Pervasive drought legacies in forest ecosystems and their implications for carbon cycle models

See allHide authors and affiliations

Science  31 Jul 2015:
Vol. 349, Issue 6247, pp. 528-532
DOI: 10.1126/science.aab1833

Drought effects on carbon cycling

The response of forest ecosystems to drought is increasingly important in the context of a warming climate. Anderegg et al. studied a tree-ring database of 1338 forest sites from around the globe. They found that forests exhibit a drought “legacy effect” with 3 to 4 years' reduced growth following drought. During this postdrought delay, forests will be less able to act as a sink for carbon. Incorporating forest legacy effects into Earth system models will provide more accurate predictions of the effects of drought on the global carbon cycle.

Science, this issue p. 528


The impacts of climate extremes on terrestrial ecosystems are poorly understood but important for predicting carbon cycle feedbacks to climate change. Coupled climate–carbon cycle models typically assume that vegetation recovery from extreme drought is immediate and complete, which conflicts with the understanding of basic plant physiology. We examined the recovery of stem growth in trees after severe drought at 1338 forest sites across the globe, comprising 49,339 site-years, and compared the results with simulated recovery in climate-vegetation models. We found pervasive and substantial “legacy effects” of reduced growth and incomplete recovery for 1 to 4 years after severe drought. Legacy effects were most prevalent in dry ecosystems, among Pinaceae, and among species with low hydraulic safety margins. In contrast, limited or no legacy effects after drought were simulated by current climate-vegetation models. Our results highlight hysteresis in ecosystem-level carbon cycling and delayed recovery from climate extremes.

Anthropogenic climate change is projected to alter both climate mean conditions and climate variability, leading to more frequent and/or intense climate extremes such as heat waves and severe drought (1). Increasing variability is likely to profoundly affect ecosystems, as many ecological processes are more sensitive to climate extremes than to changes in mean states (24). In turn, the impacts of these extremes can have major effects on ecosystem-level carbon cycling, feeding back to accelerate or limit climate change. The 2003 European heat wave, for example, led to the development of a strong anomalous carbon source, reversing 4 years of carbon uptake by terrestrial ecosystems on a continental scale (5).

Forest ecosystems store nearly half of the carbon found in terrestrial ecosystems (6), but the fate of forests under climate change and with increasing climate extremes remains uncertain and controversial. Whereas some studies contend that large regions of forest are poised on the verge of collapse into an alternate state (79), others suggest that forests are relatively resilient and likely to experience only modest changes (1012). The sensitivity of forests to climate extremes has become apparent in global patterns of widespread forest mortality (13), which highlight the possibility that the forest carbon sink could be weakened or could even transition rapidly to a carbon source in some regions (1315). Thus, the response of forest growth and mortality to extreme drought and heat constitutes a large uncertainty in projections of terrestrial carbon cycle feedbacks (16).

The treatment of drought in carbon cycle models is limited by a lack of representation of ecosystem response dynamics, such as recovery after drought and the potential for legacies or hysteresis—dynamics that are probably critical to predicting the future behavior of the system (17, 18). For example, lags in precipitation, particularly in semi-arid regions, have been shown to be important in the interannual variability of the land carbon sink (19). In current climate–carbon cycle models, plant physiological recovery from drought is often assumed to be complete and relatively fast. This is at odds with current understanding of physiological mechanisms in many ecosystems, particularly those with long-lived individual plants. Legacy effects and hysteresis after drought have been documented in stomatal conductance (20, 21), wood anatomy and density (22), xylem vulnerability to drought (23), drought-induced tree mortality (24, 25), and aboveground primary productivity (21, 26). As a biological legacy, the dynamics of recovery from severe drought can have a major influence on an ecosystem’s vulnerability to subsequent drought events, particularly if the drought return interval is shorter than the recovery time (17). The rate of recovery—for example, in the reestablishment of hydraulic function after drought—is largely unknown for the vast majority of tree species (24).

