Seasonal biological carryover dominates northern vegetation growth

Interseasonal VGC dominates peak-to-late-season growth

We first examined the VGC effect at the seasonal scale, using satellite-derived Normalized Difference Vegetation Index (NDVI, see “Methods”) for the 1982–2016 period. We defined the dormancy season (DS) and three periods of the growing season, i.e., EGS, peak growing season (PGS), and late growing season (LGS), based on phenological metrics (see “Methods”). The partial autocorrelation calculated for NDVI time series of two consecutive seasons, after factoring out concurrent and preceding climatic impacts, provides an estimate of the interseasonal VGC effects (see “Methods”). At the hemispheric scale, our analyses show a significant (p < 0.05) control of the NDVI of the preceding season (NDVIps) on the interannual variations of seasonal NDVI for all three active growing seasons (Fig. 1a). This VGC effect is also consistently positive across the majority (79%, 89%, and 94% for EGS, PGS, and LGS, respectively) of northern vegetated areas (Supplementary Fig. 1). Although the EGS NDVI is more strongly correlated with EGS temperature, the PGS and LGS NDVIs are most strongly correlated with NDVIps, rather than climate drivers (Fig. 1a). This VGC is particularly important for understanding vegetation growth in LGS, when climate is known to have a weak explanatory power^30,31.

The primary role of NDVIps in driving subsequent PGS and LGS NDVI variations is reaffirmed by conducting partial correlations with detrended anomalies of all variables (Supplementary Fig. 2), implying a robust coupling of seasonal vegetation growth within a specific calendar year. Furthermore, the robustness of the satellite-identified positive VGC effect dominating vegetation growth in PGS and LGS is also verified by examining other satellite-derived vegetation growth proxies, including leaf area index (LAI) and gross primary productivity (GPP) (see “Methods”; results in Supplementary Fig. 3). Meanwhile, we also noticed some difference between GPP- and NDVI-derived LGS-to-EGS VGC effect in some mid-to-high northern latitudes (Supplementary Fig. 4d). Negative LGS-to-EGS VGC effect has been found over eastern U.S., North China, and western and central Russia based on the GPP dataset (Supplementary Fig. 4d), suggesting greater uncertainties in LGS-to-EGS VGC effect than EGS-to-PGS and PGS-to-LGS VGC effects.

In contrast to the consistently strong and positive VGC impact, the strength and direction of climatic factors in determining the interannual variation of seasonal NDVI, including their immediate and lagged impacts^{16,17,18,19,20}, is highly variable between seasons and across regions (Fig. 1a and Supplementary Figs. 5 and 6). At the hemispheric scale, concurrent seasonal temperature is a primary and positive factor controlling the interannual variation of EGS NDVI during the last 35 years (partial correlation, r_p = 0.83, p < 0.01). The dominance of EGS temperature on EGS vegetation growth is also consistently observed when analyzing other satellite-based vegetation proxies of LAI and GPP (Supplementary Fig. 4). However, temperature has a much weaker impact during PGS (r_p = 0.42, p < 0.05) and LGS (r_p = 0.12, p > 0.05) (Fig. 1a). Although higher concurrent temperature generally stimulates vegetation activity in EGS across most of the northern vegetated areas (Supplementary Fig. 5a), it has emerged as a limit to PGS and LGS vegetation growth for most of the warm mid-latitudes and some of the high latitudes (Supplementary Fig. 5b, c). Additionally, temperature also has a negative legacy effect on vegetation growth in the subsequent season (Fig. 1a), most significantly for the DS-to-EGS legacy effect (r_p = −0.42, p < 0.05). This adverse legacy effect of DS temperature is likely due to the lower chilling accumulation required for leaf unfolding in EGS caused by DS warming³². For precipitation, we find very weak and statistically insignificant immediate and lagged impacts on vegetation growth for all the seasons at the hemispheric scale (Fig. 1a). This weak precipitation impact is likely due to a spatial cancelling-out of the positive effects at the water-limited mid-latitudes by the negative effects at high latitudes (more precipitation is often concurrent with increased cloudiness and reduced solar radiation reaching vegetation canopies) (Supplementary Figs. 5d–f and 6d–f).

