Effects of eastern vs. central Pacific El Niño on Northern Hemisphere photosynthetic seasonality

The El Niño–Southern Oscillation (ENSO) affects many climatic controls on vegetation activity, driving interannual variation in timing (phenology) and magnitude of terrestrial carbon uptake. However, the climatic effects of ENSO can differ for sea surface temperature (SST) anomalies primarily centered in the eastern vs. central Pacific (EP and CP, respectively). Here, we examine the extent to which EP and CP SST anomalies affect Northern Hemisphere photosynthetic seasonality and whether their effects differ. Using two decades of satellite near-infrared reflectance of vegetation (NIRv) and FLUXNET2015 gross primary production, we estimated annual 0.05° start and end of growing season timing corresponding to the onset and offset of photosynthetic activity, as well as peak NIRv magnitude as a proxy for peak growing season productivity. We find that correlations between Northern Hemisphere photosynthetic timing/magnitude and ENSO differ for EP- and CP-centered SST anomalies, though in many regions the effects and differences between them are quite small. Warmer SSTs generally led to an earlier start of the photosynthetic season, especially in North America and parts of Eurasia. However, the magnitude (and even direction) of the relationships between start of season and SST differed for CP- and EP-dominated events. Correlations of both peak NIRv magnitude and end of season timing with ENSO tended to be smaller in magnitude and more regionally idiosyncratic, though with strong evidence of different effects of CP and EP SSTs. In southern North America, CP SSTs (but not EP SSTs) were positively associated with peak NIRv, while in boreal regions of North America and Eurasia, CP SSTs were negatively associated with peak NIRv (despite small positive associations with EP SSTs). Differences between the effects of EP and CP SST anomalies suggest that short-term vegetation forecasts based on aggregate ENSO indices could be improved by separately considering the EP and CP components.


Introduction
The magnitude and timing (phenology) of vegetation activity affect many aspects of climate, ecology, and society, including variability of atmospheric CO 2 concentrations (Tucker et al 1986, Graven et al 2013, Forkel et al 2016, the success of biological invasions (Dech and Nosko 2004, Fridley 2012, Chapman et al 2014, availability of forage for wildlife and livestock (Vrieling et al 2016, Farzan and Yang 2018, Aikens et al 2020, and the timing and severity of seasonal allergies and asthma (Frei andGassner 2008, Chapman et al 2014). Phenological transitions are cued by seasonal environmental changes, with the cues differing by region (e.g. due to differences in energy and resource availability), provenance, plant functional type, species, and phenological stage (e.g. cues that affect budbreak differ from those that trigger senescence and dormancy). In boreal and temperate systems, the onset of the growing season typically depends on temperature and light (Zhang et al 2004, Piao et al 2019, while breakdown of chlorophyll and leaf senescence at the end of the growing season are more strongly associated with photoperiod (Bauerle et al 2012). In arid environments, vegetation activity is tightly coupled to soil moisture (Novick et al 2016): flushes of vegetation and pulses of vegetation activity occur following rainfall (Crimmins et al 2011), and activity declines during soil dry-down (Huxman et al 2004). Even in tropical rainforests, photosynthetic activity varies seasonally with changes in radiation and water availability (Guan et al 2015, Xu et al 2015. The diverse climatic cues that affect vegetation phenology and photosynthesis are embedded within and dependent upon coupled ocean-atmosphere circulation systems. In particular, the El Niño-Southern Oscillation (ENSO)-a system of coupled changes in tropical Pacific sea surface temperatures (SSTs) and the Walker circulation-is a dominant driver of global interannual climate variability, affecting both the timing (Buermann et al 2003, de Beurs et al 2018 and magnitude (Keeling et al 1995, Bastos et al 2013, Gonsamo et al 2016, Zhu et al 2017, Zhang et al 2019 of primary production. Peak ENSO variability typically occurs near boreal winter (Okumura andDeser 2010, Timmermann et al 2018), but many of the impacts on surface climate and vegetation lag several months or more behind this peak , Zhu et al 2017, Dannenberg et al 2018. Due to this strong but lagged influence on both climate and vegetation, information on the state of the ENSO system is often an important predictor in short-term seasonal forecasts of climate (O'Lenic et al 2008, vegetation activity (Tadesse et al 2014, Asoka and Mishra 2015, Hartman et al 2020, and wildfire (Harris et al 2013).
