Biophysical impacts of northern vegetation changes on seasonal warming patterns

Coupled model experiments for detecting vegetation-climate feedback

We quantified changes of near-surface (2-m) air temperature (T_a) in response to the observed NH greening for all active growing seasons during 1982–2014 using IPSL-CM. We defined the three growing seasons (spring, summer, and autumn) across the entire NH domain as periods of March-April-May (MAM), June-July-August (JJA), and September-October-November (SON), respectively. For each season, a pair of transient numerical experiments was performed by modifying LAI: a dynamic vegetation experiment (SCE) forced by annually and seasonally varying LAI from satellite observations³⁶, and three seasonal control experiments (({{{{{{rm{LAI}}}}}}}_{{{{{{rm{CTL}}}}}}}^{{{{{{rm{MAM}}}}}}}), ({{{{{{rm{LAI}}}}}}}_{{{{{{rm{CTL}}}}}}}^{{{{{{rm{JJA}}}}}}}), and ({{{{{{rm{LAI}}}}}}}_{{{{{{rm{CTL}}}}}}}^{{{{{{rm{SON}}}}}}}) for MAM, JJA, and SON, respectively) forced by annually varying LAI for all seasons, except in the season of interest when the LAI was fixed to the climatological conditions observed during 1982–2014 (Fig. S1). For all experiments, other boundary conditions, including sea surface temperature (SST), sea ice fraction (SIC), and atmospheric CO₂ concentrations, were kept consistent (Methods). Therefore, differences between SCE and the control experiments characterized the effects of the observed LAI changes on T_a (hereafter denoted as ΔT_a), both intra- and inter-seasonally. Multimember paired ensembles were generated for each coupled model experiment by performing 30 repeated runs but with different initial conditions (see Methods).

The capacity of the IPSL-CM GCM for simulating the seasonal variations and spatial patterns of T_a was assessed by comparing the SCE simulation results with the observation-based T_a data (Methods). Throughout most of the growing season (May to October), the SCE simulation well reproduced the increasing trend and interannual variability of the NH land mean T_a observed during 1982–2014 (Fig. S2). Observational data showed that the strongest NH warming occurred in early spring (March and April) and late autumn (November). However, the SCE simulation failed to capture the exceptionally strong warming during the transitional seasons, leading to the underestimation of the annual mean warming trend (SCE: 0.237 ± 0.024 °C decade⁻¹; observed: 0.362 ± 0.048 °C decade⁻¹). This underestimation stemmed from a negative bias in the increase of downwelling shortwave radiation, possibly due to an absence of short-lived forcing and bias in the cloud systems³⁷. Overall, the SCE reproduced the geographical patterns of seasonal warming reasonably well (Fig. S3), which strengthened our confidence in the model projections. Notably, it successfully captured the observed amplified warming over pan-arctic and semi-arid regions, as well as the few cases of regional cooling, such as that over northwestern North America during MAM (Fig. S3).

Intra-seasonal temperature responses to NH LAI changes

For the period from 1982 to 2014, satellite-retrieved LAI showed statistically significant increasing trends (p < 0.05) during all growing seasons, with the most substantial increase observed in JJA (0.046 ± 0.007 m² m⁻² decade⁻¹), followed by SON (0.027 ± 0.005 m² m⁻² decade⁻¹) and MAM (0.019 ± 0.005 m² m⁻² decade⁻¹)⁵ (Fig. 1a). Although forced identically by observed greening patterns, our GCM estimated highly divergent T_a responses across seasons, showing non-significant MAM warming (0.005 ± 0.018 °C decade⁻¹, p > 0.1), strong and significant JJA cooling (−0.044 ± 0.008 °C decade⁻¹, p < 0.05), and non-significant SON cooling (−0.014 ± 0.016 °C decade⁻¹, p > 0.1) (intra-seasonal feedbacks shown in Fig. 1b). The LAI-induced JJA T_a trend was equivalent to cooling of −0.15 ± 0.03 °C in JJA over the study period, offsetting the overall SCE-simulated near-surface air warming over this period by ~12.5%. This strong JJA cooling was further supported by a significant negative correlation (r = −0.64, p < 0.05) between JJA LAI and its induced T_a anomalies (Fig. S4b), with greener summers (higher JJA LAI) co-occurring with more negative ΔT_a. However, no such correlation was statistically detectable in MAM (r = −0.14, p > 0.1) or SON (r = 0.07, p > 0.1) (Fig. S4a, c), during which the LAI-induced changes accounted for only 1.3% (MAM) and −3.2% (SON) of the concurrent greenhouse warming. We also verified the robustness of our results by performing equilibrium experiments with an independent model, the NCAR Community Atmosphere Model coupled with Community Land Model (CAM-CLM, Methods). Indeed, this model generated a similarly strong LAI-induced cooling in JJA (−0.18 °C, p < 0.05 based on a two-sample t-test), which shifted to be much weaker in MAM (−0.02 °C, p > 0.1) and SON (−0.05 °C, p > 0.1) (Fig. S5).

