Regional Responses of Vegetation Productivity to the Two Phases of ENSO

The two phases of El‐Niño‐Southern Oscillation (ENSO) influence both regional and global terrestrial vegetation productivity on inter‐annual scales. However, the major drivers for the regional vegetation productivity and their controlling strengths during different phases of ENSO remain unclear. We herein disentangled the impacts of two phases of ENSO on regional carbon cycle using multiple data sets. We found that soil moisture predominantly accounts for ∼40% of the variability in regional vegetation productivity during ENSO events. Our results showed that the satellite‐derived vegetation productivity proxies, gross primary productivity from data‐driven models (FLUXCOM) and observation‐constrained ecosystem model (Carbon Cycle Data Assimilation System) generally agree in depicting the contribution of soil moisture and air temperature in modulating regional vegetation productivity. However, the ensemble of weakly constrained ecosystem models exhibits non‐negligible discrepancies in the roles of vapor pressure deficit and radiation over extra‐tropics. This study highlights the significance of water in regulating regional vegetation productivity during ENSO.


Introduction
Approximately 30% of the anthropogenic carbon dioxide (CO 2 ) emissions are absorbed by terrestrial ecosystems via photosynthesis globally, rendering them a significant carbon sink (Friedlingstein et al., 2023).Terrestrial ecosystem productivity is closely linked to Earth's climate dynamics.El Niño-Southern Oscillation (ENSO), as one of the most prominent climate variability patterns, exerts its influence on inter-annual variability (IAV) of climate globally through extensive teleconnections (Bastos et al., 2018;Singh et al., 2022;Wang et al., 2018).This phenomenon significantly impacts variability in both global and regional vegetation productivity.
ENSO-induced sea surface temperature anomalies impact temperature and rainfall patterns in the tropics as well as extra-tropics through teleconnections that span the globe (Diaz et al., 2001;Park et al., 2020).Previous studies have shown that El Niño events reduced vegetation productivity in the tropics, while La Niña increased vegetation productivity in the tropics, by controlling variation in temperature and precipitation/water availability (Liu et al., 2017;Park et al., 2020;Zhang et al., 2019).However, impacts of the two ENSO phases on the vegetation productivity exhibit considerable variations across regions in both tropics and extra-tropics (Bastos et al., 2013(Bastos et al., , 2018;;Liu et al., 2017;Zhang et al., 2019).It remains unclear whether the climate-vegetation productivity relationships as detected from the El Niño events or over certain regions are applicable to the La Niña events or other regions.Consequently, untangling the diverse effects of ENSO phases on both the tropics and extra-tropics becomes crucial for more precise estimation and prediction of terrestrial ecosystem carbon cycles.This urgency intensifies as projections indicate an anticipated increase in the frequency and magnitude of future ENSO events (Park et al., 2020).
Previous work has predominantly focused on analyzing the effects of single ENSO events or the specific ENSO phase in tropical regions, utilizing either models or observational data (Bastos et al., 2013;Fang et al., 2017;Liu et al., 2017;Park et al., 2020).However, exploration in the extra-tropics has been limited due to presumed weak teleconnections between ENSO and vegetation productivity, along with the offsetting effects at subcontinental scales that the impacts of ENSO might be alleviated/cancelled out by other climate modes/disturbances such as AMO/PDO etc. (Lu et al., 2023;Zhang et al., 2019).Recent advancements in satellite technology and observation-derived data sets have significantly improved the spatial representation as well as accuracy of vegetation productivity.Consequently, there is an emerging opportunity to robustly analyze the impacts of two phases of ENSO events on vegetation productivity in tropical and extra-tropical regions (Hu et al., 2019;Luo et al., 2018;Zhang et al., 2019).
In this study, we comprehensively assessed the impacts of two phases of ENSO on regional vegetation productivity using multiple lines of evidence during 1980-2015.Our analysis involved various methodologies, including the widely employed terrestrial ecosystem models (TEMs) used in Earth System Models (ESMs), satellite-derived vegetation productivity proxies (VPPs), and data-driven vegetation productivity assessments.Additionally and for the first time, we utilized the observation-constrained model outputs from the Carbon Cycle Data Assimilation System (CCDAS) (Rayner et al., 2005) for comprehensive inter-comparison with the other data sets.