While the general influence of ENSO on climate and vegetation is well-established, the effects of any individual ENSO event can vary substantially from the mean response (Timmermann et al 2018). This variation results partly from interactions between ENSO and other ocean-atmosphere teleconnection patterns (Wise 2010, Dannenberg et al 2018, but it also arises from excitement of different wave trains when SST anomalies are centered in the central Pacific (CP) vs. in the eastern Pacific (EP) (Paek et al 2017), which can alter the ENSO effect on regional land surface temperature and precipitation Zou 2013, Paek et al 2017). In North America, for example, CP El Niño tends to have an enhanced drying effect on the eastern United States (U.S.) (Yu and Zou 2013) and a reduced wetting effect on the western U.S. (Paek et al 2017) compared to EP El Niño. Globally, CP-dominated SST anomalies tend to have a greater and longer lasting effect on both temperature and the global carbon cycle (especially in the tropics and subtropics) than EP-dominated anomalies (Dannenberg et al 2021). Additionally, there is strong evidence that ENSO variability is increasingly dominated by CP dynamics (Yeh et al 2009, Liu et al 2017, Freund et al 2019. While it is still unclear if this represents a response to external forcing or internal variability (Maher et al 2018, Timmermann et al 2018, the frequency of CP-dominated ENSO events during the past several decades is nearly unprecedented in multi-century reconstructions (Liu et al 2017, Freund et al 2019. Here, we use two decades of remotely-sensed data and SST observations to assess the extent to which EPand CP-driven SST anomalies affect Northern Hemisphere seasonal timing and peak magnitude of ecosystem gross primary production (GPP) and whether those effects differ. We specifically examine two questions: (1) Where and how much do CP and EP SST anomalies affect photosynthetic seasonality and magnitude? and (2) Where and how much do the effects of EP and CP SST anomalies differ from each other? If EP-and CP-driven ENSO events have different effects on photosynthetic phenology, then phenology forecasts based on a single aggregate ENSO index may be less reliable during years in which ENSO events are strongly dominated by a single 'flavor' of ENSO. If land surface phenology responds more or less strongly to CP ENSO events than to EP ENSO events, for example, an increasingly CP-dominated ENSO system could also either enhance or dampen the interannual variability of land surface phenology, with cascading effects on the variability of the global carbon cycle.

Eastern and central Pacific SST anomalies
To separate the effects of EP and CP ENSO (Yu and Kim 2010, Paek et al 2017, Dannenberg et al 2021, we used monthly SST anomalies in the Niño1 + 2, Niño3.4, and Niño4 regions (figure 1) from NOAA's Climate Prediction Center (www.cpc.ncep.noaa.gov/data/indices/) and 1 • × 1 • resolution gridded monthly SST anomalies (relative to a 1981-2010 baseline) from COBE-SST2 (Hirahara et al 2014) for 1981-2019. To remove the influence of CP anomalies and isolate solely EP effects, we regressed monthly SST anomalies of each grid cell in the tropical Pacific (20 • S-20 • N, 120 • E-70 • W) on the monthly SST anomalies in the Niño4 region and then averaged the residuals of all grid cells (weighted by their respective areas) within the Niño1 + 2 region (Dannenberg et al 2021). Thus, the weighted mean regression residuals in the Niño1 + 2 region quantified the EP SST anomaly variability unexplained by (and independent from) variability in the CP ( figure 1(A)). Similarly, to isolate CP variability, we used the SST anomalies in the Niño1 + 2 region to remove the influence of EP dynamics in the Niño4 region ( figure 1(B)). We then averaged monthly EP and CP SST anomalies from September (in year t − 1) through February (in year t), the time of year when ENSO events typically reach maturity (Okumura andDeser 2010, Timmermann et al 2018) and have the greatest influence on global vegetation activity (Dannenberg et al 2018, Zhang et al 2019. This approach decomposes the overall ENSO variability into its two dominant drivers. Variability of SST anomalies in the Niño3.4 region (the typical region used to define El Niño and La Niña events) is mostly driven by CP anomalies (figure S1(A); R 2 = 0.79) with EP anomalies explaining a much smaller proportion of its variance (figure S1(B); R 2 = 0.05). However, the decomposed EP and CP SST anomalies are independent of each other (figures 1(C) and S1(D); R 2 = 0.04). Consistent with previous research (Yeh et al 2009, Liu et al 2017, Freund et al 2019, most of the 21st century El Niño events (2003,2005,2010) were dominated primarily by CP SST anomalies, with EP SSTs actually slightly cooler than normal after removing the variance common to both regions (figure 1(C)). The major exception was the extremely strong 2015-2016 El Niño, during which both CP and EP SSTs were above normal (Paek et al 2017) (figure 1(C)). The 2011 La Niña, which resulted in a remarkably strong terrestrial carbon sink (Poulter et al 2014), was also primarily driven by CP SST anomalies, while the relatively weak 2018 La Niña was driven mostly by EP SST anomalies.

