Global land carbon sink response to temperature and precipitation varies with ENSO phase

Any decrease in the global land carbon sink acts to exacerbate future climate change, but the actual impact of climate change on carbon dynamics remains highly uncertain (Ciais etal 2014, Cox etal 2000, Friedlingstein etal 2006, Friedlingstein etal 2014). Climate variability associated with the El Niño-Southern Oscillation (ENSO) and the consequent impacts on land carbon sink interannual variability have been used to investigate carbon cycle responses to climate variability, and to inform the sensitivity of the tropical carbon budget to climate change (Cox etal 2000, Cox etal 2013, Gurney etal 2012, Wenzel etal 2014). Past studies have argued that temperature drives ENSO impacts on carbon dynamics (Braswell etal 1997, Clark etal 2003, Cox etal 2013, Jung etal 2017, Kindermann etal 1996, Wang etal 2013, Wang etal 2014) and that the relationship to precipitation is relatively weak (Clark etal 2003, Wang etal 2013, Wang etal 2014). For instance, recent studies show that the correlation between the interannual variation of the atmospheric CO₂ growth rate (AGR, determined by long-term atmospheric CO₂ measurements and driven mainly by global land sink) and temperature is substantially higher (r ranges from 0.5 to 0.7 depending on the estimating methods) than that between AGR and precipitation (r < 0.2) (Wang etal 2013, Wang etal 2014). The strong temperature response was found to be largely driven by tropical ecosystems (Cox etal 2013, Wang etal 2013, Wang etal 2014). Building on this conclusion, studies have further developed a simple emergent constraint on the climate response of the global land sink by using the observation-based sensitivity of the global land carbon sink to interannual tropical temperature variations (Cox etal 2013, Friedlingstein etal 2006). Other studies, and especially modeling studies, have instead pointed to precipitation and moisture availability as key drivers (Qian etal 2011, Tian etal 1998, Zeng etal 2005). In addition, tropical droughts (including major droughts in the Amazon) have also been shown to drive the interannual variability of the global land sink (Gatti etal 2014, Hilker etal 2014). These opposing views lead to fundamentally different interpretations of how climate change and the associated changes in temperature, precipitation, and ENSO dynamics will impact the global land carbon sink. Furthermore, a recent study (Wang etal 2014) found that the temperature sensitivity of tropical ecosystem has increased over the last five decades, and may be modulated by the moisture conditions, implying that interactions between heat and moisture may be important.

Here, we examine how different climate modes, as exemplified by the phase of the El Niño-Southern Oscillation, and their corresponding prevailing climate conditions affect the sensitivity of the global land carbon sink to temperature and precipitation. To do so, we analyze the response of the global land sink to temperature and precipitation following contrasting phases of ENSO. Because heat and moisture conditions differ across ENSO phases, the analysis provides a test case for studying the influence of contrasting climate conditions. Our method is primarily diagnostic, with the aim being to observe responses across different prevailing climate conditions. Although we present high-level hypotheses about driving mechanisms, this study is not intended to replace the need for further attribution studies to deepen process-based understanding. We further use this approach as a new benchmark to evaluate the ability of ten state-of-the-art mechanistic models of terrestrial carbon fluxes to represent the observed sensitivities to climate variations.

Tropical ecosystems have been shown to be the main driver of the land carbon sink response to ENSO (Cox etal 2013, Wang etal 2013, Wang etal 2014, Zeng etal 2005). Hence, we focus here on the correlations between annual atmospheric growth rate of CO₂ (AGR) and tropical (30°S to 30°N) land mean annual temperature (TMAT) and precipitation (TMAP), for the years 1959–2010, to identify key features of the ENSO response. The time series of AGR, TMAT and TMAP are detrended before used for the correlation analyses. Normalized detrended time series of these three variables are shown in figure S1 (stacks.iop.org/ERL/12/064007/mmedia) to illustrate interannual-scale variations. To explore the influence of ENSO on the relationship between the terrestrial carbon sink and environmental drivers, we categorize years as El Niño or La Niña based on the Niño 3.4 index from NOAA's Climate Prediction Center (CPC), and analyze correlations to drivers for years immediately following El Niño/La Niña years as well as for all years. In doing so, we examine contrasts in response between post El Niño versus post La Niña conditions. We also analyze the correlations based on global land sink as simulated by an ensemble of terrestrial biosphere models participating in the Multi-scale Synthesis and Terrestrial Model Intercomparison Project, MsTMIP (Huntzinger etal 2013, Wei etal 2014b), in order to assess model capabilities in capturing the ENSO-dependence of responses.