Carbon Cycle Data Assimilation System and Simulation Experiments
Three experiments were conducted using the CCDAS: a prior simulation experiment (CCDAS_prior) with the prior model parameters, a co 2 simulation experiment (CCDAS_co2) assimilating atmospheric CO 2 concentrations from eight sampling stations of the Scripps Institution of Oceanography (SIO) network (https://scrippsco2. ucsd.edu/, last accessed 2023-06-28) (Table S1 in Supporting Information S1), and a sm+co 2 simulation experiment (CCDAS_post) simultaneously assimilating surface soil moisture data from the European Space Agency's (ESA)-Climate Change Initiative (CCI) (Dorigo et al., 2017;Gruber et al., 2017Gruber et al., , 2019) and SIO atmospheric CO 2 data.The simulation period was 1980-2015, and for computational efficiency over the 36 years, the spatial resolution was set as 8°× 10°.This assimilation process aimed to optimize numerous process parameters associated with physiology, phenology, and soil hydrology and carbon (Table S2 in Supporting Information S1), similar to the previous CCDAS work (Scholze et al., 2019;Wu et al., 2020) but at a longer time scale (i.e., 36 years).

Data
We utilized the GRACE-REC terrestrial water storage (TWS) data set (Humphrey et al., 2018) spanning from 1980 to 2015, resampled to an 8°× 10°resolution, to depict soil moisture availability.
To further validate our findings, we included two vegetation productivity proxies (VPPs) in our analysis, that is, the long-term solar induced fluorescence (SIF) product (LT_SIF) (Wang et al., 2022) and the NIRv_GPP product (Wang et al., 2021).
Since ENSO events have lagged impacts on the vegetation productivity, we took this time-lag into account by using a 7-month moving-average mean for the GPP/VPPs data sets based on our previous analysis (Wu et al., 2020;Xing et al., 2023).
Then, different types of ENSO events were defined for both ENSO indices to detect the event-based characteristics of the carbon cycle.These two indices were de-seasonalized by subtracting the climatology and normalized by the standard deviations.For each standardized ENSO index, the moderate-to-strong El Niño events (anomaly > 1) and moderate-to-strong La Niña events (anomaly < 1) were defined.The neutral ENSO events were then defined as when the absolute value of the ENSO index anomaly was <1.

ENSO Impacts on Vegetation Productivity
Similar to Kim et al. (2017), we conducted the sensitivity analysis on ENSO impacted carbon cycle by using the GRACE-REC TWS, and surface temperature (T), vapor pressure deficit (VPD), and net radiation (Rn) derived from the CRU_TS 4.3 (https://crudata.uea.ac.uk/cru/data/hrg/cru_ts_4.03/, last accessed 2023-06-28) and CRUNCEP_v7.We employed 4 factors in our analysis, which extended the approaches in Kim et al. (2017) who only analyzed impacts from temperature and precipitation induced by ENSO, by unveiling the ENSO impacts on multiple environmental factors: The partial derivative terms in Equation 1were calculated based on the multiple-linear regression coefficients between GPP and T, TWS, VPD, and Rn; while the ordinary derivative terms were derived as the univariate linear regression coefficients between T, TWS, VPD, Rn, and the ENSO indices.We further conducted the calculation for El Niño and La Niña events separately to detect their distinct impacts.

Contributions of Climatic Variables to Vegetation Productivity
To further disentangle the contributions to vegetation productivity from different climatic variables, we defined the contribution of a specific variable (var m ) at grid-scale under spatial resolution n as: where C var m is the contribution of variable m (e.g., T, TWS, VPD, or Rn) (%); M is the number of variables, M = 4 here for the variables T, TWS, VPD, and Rn; and γ var m is the sensitivity of variable m (T, TWS, VPD, Rn).

The Temporal and Spatial Impacts of ENSO
We first evaluated the performance of CCDAS outputs with detailed validation of CCDAS depicted in Text S1 in Supporting Information S1.The CCDAS_post experiment shows good performance in representing the impacts of periodic ENSO events on soil moisture as well as on the physiology of ecosystems from different regions, with higher correlations to the ENSO cycle period for the CCDAS_post experiment than the prior experiment (Figure S1 in Supporting Information S1).Specifically, the assimilation of soil moisture alongside atmospheric CO 2 concentrations has profoundly improved the model performance, notably reducing the global mean root-meansquare-error (RMSE) of soil moisture by 1.28 mm and of net ecosystem productivity (NEP) by 0.51 gC/m 2 / month (Figure S2 in Supporting Information S1).Additionally, the parameters related to physiology, phenology, soil hydrology as well as energy balance and carbon cycle (e.g., Vcmax, T ϕ , C w0 , s rdepth , s b , emis0, Offset) experienced substantial constraints (Table S2 in Supporting Information S1).