Northern Hemisphere photosynthetic seasonality
We estimated Northern Hemisphere photosynthetic timing and peak magnitude using two decades (2000-2019) of 0.05 • resolution daily surface reflectance from the Moderate Resolution Imaging Spectroradiometer (MODIS) Nadir Bidirectional Reflectance Distribution Function (BRDF)-Adjusted Reflectance product, or MCD43C4 (Schaaf et al 2002). After screening out observations for which the BRDF could not be retrieved or where snow cover exceeded 25% of the pixel area, we calculated the near-infrared reflectance of vegetation (NIR v ) (Badgley et al 2017): where NIR and R are near-infrared (MODIS band 2) and red (MODIS band 1) surface reflectance, respectively. NIR v effectively tracks GPP of the land surface (Badgley et al 2017(Badgley et al , 2019, including in arid and semiarid ecosystems (Wang et al 2022), where phenology estimation has traditionally been challenging due to high soil background reflectance (which NIR v effectively removes), relatively low leaf area and photosynthetic rates, and phenology curve fitting procedures that are not well adapted to 'flashy,' precipitation-driven growing seasons (White et al 2009, Ganguly et al 2010, Smith et al 2019. To ensure reasonable estimates of dormant-period NIR v , we filled snow-contaminated observations with the mean snow-free NIR v of that same day of year for that pixel. We then filtered out noise-contaminated observations using a seasonal outlier filter (Hwang et al 2011a, 2011b) and a modified best index slope extraction (BISE) filter (Viovy et al 1992, Hwang et al 2011a, 2011b, which detects sudden but temporary changes in land surface greenness that are unlikely to arise from real changes in vegetation cover or activity. We gap-filled the resulting 20 year NIR v time series for each pixel using linear interpolation, and smoothed it with a Savitzky-Golay filter (e.g. Jönsson andEklundh 2004, Wang et al 2020). Finally, we defined the start and end of the photosynthetic season (SOS and EOS, respectively) for each pixel in each year using local thresholds based on a percentile of the pixel's seasonal NIR v amplitude in that year (White et al 2009), excluding years where ⩾75% of the NIR v observations were invalidated due to unsuccessful BRDF inversions, excessive snow cover, or failure to pass the outlier and BISE filters.
We determined the SOS and EOS thresholds by comparison of NIR v to daily GPP from FLUXNET2015 (Pastorello et al 2020). For each of 55 non-cropland eddy covariance sites with at least 10 years of flux records (table S1), we filtered and smoothed the daily GPP record using the procedures described above, calculated the first and last day of year when GPP exceeded 20% of its annual amplitude, and then determined the percent of the annual NIR v amplitude (using the 500 m MCD43A4 surface reflectance for each tower site) that had been reached on those days (figure S2). NIR v usually tracked GPP closely during green-up but lagged GPP during brown-down, so the median SOS and EOS thresholds were ∼20% and ∼35%, respectively, of the annual NIR v amplitude, which we used as the thresholds for global SOS and EOS detection. In addition to the annual SOS and EOS of each pixel, we also calculated the annual peak NIR v during the growing season as a proxy for the peak rate of GPP. Across all eddy covariance sites, the approach captured variability in peak GPP and SOS quite well (figures S2(A) and (B); Spearman's ρ = 0.79 and ρ = 0.67, respectively). Estimating EOS timing is considerably more challenging (Berra and Gaulton 2021), with Spearman's ρ = 0.54 between EOS NIRv and EOS GPP (figure S2(C)).