Atmospheric CO₂ growth rate (AGR)

ENSO index and ENSO event classification

Climatic drivers (TMAT and TMAP)

Terrestrial biosphere models

Statistical analysis

The partial correlations between AGR and TMAT and between AGR and TMAP indicate that the response of the AGR to ENSO varies by ENSO phase (figure 1). The AGR post El Niño is significantly and positively correlated to TMAT (r_TG,P = 0.59, p < 0.01). This response is similar to the average responses across all years identified both in this study (r_TG,P = 0.54, p < 0.01) and in earlier studies (Cox etal 2013, Wang etal 2013, Wang etal 2014). Post La Niña, however, the coupling between TMAT and AGR disappears (r_TG,P = −0.02, p = 0.95). The post La Niña response of AGR to TMAT (b_TG,P = −0.1 ± 0.8 PgC yr⁻¹ K⁻¹) is not significantly different from zero and is significantly lower than for post El Niño (b_TG,P = 3.1 ± 1.1 PgC yr⁻¹ K⁻¹) and for all years (b_TG,P = 2.6 ± 0.5 PgC yr⁻¹ K⁻¹). Note that the result across all years obtained here is consistent with previous estimates (Cox etal 2013, Wenzel etal 2014) if we account for the fact that these earlier studies used tropical land as well as ocean temperatures, yielding a higher regression coefficient (b_TG,P = 4.1 ± 0.7 PgC yr⁻¹ K⁻¹). On the other hand, a negative correlation between AGR and TMAP that is weak across all years (r_PG,T = −0.03, p = 0.81, b_PG,T = − 0.01 ± 0.03 PgC yr⁻¹ (cm yr⁻¹)⁻¹) and post El Niño (r_PG,T = −0.29, p = 0.24, b_PG,T = − 0.05 ±0.04 PgC yr⁻¹ (cm yr⁻¹)⁻¹), becomes stronger and statistically significant (r_PG,T = −0.46, p = 0.04, b_PG,T = −0.11 ± 0.05 PgC yr⁻¹ (cm yr⁻¹)⁻¹) post La Niña.

Zoom In Zoom Out Reset image size

This analysis implies substantial differences in carbon dynamics driving AGR interannual variability dependent on ENSO phase. The post El Niño partial correlation between AGR and TMAT is significantly stronger than that post La Niña (p = 0.02), suggesting a robust shift in the response to temperature. Meanwhile, the significant partial correlation between AGR and TMAP post La Niña indicates that focusing on the AGR response across all ENSO conditions obscures the role of precipitation. These findings are robust to the alternate approaches for calculating the AGR (see Data And Method, figure S2).

An analysis of the variability of fossil fuel emissions, land use and land cover change, oceanic uptake, and fire emissions indicates that the response of AGR to TMAT and TMAP is likely driven by terrestrial ecosystems (supplementary text 1). A lack of significant correlations between AGR and temperature/precipitation in extra-tropical regions also further confirms the key role of tropical ecosystems (supplementary text 1). Additional sensitivity analyses also suggest that the observed responses of AGR to TMAT and TMAP are not driven by extreme events (supplementary text 2).

The timing of the AGR response to ENSO further supports the dependence of the tropical land carbon flux response on the phase of ENSO. The AGR for a given post El Niño year is significantly correlated to the Niño 3.4 index as far back as September of the preceding year, while a significant correlation only appears in January in post La Niña conditions (figure 2(a-b), black). This ENSO response of AGR reflects both the response of TMAT/TMAP to ENSO (figure 2(a-b)), magenta and green) and the response of AGR to TMAT and TMAP (figure 1). The timing of these AGR correlations with the Niño 3.4 index is consistent with the timing of the correlations between TMAT/TMAP and Niño 3.4, where TMAT post El Niño is significantly correlated to Niño 3.4 as far back as September of the previous year (figure 2(a), magenta) while TMAP post La Niña is correlated to Niño 3.4 back starting in January (figure 2(b), green). More broadly, the timing of the correlation between AGR and the Niño 3.4 index (figure 2(a-b), black) closely tracks that of temperature following El Niño conditions (figure 2(a), magenta), and precipitation following La Niña conditions (figure 2(b), green). This result is again consistent with temperature playing a strong role in driving the carbon flux response following El Niño conditions, and precipitation being important in determining the response following La Niña conditions. Similar to what is seen in figure 1, the aggregated response across all years masks the role of precipitation post La Niña, with the AGR response mirroring that of TMAT only (figure 2(c)). These findings are also robust to the choice of a different ENSO index (MEI) (see Data and Method, figure S3).

