Worldwide increases in plant mortality events have been associated with extreme stress conditions caused by intensifying climate variations1–3. In turn, severe mortality events affect regional carbon, water, and energy cycling by altering biosphere-atmosphere exchange processes, potentially perturbing regional to global climate systems, exacerbating the severity, frequency, and extent of climatic extremes4,5. Indeed, no specific vegetation type or climate zone is safe from the effects of predicted future warming and droughts, even in regions historically endowed with ample water6,7, yet some may be better equipped to withstand such pressures. However, current specifications of dynamic global vegetation models do not account for the wide variance among plant hydraulic properties and their water use strategies, rendering plant mortality predictions uncertain8,9. Therefore, understanding the coupled response of carbon sequestration and water consumption to climate-induced stress and identifying species and ecosystems particularly vulnerable to climate extremes are essential for accurate predictions of future vegetation dynamics and distributions10,11.
Drought has been, and will remain one of the major threats to humans and the natural systems12,13. Observational and modeling studies have demonstrated that increasing atmospheric aridity, indicated by vapor pressure deficit (VPD), is chief among drought-related constraints to terrestrial carbon sinks and ecosystem function14,15. When climate regimes are localized, less affected by air masses from drier or wetter regions, high VPD and low soil moisture are correlated over the range of soil moisture below which stomatal conductance is affected16. In the context of climate change, ecosystem-scale stomatal response to decreasing soil moisture and increasing VPD is much more pronounced than observed responses to increasing atmospheric [CO2]17. Indeed, progressive change in climate, during which the feedback between soil water and VPD drives increasingly drying terrestrial surfaces and more severe plant water stress, could cause abrupt and irreversible shifts in biotic communities, affecting their biodiversity, ecosystem functions, and carbon storage4,18. Correspondingly, the way plants respond to the variation of atmospheric demand for moisture, especially under droughts, determines community resilience and stability.
Plant water-use strategies largely determine the tradeoff between carbon uptake and water use19,20. At the ecosystem scale, there is no definitive evidence that water-use strategy (WUSe) is a vital mechanism of drought response strategy. However, the role of WUSe in the maintenance of ecosystem functions during extreme droughts and the recovery following droughts has not been well assessed24. Furthermore, hydraulic behavior is largely considered fixed to a species21,22. However, evidence is mounting that plant hydraulic behavior is the product of the interaction of biology and growth environment20. We propose that the sensitivity of canopy stomatal conductance (Gc) to VPD (Sc) is a factor contributing to the sustainability and resilience of ecosystems encountering droughts14. Along a continuum of Sc, ecosystems may experience varying degrees of carbon starvation or hydraulic failure under water-limited conditions23. Here, we used Sc to explore how WUSe regulates carbon sequestration during droughts and predict the potential drought-induced risk of ecosystem dysfunction worldwide.
Canopy stomatal sensitivity regulated by water availability in arid regions
We evaluated WUSe based on canopy stomatal response to VPD relative to a theoretical response, one consistent with the role of stomata in regulating the xylem water potential (Ψx), thus protecting its hydraulic function25. The normalized deviation from theoretical canopy stomatal sensitivity to VPD25, Sc (see Fig. 4 in Methods) was used to classify ecosystems as hypersensitive when positive (>1.05 the theoretical) and hyposensitive when negative (<0.95 the theoretical), the latter consistent with anisohydric behavior. We consider Sc in the intermediate range as consistent with isohydric behavior, consistent with conservative regulation of Ψx. The average empirical ratio of dGc/dln(VPD) (m) to reference canopy stomatal conductance (Gcref, i.e., conductance at VPD = 1 kPa) was 0.62 ± 0.01 ln(kPa)–1 (mean ± standard error) across all 165 sites, smaller than expected (0.64 ± 0.01; p < 0.001, paired t-test), yet larger than the ratio determined by porometric and sap flow methods, and based on hydraulic theory25 (0.60; p = 0.003, one-sample t-test). The distribution of Sc was fairly uniform for each ecosystem type, ranging mostly between –0.2 to 0.2 (Fig. 1a). For wetlands and croplands, the empirical m/Gcref was significantly lower than the theoretical ratio (p = 0.001 and 0.003, respectively), and Sc was lower than that of other ecosystems (p < 0.05). Although this may suggest highly anisohydric behavior (i.e., weak regulation of Ψx), reflected in hyposensitivity of stomatal conductance to VPD, it is more likely that evaporation of free water from open water, and wet soil and canopy (even though precipitation effects were removed), introduced a VPD-dependent bias in estimates of canopy-scale stomatal conductance based on eddy covariance data. Thus, we excluded these ecosystems from further analyses. Among the remaining datasets, there was no difference in Sc between angiosperm- and gymnosperm-dominated ecosystems (p = 0.102; Fig. 1b).