ENSO response regions
To summarize regional responses of photosynthetic seasonality to EP and CP SST anomalies, we defined ten regions based on their climatic responses to ENSO (figure S3). We first calculated Spearman's correlation coefficients between EP and CP SST anomalies and 0.5 • gridded CRU TS4.04 (Harris et al 2014) minimum temperature, maximum temperature, precipitation, and vapor pressure deficit in four seasons (Dec-Feb, Mar-May, Jun-Aug, and Sep-Nov) during the period 2000-2019, resulting in 32 correlations per grid cell (figures S4-S7). We then used k-means clustering based on those 32 correlation coefficients to group the Northern Hemisphere into regions of similar ENSO response, which we used to guide selection of latitude and longitude bounds for our ten regions. We did not directly use the k-means clusters as regions because: (i) many clusters were either not geographically contiguous or spanned regions with similar climate responses to ENSO but very different ecological characteristics (e.g. a cluster spanning the eastern boreal forests of North America to the arid and semiarid U.S. Southwest); (ii) the precise grid cells assigned to each k-means cluster are sensitive to the number of clusters chosen and the random initialization of cluster centroids (figure S3); and (iii) the CRU climate data are two orders of magnitude coarser than the MODIS NIR v data.

Statistical analysis
We assessed responses of Northern Hemisphere photosynthetic phenology to mean September-February EP and CP SST anomalies using Spearman's rank correlation coefficients (ρ), excluding any pixels where fewer than 15 years of phenology estimates could be successfully retrieved. We tested significance of the pixel-wise photosynthetic timing/magnitude responses to EP and CP SST anomalies (i.e. null hypothesis that ρ = 0) in each region using false discovery rate (FDR)-corrected 'field significance' testing at the α global = 0.1 level (Wilks 2006(Wilks , 2016, which tests the distribution of p-values from multiple hypothesis tests against the theoretical distribution if all null hypotheses (ρ = 0) were true. A positive FDR test thus indicates whether at least one local hypothesis test in the spatial 'field' is significant when adjusting for expectations arising from false discovery. The FDR field significance test is robust to spatial autocorrelation (Wilks 2006) and has gained popularity for significance testing on gridded data in both climate sciences (e.g. Johnson 2013, Wise andDannenberg 2014) and remote sensing (e.g. Zhu et al 2017, Dannenberg et al 2018. We note, however, that these significance tests are performed on short (15-20 year), noisy time series, so 'statistical significance' is difficult to establish. Consistent with recommendations regarding use of p-values (Wasserstein and Lazar 2016, McShane et al 2019, Wasserstein et al 2019, we therefore use these significance tests to guide discussion of where confidence in the ENSO connection to photosynthetic seasonality is highest. We also examined asymmetries in the effects of EP vs. CP SST anomalies (i.e. shifts of the distribution of pixel responses to SST anomalies away from 0) within each of the ten regions using a two-step filtering and asymmetry test Henebry 2018, Liang et al 2021). For each region, each photosynthetic timing and magnitude metric, and each ENSO 'flavor,' we calculated the percentage of pixels with significant (p < 0.05) positive and negative correlations and then calculated an asymmetry ratio (AR) as the percentage of significant positive correlations divided by the percentage of significant negative correlations. We focus especially on regions with AR > 2 (twice as many positive as negative correlations) and AR < 0.5 (twice as many negative as positive correlations) Henebry 2018, Liang et al 2021), as well as on regions with AR mismatches between effects of EP and CP SST anomalies. The FDR and AR tests thus provide complementary information: while the FDR test examines whether the distribution of p-values is significantly different from that expected if all null hypotheses were true (but is agnostic to the direction of the relationship), the AR quantifies systematic shifts towards positive or negative relationships.

Results
Across the Northern Hemisphere as a whole, both EP and CP SST anomalies were negatively correlated with SOS timing (i.e. earlier SOS with warmer SSTs; figures 2(A)-(D); table S2), though hemispheric median correlations were quite low (|ρ| < 0.1) (table S2). However, the hemispheric distribution showed clear evidence of an asymmetric response of SOS to ENSO, with roughly four times as many significant (p < 0.05) negative correlations as positive correlations with EP SST anomalies (AR = 0.26; figure 2(B); tables 1 and S2) and more than twice as many negative correlations as positive correlations with CP SST anomalies (AR = 0.47; figure 2(D); tables 1 and S2).