We tested the occurrence, prevalence, and magnitude of legacy effects after severe drought using tree growth (i.e., tree-ring width) stand-level chronologies from 1338 sites across the globe, primarily in Northern Hemisphere extra-tropical forest ecosystems, which collectively represented 49,339 site-years. We selected tree-ring master chronologies (typically of 10 to 20 trees per site) from the International Tree Ring Data Bank (27) that contained at least 25 years of data between 1948 and 2008. We defined drought legacy as a departure of observed tree growth (ring-width index) from expected growth (based on the relationship between growth and climate) in the period after a drought episode. Wood growth is ideal to test for drought legacy effects, because it provides a long temporal record and has major implications for the carbon cycle. Wood is a carbon pool with slow turnover that stores immense amounts of ecosystem carbon (6), and wood growth is tightly correlated with net primary productivity (28). We further examined the extent to which observed legacy effects are simulated in current climate-vegetation models from the Coupled Model Intercomparison Project, Phase 5 (CMIP5). We asked: (i) Are legacy effects after extreme drought pervasive in tree growth? (ii) Are legacy effects more prominent in wet or dry environments? (iii) Do legacy effects vary among species with different hydraulic safety margins (29) [a measure of how closely a tree approaches catastrophic damage to its xylem during drought (25)]? (iv) Are the legacy effects simulated in CMIP5 climate-vegetation models similar to those observed in tree rings?

We quantified legacy effects in tree-ring width chronologies using two methods: (i) the departure of observed from predicted growth recovery after drought based on correlations with climate and (ii) partial autocorrelation coefficients. We focused primarily on sites where ring-width anomalies exhibited significant correlations (r > 0.3; mean correlation r = 0.51) with drought [climatic water deficit (30)], because our aim was to quantify the duration of growth suppression or enhancement after drought episodes. We found significant legacies in radial growth after severe drought (>2 SD from the mean climatic water deficit) that lasted 2 to 4 years (Fig. 1A and fig S1). These effects were substantial in magnitude: a ~9% decrease in observed versus predicted growth in year 1 and 5% in year 2 after drought (Fig. 1A). Legacy effects were observed regardless of the minimum climate correlation cutoff (fig. S1) or the drought variable used (figs. S2 and S3). Legacy effects were also observed in the partial autocorrelation analysis (fig S4). There did not appear to be a strong link between the magnitude of the legacy effect and the peak intensity of the observed drought [coefficient of determination (R2)= 0.01, P = 0.08] (fig S5).

Fig. 1 Legacy effects are substantial and persist for 3 to 4 years.

Legacy effects are quantified as the difference between observed and predicted growth (unitless index) after a 2-SD dry anomaly in the climatic water deficit (drought). (A) Legacy effects observed across all 1338 tree-ring chronologies (dashed line) and across 695 tree-ring chronologies at sites that correlate significantly with the climatic water deficit (solid line and red shaded region). (B) Legacy effects at sites from among the above 695 that were categorized as arid (mean annual precipitation <500 mm) or wet (mean annual precipitation >1000 mm). (C) Legacy effects at sites from among the above 695 that support either of the two main families represented, Pinaceae and Fagaceae. Shaded regions in all panels represent the 95% confidence interval around the mean from bootstrapping (n = 5000 resamplings).

Legacy effects were most pronounced in arid ecosystems (Fig. 1B). Mean annual precipitation was the only significant predictor of the magnitude of drought legacy effects in tree growth; it explained a low proportion of the variance (R2 = 0.05, P = 0.0003) (fig. S6). Correlations with mean annual temperature and potential evapotranspiration were both insignificant (P > 0.05). Strong legacy effects also tended to occur in semi-arid regions in the Northern Hemisphere (Fig. 2A) and where correlations between growth and drought were higher (Fig. 2B). Tree-ring chronologies in the southwestern and midwestern United States and in parts of northern Europe exhibited particularly strong legacy effects (Fig. 2A). Positive legacy effects, where observed growth was higher than predicted after drought, were most frequent in California and the Mediterranean region (Fig. 2A).