At several flux sites, the normalized slope of Gc to VPD changed with soil moisture14; we found that Sc decreased with multi-year mean ecosystem water availability (α) only in arid regions (Fig. 1c). Although the range of data from humid regions was shifted to slightly greater water availability, ecosystems in both arid and humid regions may experience α ranging from ample to limiting. Sc exponentially varied with α in arid regions (p < 0.001), increasing from –0.044 at α = 0.8, to 0.061 at α = 0.3 (Fig. 1c). In contrast, Sc was insensitive to the range of α in humid regions (p = 0.925), remaining stable at around –0.025. Thus, ecosystems in arid and humid regions shared Sc at high α, but clearly separated as water availability decreased.
The prevailing water conditions control not only ecosystem functions but also plant xylem structure and hydraulic conductivity26, jointly determining WUSe. In relatively arid sites, Sc varied with α within a narrow range of soil moisture27. Thus, for a given site, WUSe depends on plant hydraulic characteristics and long-term environmental conditions19,20, changing relatively little even in drought years, as indicated by m and Gcref decreasing proportionally during droughts (Supplementary Fig. 1). Although we found no evidence that WUSe was affected by ecosystem type (Fig. 1a) or ecosystem-scale hydraulic traits (as reflected in plant xylem type; Fig. 1b), WUSe was considerably influenced by the prevailing habitat water availability in arid regions only (Fig. 1c). For ecosystems in humid regions, permissive WUSe prevails regardless of available moisture, perhaps reflecting acclimation to lesser atmospheric demand for moisture. It is notable that in arid regions, where habitat moisture was not very limiting (α > 0.5), hyposensitive ecosystems showing permissive WUSe were ~2.2-fold more common than hypersensitive ecosystems of conservative WUSe, while the ecosystems with the conservative behavior were ~1.7-fold more common in drier habitats.
Ecosystem water-use strategies under droughts
Water-deficit stress lowered Gcref (z-score) in moderate and extreme droughts (Fig. 2a), independent of the climatic region, suggesting that some loss of hydraulic conductivity in the soil-plant pathway occurs under drought. Although Gcref anomaly was linearly related to the intensity of droughts regardless of WUSe, hyposensitive ecosystems showed greater sensitivity than hypersensitive ecosystems (p = 0.038, ANCOVA; Fig. 2a). Thus, although there was no difference in the Gcref anomaly between the two types of ecosystems during moderate droughts (p = 0.695), hyposensitive ecosystems showed more negative Gcref anomaly than hypersensitive ecosystems during extreme droughts (p = 0.047). Consistent with the canopy stomatal conductance response to drought, the annual net ecosystem productivity (NEP) of most ecosystems decreased with drought intensity (Fig. 2b). Focusing on hypersensitive and hyposensitive ecosystems, the slope of the linear relationship between NEP anomaly (z-score) and drought severity was 1.17 and 1.11, respectively, showing similar sensitivity (p = 0.899, ANCOVA). Too few data preclude a firm conclusion, but extreme droughts seem to impact NEP of hyposensitive ecosystems less than hypersensitive ecosystems during the drought year. Drought intensity noticeably affected the ratio of leaf area index (LAI) to NEP (Fig. 2c). The LAI/NEP anomaly of the two types of ecosystems was similar at low drought intensity and very different under extreme drought conditions. Specifically, under extreme drought conditions, the mean LAI/NEP anomaly for hyposensitive ecosystems was 0.64 ± 0.15 and remained high as drought intensified, but declined in hypersensitive ecosystems to an average of –0.08 ± 0.07.
