Examining the sensitivity of the terrestrial carbon cycle to the expression of El Niño

expression of El Niño Lina Teckentrup1,2, Martin G. De Kauwe1,2,3, Andrew J. Pitman1,2, and Benjamin Smith4,5 1ARC Centre of Excellence for Climate Extremes, Sydney, NSW, Australia 2Climate Change Research Centre, University of New South Wales, Sydney, NSW, Australia 3Evolution & Ecology Research Centre, University of New South Wales, Sydney, NSW 2052, Australia 4Hawkesbury Institute for the Environment, Western Sydney University, Penrith, NSW, Australia 5Department of Physical Geography and Ecosystem Science, Lund University, Lund, Sweden Correspondence: Lina Teckentrup (l.teckentrup@unsw.edu.au)

and the original base periods to create the manipulated climate forcing.
This approach only isolates the effect of different expressions of El Niño to a limited extent since the calculated anomalies can also be influenced by other climate modes of variability. For example in Australia and Indonesia, different expressions of El Niño and different phases of the IOD can combine to drive the fire season (e.g. Pan et al., 2018). Given we create two synthetic forcings with respectively 15 CP (nine events replaced) and EP (eight events replaced) El Niño events, we assume 130 that the emerging signal in the model results will be representative of the effect of different expressions of El Niño on the carbon balance.
We use the identical approach for the GSWP3 dataset. The only difference is the shorter length of the time period covered:  [1976][1977][1978][1979][1980][1981][1982][1983][1984][1985][1986] baseperiod [1976][1977][1978][1979][1980][1981][1982][1983][1984][1985][1986] forcing 1976-1986 -baseperiod 1976-1986 = anomaly [1976][1977][1978][1979][1980][1981][1982][1983][1984][1985][1986]   We processed the data with netCDF Operators (NCO; version 4.7.7. http://nco.sf.net) and climate data operators (CDO; version 1.9.5. http://mpimet.mpg.de/cdo). The data analysis is conducted with python version 3.   However, the IAV does not lead to sustained trends in the ecosystem fluxes. The spatial distribution of the flux anomalies in the final year of the experiment (2013) displays spatial variability rather than systematic patterns, implying that the imposed changes also did not lead to long-term shifts in ecosystem processes at regional scales (see appendix figure B1). Only EP vs CTRL (cumulative) Only CP vs CTRL (cumulative) Figure 2. Total net biome production (NBP) (a-c), cumulative NBP (d-f) as well as absolute difference and cumulative sums of the difference between CP-only-scenario and control climate and EP-only-scenario and control climate (g-i).    fig. 3c). Over the 45 years, the accumulated GPP leads to a differences of 54.6 PgC and 28.8 PgC for the CP-and the EP-only-scenario, respectively, and this is largely balanced by the accumulated TER, 47.5 PgC and 22.3 PgC for the CP-and the EP-only-scenario, respectively (see fig. 3a, b). For NBP, GPP and TER, a CP-only-scenario leads to stronger increases compared to an EP-only-scenario both globally and for tropical regions (see fig. 2g and h; see fig. 3a, b, d, e). In Australia, this ratio shifts for GPP and TER so that the cumulative sums of GPP (7 PgC) and TER (5.6 PgC) anomalies in an EP-only-165 scenario exceed those in a CP-only-scenario (5.8 PgC for GPP and 5.2 PgC TER) in 2013 (see fig. 3g, h). The cumulative carbon lost through fires declines in an CP-only-scenario globally and in the tropical regions and is close to zero for Australia (see fig. 3c, f, i). In contrast to the absolute differences in the fluxes in the year 2013 (see above), cumulative GPP and TER show a clear(er) pattern with increases for both fluxes in southern South America and over Australia (see appendix figure B3).

Results
The accumulated increases in GPP however are balanced by increases in TER so that cumulative NBP shows strong spatial   In both CP-and EP-only-scenario, the total differences in terrestrial carbon pools are largely the result of the responses of tropical ecosystems. Similar to the carbon fluxes, no clear patterns in the spatial distribution of the carbon pool anomalies 180 emerge (see appendix figure B4).