Fig. 2 Legacy effects are most prevalent in the southwestern and midwestern United States and parts of northern Europe.

Legacy effects are quantified as the difference between observed and predicted growth (unitless index) after a 2-SD dry anomaly in the climatic water deficit across 1338 sites. (A) Site-level legacy effect summed over the first 4 years after drought. (B) Average correlation between tree growth (ring-width index) and the climatic water deficit (soil moisture from 0 to 100 cm minus potential evapotranspiration).

Gymnosperms exhibited legacy effects that were slightly but significantly larger (in terms of magnitude and duration) than those exhibited by angiosperms (t = 2.25, P = 0.02) (fig. S7). Among families, Pinaceae (pines) and Fagaceae (mostly oaks) were best represented in the data set, accounting for >90% of chronologies analyzed. Pines exhibited substantially larger legacies than did oaks (Fig. 1C). Although pines were typically found at drier locales than oaks (average mean annual precipitation for pines = 660 mm/year; average for oaks = 760 mm/year), a model allowing for interactions between precipitation and family was highly significant (t = 2.55, P = 0.01), indicating that such interactions were important. Both wet and dry pine sites exhibited strong negative legacy effects, whereas wet oak sites exhibited slightly negative legacy effects, and dry oak sites had strong positive legacy effects (fig. S8). Pines also had stronger negative legacy effects than the other main gymnosperm family in the database, Cupressaceae (fig. S9). This result is consistent with Cupressaceae’s generally higher drought tolerance relative to Pinaceae (31) and is supportive of a hydraulic damage mechanism underlying legacy effects.

Several physiological mechanisms may underlie the observed legacy effects of reduced growth after drought. Loss of leaf area and/or stored nonstructural carbohydrates during drought may impair growth in subsequent years (25). Pest and pathogen impacts may lag drought or accumulate in drought-stressed trees, thereby lowering growth rates (25). Finally, stress-induced shifts in xylem anatomy and associated vulnerability to hydraulic dysfunction, or remnants of drought-induced xylem cavitation, could impair water transport and, therefore, growth (25). Although data that could test the first two hypotheses are not available, testing the third hypothesis is possible with an existing global hydraulic trait database (29). We found that species with lower hydraulic safety margins, defined as the water potential (Ψ) at which 50% conductivity is lost minus the minimum measured water potential (Ψ50 – Ψmin), exhibited larger legacy effects (R2 = 0.33, F = 4.95, P = 0.04) (Fig. 3 and table S1). This indicates that the species most at risk of hydraulic damage are also those that have the slowest growth recovery after drought. Previous studies at individual sites have observed drought-induced shifts in plant hydraulics, especially in the first 3 to 4 years after drought in oaks and poplars (22, 25), and our results generalize these findings across many taxonomic groups and a broad geographic range.

Fig. 3 Higher legacy effects are associated with species with low hydraulic safety margins.

Integrated legacy effects are quantified as the difference between observed and predicted growth (unitless index) after a 2-SD dry anomaly, summed over 1 to 4 years, averaged across all droughts within a chronology, and averaged across all chronologies for a given species. Each point represents a species where legacies and hydraulic traits were both available. Error bars represent ± 1 SE. The regression line is in red.

The CMIP5 models captured few to no detectable legacy effects from severe drought in grid cells where the tree-ring chronologies were located (Fig. 4). In many cases, interannual variability of wood carbon growth was low and more weakly correlated with water limitation or drought (mean correlations of R = 0.01 to 0.09) than were the observed tree-ring widths at the same locations (mean correlation R = 0.25). Only the Geophysical Fluid Dynamics Laboratory Earth System Model 2G (GFDL ESM2G) exhibited significant legacy effects of 1 to 2 years (Fig. 4A), and these were of lower magnitude than the observed legacies (Fig. 1A). GFDL ESM2G and CanESM (Canadian Centre for Climate Modeling and Analysis Second Generation Earth System Model) both use a dynamic carbon allocation scheme, but they use different approaches to allocate carbon, particularly under drought conditions. GFDL ESM2G’s scheme (32) is based on the pipe model for the relationship between sapwood area and leaf area (33) and allows drought-induced loss of living carbon, including from the sapwood pool, which may allow it to capture legacy effects. Most CMIP5-class models use constant fractional allocation among the vegetation pools and do not simulate plant hydraulic damage during drought, and these appears to be crucial limitations to capturing legacy effects of drought.