For each season, we further derived the individual contributions of VGC, as well as the immediate and lagged climatic effects, to the 35-year NDVI trends (see “Methods”). The effects of temperature and precipitation were here combined as a single variable of climatic effects. As expected, the strong observed EGS greening trend (0.0012 yr⁻¹, p < 0.05) is predominately attributed to the concurrent climate change (77%), particularly EGS warming that stimulates earlier phenology, followed by smaller but non-negligible contributions from climate (8%) in the preceding DS and vegetation growth in the preceding LGS (17%) (Fig. 1b). However, for PGS and LGS, about half of the observed greening trends (48% and 54%, respectively) are attributed to greening in the preceding season, supporting the notion of a strong positive biological carryover between seasons. In comparison, climate, including its immediate and lagged effects, plays a much smaller role in PGS and LGS greening (EGS climate may even cause a negative lagged impact on the PGS greening trend; Fig. 1b). Hence, warming-induced greening in EGS persists into the mid-to-late growing season, and has been the primary source for the overall satellite-observed NH growing-season greening over the last few decades^33,34. It is interesting to note that PGS is the season whose interannual productivity variations most strongly correlate with that of the growing-season mean³⁵, and trend of vegetation growth most similar to that of the growing-season mean (Fig. 1b). However, our results demonstrate that the higher peak growth rate in PGS is primarily inherited from greening of the preceding EGS (48%) rather than from direct contributions of PGS climate change (20%) (Fig. 1b).

Considering the substantial fraction of unexplained variance of observed vegetation growth in PGS and LGS after accounting for the climate and VGC effect of the concurrent and immediate precedent seasons (residuals in Fig. 1b), we further investigated the residual changes of PGS and LGS NDVI with vegetation and climatic factors in the previous year (see “Methods”). At the hemispheric scale, about 58% of the residuals of PGS NDVI changes can be explained by all the factors collectively (Supplementary Fig. 7c). NDVI of the previous LGS significantly correlates to PGS NDVI residuals (r_p = 0.55, p < 0.05), and contributes the most to the variance of residuals (Supplementary Fig. 7a). Among the climatic factors, PGS precipitation of the previous year shows the strongest correlation (r_p = 0.31, p = 0.09) with PGS NDVI residuals (Supplementary Fig. S7a), indicating a strong legacy effect of precipitation anomalies (such as droughts) on PGS vegetation growth. None of the considered factors shows a significant (p > 0.05) correlation with LGS NDVI residuals, and collectively they explain about 20% of LGS NDVI variance (Supplementary Fig. S7c).

We further examined the VGC effect and vegetation–climate connections with seasonal GPP data from the global FLUXNET EC network. The short temporal coverage of EC records prevents calculating temporal correlations, we hence analyzed the relationship between the trend of GPP and that of its potential drivers across 50 available flux-tower sites (“Methods”). Consistent with the satellite-based findings, we discovered strong positive cross-site correlations between the trend of GPP and that of its preceding values for all the growing seasons (Pearson correlation, r = 0.42, 0.70, and 0.82 for EGS, PGS, and LGS, respectively, p < 0.01 in all cases; Fig. 2a). However, temperature and precipitation changes cannot account for the cross-site variation of GPP trends for any season (Supplementary Fig. 8), even though EGS temperature is identified as the primary driver of satellite-based EGS NDVI changes (Fig. 1a). The weak cross-site correlation with climatic variables may be overshadowed by the biome-dependent sensitivity of GPP to climate^16,17. To test this, we examined the cross-site relationship between the GPP trend and its climatic sensitivities (“Methods”). We found a significant positive correlation between the EGS GPP trend and its sensitivity to temperature (r = 0.29, p < 0.05) (Fig. 2b), supporting that EGS warming controls EGS greening patterns. For all the other cases, the insignificant (p > 0.05) relationship between the GPP trend and its climatic sensitivity (Fig. 2b, c) supports a weak climatic impact.

