Tree mortality predicted from drought-induced vascular damage

The projected responses of forest ecosystems to warming and drying associated with twenty-first-century climate change vary widely from resiliency to widespread tree mortality1–3. Current vegetation models lack the ability to account for mortality of overstorey trees during extreme drought owing to uncertainties in mechanisms and thresholds causing mortality4,5. Here we assess the causes of tree mortality, using field measurementsof branchhydraulic conductivityduringongoing mortality in Populus tremuloides in the southwestern United States and a detailed plant hydraulics model. We identify a lethal plantwater stress threshold that correspondswith a loss of vascular transport capacity from air entry into the xylem. We then use this hydraulic-based threshold to simulate forest dieback during historical drought, and compare predictions against three independent mortality data sets. The hydraulic threshold predicted with 75% accuracy regional patterns of treemortalityas found infieldplotsandmortalitymapsderived from Landsat imagery. In a high-emissions scenario, climate models project that drought stress will exceed the observed mortality threshold in the southwestern United States by the 2050s. Our approach provides a powerful and tractable way of incorporating tree mortality into vegetation models to resolve uncertainty over the fate of forest ecosystems in a changing climate. Forests play a central role in global water, energy and biogeochemical cycles and provide substantial ecosystem services to societies around the globe6. Yet the fate of forest ecosystems in a changing climate is highly uncertain. Rising atmospheric CO2 concentrations may benefit trees, particularly through increasing water-use efficiency7, but concomitant increases in temperature and drought stress could potentially overwhelm these benefits, leading to widespread forest dieback in many ecosystems globally8. Although precipitation projections under climate scenarios are more variable and uncertain, general circulation models project consistent increases in air temperature and thus evaporation over much of the world and resulting decreases in soil moisture in many regions, leading to more intense and frequent droughts9. Recent studies have indicated resilience in forest biomes in response to early twenty-first-century droughts through inter-annual modulations in water-use efficiency10 and long-term increases in forest water-use efficiency7. In contrast, severe regional droughts have strongly decreased the carbon sink of key forest ecosystems11–13 and widespread, climate-induced tree mortality has been observed around the globe8,14. The balance of resiliency versus the potential for widespread forest dieback due to climatic extremes hinges in large part on poorly understood demographic processes, which are not well represented in most dynamic global vegetation models (DGVMs). An increase in mortality rate can be as important as a change in productivity for carbon sinks15. Application of DGVMs to known severe drought stress in controlled rainfall exclusion experiments reveals that they do not accurately capture drought-induced forest dieback5. Thus, there is a compelling need for an approach to simulate spatial and temporal patterns of tree mortality and to test model predictions against regional mortality data sets, such as remote-sensing estimates16. This can then be a foundation for incorporating mortality algorithms into vegetation models that can be trusted for future projections of change, for instance when coupled to global circulation models. Trees die from drought and temperature stress through complex and poorly understood pathways of interrelated physiological failures that often interact with biotic agents4,17. Hydraulic failure through xylem cavitation has been shown to be a major mortality mechanism across a number of angiosperm species18–20. Using a combination of field physiological measurements, a plant hydraulic model21, a hydrologic model22, climate projections and multiple mortality data sets (Supplementary Fig. 1), we tested: does a hydraulic threshold as a function of drought stress emerge across forest sites during ongoing tree mortality; can this mechanistic-based threshold be used to hindcast patterns of mortality with reasonable accuracy compared to multiple mortality data sets; and what do global circulation model projections suggest for future trajectories of this drought stress and exceedance frequency of a mortality threshold? We examined a recent widespread, climate-induced forest die-off of trembling aspen (Populus tremuloides; hereafter aspen) in 91,500 Ha in the southwestern United States. As the most widely distributed tree species in North America with major economic and ecological value23, aspen presents an ideal test case for modelling tree mortality. We quantified drought stress as climatic water deficit24 (CWD)— the difference between plant water demand, which is determined by atmospheric water demand (here, potential evapotranspiration, PET), and plant water supply, which is determined in part by available soil moisture (here, actual evapotranspiration, AET). Owing to low precipitation and high temperatures, CWD reached peak values around 2000–2003, representing the severe drought that initiated ongoing tree mortality (Fig. 1).