Zoom In Zoom Out Reset image size

We also estimate the partial correlations between AGR and local temperature, and between AGR and local precipitation, to identify regions that are likely to be driving the observed global relationships. Given that AGR represents a global average, however, this heuristic approach can only be interpreted qualitatively. We find that the significant positive correlation between AGR and local temperature post El Niño occurs across all the major ecosystems (Flexas etal 2014) in the tropics, including forests, grasslands, and semi-arid regions (figure 3 and S4). Conversely, TMAT is not related to AGR in any biome post La Niña (figure S4). The areas with much stronger local temperature correlations post El Niño relative to post La Niña generally overlap with regions showing high temperature anomalies post El Niño (figure 3 and S5).

Zoom In Zoom Out Reset image size

We hypothesize that the differing temperature responses of AGR across ENSO phase are due to the impact of differences in prevalent temperature and moisture conditions on Gross Primary Production (GPP). The hot and dry conditions following El Niño (figure S5(a), (c)) result in a soil moisture deficit and lead to greater water stress (Jung etal 2010, Miralles etal 2014). Under strong water stress, GPP tends to decline with increases in temperature, as observed for a variety of ecosystems (Flexas etal 2014, Nobel etal 1978, Osonubi and Davies, 1980, Tenhunen etal 1985). This is in line with earlier studies showing that soil moisture conditions can control the response of terrestrial ecosystems to temperature fluctuations (Wang etal 2014). The cooler and wetter conditions following La Niña (figure S5(b), (d)), on the other hand, represent more favorable conditions for GPP in many tropical ecosystems (Miralles etal 2014, Poulter etal 2014). With reduced water stress and cooler temperatures, the impact of temperature on vapor pressure deficit is reduced, and thus tropical ecosystem GPP is likely to decrease with decreasing temperature, leading to little temperature influence on net carbon uptake.

Zoom In Zoom Out Reset image size

Although a recent study (Anderegg etal 2015) has suggested that total respiration dominates the interannual variability in the tropical land sink, we argue that respiration responses to temperature are unlikely to explain the different correlations between AGR and temperature across ENSO phases. Respiration is typically positively correlated with temperature, but the temperature sensitivity of respiration has been found to be affected by temperature and soil moisture. Under hotter and dryer conditions, the apparent temperature sensitivity of respiration is expected to be weaker (Moyano etal 2012, Song etal 2014). Therefore, respiration cannot explain the stronger correlation between AGR and TMAT post El Niño (when conditions are dryer and hotter) relative to that post La Niña (when conditions are wetter and cooler), as shown in figure 1. Respiration also cannot explain the increases in the sensitivity of AGR identified over the last five decades (Wang etal 2014).

The significant correlation between AGR and TMAP post La Niña, but not post El Niño (figure 1), is counter-intuitive at first because ecosystems are more strongly water-limited post El Niño. Under such conditions, precipitation variability would be expected to lead to variability in carbon uptake. Following El Niño conditions, however, our results imply that the water stress experienced by tropical ecosystems may be so severe that incremental changes in precipitation will not elicit a net response. This implies that while GPP and respiration may both increase with increased precipitation (Meir etal 2008, Nemani etal 2003, Wang etal 2013), the GPP response does not substantially exceed that of respiration. The weak precipitation post El Niño may reflect buffering mechanisms, such as plant physiological adjustment to water stress, soil water storage, vegetation structural constraints, which have been shown to reduce the impact of precipitation on net primary productivity (NPP) of grasslands and tropical forests under dry conditions (Doughty etal 2015, Knapp and Smith, 2001).

Following La Niña conditions, conversely, certain regions in the tropics are still likely to be water limited, but less severely than Post El Niño, leading to a stronger response of GPP than respiration and hence a significant response of net vegetation growth. The stronger precipitation correlation post La Niña is in line with literature that show that, in the most dynamic biomes, biome-level pattern of aboveground NPP responds more strongly to precipitation during wet than dry years (Doughty etal 2015). In the analysis presented here, the correlation to precipitation post La Niña is stronger over many areas, including tropical forests in the central Amazon and in Southeast Asia, in semi-arid ecosystems in South Africa and Australia, and in grasslands in South Asia (figure 4). On a biome level, the strongest post La Niña correlation between AGR and precipitation occurs for tropical forests (figure S4), which experience higher precipitation relative to other tropical regions. Looking at long-term-average precipitation in more detail, we find that the post La Niña correlation between AGR and annual total precipitation is strongest for regions with a long-term-average precipitation in the range of 2000 to 3000 mm yr⁻¹ (r = −0.60, p = 0.007), and regions that fall into this range are mainly tropical forests in the Amazon and Southeast Asia.

