Main

Rapid urbanization driven by human activities has directly and indirectly resulted in substantial modifications to landscapes, regional environments and ecosystem1. For example, the urban heat island (UHI) effect, characterized by higher temperatures in urban areas compared with rural areas2, and artificial light at night (ALAN)3,4 profoundly change the temperature and light regimes in cities. Urban surface warming increased by 0.50 ± 0.20 K per decade, 29% faster than rural areas during 2002–20215. Meanwhile, ALAN increased by 3–6% annually in the late twentieth century and surged to nearly 10% per year over the past decade6,7. The warmer and brighter urban environments may have a large influence on urban ecosystems, particularly extending to vegetation phenology—the timing of biological events such as the start of season (SOS) and the end of season (EOS)3,8. Vegetation phenology exerts a profound influence on the energy budget, water and carbon cycles, biodiversity conservation of ecosystems and feedback on climate change9,10,11. Therefore, quantifying the joint effects of increased air temperature (Ta) and ALAN on vegetation phenology is essential to understanding the anthropogenic effects on ecosystem dynamics, especially under continuous urbanization and climate change.

Studies have shown earlier onset in spring and later leaf coloring in autumn in urban areas than in rural areas12,13. Such shifts in urban phenology have been attributed solely to the higher Ta prevalent in cities14, overlooking the potentially substantial influence of ALAN on phenology (but see ffrench-Constant et al.15 and review in Briggs16). Nevertheless, plant phenology is intricately regulated by both temperature and light—a well-established understanding gained from controlled experiments17,18. Without interannual variation in natural conditions, the photoperiod serves as a reliable cue for seasonal progression19. Therefore, seasonal changes in light duration and intensity play a crucial role in inducing phenological events at the optimal time20,21. By altering natural light levels (ranging from a surface illuminance of <0.6 mlx on a cloudy night to 0.3 lx with a full moon at zenith), ALAN in urban areas (ranging from 7 to 65 mlx on a clear night, up to 150 lx near street lights) largely amplified both the duration and the intensity of nocturnal light22. Experimental studies have shown that increased ALAN intensity caused earlier bud swelling dates up to 4–20 days18,23. A recent study further showed ALAN advanced breaking leaf buds by 9 days and delayed leaf color change by 6 days across the USA8, highlighting the consistent and widespread effect of ALAN on phenological changes in cities.

The relative impact of ALAN compared with Ta in driving phenological shifts remains largely unknown, leading to large uncertainties in predicting future phenological changes under the combined impact of increasing ALAN, UHI and climate change24. Disentangling the effects of ALAN and Ta is inherently challenging because both increase from rural to urban areas. Previous studies often used a binary rural–urban division8,14, overlooking the spatial heterogeneity of ALAN and UHI, especially in cities with multiple subcenters25. Furthermore, these studies have yet to evaluate the vegetation response across climate gradients, limiting our understanding of the effects of light and temperature on phenological changes at regional to global scales.

This study investigated the influence of ALAN and Ta on phenology along rural-to-urban gradients across 428 cities in the Northern Hemisphere. We asked the following research questions: What are the spatial patterns of ALAN, Ta, SOS and EOS from rural to urban areas across study cities? To what extent do ALAN and Ta affect changes in SOS and EOS? Does the magnitude of ALAN and Ta effects on phenology differ across climatic zones? To answer these questions, we developed a buffer-based method to quantify phenology, ALAN and Ta across rural–urban gradients for each city by defining ten buffer zones with incremental changes in the impervious surface area index (β; indicating the proportion of impervious surface area that ranges from 0 to 1) at the interval of Δβ = 0.1 (the first buffer represents the outermost rural area, and the tenth buffer represents the urban center; Methods). We examined the spatial patterns of ALAN, Ta, SOS and EOS for each city with the increasing β using ALAN measured by the Visible and Infrared Imaging Radiometer Suite (VIIRS) Day Night Band (DNB)26, gridded daily Ta dataset27 and Moderate Resolution Imaging Spectroradiometer (MODIS) phenology product from 2014 to 2020 (Fig. 1). We further validated the spatial pattern of phenology shifts along rural and urban areas and isolated the grass effect on urban phenology using a 30-m urban phenology dataset. We then disentangled the effects of ALAN and Ta on phenology using partial correlation analysis and examined the strength of ALAN and Ta effects across climate zones and latitudes using the Köppen climate classification28. This study helps us better understand how altered urban environments affect ecosystems in previously unrecognized ways, thus having critical implications for sustainable urban development policies and practices aimed at preserving and enhancing the stability of urban ecosystems.