In addition to these global datasets, we also used long-term GPP measurements from 11 AmeriFlux EC sites (“Methods”) to characterize temporal relationships between vegetation growth and climate. Results of this analysis again confirm our main findings: EGS temperature is the primary determinant of EGS GPP (cross-site median correlation: r_p = 0.52, p < 0.05), which is then carried over to dominate the variance of GPP in PGS (r_p = 0.43, p < 0.05) and LGS (r_p = 0.62, p < 0.05) (Fig. 2d). GPP of PGS and LGS also show varied signs of correlation with other climate factors in the previous year (Supplementary Fig. 7), which collectively explain 30–72% and 16–81% of the GPP residual variance, respectively.

Our observed interseasonal connection in vegetation activity may also be modulated by indirect non-biological pathways, for example, soil moisture anomalies caused by vegetation changes persisting into the next season^19,22,25,36. These different mechanisms imply potentially multiple simultaneous pathways for the interseasonal interactions between vegetation, climate, and soil moisture status. To quantify the complex pathways underpinning interseasonal vegetation–climate–soil interactions, we constructed structural equation models (SEMs), forced with satellite-based NDVI and soil moisture, and climatic variables (see “Methods”). We allowed for broad biome differences by grouping northern vegetation into three main vegetation types of temperate grassland, forest, and arctic tundra and shrubland, based on satellite-derived land-cover maps (“Methods”; Supplementary Fig. 9). Figure 3 shows all pathways of the EGS–PGS connection (for other interseasonal linkages see Supplementary Fig. 10). The SEM analysis identifies the significant positive influence of EGS vegetation growth on that of PGS, explaining the largest fraction of PGS NDVI variations for all vegetation types (Fig. 3). This result provides further support for strong VGC between EGS and PGS vegetation growth. This EGS-to-PGS VGC effect is robust by further demonstrating that for all vegetation types, years with greener EGSs (under favorable climates) generally have greener PGSs, and accordingly, years with browner EGSs (under unfavorable climates) tend to have browner PGSs (Supplementary Fig. 11).

In parallel to the interseasonal vegetation growth carryover, we also diagnosed a strong interseasonal carryover effect of soil moisture, where local soil moisture status in PGS links tightly to that in EGS (Fig. 3). However, the indirect impact of EGS vegetation growth on PGS vegetation via this soil moisture pathway may be weaker than previously thought²². For grassland where water is often the dominant limiting factor, PGS soil moisture does significantly influence PGS productivity, yet the amount of soil moisture in EGS is controlled predominantly by EGS climate rather than vegetation (Fig. 3a). For forest-dominated ecosystems, EGS greening does significantly dry out the soil, causing a soil moisture deficit that is further carried over to the PGS²⁵. However, this moisture deficit has limited impacts on restraining PGS forest growth (Fig. 3b), likely due to the deep root system that can access water reservoirs in deep soil layers³⁷. For arctic tundra and shrubland, temperature is a key limiting factor³⁸, and thus the vegetation–soil moisture interaction is relatively weak for both EGS and PGS seasons (Fig. 3c). Furthermore, we also find that the VGC effect is more dominant than soil moisture carryover effect for both EGS (from DS or the preceding LGS; Supplementary Fig. 10a, c, e) and LGS NDVI (from PGS; Supplementary Fig. 10b, d, f).