Figure 2 shows the spatial distribution of the intra-seasonal T_a responses to LAI changes based on IPSL-CM. In the JJA experiment (SCE − ({{{{{{rm{LAI}}}}}}}_{{{{{{rm{CTL}}}}}}}^{{{{{{rm{JJA}}}}}}})), northern areas where ΔT_a decreased the most corresponded well to those where LAI showed the strongest greening, such as Europe, Russia, and the central U.S. (Fig. 2b, e). Similarly, increasing ΔT_a occurred in land areas that experienced the strongest browning, such as the southern U.S., Alaska, and semi-arid Asia (Fig. 2b, e). However, this spatial consistency of greening-cooling (or browning-warming) was not detected in most northern areas during MAM or SON (Fig. 2a, c, d, f). During these two seasons, though widespread areas showed significant greening (MAM: 42.0%; SON: 40.3%), only a small fraction (<5%) of northern lands exhibited significant trends of extra cooling or warming (Fig. 2a, c, d, f). In regions of Europe, East Asia. and the eastern U.S., the regional surface cooling in MAM and SON was also insignificant and discernably weaker than that in JJA, albeit with comparably strong greening signals across the three seasons. CAM-CLM also broadly agreed with the spatial patterns of signs despite varying magnitudes (Fig. S6). On one hand, these results imply a lower capacity for spring and autumn vegetation changes to influence near-surface climate, with potentially feedback processes different from summer periods. On the other hand, process-based uncertainties are more substantial in these relatively cold seasons associated with a model deficiency in representing vegetation–snow–moisture interactions.

Inter-seasonal temperature responses to NH LAI changes

The IPSL-CM model also produced particularly strong time-lagged T_a responses to greening in previous growing seasons. For example, MAM greening-induced cooling in late spring carried over to early summer; JJA greening-induced cooling extended throughout the entire growing season, although cooling in autumn and spring of the next year was not as strong as that in summer; and SON LAI changes caused anomalously strong cooling in the following spring (Fig. 1c). In all growing seasons, NH greening consistently resulted in net warming during winter (December to February, DJF). For each focused season, the overall time-lagged response of T_a to previous-season greening was calculated by summing up the GCM-simulated T_a changes in response to greening in the two other growing seasons. At the scale of the NH, the GCM estimated ΔT_a trends of −0.057 ± 0.029 °C decade⁻¹ (p = 0.06; MAM), −0.016 ± 0.013 °C decade⁻¹ (p = 0.21; JJA), and −0.029 ± 0.019 °C decade⁻¹ (p = 0.13; SON) in response to previous-season greening (Fig. 1b), corresponding to cooling of 0.19 ± 0.10 °C, 0.05 ± 0.04 °C, and 0.10 ± 0.06 °C, respectively during 1982−2014. In JJA, this cooling signal caused by previous greening amplified the transient cooling caused by JJA greening by ~36%. In MAM and SON, this inter-seasonal signal exceeded that caused by LAI changes within the same season (Fig. 1b), as also confirmed by the CAM-CLM results (Fig. S5), thereby dominating their overall T_a responses to LAI changes for the entire growing season. In both models, however, we did not detect the inter-seasonal signal with a very high statistical confidence level, due to trade-offs among regions and processes driving this carry-over effect.

Examination of the intra- and inter-seasonal biophysical feedbacks together revealed that growing season vegetation greening had a discernable effect on the seasonal trajectory of the NH land mean T_a (Fig. 1c, bottom panel). Seasonal greening overall contributed to net warming in winter and net cooling throughout the entire growing season. By simultaneously causing warmer winters and cooler summers, NH greening reduced the amplitude of the T_a annual cycle (SAT, defined as July T_a minus January T_a) at an average rate of −0.14 ± 0.07 °C decade⁻¹ (p < 0.1) (Fig. 3), signaling to weaken T_a seasonality. Several previous studies reported that land SAT decreased throughout the twentieth century^26,27,28. However, during the more recent 1982–2014 period, real-world observations and SCE results consistently showed a trend of leveling-off or a non-significant increase (+0.11 °C decade⁻¹) (Fig. 3), indicating that greening diminished the recent increase in the seasonal amplitude of T_a in a warmer climate. In other words, in the absence of vegetation-climate feedbacks, northern lands would have experienced a much faster SAT increase, by about +0.25 °C decade⁻¹, over the past 33 years.