Although recent studies have suggested that semi-arid ecosystems dominate the trend and variability of the land carbon sink (Ahlström etal 2015, Poulter etal 2014), no significant correlation between AGR and precipitation is found on biome level under either ENSO phase (figure S4), suggesting that some regions of relatively strong negative correlation may be getting averaged out by other regions with positive correlation. We also explore whether the longer duration of La Niña relative to El Niño conditions could account for the observed correlation with TMAP, but this is not the case (supplementary text 3). Rather, our results suggest that the wetter conditions post La Niña represent preferable growth conditions and thus lead to substantial additional net terrestrial carbon uptake in tropical ecosystems, especially those with relatively high average precipitation.

Overall, the asymmetrical ENSO response of the global land sink suggests that the hotter and drier climate post El Niño leads to stronger water stress and conditions that are less conducive to carbon uptake. As a result, GPP tends to decrease with higher temperatures while the GPP negative response to lower precipitation is comparable to that of respiration. Under La Niña conditions, on the other hand, GPP increases with cooler temperature, and the positive sensitivity of GPP to higher precipitation is greater than that of respiration. Together, these behaviors lead to the asymmetric response of AGR to TMAT and TMAP observed across ENSO phases. Our findings imply that simply extrapolating the average historical sensitivity of global land sink to temperature and precipitation is insufficient to understand the impacts of climate change on the global land sink, as this misses the impact of climate modes and prevalent climate conditions on temperature and precipitation sensitivities.

To examine if state-of-the-art terrestrial biosphere models (TBMs) capture the observed relationship between AGR and TMAT/TMAP, we apply the same analytical framework to 10 models participating in MsTMIP (table S1) (Huntzinger etal 2013, Wei etal 2014b). The analysis parallels that described above for AGR, but uses modeled tropical net ecosystem exchange (NEE) of CO₂ instead. Here, positive NEE is defined as CO₂ release to the atmosphere.

We find that none of the individual TBM nor the ensemble mean model capture the phase-dependent response of the land biosphere to ENSO (figure 1, S6). For all models, the correlations of tropical NEE to TMAT and TMAP remain largely unchanged across ENSO phases. The fact that the models do not capture the lack of correlation between AGR and TMAT post La Niña suggests that models do not capture the comparable responses in GPP and respiration to temperature under wetter and cooler conditions. In addition, we find that all TBMs erroneously predict a significant correlation with TMAP across ENSO conditions (figure 1, S6), which may explain why previous model-based studies emphasized precipitation as an important driver of carbon sink interannual variability (Qian etal 2011, Tian etal 1998, Zeng etal 2005). We find that overestimates of the sensitivity to precipitation occur across all biomes and across regions spanning the full range of long-term-average precipitations. Earlier site-level studies have suggested that models overestimate the response of GPP to precipitation (Piao etal 2013), consistent with the results presented here. We find that the strong association with precipitation is also evident when looking at tropical GPP, at terrestrial ecosystem respiration, or at global NEE (figure S7(a-b)).

Taken together, the persistent stronger correlations in modeled NEE with both TMAT and TMAP than those derived from AGR observations also lead to an incorrect timing in the response of modeled tropical NEE to ENSO. Following El Niño conditions, the response of modeled NEE to the Niño 3.4 index is delayed relative to the AGR response (figure 2(a)). Following La Niña conditions, conversely, the response of modeled NEE to the Niño 3.4 index occurs earlier than observed in the AGR response (figure 2(b)). In both phases, the range of modeled tropical NEE responses to ENSO falls between the observed TMAT and TMAP responses to ENSO, rather than tracking the response of TMAT/TMAP during post El Niño/La Niña conditions, as observed from the AGR. Analogous analyses of global, rather than tropical, NEE show the same result.

Furthermore, although including nitrogen dynamics in TBMs has been thought to improve the models' representation of carbon cycle responses to temperature (Piao etal 2013), it does not improve model performance in representing correlations with climatic drivers or ENSO for the suite of TBMs examined here (figure S7(c) and S8).