Fig. 1: ALAN and UHI effects on vegetation phenology.
figure 1

Increasing impervious fractions results in a concave quadratic increase of air temperature (Ta) and exponential increase of ALAN from rural to urban areas. The longer exposure to ALAN and higher Ta collectively contributes to a longer growing season, represented by the advancement of the SOS and delay of EOS in urban areas compared with rural areas.

Source data

Full size image

Results

Spatial patterns of phenology, ALAN and T a

From rural to urban areas (represented by ten buffers where β increased from 0 to 1), we observed a consistently nonlinear trend of earlier SOS, later EOS and increasing ALAN and Ta in 378 (88.3%) of 428 Northern Hemisphere cities (Fig. 2 and Extended Data Figs. 1 and 2). Specifically, both SOS and EOS exhibited a quadratic change, that is, SOS advanced from 94.4 ± 0.4 day of year (DOY) (mean ± standard error, identical hereafter) in the outermost rural area (first buffer) to 81.8 ± 0.3 DOY in the innermost urban center (tenth buffer), and EOS extended from 227.6 ± 0.3 DOY to 238.8 ± 0.3 DOY (P < 0.001) on average across all study cities (Fig. 2a,c). Spring ALAN increased exponentially from rural to urban areas (3.1 ± 0.1 nW cm−2 sr−1 in the first buffer and 53.3 ± 0.6 nW cm−2 sr-1 in the tenth buffer) (P < 0.001), with approximately 50% increase occurring from the first to the seventh buffer, followed by a dramatic increase from the eighth to the tenth buffer (Fig. 2a,b). By comparison, spring Ta followed a concave quadratic increase from rural to urban areas (P < 0.001), with approximately 50% rise occurring only between the first and the second buffer, followed by a minor and gradual increase toward the urban center (10.7 ± 0.1 °C in the first buffer and 11.5 ± 0.1 °C in the tenth buffer). Autumn ALAN and Ta showed similar changes as spring ALAN and Ta, although autumn ALAN was lower than that in spring and autumn Ta was higher than spring Ta (Fig. 2c,d). Furthermore, we examined the spatial patterns of SOS, EOS, ALAN and Ta for 20 cities with the best models fit and found consistently nonlinear changes along rural–urban gradients for individual cities (Methods and Supplementary Figs. 2 and 3). To validate the spatial pattern of phenology at finer resolution, we used a 30-m urban phenology dataset and examined spatial pattern among 62 cities in the continental USA (Methods and Extended Data Fig. 3). It showed consistent earlier SOS, ranging from 124.8 ± 1.20 DOY in the fourth buffer to 121.3 ± 1.1 DOY in the tenth buffer, and delayed EOS from 276.2 ± 1.0 DOY in the first buffer to 281.3 ± 1.1 DOY in the tenth buffer along rural–urban gradients. This spatial pattern remained robust even after excluding developed open space areas, which are predominantly covered by lawn and grass, further supporting our findings (Extended Data Fig. 3e,f).

Fig. 2: Spatial patterns of vegetation phenology, ALAN and Ta along rural to urban gradients in 428 cities in the Northern Hemisphere.
figure 2

a,c, β is the impervious surface areas index, with 0 indicating utter permeable area in rural areas and 1 representing absolute impervious areas in urban areas. The dots and error bars indicate the mean values and s.e.m. among 428 cities during 2014‒2020 in each buffer. Solid lines represent the fit curves. The fit regression models and P values (using one-sided F test with 99% confidence intervals) are shown. Spring (a) and autumn (c) ALAN and Ta are presented. b,d, Ridge maps for the variance of SOS, EOS, ALAN and Ta across ten buffers. Buffer 1 represents the outermost rural areas, and buffer 10 represents the innermost urban areas. Spring (b) and autumn (d) ALAN and Ta are presented.