The Dominant Role of Soil Moisture in Controlling Vegetation Productivity
Previous studies have argued that the impacts of soil moisture on vegetation productivity are compensated by temperature on large scales (Humphrey et al., 2021;Jung et al., 2017).It remains unclear if this finding applies over different regions during the ENSO events.We comprehensively evaluated the relative contributions of various climatic variables on GPP during ENSO events across different spatial scales using multiple lines of evidence.The spatial evolution of the contributions of four climatic drivers, that is, TWS, T, VPD, and Rn, to the IAV of GPP CCDAS shows that the TWS plays a consistently dominant role in controlling the IAV of the multiscale GPP (Figure 1f; Figure S7 in Supporting Information S1).The TWS dominantly contributes approximately 40% of the IAV of the GPP at multiple scales.Meanwhile, T and VPD contribute approximately 20% each, whereas Rn contributes around 10%.
Consistent findings were observed in satellite-derived VPPs and GPP from the TRENDY model ensembles, emphasizing the dominant role of TWS (40% or more) in controlling vegetation productivity across spatial scales during ENSO events, albeit with contrasting contributions from other climatic variables (Figure 2; Figure S8 in Supporting Information S1).FLUXCOM_GPP depicts T as a secondary driver (contributing around 20%-30%), whereas NIRv_GPP shows comparable contribution from T and VPD.LT_SIF notably indicates substantial contribution from Rn at certain spatial scales (reaching a maximum of ∼45%).In contrast, the TRENDY models consistently position VPD as the second-largest contributor (averaging ∼20% contribution) across various spatial scales.Additionally, within the observation-constrained CCDAS, an augmented correlation between GPP and VPD was observed across most regions (Figure S9 in Supporting Information S1).However, this enhanced correlation did not alter the dominant role of T over VPD in controlling large-scale GPP (Figure 1f), indicating the robustness of the CCDAS results.

The Asymmetry in the Climatic Impacts During El Niño and La Niña
We employed the partial correlations between vegetation productivity and four climatic factors during the ENSO or El Niño and La Niña events separately to linearly interpret the apparent impacts of two phases of ENSO events, as shown in Figure 3 (see also Figures S10 and S11 in Supporting Information S1).Across data sets, unanimous agreements were observed regarding the impacts of TWS in tropics (TRO) during ENSO, with four out of five data sets concurring on the impacts of T and VPD in TRO.Discrepancies in the impacts of Rn between the two phases of ENSO events were identified specifically by LT_SIF, NIRv_GPP and TRENDY_S3 in TRO.Furthermore, NIRv_GPP and TRENDY_S3 show opposing signs in partial correlations with Rn compared to the other three data sets.
Generally, we noticed that in Europe (EU), the majority (≥3) of the data sets depict the asymmetric relationships between vegetation productivity and TWS, with the ratios of partial correlations for El Niño to La Niña being 0.60 to 2.58 (Figure S11a in Supporting Information S1).These asymmetric relationships hinder the comprehensive understanding of ENSO's regional impacts by portraying insignificant partial correlations (Figure 3; Figure S10 in Supporting Information S1).Similarly, the asymmetric relationships between vegetation productivity and T were also observed in Africa (AF), with the partial correlation ratios for El Niño to La Niña ranging from 0.80 to 2.11 (Figure S11b in Supporting Information S1).While for AF and Oceania (OC), we detected the asymmetric relationships between vegetation productivity and VPD, though we found more regions like EU, OC and TRO with asymmetric relationships in vegetation productivity and Rn (Figures S11c and S11d in Supporting Information S1).