While hemispheric-scale SOS responses to both EP and CP SST anomalies were relatively small, responses differed substantially from region to region in both magnitude and relative importance of EP vs. CP SST anomalies (figures 2(E)-(N); tables 1 and S2). Across the Northern Hemisphere tropics (figure 2(E)), there was some evidence of a significant response to CP SST anomalies (at the α global = 0.1 level in the FDR test) but no clear asymmetry in the distribution of correlations (0.5 < AR < 2) for either EP or CP SSTs. Three regions of North America responded to one or both 'flavors' of ENSO (figures 2(F)-(H)), but with regional differences in which was more important. In Southwest North America (figure 2(F)) and the North Pacific (figure 2(H)), responses of SOS to both EP and CP SSTs were strong enough to exceed expectations from false discovery (at the α global = 0.1 level in the FDR test) with a clear shift towards earlier SOS with warmer EP and CP SSTs (AR < 0.5), especially in North Pacific responses to CP SST anomalies (median ρ = −0.23; AR = 0.02). Southeast and South-Central North America (figure 2(G)) showed the clearest divergence between the effects of EP and CP SST anomalies on SOS: EP SSTs were negatively and significantly correlated with SOS (median ρ = −0.25; AR = 0.02), but CP SSTs were weakly correlated with SOS (median ρ = −0.04; AR = 1.58). Only one other region (Central Boreal Eurasia; figure 2(M)) showed strong evidence of significant and asymmetric relationships between SOS and EP or CP SST anomalies (at the α global = 0.1 level for CP and AR = 0.04 for both EP and CP), and only two regions (Central and East Asia and Eastern Boreal Eurasia; figures 2(L) and (N), respectively) showed strong evidence of a shift towards positive correlations (AR > 2) between SOS and EP or CP SST anomalies.
At the hemispheric scale (figures 3(A)-(D); tables 1 and S3), there was little evidence of a significant or systematic relationship between peak NIR v and either EP or CP SST anomalies (median |ρ| < 0.05; 0.5 < AR < 2). However, as with SOS correlations, the strength and direction of peak NIR v responses to EP and CP SST anomalies differed among regions. In the FDR test, the Tropics (figure 3(E)), Southeast and South-Central North America (figure 3(G)), and Central and East Asia (figure 3(L)) all showed evidence of significant relationships (at the α global = 0.1 level) with CP SST anomalies, and all three (along with Southwestern North America; figure 3(F)) showed evidence of strong increases in peak NIR v following warmer CP SST anomalies (AR > 2) but little systematic response to EP SST anomalies (0.5 < AR < 2). By contrast, we found decreased peak NIR v with increasing CP SSTs (AR < 0.5) coupled with either increased peak NIR v (AR > 2) or little change in peak NIR v (0.5 < AR < 2) with increasing EP SSTs in four regions: the North Pacific ( figure 3(H)), Eastern Boreal North America (figure 3(I)), Northern Europe (figure 3(K)), and Eastern Boreal Eurasia ( figure 3(N)). Hemispheric scale relationships of EOS to EP and CP SST anomalies were small (median |ρ| ≈ 0 and AR ≈ 1; figures 4(A)-(D); tables 1 and S4). Several regions, however, showed coherent (albeit relatively small) relationships between EOS and either EP or CP SST anomalies (figures 4(E)-(N)). The region with the strongest overall relationship between EOS and either ENSO 'flavor' was the North Pacific (figure 4(H)), where there was evidence of both a significant effect of CP SST anomalies that exceeded expectations of false discovery (at α global = 0.1) and a very large asymmetry in the distribution of correlations (AR = 0.05), indicating a general tendency towards earlier EOS following warmer-than-normal CP SSTs. Eastern Boreal North America (figure 4(I)) also had a significant (at α global = 0.1) and asymmetric (AR = 0.25) relationship to CP SST anomalies and a significant but not strongly asymmetric (AR = 0.60) relationship to EP SST anomalies. The Mediterranean (figure 4(J)) had both a significant and asymmetric (AR = 0.36) relationship with EP SST anomalies but not with CP SSTs. The region with the largest difference between EP and CP effects (though not significant in the FDR tests), was Eastern Boreal Eurasia (figure 4(N)), where there was a general tendency towards later EOS following warmer-than-normal EP SSTs (AR = 5.22) and earlier following warmer-than-normal CP SSTs (AR = 0.49). The remaining regions did not have strong evidence of 'field significance' in the FDR tests or did not have large systematic shifts in the correlation distribution (0.5 < AR < 2).