The persistence of the VGC effect into the subsequent year

In order to examine whether this VGC effect operates at longer time scales of multiple years, we next performed lagged partial autocorrelations with interannual anomalies of satellite-observed NDVI and 2739 standardized tree-ring width (TRW) records (see “Methods”). For a time lag of 1 year, a positive interannual VGC is present across northern lands, with 75.6% of vegetated areas (for NDVI) and 82.9% of the tree-ring samples (for TRW) showing positive lagged correlations (Fig. 4a and Supplementary Fig. 12). This positive interannual VGC indicates that a greener year is often followed by another greener year. The positive VGC is statistically significant (p < 0.05) for 18.3% of northern areas based on NDVI, but noting it is significant for 46.4% of the tree-ring samples that cover much longer periods (Fig. 4a). The positive interannual VGC effect is most significant at high latitudes, particularly over northern North America and East Siberia (Supplementary Fig. 12). Interestingly, by further grouping tree species into ring-porous, diffuse-porous and non-porous species (Supplementary Table 1), we found stronger interannual VGC effect for diffuse-porous species (95.0% positive) than for ring-porous (85.3% positive) and non-porous species (81.9% positive) (Fig. 4b). This observation suggests substantial influence of wood phenology on the strength of vegetation growth carryover, and diffuse-porous species whose woody growth is more concentrated in later growing season are more likely to carry transient growth anomalies over to the subsequent year. By contrast, only a few locations (including central Siberia, eastern Europe, and some semi-arid regions) have a negative yet generally insignificant (p > 0.05) interannual VGC (Supplementary Fig. 12). If the time lags are extended to 2 years, the positive correlation between current-year NDVI (or TRW) and that of 2 years earlier is significant for only 14% of tree-ring samples or 5% of the total vegetated area (for NDVI). If time lags of 3 years are considered, the lagged correlation is found to be close to zero (Fig. 4a). Previous studies have reported stronger legacies of severe drought episodes (e.g., >2 SD from the mean climatic water deficit) lasting 2–4 years^20,21. However, for interannual anomalies (=1 SD) of vegetation growth that is much less deviated from the multi-year average than severe drought anomalies, the VGC effect can be carried over to the next year but rarely to years after that.

Terrestrial biosphere models underestimate the VGC effect

Process-based terrestrial biosphere models are a useful tool for predicting vegetation growth and examining the associated complex mechanisms^29,31. We next assessed 16 terrestrial biosphere models participating in the TRENDY intercomparison project (“Methods” and Supplementary Table 2) for their ability to capture the dominant factors contributing to the satellite-observed greenness changes in each season. By comparing modeled GPP (Fig. 5b, of multi-model mean) against satellite-observed greenness (Fig. 5a), we found that the models correctly identified EGS temperature as the primary factor controlling interannual variations of EGS vegetation activity for most northern areas. In 15 out of the 16 models, areas where EGS temperature is the primary driver of concurrent GPP variations were found to have the largest spatial coverage among all potential driving factors (Fig. 5c). However, the satellite-identified dominance of VGC effects in PGS and LGS vegetation growth for much of the northern lands (42% for PGS and 58% for LGS; Fig. 5d, g and Supplementary Fig. 13b, c) is significantly underestimated by models (19% for both PGS and LGS; Fig. 5e, h and Supplementary Fig. 13e, f). Multi-model averaged results indicate an overwhelming fraction of vegetated land is instead dominated by the immediate climatic effects for both PGS (75%) and LGS (78%). Specifically, 10 out of the 16 models significantly underestimate the proportion of VGC-dominated areas for PGS vegetation greening, and nearly all (15) of the models significantly underestimate the proportion for LGS vegetation growth, despite the strong intermodel discrepancy in the proportion of projected VGC-dominated areas (from 14% in LPX-Bern to 72% in SDGVM for PGS and from 9% in LPJ-wsl to 75% in SDGVM for LGS) (Fig. 5f, i).