Combined with the responses of Gcref, LAI/NEP, and NEP to drought intensity, hypersensitive and hyposensitive ecosystems achieve similar net carbon uptake through very different responses: Hypersensitive ecosystems respond to droughts through a reduction of LAI, allowing the residual foliage to maintain higher conductance, thus reducing the impact on NEP, while hyposensitive ecosystems show a permissive WUSe, hanging onto their inefficient leaves, potentially with a lesser impact on NEP. Thus, both ecosystem types meet increasing drought severity with decreasing rate of carbohydrate production, yet in very different states: Hypersensitive with lower LAI potentially at its maximum efficiency; hyposensitive maintaining high LAI of very low efficiency.
Moreover, Gcref of most hypersensitive ecosystems (91%) recovered to its multi-year non-drought mean the following year, but that of the majority (67%) of hyposensitive ecosystems, including all those experiencing extreme droughts, did not (Supplementary Fig. 2). Indeed, the ratio of Gcref following droughts to that of non-drought years across all hypersensitive ecosystems (1.02 ± 0.03), similar to that of isohydric systems (1.00 ± 0.07), was not different from unity (p = 0.357), while that of the hyposensitive ecosystems (0.90 ± 0.02) was lower (p = 0.001; Fig. 3a), indicating that hyposensitive ecosystems likely experienced a substantial loss of hydraulic conductance (e.g., cavitation, embolism, root surface-area decline) during droughts. Although LAI/NEP of the year following a drought year recovered to a normal state, NEP of hyposensitive ecosystems was still less than the multi-year non-drought level (0.84 ± 0.07; p = 0.040; Fig. 3a). Consequently, few hypersensitive ecosystems show carryover effects of droughts, while the cost of permissive WUSe was a failure of most hyposensitive ecosystems to recover fully from droughts.
Consistent with these findings, moving from ecosystems with conservative WUSe to those less regulated, as indicated by lower Sc (see Fig. 1c), ecosystem resistance (Rt) and recovery (Rc) displayed divergent trends (Fig. 3b). Specifically, Rc increased (R2 = 0.27, p < 0.001) while Rt sharply decreased (R2 = 0.23, p = 0.002) with Sc across all ecosystems. When Sc is between 0.04 and 0.16, the difference between Rt and Rc is the smallest, delineating the range in which the ability of an ecosystem to both persist and recover from droughts to its long-term quasi-stable equilibrium is balanced (Fig. 3b). We note that this region is occupied by hypersensitive ecosystems, while hyposensitive ecosystems display higher Rt but lower Rc. Therefore, the results suggest that hyposensitive ecosystems, behaving consistently with anisohydric hydraulics, recover slower from a drought year due to the increased hydraulic resistance of the soil-plant pathway.
Hyposensitive ecosystems at risk of mortality under extreme droughts
Although drought can cause substantial stress to plants, with potential physiological, biochemical, and morphological effects11, their impact on ecosystem functions ranges from none to severe depending on drought intensity and duration as well as WUSe. We found that moderate-intensity drought did not greatly impact ecosystem carbon sequestration and hydraulic functions, and the responses were similar in hyposensitive and hypersensitive ecosystems (Fig. 2). However, when ecosystems experienced extreme drought stress, hyposensitive and hypersensitive ecosystems exhibited different structural and functional strategies despite a relatively similar decline in carbon sequestration (Fig. 2). Specifically, hypersensitive ecosystems decreased the degree of carbon starvation caused by extreme droughts, reducing the proportion of structural carbohydrates aboveground (e.g., LAI) while increasing leaf carbon sequestration efficiency (Fig. 2c). Reducing LAI by shoot and leaf dieback are strategies that improve the water supply to the residual foliage28, maintaining conductance and photosynthetic rates while permitting higher partitioning of photosynthates to non-structural carbohydrates (NSC, e.g., sugars and starch) in support of plant metabolism and osmolality29. Such loss of canopy elements may reflect a conservative behavior when biochemical regulations are insufficient to withstand extreme droughts. This behavior, reflected in yellowing foliage following a few weeks of severe drought, is commonly observed in ecosystems composed of fast-growing species of high gas-exchange leaves30, representing a preference for cutting losses short, giving up return on carbon invested in a portion of foliage rather than risk the entire hydraulic system and mortality if droughts persist. In contrast, hyposensitive ecosystems kept their LAI as drought severity increased, thus keeping photosynthetically less efficient foliage, a strategy that may allow recouping the investment in foliage over the short term, but risking the hydraulic system and future carbon uptake.