Driving processes of the seasonal vegetation-climate feedback

An intriguing question is why the detected vegetation biophysical feedbacks on T_a vary in sign and magnitude among seasons. This variation is partly explained by the differing greening trends among seasons, with the most significant greening and cooling co-occurring in summer (Fig. 1a, b). Moreover, the major biophysical processes that govern the surface energy budget may also diverge seasonally. To fully understand the mechanisms of seasonal dependence among LAI-T_a feedbacks, we decomposed the modeled T_a response to LAI changes into contributions from different surface physical properties/processes, including surface albedo (α), evapotranspiration (λE or ET), surface incoming shortwave radiation (S_dn), air longwave radiative emissivity (ε_a), and aerodynamic resistance (r_a) (see Methods). Together, these processes well reproduced the modeled LAI-induced net change in surface radiative forcing related to changes in T_a (Fig. 4a–c).

During all growing seasons, the ET-related change in surface radiative forcing was the greatest among the surface processes considered (Fig. 4a–c). The ET enhancement was paralleled with a proportional decrease in sensible heat (Fig. S7a–c). The less available surface energy dissipated as latent heat (Fig. S7d) favors the evaporative cooling of the surface. Geographically, regions with ET enhancement (Fig. 4d–f and S8b, h, n) also corresponded well to those of satellite-observed greening (Fig. 2a–c), confirming localized vegetation-temperature feedbacks through this key process^13,29. Among the three growing seasons, greening generated a greater increase in the fraction of net solar radiation partitioned into ET (Fig. S7d), which resulted in a greater cooling in JJA (−0.64 ± 0.09 W m⁻² decade⁻¹, p < 0.05) than in MAM (−0.23 ± 0.05 W m⁻² decade⁻¹, p < 0.05) and SON (−0.15 ± 0.03 W m⁻² decade⁻¹, p < 0.05) (Fig. 4a–c). In JJA, plants often consume a greater amount of soil water to keep up with the greater water demand. For example, the ratio of transpiration to total ecosystem ET (T/ET)—a measure of vegetation control on surface climate through evaporative cooling³⁸—was ~50% greater in JJA than in MAM/SON (Fig. S9a), which resulted in larger positive sensitivity of land ET to LAI translated to stronger surface cooling in JJA (Fig. S9a). The stronger evaporative cooling capacity of JJA greening than other growing seasons was also supported by 33 state-of-the-art Earth system models (Fig. S9b), suggesting that the seasonally varying LAI-T_a feedbacks are not dependent on our specific model use. Interestingly, within both MAM and SON, the LAI-T_a feedbacks shifted progressively from warming-dominated in colder months (i.e., March and November) to cooling-dominated in warmer months (i.e., May and September) (Fig. 1c). This phenomenon can be partly explained by the land–atmosphere theory that latent heat of vaporization is a more efficient cooling mechanism at higher temperatures³⁹.

In addition to affecting how the land surface dissipates absorbed energy, greening also affects how much energy the surface absorbs (i.e., due to changes in α, S_dn, and ε_a). These radiative processes played a secondary, but often non-negligible, role, together increasing the surface radiative forcing at an average rate of 0.25 W m⁻² decade⁻¹ (p < 0.05; MAM), 0.18 W m⁻² decade⁻¹ (p < 0.05; JJA), and 0.05 W m⁻² decade⁻¹ (p < 0.05; SON) (Fig. 4a–c). During MAM when ET is constrained by relatively low temperatures, this radiative warming (as the combined effect of α, S_dn, and ε_a) overrode evaporative cooling and caused net surface warming (Fig. 4a). Increased ε_a contributed the most to this warming (Fig. 4a, d), as greening increased atmospheric water vapor content, which warmed the air by trapping extra longwave radiation^13,21. Decreased α also significantly increased surface radiative forcing, particularly over greening and boreal/arctic regions, and throughout the seasonal cycle, such albedo-related warming peaked during MAM in the presence of snow cover^11,18 (Fig. 4a, d and S8c). In JJA, radiative warming, as a result of decreased α and increased ε_a, offset only 26% of evaporative cooling, causing net surface cooling. In SON, radiative warming due to decreased α and increased S_dn offset ~35% of the concurrent evaporative cooling, such that the surface cooling became insignificant.