Discussion
The ENSO is the dominant driver of interannual variability in the global climate system (Timmermann et al 2018) and is therefore an important tool in seasonal temperature and precipitation forecasts (O'Lenic et al 2008). Since vegetation phenology and productivity depend at least in part on temperature and water Figure 3. Spearman's rank correlation coefficients (ρ) between EP and CP SST anomalies and Northern Hemisphere peak NIRv magnitude. Spatial patterns (A) and histogram (B) of peak NIRv correlations with EP SST anomalies. Spatial patterns (C) and histogram (D) of peak NIRv correlations with CP SST anomalies. (E-N) Regional distributions of correlations between peak NIRv and EP SST anomalies (left) and CP SST anomalies (right), with dots indicating the median correlation across all pixels in the region. Boundaries of each region are shown in A and C. Indicators above the histograms indicate the asymmetry ratio (AR) and whether responses to EP and CP SST anomalies are significant ( * ) or not significant (ns) when accounting for the FDR (at α global = 0.1). availability in many regions, global-scale variation in vegetation activity has been strongly linked to ENSO (Keeling et al 1995, Buermann et al 2003, Bastos et al 2013, Poulter et al 2014, Gonsamo et al 2016, Zhu et al 2017, Zhang et al 2019, Dannenberg et al 2021. However, ENSO dynamics are spatially and temporally complex (Timmermann et al 2018), and the climatic and ecological effects of ENSO in any given year can differ substantially from the mean response, partly due to variation in the location of tropical Pacific SST anomalies (e.g. EP vs. CP anomalies).
Regardless of ENSO 'flavor,' we found that El Niño events were associated with earlier SOS at both the hemispheric scale and for most regions (figure 2; table 1). The negative association between tropical Pacific SSTs and SOS likely arises from the warming effect of El Niño on global temperatures (Zhang et al 2019) and the cooling effect of La Niña (Bastos et al 2013, Poulter et al 2014, especially in the months preceding and during the early growing season (figures S4 and S5). Despite this general tendency towards earlier SOS following El Niño, the magnitude (and in some cases direction) of the effect depended on whether the SST anomalies were centered in the EP or the CP (table 1; figure 2). The difference in SOS response to EP vs. CP SST anomalies was clearest in Southeast and South-Central North America (figure 2(G)): while EP El Niño was associated with a strong shift towards earlier SOS, CP El Niño had little effect on SOS timing, likely due to differences in how these two ENSO flavors affect temperature in the region (figures S4 and S5). Phenological forecasts built on an aggregate ENSO index would therefore likely underperform in this region compared to forecasts that consider the effects of EP and CP SST anomalies separately.
While the overall hemispheric-scale relationships between peak NIR v magnitude and ENSO (both EP and CP) were weaker than those for SOS timing, the differences between the effects of EP and CP SST anomalies were generally greater (tables 1 and S3; figure 3). These differences could arise from the importance of  Table 1. Summary of asymmetry ratio (AR) strength and direction for EP (left) and CP (right) relationships with SOS, peak NIRv, and EOS for each region and for the whole Northern Hemisphere. The symbols indicate very strong positive asymmetries (++, AR ⩾ 10), strong positive asymmetries (+, AR ⩾ 2), no strong asymmetry (n, 0.5 < AR < 2), strong negative asymmetry (−, AR ⩽ 0.5), and very strong negative asymmetries (−−, AR ⩽ 0.1).