Discussion
The El Niño Southern Oscillation (ENSO) strongly influences global and regional climate and has the potential to modify the regional and global carbon balance. Here, we examine whether two expressions of El Niño (CP and EP), as distinct from the El Niño phenomenon itself, modifies the regional and global carbon balance. This is timely: EP El Niño events might become 185 more extreme in the future (e.g. Wang et al., 2019;Cai et al., 2018) and the occurrence of CP El Niño events seems to have increased over the later half of the 21st century and may increase further in the future (e.g. Yeh et al., 2009;Ashok et al., 2009). While the impact of more extreme (EP) El Niño events has been examined (e.g. Kim et al., 2017), there are few studies exploring the impact of different expressions of El Niño on the terrestrial carbon cycle. Previous work has focused on short time scales and explored either time lag effects on the carbon growth rate (Chylek et al., 2018), single regions and/ or single 190 events (Amazonia, Li et al. (2011); Indonesia, Pan et al. (2018)), or the composite anomalies in the carbon fluxes (Wang et al., 2018) in a larger spatial context. In effect, the response of ecosystems to different expressions of El Niño on longer timescales is not well understood.
In this study we show that, in line with previous studies (e.g. Wang et al., 2018;Chylek et al., 2018) climate anomalies associated with different expressions of El Niño have a strong impact on the IAV of ecosystem carbon fluxes. The El Niño-195 associated climate anomalies in our experiments do not show a consistent pattern but rather display high temporal variability between individual El Niño events (see appendix figure B5). Wang et al. (2018) showed that between El Niño events, the atmospheric CO 2 growth rate varied by 4 PgC yr −1 at the peak for EP events and ∼2 PgC yr −1 for CP events. Consequently, the ecosystem fluxes vary strongly in their response to different expressions of El Niño for individual years. Overall, the high spatial and temporal variability in the changes suggest that the effect of different expressions of El Niño on the terrestrial carbon cycle are important for predicting responses on interannual timescales (e.g. the atmospheric CO 2 growth rate) but are unlikely to affect the terrestrial carbon balance on longer timescales. Our model results imply that the anomaly patterns in the El Niño expression on climate forcing were too variable (and short-lived) to result in systematic shifts in 210 species composition. Nevertheless, the marked IAV of carbon fluxes implies an underlying sensitivity that may be particularly important for predictability of the carbon balance in drier ecosystems and/or water-limited agricultural regions.

Future directions
In our study we used a single model, so we cannot quantify uncertainties associated with alternative representations and/or missing processes. For example, LPJ-GUESS, similar to many land surface and dynamic global vegetation models, does 215 not account for acclimation of plant respiration to increased temperature, and may consequently overestimate the carbon sensitivity to temperature changes on short time scales (e.g. Wang et al., 2020;Huntingford et al., 2017;Smith et al., 2015).
Future experiments will further need to explore how rising CO 2 and temperature may change the relative balance of GPP uptake and carbon losses via respiration during El Niño events. Wang et al. (2018) showed that the TRENDY model ensemble (which includes an LPJ family member) generally captured the NBP anomalies associated with CP El Niño events and only 220 underestimates the anomalies associated with extreme EP El Niño events. This suggests results obtained with LPJ-GUESS would be broadly typical across other similar DGVMs.
To place our results into a broader context, we examined whether LPJ-GUESS captures anomalies associated with different expressions of El Niño in the carbon cycle similarly to other models. We used the TRENDY V5 S2 run with transient CO 2 forcing and climate, but no imposed land use change. We choose the seven state of the art DGVMs CLM4.5 (Oleson et al., 2013), JSBACH (Reick et al., 2013), JULES (Clark et al., 2011), LPX (Keller et al., 2017, ORCHIDEE (Krinner et al., 2005), VEGAS (Zeng et al., 2005) and VISIT (Kato et al., 2013) to calculate the TRENDY composite. LPJ-GUESS matches the TRENDY composite well for GPP and TER for the global, tropical and Australian averages with high correlation coefficients for the global and Australian averages (0.78-0.85) and low to moderate correlation coefficients for the tropics (0.13-0.55) except for the GPP anomaly associated with EP El Niño events (0.84) (see appendix figure B10). Similarly, the R 2 -values are 230 low for all tropical anomalies (0.05-0.3) except the GPP anomaly for EP El Niño events (0.76), and moderate to high for the remaining regions (0.59-0.78). In general, LPJ-GUESS displays greater variability than the TRENDY composite but is mostly within the model range (except for the GPP anomaly for CP El Niños; see appendix figure B10). Based on this analysis we suggest that our model sensitivity would likely be similar to that displayed by the other TRENDY models.
A further research path may consider driving a model with a larger ensemble of meteorological forcing to account for 235 uncertainties associated with global climate reanalysis products. We conducted the same experiment based on the GSWP3 climate forcing and found that the overall variability in all terrestrial ecosystem flux and carbon pool anomalies is similar compared to the experiment based on the CRUNCEP dataset but with a smaller magnitude (see appendix figures B8 and B9). Wu et al. (2017) showed that the simulated GPP by LPJ-GUESS could vary by as much as 11 PgC yr −1 globally, due to the use of alternative climate forcing data sets. Nevertheless, in their analysis Wu et al. (2017) showed that overall, the magnitude 240 of tropical GPP was largely robust to the use of different precipitation forcing, although there was variation regionally.
In this study, we run LPJ-GUESS with active stochastic and fire disturbance. Including these two types of disturbance contributes significantly to the spatial variability (compare appendix figures B1 and B2). Our results show that the fire patterns in LPJ-GUESS are largely insensitive to the imposed changes due to the expression of El Niño, which is in contrast with observational studies that suggest that El Niño events themselves are strongly linked to fire activity on regional scales (e.g. Pan Finally, isolating the effect of El Niño on the atmosphere and terrestrial biosphere is not trivial for individual events. Individual El Niño events vary in location, timing and magnitude and teleconnections are influenced by the background climate and climate variability (e.g. the Indian Ocean Dipole). In our study, we assume that replacing a CP event with an EP event, or vice 255 versa, did not modify the role played by other modes of variability. By generating two experiments with either 15 (nine manipulated events) CP or 15 (eight manipulated events) EP El Niño events, we assume that the signal observed at the end of the time period is driven by the respective expression of El Niño. In order to test the validity of our results, we applied a different approach where we replaced the climate forcing of CP El Niño years with the climate forcing of the EP El Niño events closest in time and found even smaller changes in carbon fluxes (see appendix figure B6) and in carbon sequestration (see appendix figure B7). An alternative approach could be to calculate composite anomalies for both CP and EP El Niño events and use those for replacement, but this would dampen variability in the forcing and introduce a different bias. Alternatively, generating a sea surface temperature forcing representing the different expressions of El Niño and using an atmospheric model to generate the climate anomalies that result from the changes in sea surface temperatures could help quantify the effect of the expression of El Niño on the carbon sequestration. However, given the changes we found are very small and spatially variable we doubt 265 this would lead to different conclusions.