Source data

Full size image

Cities in Asia (200 cities), Europe (105) and North America (123) displayed similar nonlinear patterns of phenology, ALAN and Ta across rural–urban gradients (P < 0.001; Extended Data Table 1 and Extended Data Fig. 2), although with different magnitudes. Europe displayed the earliest SOS (81.8 ± 0.6 DOY) and EOS (228.4 ± 0.5 DOY), followed by Asia (88.5 ± 0.6 DOY for SOS and 233.1 ± 0.4 DOY for EOS) and North America (89.5 ± 0.7 DOY for SOS and 236.5 ± 0.5 DOY for EOS). ALAN in Asia (20.8 ± 0.6 nW cm−2 sr−1 in spring and 15.2 ± 0.3 nW cm−2 sr−1 in autumn) was lower than that in Europe (23.6 ± 0.5 nW cm−2 sr−1 in spring and 21.6 ± 0.3 nW cm−2 sr−1 in autumn) and in North America (30.1 ± 0.5 nW cm−2 sr−1 in spring and 20.7 ± 0.3 nW cm−2 sr−1 in autumn). North American cities are the brightest among all three continents. Ta in Europe was slightly lower in spring (10.3 ± 0.1 °C) and autumn (25.9 ± 0.1 °C) than in Asia in spring (11.3 ± 0.2 °C) and autumn (30.7 ± 0.1 °C) and North America (11.0 ± 0.2 °C in spring and 30.2 ± 0.1 °C in autumn).

To accurately measure the magnitude of the rural–urban difference, we calculated the multiyear average difference in phenological metrics, ALAN and Ta during 2014–2020 in 428 cities. SOS was 6.4 ± 0.2 days earlier on average (Fig. 3a and Extended Data Table 2), and EOS was 7.1 ± 0.2 days (Fig. 3b) later in urban areas, compared with rural areas. Similar differences in SOS and EOS were observed using the 30-m urban phenology dataset, although with a smaller magnitude (Extended Data Fig. 4). Most cities with greater SOS difference (<−10 days) between urban and rural areas were located in the southeast of Asia and North America, as well as the northwest coast of Europe. Cities with a greater EOS difference (>10 days) were mainly in southern Asia, central Britain in Europe and west of North America. ALAN in urban areas was 25.5 ± 0.3 nW cm−2 sr−1 higher in the spring and 20.1 ± 0.2 nW cm−2 sr−1 higher in the autumn compared with nearby rural areas (Fig. 3e,f). Most cities that had a greater ALAN difference were in eastern coastal Asia, North America and Europe. Moreover, spring Ta was 0.3 ± 0.004 °C higher and autumn Ta was 0.5 ± 0.005 °C higher in urban areas than in rural areas (Fig. 3c,d). Cities with stronger UHI effects (>0.5 °C) converged on the east coast of Asia, the southern regions of Europe and the western coast of North America. Among Asia, Europe and North America, we observed that the urban–rural difference in SOS (−5.3 ± 0.3 days; Extended Data Table 2 and Supplementary Figs. 4–6) was the smallest in Europe and EOS (6.6 ± 0.2 days; Extended Data Table 2 and Supplementary Figs. 4–6) was smallest in North America. ALAN differences in Asia (20.1 ± 0.5 nW cm−2 sr−1 for spring and 15.6 ± 0.3 nW cm−2 sr−1 for autumn) were smaller than those in Europe and North America (Extended Data Table 2 and Supplementary Figs. 4–6). The urban–rural difference in spring Ta (0.4 ± 0.01 °C) and autumn Ta (0.6 ± 0.01 °C) in Europe were greater than the Ta difference in Asia and North America (Extended Data Table 2). In addition, we found a greater Ta difference in autumn than in spring, indicating a stronger UHI effect in autumn.

Fig. 3: Earlier SOS and later EOS in brighter and warmer cities during 2014–2020 in the Northern Hemisphere.
figure 3

a–f, The difference in SOS (a), EOS (b), spring ALAN (c), autumn ALAN (d), spring Ta (e) and autumn Ta (f) between urban and rural areas. The inserted figures are histograms of SOS, EOS, ALAN and Ta differences, with average differences shown as vertical lines and bold numbers. Orange represents positive, and purple indicates negative differences.