SOS
Peak −/− n/n n/n moisture availability for growing season peak productivity and the strong regional dependence of precipitation (figure S6) and vapor pressure deficit (figure S7) on EP vs. CP centers of action. For example, in both regions of southern North America, there was a strong positive association between peak NIR v and CP SSTs but very little association with EP SSTs (tables 1 and S3; figures 3(F) and (G)). In those same regions, warmer CP SSTs were more strongly associated with increased precipitation and lower vapor pressure deficit than EP SSTs during the study period (figures S6 and S7). There were also very large differences in the effects of EP and CP SSTs on peak NIR v in the boreal regions of eastern Eurasia and North America, with relatively consistent negative associations of peak NIR v with CP SSTs and slight positive associations with EP SSTs (tables 1 and S3, figure 3(I) and (N)), though the likely climatic drivers are less clear. Weaker relationships between ENSO and peak NIR v in other regions could be due either to strong dependence on other circulation systems (Aasa et al 2004, de Beurs andHenebry 2010) or to the dominant role of land surface characteristics (e.g. topography) on vegetation phenology (Tomaszewska and Henebry 2020). Generally and unsurprisingly, the relationships between EOS and EP vs. CP ENSO tended to be relatively weak and inconsistent (figure 4; table 1), though there was some evidence of intra-regional differences in the effects of EP and CP SST anomalies (e.g. in the Tropics and Eastern Boreal Eurasia; tables 1 and S4). This weak relationship between EOS and ENSO (both CP and EP) is likely associated with a combination of insufficient or noisy data and biological drivers of senescence and dormancy. First, satellite EOS retrievals are notoriously noisier than SOS retrievals (Berra and Gaulton 2021), possibly associated with temporal asynchrony in end-of-season changes in leaf area, chlorophyll and other pigment contents, and leaf water content (Wang et al 2020). Second, ENSO events tend to mature in boreal winter and decay in boreal spring and summer (Timmermann et al 2018), and the effects of ENSO on surface climate also typically diminish as boreal summer progresses (e.g. Dannenberg et al 2015Dannenberg et al , 2021. Third, the September-February SST composites used here may not best reflect the time during which ENSO exerts the greatest influence on the late-season climatic conditions that are most relevant for EOS timing, though Dannenberg et al (2018) showed that EOS timing was generally insensitive to a range of ENSO composite periods in North America. Finally, while temperature and/or moisture have well documented effects on early-to-peak timing of land surface greenness (Buermann et al 2003, Zhang et al 2004, de Beurs and Henebry 2005, 2010, Piao et al 2019 and growing season vegetation productivity (Humphrey et al 2018, Stocker et al 2019, Dannenberg et al 2022, EOS timing is often strongly associated with photoperiod (Bauerle et al 2012, Richardson et al 2013, which is not affected by ENSO.

Conclusions
We examined the influences of EP and CP SST anomalies on the timing and peak magnitude of GPP in the Northern Hemisphere using two decades of daily NIR v from MODIS. While the effects of ENSO tended to be region-specific, both EP and CP SSTs were typically negatively correlated with SOS timing (earlier start of the photosynthetic season in El Niño years than La Niña years), likely due to the overall warming effect of El Niño. These effects were generally greatest in North American ecosystems, though there was also a clear hot spot of SOS sensitivity to ENSO in Eurasian boreal forests. The clearest differences between the effects of EP and CP SST anomalies were in Southeast and South-Central North America, where only EP SST anomalies had strong and significant effects on SOS timing while CP SST anomalies had little effect. Peak NIR v and EOS timing were generally less coupled than SOS timing to EP or CP SST anomalies at the hemispheric scale, but regional responses to EP vs. CP SSTs often diverged. This situation was especially the case in southern North America, where warmer-than-normal CP SSTs (but not EP SSTs) were associated with a likely moisture-driven increase in peak productivity, and in the boreal regions of North America and eastern Eurasia, where CP SSTs were negatively associated with peak productivity while EP SSTs had either no effect or a small positive effect. These differences in sign between the effects of EP and CP SST anomalies suggest that empirical forecasts of vegetation activity built on a single ENSO index may not adequately capture the diversity of ENSO effects on climate and vegetation. Given the increasingly CP-dominated nature of ENSO (Liu et al 2017, Freund et al 2019, the differences in phenological and photosynthetic responses to EP vs. CP SST anomalies could lead to a non-stationary connection between ENSO and vegetation activity, with the EP-dominated dynamics of the late 20th century being unrepresentative of 21st century conditions.

Data availability statement
No new data were created or analysed in this study.