The consistency in the observed biases across all models is especially striking given the vast differences in model structure across the ensemble (Huntzinger etal 2013). It also precludes the possibility of easily teasing out potential mechanisms based on differences in model structure. Instead, it suggests that interactions between water availability and heat are generally not captured in state-of-the-art terrestrial biosphere simulators.

The inability of models to represent how the sensitivity of the global land carbon sink to temperature and precipitation is impacted by prevalent climate conditions points to a key uncertainty in our understanding of climate impacts on the future of the global land carbon sink (Cox etal 2013, Wang etal 2013, Wang etal 2014, Wenzel etal 2014). One potential issue is that existing models may not capture the impact of water availability and stress on the temperature response of tropical ecosystems. With current tropical temperatures already close to the optimal photosynthetic temperature (Clark etal 2003, Corlett, 2011, Doughty and Goulden, 2008), the response of tropical ecosystems to climate change may be highly sensitive to the impact of water stress on the temperature response of GPP. In addition, consistent model overestimation of the precipitation impact suggests that models may misrepresent processes related to how precipitation affects moisture availability, how tropical ecosystems respond to moisture availability, or how water stress impacts this response. Furthermore, because models do not capture the ENSO phase dependency in the response of the tropical carbon cycle to temperature and precipitation, they are unlikely to correctly simulate the influence of future climate variability on the global land carbon sink.

Overall, our findings provide new insights into the roles of, and the interactions between, temperature and precipitation in controlling the global terrestrial carbon sink, and therefore into the future response of the global carbon cycle to climate variability and change. Specifically, we find that the coupling between AGR and both tropical temperature and precipitation varies strongly with prevalent climate conditions, as exemplified by ENSO phase, which has substantial implications for projecting the future land carbon balance. The average linear relationship between AGR and temperature across all ENSO conditions has been used to constrain the sensitivity of tropical land carbon uptake to climate change (Cox etal 2013, Wenzel etal 2014). Results shown here suggest that this approach may be misguided and likely leads to an overestimate of tropical land carbon responses to temperature during post La Niña years, when AGR is insensitive to temperature.

In addition, as climate continues to change in the future, tropical ecosystems may experience a climate shift that will affect their responses to climate variations. For example, if the tropical lands are to be drier and warmer (i.e. more El Niño-like), a stronger mean temperature response of the global land carbon sink is expected than under current estimates. This possibility highlights the importance of studies exploring the joint impacts of heat and water availability on tropical ecosystems, impacts that have rarely been examined (Busch, 2015). We argue that contemporary models need to correctly include mechanisms related to such joint impacts to enable the simulation of the response of tropical ecosystems to climatic drivers as influenced by differences in prevailing climate conditions, and hence to capture the impact of future climate change on the global land carbon sink.

This work was supported by NASA under Grant #NNX14AI54G and by NSF under Grant #1342076. We acknowledge atmospheric CO₂ data from the US National Oceanic and Atmospheric Administration Earth System Research Laboratory (NOAA/ESRL) and the CO₂ program in Scripps Institution of Oceanography (SIO). Estimates of fossil fuel emission, ocean sink, CO₂ release from Land use land coverage, and residual land sink were obtained from the Global Carbon Project (GCP); fire emission of CO₂ was obtained from Global Fire Emission Database version 4.1 (GFED 4.1s); Niño 3.4 index was obtained from National Weather Service Climate Prediction Center; MEI index and the GPCC precipitation data were obtained from the Physical Science Division (PSD) of NOAA/ESRL; CRU-NCEP climatological drivers (precipitation, surface air temperature, etc.), along with simulated terrestrial carbon fluxes, were provided by the Multi-scale synthesis and Terrestrial Model Intercomparison Project (MsTMIP; http://nacp.ornl.gov/MsTMIP.shtml). Funding for the MsTMIP activity was provided through NASA Grant #NNX10AG01A. Data management support for preparing, documenting, and distributing model driver and output data was performed by the Modeling and Synthesis Thematic Data Center at Oak Ridge National Laboratory (ORNL; http://nacp.ornl.gov), with funding through NASA Grant #NNH10AN681. J. B. F contributed from the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. Finalized MsTMIP data products are archived at the ORNL DAAC (http://daac.ornl.gov). We also acknowledge the modeling groups that provided results to MsTMIP and their funding sources. This is MsTMIP contribution #8.