Source data

Full size image

Artificial light and temperature effects on phenology

We found that both earlier SOS and later EOS were associated with higher ALAN (partial correlation coefficients RALAN_SOS = −0.37 ± 0.31 (mean ± S.D) and RALAN_EOS = 0.38 ± 0.28) and higher Ta (RTa_SOS = −0.33 ± 0.35 and RTa_EOS = −0.03 ± 0.39) in 428 Northern Hemisphere cities using partial correlation analysis (Fig. 4). SOS was negatively correlated with ALAN in 300 (70.4%) cities, mainly in the north of Eurasia and both west and east of North America, and negatively correlated with Ta in 62.4% of cities (P < 0.05; Fig. 4a). A total of 279 cities (65.2% of cities) indicated a significant difference between ALAN and Ta coefficients on SOS change (P < 0.05) with Fisher Z-transformation testing (Extended Data Fig. 5a). In these 279 cities, ALAN was the dominant factor in regulating SOS in 135 (31.4%) out of 428 cities, while Ta played a predominant role in 144 cities (33.6%).

Fig. 4: Stronger impacts of ALAN than Ta on SOS and EOS.
figure 4

a–d, Spatial patterns of partial correlation coefficients between SOS and ALAN (a), between SOS and Ta (b), between EOS and ALAN (c) and between EOS and Ta (d) in 428 cities during 2014‒2020. The inserted figures represent the percentage of cities with each correlation category (Neg*, significant negative; Neg, negative; Pos, positive; Pos*, significant positive; P < 0.05 for significance). e,f, Histograms of the partial correlation coefficients in 428 cities for SOS (e) and EOS (f). The dashed lines indicate the mean partial correlation coefficients of ALAN and Ta. Mean ± s.d. of correlations corresponding to ALAN and Ta are also shown in the figures. ***Significant difference between RALAN and RTa for EOS across all cities (P < 0.001 using two-sided t-test).

Source data

Full size image

Notably, the magnitude of the ALAN effect surpassed the influence of Ta on EOS (Fig. 3f). EOS was positively correlated with ALAN in 310 (72.7%) cities (Fig. 4c,d and Extended Data Fig. 5b; P < 0.05) but exhibited an inconsistent relationship with Ta. EOS was negatively correlated with Ta in 134 (31.5%) out of the 428 cities, mainly in the west of Europe and North America (Supplementary Figs. 8c and 9c), and positively correlated with Ta in 118 (27.7%) cities, predominantly observed in cities located in in the east of Asia and North America (Fig. 3c and Supplementary Figs. 7d and 9d; P < 0.05). A total of 334 cities (78.0% of 428 cities, P < 0.05) showed a significant difference between ALAN and Ta coefficients for EOS. In these 338 cities, a stronger relationship between EOS and ALAN change was observed in 48.8% of cities (P < 0.05), located primarily in southeastern Asia, Europe and North America (Extended Data Fig. 5). By contrast, only 125 cities (29.2%) showed greater EOS sensitivity to Ta along rural–urban areas. Given the inherent collinearity between light and temperature, we incorporated an interaction term between ALAN and Ta in our analysis (Supplementary Fig. 11). The results showed that ALAN exhibited a stronger effect than Ta on both SOS and EOS, further supporting our findings. To investigate the impact of daily extreme temperature, we replaced daily average temperature with daily maximum and minimum air temperature and obtained a consistent finding (Extended Data Fig. 6). Similar analysis using the 30-m urban phenology dataset, across the 62 cities in the continental USA, further confirmed ALAN’s role in extending the growing season, particularly by delaying EOS (Supplementary Figs. 12 and 13).