We also evaluated how the widespread NH greening triggered inter-seasonal T_a responses. During those seasons when LAI was kept the same in the experiment pair (Table S1), any detected climate anomalies were linked to perturbations of soil and atmospheric conditions arising from LAI variations in the preceding season. At the NH scale, direct surface biophysical processes had poor explanatory power with respect to inter-seasonal changes in T_a (Fig. S10), signaling weak legacy effects through coupling between soil and surface boundary layer. However, regionally, soil moisture was a key factor connecting MAM LAI changes to JJA T_a responses, often showing regional warming in areas with soil drying (Figs. S11a, S12a). This occurred because MAM greening enhanced ET, causing a soil moisture deficit that was carried over to JJA; drier soils further enhanced JJA warming by dissipating absorbed solar radiation to a greater extent as sensible rather than latent heat²³. In other cases, we found similar spatial patterns between trends of 500 hPa geopotential height (z500, an indicator of synoptic circulation changes impacting regional climate) and inter-seasonal ΔT_a, with negative (positive) z500 trends corresponding to cooling (warming) (Figs. S11,S13). These patterns showed little similarity to greening patterns, suggesting that atmospheric circulation response connected the direct forcing of LAI changes on the local atmosphere to other regions and seasons. Via atmospheric teleconnections, SON NH greening triggered widespread MAM cooling centered in western Asia, as well as local MAM warming centered in northern Canada (Fig. S11g). This MAM cooling matched well with the SON LAI-induced MAM anomalous low, with the western Asia cooling center controlled by a cyclonic anomaly structure (Figs. S11g, S13g). By analyzing every repeated experiment with different initial conditions (Methods), we found similar teleconnection patterns in 19 out of 30 ensemble members (Fig. S14), suggesting that this pattern is more likely connected to vegetation greening than internal variability in the climate system. During DJF, an anomalous high prevailed over land forced by greening in all growing seasons (Fig. S13c, f, i), resulting in widespread NH warming (Fig. S11c, f, i), which implies that greening distributed more available solar energy toward the coolest periods of the seasonal cycle. By summarizing the ΔT_a responses in the space of soil moisture and z500 changes, we found that the inter-seasonal ΔT_a increases mostly followed a positive gradient of z500 trends (Fig. 5). Therefore, indirect atmospheric processes, such as warm/cold air mass advection and large-scale atmospheric circulation^25,40, have played a key role in propagating the climatic effects of greening to other seasons.

In summary, this model-based study shows that LAI-T_a feedbacks through biophysical processes have apparent seasonal inconsistency regarding the signs and magnitudes, resulting from a trade-off between radiative warming and evaporative cooling. This is in contrast to the well-documented carbon-cycle feedbacks, for which NH greening exerts a seasonally consistent negative forcing on T_a through enhancing plant photosynthetic rates. Our results indicate that, in summer, the overwhelmingly strong ET enhancement forces net surface cooling, whereas in spring and autumn, radiative warming due to decreased albedo and higher atmospheric water vapor content nullify the evaporative cooling, producing weak and insignificant T_a changes. Compared with the straightforward response to greening, spring and autumn T_a changes are impacted to a greater extent by atmospheric circulation anomalies generated by greening in the preceding seasons. We however note potential uncertainties in our model-based LAI-T_a feedback results, particularly in terms of the inter-seasonal signal, for which solid observational constraints on the atmospheric circulation processes are presently lacking. We encourage other climate model centers to perform similar experiments to refine the exact values of vegetation biophysical feedbacks on seasonal T_a changes.

Our study achieves a holistic understanding of the seasonal responses of air temperature to NH greening over the growing season. We illustrate complex trade-offs and linkages of vegetation-climate feedbacks between different growing stages, which tend to be concealed within annually aggregated values¹³. This highlights the need to better understand these biophysical processes operating both within and across seasons so that their potential time-lagged climate benefits and/or counterproductive consequences will not be overlooked. The regulatory role of NH greening on seasonal climate also has implications for adaptation planning and decision-making, as greening is now increasingly shaped by human land-use practices such as afforestation and reforestation⁴¹.

Source: Ecology - nature.com