we need to assess the vulnerability of Australian trees to drought. Here we implemented a model of plant hydraulics into the Community Atmosphere Biosphere Land Exchange (CABLE) land surface model. We parameterized the drought response behaviour of five broad vegetation types, based on a common garden dry‐down experiment with species originating across a rainfall gradient (188–1,125 mm/year) across South‐East Australia. The new hydraulics model significantly improved (~35%–45% reduction in root mean square error) CABLE’s previous predictions of latent heat fluxes during periods of water stress at two eddy covariance sites in Australia. Landscape‐scale predictions of the greatest percentage loss of hydraulic conductivity (PLC) of about 40%–60%, were broadly consistent with satellite estimates of regions of the greatest change in both droughts. In neither drought did CABLE predict that trees would have reached critical PLC in widespread areas (i.e. it projected a low mortality risk), although the model highlighted critical levels near the desert regions of South‐East Australia where few trees live. Overall, our experimentally constrained model results imply significant resilience to drought conferred by hydraulic function, but also highlight critical data and scientific gaps. Our approach presents a promising avenue to integrate experimental data and make regional‐scale predictions of potential drought‐induced hydraulic failure.

Land surface models (LSMs) must accurately simulate observed energy and water fluxes during droughts in order to provide reliable estimates of future water resources. We evaluated 8 different LSMs (14 model versions) for simulating evapotranspiration (ET) during periods of evaporative drought (Edrought) across six flux tower sites. Using an empirically defined Edrought threshold (a decline in ET below the observed 15th percentile), we show that LSMs simulated 58 Edrought days per year, on average, across the six sites, ∼3 times as many as the observed 20 d. The simulated Edrought magnitude was ∼8 times greater than observed and twice as intense. Our findings point to systematic biases across LSMs when simulating water and energy fluxes under water-stressed conditions. The overestimation of key Edrought characteristics undermines our confidence in the models' capability in simulating realistic drought responses to climate change and has wider implications for phenomena sensitive to soil moisture, including heat waves.