Fig. 4 Legacy effects after drought are not captured in predictions of woody biomass by Earth system models.

(A to F) Legacy effects after a 2-SD dry anomaly in grid cells that correlated significantly with drought and overlapped with the real-world locations of the 1338 tree-ring chronologies. Shaded regions represent the 95% confidence interval around the mean from bootstrapping (n = 5000 resamplings). Models used included GFDL ESM2G (A), the Norweigan Earth System Model (NorESM) (B), the Community Earth System Model (CESM) (C), CanESM (D), the Beijing Climate Center Climate System Model (BCC-CSM) (E), and the Hadley Centre Global Environmental Model (Had-GEM) (F).

The response of terrestrial ecosystems to drought has been reported to be one of the largest uncertainties in the carbon cycle (34) and is not well represented in current climate-vegetation models, as evidenced by our model-data comparison. Current models lack representation of some basic physiological and structural properties of plants, such as the vulnerability of xylem transport to hydraulic water stress, that lead to growth suppression, legacy effects, and drought-induced mortality (35). Mortality is generally not measured or reported at these sites, so our analysis does not examine drought-induced mortality; however, mortality or canopy dieback of surrounding trees could generate some of the positive legacy effects in surviving trees that we observed via increased resource availability. Although the impacts of climate extremes on plant mortality and species turnover will also influence carbon cycling (14), we detected a strong, pervasive, and previously undocumented legacy effect of drought on tree growth, especially in dry regions. That is, even when climatic conditions return to normal, surviving trees do not recover their expected growth rates for an average of 2 to 4 years. Given that (i) woody plant growth is a central component of carbon storage and often correlated with productivity and (ii) semi-arid regions play a prominent role in the variability of the global carbon cycle (19), these legacy effects have potential ramifications for the interannual variability of ecosystem-level carbon cycling and for long-term carbon storage. For example, a simple conservative estimate based on forests in the southwestern United States revealed that legacy effects could lead to 3% lower carbon storage in semi-arid ecosystems over a century, equivalent to 1.6 metric gigatons of carbon when considering all semi-arid ecosystems across the globe (30).

Drought could lead to changes in carbon allocation by trees, with less being allocated to bole growth and more to roots or leaves (36, 37), which would mean that growth declines might not immediately reflect decreases in carbon uptake by forests. The fast turnover of leaves and roots, however, would still result in overall decreases in ecosystem-level carbon storage relative to ecosystems without legacy effects (37). The prominence of legacy effects in tropical forests, where tree-ring analyses are challenging, is a major remaining question. There are some indications of legacy effects in the Amazon rainforest in satellite (38) and time-series inventory plot analyses (39) after the severe 2005 and 2010 droughts. The lack of legacy effects in CMIP5 models indicates that drought impacts and their effect on carbon cycling are not accurately captured. These findings reveal the critical roles of contingency and hysteresis in ecosystem response after climate extremes.

Supplementary Materials

Materials and Methods

Figs. S1 to S10

Table S1

References (4067)


  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: Funding for this research was provided by NSF (grant no. DEB EF-1340270). W.R.L.A. was supported in part by a NOAA Climate and Global Change Postdoctoral Fellowship, administered by the University Corporation for Atmospheric Research. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the opinions or policies of the funding agencies. All tree-ring data are available at We thank all data contributers at the International Tree-Ring Data Bank. All CMIP5 data are available at We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in the supplementary materials) for producing and making available their model output. For CMIP,the U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led the development of software infrastructure in partnership with the Global Organization for Earth System Science Portals.
View Abstract

Navigate This Article