Varying effects across climate zones

We found that the magnitudes of ALAN and Ta effects on vegetation phenology differed across climate zones and latitudes (Fig. 5). The ALAN effect on advancing SOS was stronger than the Ta effect in 4 out of 7 climate zones (Fig. 5b), including the dry summer (Cs) zone with 46 (10.7%) cities and dry winter (Cw) zone with 17 (4.0%) cities in temperate regions and the dry winter (Dw) zone with 57 (13.3%) cities and no dry season (Df) zone with 102 (23.8%) cities in cold regions (Fig. 5a and Supplementary Table 1). Significant stronger effects of ALAN on SOS were observed in the Df zone (P < 0.05), where these cities were located mainly in the northern regions of continents. By contrast, EOS showed a consistent positive correlation with ALAN across all climate zones, with the ALAN effect (RALAN = 0.49) significantly outweighing the Ta effect (RTa = 0.003) on EOS in all zones (P < 0.01) (Fig. 5c). The relationship between EOS and Ta was, however, less solid and varied with different climatic zones. In general, higher Ta led to earlier EOS in temperate and cold regions without dry seasons and later EOS in arid zones as well as temperate and cold regions with dry seasons. Higher Ta slightly induced an earlier EOS in 58% (250 out of 427) cities in Cw, Cf and Df zones. These climate zones are in Southeast Asia, North America and most parts of Europe with mild climate conditions (Fig. 5a). EOS showed a slightly positive relationship with Ta in arid areas in both arid desert (BW) and arid steppe (BS) zones, Cs and Dw zones.

Fig. 5: Varying effects of ALAN and Ta on phenology across climate zones.
figure 5

a, The spatial distribution of study cities classified by the Köppen climate zones. The inserted bar plot represents the number of cities in each climate zone. b,c, Box plots showing the partial correlation coefficients (R) of ALAN and Ta for seven climate zones for SOS (b) and EOS (c) across 427 cities from 2014 to 2020, as climate zones with fewer than ten cities are excluded (Methods). The middle solid line indicates the median partial coefficients of ALAN (in purple) and Ta (in orange) in each group. The bottom bounds indicate the 25th quantile and the top bounds indicate the 75th quantile for each box. The lower whisker represents the minimum, and the upper whisker represents the maximum. The black dots indicate the significant difference between RALAN and RTa in each group using a two-sided t-test. *P < 0.05, **P < 0.01.

Source data

Full size image

Meanwhile, the effects of Ta and ALAN on vegetation phenology varied with latitude (Extended Data Fig. 7). The correlation between EOS and Ta shifted from positive to negative with increasing latitude. Both SOS and EOS were more sensitive to ALAN than Ta in low latitudes (<48° N, RTa = −0.20 and RALAN = −0.48 for SOS, and RTa = 0.10 and RALAN = 0.46 for EOS), but SOS was more responsive to Ta than ALAN in high latitudes (>48° N, RTa = −0.49 and RALAN = −0.44 for SOS, and RTa = −0.18 and RALAN = 0.54 for EOS). SOS in cold-climate cities is more sensitive to warming, while ALAN’s effect on advancing SOS weakened in warmer cities (Supplementary Fig. 14), possibly due to the decrease in chilling20. ALAN played a more important role in delaying EOS than Ta, but its effect diminished as city temperatures increased.

Discussion

The concurrence of increasing ALAN and rising air temperature along rural–urban gradients posed a challenge to determine the independent effects of ALAN and air temperature on vegetation phenology. To address this, we used a systematic approach to capturing the nonlinear spatial distribution of vegetation phenology, ALAN and Ta along the rural–urban gradients. Our analysis revealed an exponential increase in ALAN and a quadratic increase in Ta with increasing impervious fractions. In addition, we found that ALAN in rural and suburban areas was not only lower but also more homogeneous than in urban areas, according to the increasing variation of ALAN from rural to urban buffers. This can be attributed to extensive lighting infrastructure within high-density urban and commercial areas, leading to a more pronounced, rapid and heterogeneous increase in ALAN7,29. Based on the spatial pattern of ALAN and Ta from rural to urban areas, phenological changes from rural to suburban areas were probably predominantly driven by Ta change, while phenological changes from suburban to urban cores were probably driven by the change in ALAN. These findings provide valuable insight into the urban-induced impact on phenology and the further accurate assessment of the vegetation feedback on mitigating UHI effect9. With rapid worldwide urbanization, we highlight the importance of considering both ALAN and Ta in evaluating the impact of urbanization on urban ecosystem service and various ecological processes, especially for the profound but often underestimated effect of ALAN.