Contributions of Climatic Factors to Vegetation Productivity During ENSO Events
The spatial analysis of the contributions of multiple climatic variables to GPP CCDAS reveals distinct regional differences in their impacts on vegetation productivity (Figure 4; Figures S12 and S13 in Supporting Information S1).Notably, TWS shows larger contributions to vegetation productivity compared to T across most of the tropics and the Asian extra-tropics (Figure 4c).TWS shows predominantly negative contributions across most regions, consistent across all five data sets, except in Southern Extra-tropics (SET).Similarly, negative T contributions to vegetation productivity were observed across most regions, except in South America (SA), OC, and SET, consistent across the five data sets.The negative T sensitivity identified in our study primarily stemmed from divergent sensitivities of regional vegetation productivity to climatic factors, as well as the distinct regional sensitivities of climatic variables to ENSO events (Figure S14 in Supporting Information S1) (Kim et al., 2017).
Generally, consistency was observed from FLUXCOM_GPP, LT_SIF, NIRv_GPP, and CCDAS_post, highlighting the robust role of TWS versus T in influencing regional vegetation productivity changes during ENSO events across different data sets.However, TRENDY_S3 shows contrasting results over certain regions, such as EU, North America (NA) and Northern Extra-tropics (NET) (Figure 4f).Notably, both CCDAS_post and TRENDY_S3 show the contrasting outcomes in SA (Figure 4f), indicating the possible limitations of TEMs in detecting the vegetation productivity changes over the world's largest tropical forests.Furthermore, we observed more than half of the data sets showcasing that the El Niño events resulted in contrasting and larger TWS contributions than La Niña to vegetation productivity in NA and AF, with the ratios of contributions from El Niño to La Niña being 2.33 to 14.17 and 1.82 to 69.99, respectively (Figure 4d; Figures S12d and S15 in Supporting Information S1).Similarly, the El Niño events led to contrasting and larger T contributions in EU, compared to La Niña events with ratios of contributions ranging from 1.07 to 1.30 (Figure S15 in Supporting Information S1).These robust estimates revealed that the sensitivity of vegetation productivity to the water and temperature changes during El Niño in these regions are asymmetrically reversed during La Niña events.

Role of Environmental Factors Controlling Vegetation Productivity During ENSO
This study revealed a robust dominance of TWS in controlling vegetation productivity across various spatial scales during ENSO events, elucidated through more mechanistic analysis using multiple data sets.This finding contradicts the previous understanding that favored temperature as the dominant factor (Park et al., 2020).However, a recent investigation by Liu et al. (2023) indicates an increasing influence of soil moisture on vegetation productivity in tropics in recent decades.This supports our observation of the predominant contribution of TWS to GPP/VPP in TRO during ENSO events.Therefore, we suggest that more data sets are necessary to better quantify the ENSO impacts both in tropics and extra-tropics.Moreover, previous studies have primarily relied on correlating vegetation productivity with climatic variables during specific ENSO events (Bastos et al., 2018;Fang et al., 2017;Luo et al., 2018;Zhang et al., 2019).However, they did not isolate the impacts of ENSO from other factors such as human activities and alternate climate patterns (e.g., AMO, PDO).We illustrate that the simplistic correlation approach can lead to misinterpretations regarding ENSO's effects in various regions.This also clarifies the significant disparities observed in Figure 3 and Figure S10 in Supporting Information S1 when examining partial correlations during ENSO events.

ENSO Impacts Over Extra-Tropical Regions
Interestingly, we found that the data sets have shown consistencies in representing the TWS control on the vegetation productivity across regions, even with exceptions for some products, for example, CCDAS_post in EU, LT_SIF in NA, CCDAS_post and FLUXCOM_GPP in SA, TRENDY_S3 in NET, and TRENDY_S3 and LT_SIF in SET.These discrepancies may arise from the uncertainties in the models or satellite products (Table S2; Figure S2 in Supporting Information S1), as shown in previous CCDAS work (Scholze et al., 2016(Scholze et al., , 2019;;Wu et al., 2020) and the variations detected in individual TENDY models (see Figures S16-S21 in Supporting Information S1), as well as the comparison of multiple data sets over certain regions (He et al., 2023;Luo et al., 2018;Stocker et al., 2019).Furthermore, we robustly identified the asymmetry in the contributions of El Niño and La Niña-induced changes in TWS or T to the variability in regional vegetation productivity variability over NA, AF or EU.The contributions from El Niño events consistently surpassed those from La Niña events over these regions, primarily attributed to the asymmetry in the TWS and T changes during the two phases of ENSO (Park et al., 2020;Wang, 2018).Our findings challenged the traditional notions that El Niño-induced drought reduces vegetation productivity while La Niña-induced wetness enhances it, primarily focusing on tropical regions (Bastos et al., 2013(Bastos et al., , 2018;;Liu et al., 2017).Given that the two phases of ENSO create the contrasting water and temperature extremes (e.g., drought vs. flooding, heatwave vs. frost), the divergent contributions of TWS and T during these phases imply a shift in vegetation productivity response regimes and might impact the future global and regional carbon cycles.Therefore, it is recommended that Earth system models carefully consider the mechanisms of vegetation responses to different climate extremes induced by distinct ENSO phases.