With rising atmospheric CO₂ concentrations and anticipated warmer climate, Earth system models that simulate stronger VGC effects tend to project higher carbon uptake potentials over northern ecosystems (Supplementary Fig. 14). To improve estimates of how the global carbon cycle will evolve in the decades ahead, it is critical to diagnose the causes of this underrepresentation of modeled VGC effects. We therefore compared the three models that best identify the areas identified by satellite where VGC dominates vegetation growth versus the three models that least capture it (based on Fig. 5f, i). As expected, we find that the models with the best representation of the VGC effect produce better estimates of PGS and LGS levels of greenness based on EGS growth levels, for all the three major biomes (Supplementary Fig. 15a, c, e). Conversely, for the models that fail to replicate the VGC effect, modeled years with the greenest EGSs do not necessarily imply a greener PGS or LGS, especially for temperate grasslands and forests (Supplementary Fig. 15b, d).

Guided by the identified drivers from our empirical analyses (Figs. 1–3), we tested the hypothesis that the interseasonal inconsistency in modeled greening trends is related to sensitivity biases of vegetation productivity responses to climate variation (Fig. 6). Comparison of satellite-based and modeled sensitivities of PGS and LGS vegetation productivity to climatic variables confirms this hypothesis. For both PGS and LGS, models with better VGC representation show very similar spatial patterns of productivity sensitivities to temperature and precipitation as that derived from observations (Supplementary Figs. 16a, b, d, e and 18a, b, d, e). However, models underestimating the VGC effect broadly overestimate the climate sensitivity. In specific, for PGS, models underestimating the VGC effect severely overestimate the magnitude and extent of the negative impact of PGS warming and precipitation decrease on vegetation productivity in temperate regions (Supplementary Fig. 16c) and some semi-arid areas (Supplementary Figs. 16f and 17b), respectively. Similarly, for LGS, models that underestimate the VGC effect tend to overestimate the positive effects of LGS temperature on vegetation productivity in cold areas (>45°N) (Supplementary Fig. S18a, c, d, f).

Conclusions and implications

In summary, our analyses reveal strong biological carryover effects of vegetation growth and productivity across succeeding seasons and years, providing new insights into how vegetation changes under global warming. The VGC effect represents a key yet often underappreciated pathway through which warmer EGSs and associated earlier plant phenology subsequently enhance plant productivity in the mid-to-late growing season, which can further persist into the following year (Fig. 6). Without considering this biological carryover of vegetation growth, some previous studies suggest an overly negative impacts of EGS warming on PGS/LGS vegetation growth, in particular, through triggering earlier transpiration and associated soil moisture deficits^19,23,36. Yet, despite the potential for raised soil moisture deficits, we find the strong VGC effects can override this negative abiotic legacy impacts, with greener EGSs ensuring lush PGS vegetation (Fig. 6). Hence, warming in EGS not only augments concurrent vegetation growth and carbon uptake but also has a positive legacy impact on the following PGS and LGS vegetation carbon sequestration, ultimately enhancing the annual carbon sink. However, it is important to bear in mind that, while the beneficial VGC effect of EGS vegetation growth can override immediate and time-lagged climatic impact under the present climate, whether this stronger VGC effect will continue with future warmer climate remains an open question (Fig. 6). Processes involved in the lagged vegetation responses to precedent climate, soil, and growth conditions are highly complex and often non-linear^6,39,40. For example, summer climate extremes, which are often associated with large-scale climate oscillations and partly contributed by enhanced EGS vegetation growth²⁵, could trigger severe tree mortality and fires that nullify any positive carryover effect from EGS (Fig. 6). Recent advances in statistical modelling and machine learning^6,39,41 may provide useful tools for a better understanding of such non-linear vegetation responses.

We also find poor representation of the VGC effect in dynamic vegetation models, and as this likely influences predictive capacity of future global carbon cycle changes, a major research challenge is to better simulate biological processes related to this carryover effect. Tackling this challenge requires not only using satellite and ground measurements to refine existing parameterizations, but also using leaf-level measurements to understand the physiological mechanisms controlling VGC patterns and to incorporate new process representation in model components. Long-term manipulative field experiments will also be useful to better characterize VGC features under different imposed meteorological regimes and to provide key process parameters for future model improvements.

Source: Ecology - nature.com