In contrast to previous studies that attributed longer urban growing seasons to higher Ta (refs. 30,31), our study accentuated that ALAN also plays a critical role in altering urban vegetation phenology, especially for EOS. The negative relationship between SOS and Ta in 75.3% of cities aligns with research showing that climate warming drives earlier spring phenology. More importantly, the significantly greater effect of ALAN on SOS in 31.5% of cities and on EOS in 48.8% cities suggests that ALAN has been underestimated as a key driver of earlier spring phenology and later autumn phenology (Fig. 1). This aligns with prior in-site observations in the UK that showed spring bud break shifted earlier under ALAN by about 4‒7.5 days15, and with remotely sensed data showing prominent ALAN effects on phenology in South Korea32 and in the USA8. Temperature influences SOS in more cities than ALAN, but ALAN has a stronger overall effect on SOS, suggesting that, whereas temperature governs broad-scale pattern, ALAN may have a more pronounced effect in specific regions or under particular conditions. The interaction effect of light and Ta effects on phenology shifts has been reported in previous studies. For instance, a study showed that light and temperature jointly modulate the spring phenology to mitigate the risk of frost damage in spring20. In autumn, the combination of sufficient light and adequate temperature may increase productivity, thereby leading to an earlier date of leaf senescence33. This interaction may create more complex synergistic effects, leading to greater asynchrony in phenology in cities. We also observed that the effects of both ALAN and temperature on SOS diminished with decreasing latitudes and increasing temperatures, potentially resulting from insufficient chilling accumulation in winter12,34. Furthermore, EOS was found to be more depended on ALAN than on Ta compared with SOS, supported by previous studies that showed a stronger day length effect on phenology than temperature effect19,24. ALAN acts as an artificial extension of natural daylight, disrupting photoperiod cues that regulate seasonal transitions in plant35. This prolonged exposure to light can delay the accumulation of senescence-inducing signals, such as reduced photosynthetic efficiency and hormonal changes (for example, decreased auxin and increased ethylene and abscisic acid levels), thereby postponing leaf senescence and extending the growing season. The increased dependence of EOS on ALAN with higher latitudes indicates stronger sensitivity of plant phenology to natural day length, that is, EOS at higher latitudes, with larger seasonal variation of day length, may experience greater delays compared with lower latitudes under the same ALAN enhancement36.

The regional variations in the effects of ALAN and Ta on phenology are probably driven partly by differences in climate conditions. For example, our findings indicate that Ta had a greater influence on SOS than ALAN in arid areas, possibly due to higher temperatures inducing overwhelming water stress on vegetation37,38. This water stress may also explain the earlier, instead of later, EOS observed under higher temperatures in European cities. While higher temperatures in summer and early autumn could lead to extended growing seasons, they can paradoxically reduce water availability, leading to premature leaf senescence and accelerating autumn phenology33,37. In arid cities within central Eurasia, the cooling island effect39 may explain the negative relationship between temperature and EOS across rural and urban gradients in our study, such as in Toshkent and Andijan, Uzbekistan. Our study also suggests that phenology changes in temperate forests are more sensitive to ALAN and Ta than in tropic or subtropic forests, as indicated by the comparison across Koppen climate groups and latitude intervals.

Different plant functional types and plant species compositions may contribute to varying urban phenological patterns, as cities often contain a higher proportion of nonnative species selected for their resilience to urban stressors such as heat, drought and soil salinity40,41,42. These species can exhibit distinct phenological responses compared with native vegetation43,44, potentially influencing the observed effects of temperature and ALAN on phenology. At the same time, many cities are increasingly restoring native species for urban landscaping and expanding urban greenness and biodiversity, which may help to homogenize species composition and reduce species-specific effects on phenological responses. Although differentiating ALAN and Ta effects on phenology across PTFs or species is challenging due to the predominance of impervious surfaces in cities and the moderate resolution of remote-sensed datasets on global scale, our analysis using both 500-m MODIS and a 30-m urban phenology product reveals consistent spatial patterns. After excluding developed open space, primarily lawns and grasses, our findings further confirmed the robustness of our main conclusion to plant functional type variations. By averaging phenology across buffer zones, our satellite analysis minimizes species-specific influence, allowing observed patterns to primarily reflect environmental gradients such as ALAN and temperature. Future studies incorporating finer-scale species composition data could help to disentangle species effects and enhance our understanding of urban phenology. This is particularly relevant for applied ecology and urban planning, where plant selection and management decisions shape ecosystem functioning and trophic interactions42,43,44.