The question remaining is, which strategy may be more detrimental to plant survival and ecosystem stability not only during droughts, but in subsequent years. We found no difference between hyposensitive and hypersensitive ecosystems in ecosystem-scale P50 (i.e., Ψx at which 50% loss of conductivity) and hydraulic safety margin (HSM) (Supplementary Fig. 3). Although NEP of the ecosystems with different WUSe responded to droughts in a similar way, the consequences of their strategies reflect diverging effects on ecosystem stability and resilience, at least during the year following droughts. A mechanism advanced in support of less strict stomatal regulation is the prevention of drought-induced carbon starvation, but this may result in hydraulic failure and a cascade of consequences23. A number of studies documented drought-induced tree mortality often associated with substantial loss of hydraulic conductivity, yet plant death was not uniformly caused by insufficient NSC because carbohydrate distribution and allometric growth changed with drought2,31. Indeed, these two attributes may be feeding back on one another21,32: decreased NSC in aboveground woody tissues (e.g., twig, branch, and bole) during extreme droughts was closely tied with reduced xylem resistance to embolism, impeding the maintenance of turgor and cell metabolism; in turn, reduced hydraulic conductance slowed xylem refilling and regrowth following osmotic regulation and drought relief32,33. Our analysis shows that ecosystems displaying hyposensitive canopy stomatal control had lower stability (i.e., the absolute value of Rc–Rt) during droughts than hypersensitive ecosystems (Fig. 3b), suggesting that they might be more susceptible to mortality from extreme, prolonged droughts. This was further supported by the lesser recovery of Gcref and NEP to normal the year following droughts (Fig. 3a and Supplementary Fig. 2). Utilizing permissive WUSe under extreme droughts, thus escaping the consequences of a marked decrease in NEP at the cost of greater risk to the hydraulic system, may be a riskier WUSe.
These findings suggest that ecosystems can be divided into those which display hyposensitive sensitivity of canopy stomatal conductance to VPD and others showing hypersensitive behavior from the perspective of WUSe, with the majority of ecosystems behaving in a manner consistent with isohydric regulation of Ψx. Ecosystems showing drastically different behavior may facilitate understanding processes responsible for ecosystem dysfunction and allow modeling and predicting ecosystem stability under droughts.
Long-term additional water supply (e.g., irrigation) in arid and semi-arid regions is widely applied, enhancing ecosystem services by promoting plant growth and yield34. Although such management increases plant photosynthetic productivity to some extent, long-term excessive irrigation will result in permissive WUSe (Fig. 1c), decreasing plant water-use efficiency and further exacerbating regional water deficit. For instance, a large proportion of forest plantations in China’s Three-North Shelter Regions are intensively irrigated to ensure survival and enhance multiple ecosystem functions35. Such irrigation is conducive to accelerating tree growth by eliminating drought stress in the early stages of stand development. However, the permissive WUSe developed by additional water supply in these plantations will deplete groundwater, leading to water shortages, while the resulting WUSe may lead to the death of trees unprepared to withstand extreme droughts36. In contrast, unirrigated, water-deficit ecosystems in arid regions are adapted to their growth environment and survive with a relatively conservative WUSe (Fig. 1c). Furthermore, as suggested by our results, WUSe in humid regions is generally permissive; a continuous increase in VPD may result in higher water use with consequences to water resources and ecosystem stability in the future.