Conclusions
We explored the impact of the expression of El Niño on the terrestrial carbon cycle on multi-decadal timescales using LPJ-GUESS. The changes in anomalies reflecting the two expressions of El Niño in NBP accumulate to around 7.7 PgC (CPonly-scenario) and 2.6 PgC (EP-only-scenario). However, this accumulation period cover more than 45 years and is therefore 270 negligible compared to annual anthropogenic emissions of 9.4 +-0.5 PgC yr -1 (Le Quéré et al., 2018). Our results therefore suggest that the impact of different expressions of El Niño on the carbon cycle on long time scales is negligible.
The results imply that simulations of the terrestrial carbon cycle over the recent past and into the future using global climate models may not require the expression of El Niño events to be well captured. There are major challenges in capturing El Niño -La Niña cycles and the teleconnections associated with El Niño events well with existing climate models. Had we 275 found the expression of El Niño to be critical in simulating the long term terrestrial carbon balance, this would have added a very significant additional uncertainty to projections of the future role of the land in storing carbon. Our results suggest that the expression of El Niño, as distinct from whether there is an El Niño or a La Niña, is relatively unimportant over the long term. We note that our results do agree with earlier studies (Chylek et al., 2018;Wang et al., 2018;Pan et al., 2018) that the expression of El Niño is important to terrestrial carbon fluxes on annual and interannual timescales. Overall, in the context 280 of the long-term global and regional terrestrial carbon balance, our results imply that model development should prioritise simulating El Niño -La Niña cycles and the associated teleconnections, without needing to consider the additional challenge of resolving the expression of individual El Niño events.
Code and data availability. The analysis codes are available at https://github.com/lteckentrup/nino_experiment). The model code is available upon request from http://web.nateko.lu.se/lpj-guess/contact.html. The model outputs will be shared in line with UNSW's open-access policy 285 on publication. The TRENDY version 5 model output is available upon request (https://sites.exeter.ac.uk/trendy) and the CRUNCEP climate forcing is available from https://rda.ucar.edu/datasets/ds314.3/. Table A1. El Niño events from 1968-2010 identified by the NOAA Oceanic Niño Index (ONI) and their different expressions derived by four methods according to Yu and Kim (2013): pattern correlation method ('PTN'; Yu and Kim, 2013); central location method ('Niño'; Kug et al., 2009;Yeh et al., 2009), the El Niño Modoki Index ('EMI'; Ashok et al., 2007) and the cold tongue/ warm pool index ('CT/WP'; Ren and Jin, 2011). We define a CP or EP El Niño where three out of the four indices agree on the same El Niño type. The remaining events are defined as mixed events ('MIX').
Year Dominant El Niño type 1968-1969CP 1969-1970EP 1972-1973EP 1976-1977EP 1977-1978       Extremes (CE170100023). MDK and AJP acknowledge support from the ARC Discovery Grant (DP190101823). MDK was also supported from the NSW Research Attraction and Acceleration Program. We further acknowledge the TRENDY DGVM community, as part of the Global Carbon Project, for access to their model outputs. We thank the National Computational Infrastructure at the Australian National University, an initiative of the Australian Government, for access to supercomputer resources. Finally, we thank Matthew Forrest for his support in running LPJ-GUESS.