Uncertainties and Implications for Carbon Cycle Modeling
We noticed that from the seven of the TEMs in the TRENDY model ensemble, certain models consistently exhibited divergent results from the others in representing the impacts of climatic variables during ENSO events, for example, .This discrepancy might stem from these models' more detailed considerations of water and temperature effects, especially regarding plant hydraulics on vegetation productivity (Hickler et al., 2006;Yao et al., 2022).Hence, there is an urgent need for the TEMs group to enhance model development, aiming for more realistic representations of ecosystem processes.This refinement will yield more robust estimates of ENSO impacts across multiple scales.
Our analysis primarily focused on regional and continental scales, limiting the ability to reflect responses at the biome-scale due to the high computational demands of conducting data assimilation over an extensive 36-year period.While our study centered on examining the impacts of ENSO events on regional and global vegetation productivity over the last 36 years, analyzing finer scales was beyond our current scope.We anticipate that future investigations will delve into event-specific analyses by assimilating high-resolution data over single ENSO event periods, enabling an exploration of spatial differences in vegetation productivity responses at the biome-scale.

Conclusions
In summary, our analysis across multiple data sets underscored the significant role of soil moisture in regulating regional and continental vegetation productivity during ENSO events, particularly in the pan-tropics.Furthermore, we revealed the asymmetric roles of El Niño and La Niña in controlling the regional vegetation productivity due to the shifted response regimes.Generally, the observation-constrained GPP product showed consistency with satellite-derived and data-driven products in understanding the impacts of ENSO events on vegetation productivity across regions.This study extended the analysis of the ENSO impacts from single region (e.g., pan-tropics) with limited data sets to extra-tropical regions, taking into account multiple climatic factors.Future efforts, leveraging the increasing availability of Earth observations and a process-oriented modeling framework, are crucial in consistently elucidating the contributions of individual environmental drivers and their complex interactions at higher spatial resolutions.

Figure 1 .
Figure 1.Inter-annual variability of ENSO indices, CCDAS_post output of surface soil moisture (SSM, 0-4 cm), GPP, and atmospheric CO 2 concentration growth rate (CGR), spatial correlations between GPP and GRACE_REC TWS, and impacts of spatial resolution on the contributions of the different variables.(a) Normalized anomalies for two ENSO indices, MEI (Multivariate ENSO Index, black solid line) and NINO3.4 (black dashed line).(b) SSM CCDAS anomalies calculated from the mean surface soil moisture in the mid-latitude region (30-60°N).(c) GPP CCDAS anomalies calculated from the mean GPP in the mid-latitude region (30-60°N).(d) The CGR CCDAS anomalies were calculated from the global mean CO 2 concentrations (taken as weighted average of CO 2 concentrations from MLO and SPO stations, i.e., 0.75 × MLO + 0.25 × SPO).Asterisk denotes significant correlation ( p < 0.001).(e) The spatial distribution of the correlation between the GPP CCDAS and GRACE_REC TWS (ρ(GPP CCDAS , TWS)), the dot denotes a significant correlation ( p < 0.05), and the inset shows the distribution of ρ(GPP CCDAS , TWS) for all of the grid points (light blue) and for the significantly correlated grid points (pink).(f) Spatial resolution effects of the different variables (i.e., the TWS, T, VPD, and Rn) on the contributions to the inter-annual variability of the GPP.Contributions are shown with their absolute values and summed up to 100.

Figure 2 .
Figure 2. Contributions of TWS, T, VPD and Rn to the inter-annual variability of GPP at different spatial scales.(a) FLUXCOM_GPP, (b) LT_SIF, (c) NIRv_GPP, (d) TRENDY_GPP from multi-model mean (TRENDY_S3) using MEI as the ENSO index.The contribution (C var ) is expressed in absolute value and summed up to 100.

Figure 4 .
Figure 4. Regional contributions of TWS and T to vegetation productivity.(a) Contribution of TWS to GPP CCDAS (CCDAS_post), (b) Contribution of T to GPP CCDAS , (c) The difference in the contributions of TWS and T (absolute values), (d) Contribution of TWS to GPP/VPPs from FLUXCOM_GPP, LT_SIF, NIRv_GPP, CCDAS_post, and TRENDY_S3 in different regions, (e) Same as (d) but for T, (f) the difference in the absolute values from (d) and (e).Colorbar denotes the contribution value or the difference in absolute contribution values (%).Gray slash in (d)-(e) denotes where the contributions show contrasting signs from the El Niño events and the La Niña events, that is, asymmetric contributions.