The rapid increase in nighttime artificial lighting emissions is a global phenomenon7, driven by the transition from traditional light technologies, for example, high-pressure sodium lamps, to energy-efficient white light-emitting diodes (LEDs)45. While this transition reduces energy consumption and carbon emissions, it paradoxically increases ALAN and alters spectral compositions, which has far-reaching implication on urban ecosystem but is less understood. Plants sense different wavelengths through different photoreceptor pathways, for example, phytochrome is sensitive to red (620‒700 nm) and far-red light (700–1,000 nm) whereas phototropins and cryptochromes respond mainly to blue light46,47. The shift toward LED lighting, with its distinct blue-light emissions, may fundamentally alter how ALAN affects phenology—a question yet to be investigated. However, current satellite technology, such as VIIRS DNB, is designed to detect upward panchromatic light emitted by cities and other human activities in the 500–900 nm range, which is highly correlated with the light intensity that plants are exposed to in cities48. This spectral limitation underscores the imperative need for an enhanced nighttime satellite sensor capability of accurately capturing a broader spectrum of ALAN emissions, particularly in the blue range4. However, because we assume that the transition to LED technology took place relatively simultaneously in the individual urban regions, the uncertainties in our analyses of the VIIRS data along rural–urban gradients can be expected to be relatively small.

Increased light will continue to elevate nighttime light levels in both urban areas and nearby remote rural areas through skyglow. Our study demonstrating a significant ALAN effect on extended growing seasons has critical implications for urban ecosystem conservation and sustainable development. For example, longer growing seasons of urban trees may reduce the UHI effect and enhance carbon sequestration49. However, earlier leaf-out and later leaf senescence driven by ALAN may increase the risk of frost damage in early spring and late autumn, disrupt plant–pollinator interactions and lead to an earlier occurrence of pollen allergy symptoms50,51,52. The key to mitigating these impacts is to develop urban lighting strategies that minimize the impact on the urban ecosystem while concurrently fulfilling essential functional requirements. With increasing urbanization and climate change, it is imperative to implement sustainable lighting solutions that better protect trees and their associated communities through effective conservation measures, ultimately fostering resilient and healthy urban ecosystems.

Methods

Dataset

Artificial light at night

We selected the monthly ALAN images from the VIIRS DNB with Joint Polar-orbiting Satellite System from 2014 to 2020 with a spatial resolution of 500 m (https://eogdata.mines.edu/products/vnl/)53. The VIIRS DNB dataset has been calibrated by removing the data polluted by stray light and the impact of cloud cover. Due to the effect of background radiance for nighttime light, a logarithmic scaling method was applied to remove outlier pixels54. Furthermore, the VIIRS DNB captures the dynamic between 0.5 μm and 0.9 μm light spectra, which is profound to vegetation phenology. Following the preceding experimental and observation research8, we defined the fixed period of ALAN as 3 months in terms of the vegetation phenology date when we examined the effect of ALAN on vegetation phenology. We calculated the mean ALAN from January to March as an indicator of the SOS effect and from July to September to represent the EOS effect.

Near-surface air temperature

We used the global daily near-surface air temperature (Ta) data from Zhang et al.27 with a high resolution of 1 km for 2014–2020 (https://iastate.figshare.com/collections/A_global_1_km_resolution_daily_near-surface_air_temperature_dataset_2003_2020_/6005185). We reprojected the dataset to 500 m resolution to keep consistent with the vegetation phenology dataset. We calculated the daily mean Ta from Tmax (air temperature at mid-daytime) and Tmin (air temperature at midnight) to quantify the temperature difference across the rural–urban gradients. We also used the Tmax and Tmin in partial correlation analysis to test the robustness of the results. We chose Ta rather than the land surface temperature to represent the UHI phenomenon, because the atmospheric temperature 1.5–2 m above the surface was more appropriate to indicate the temperature impact on the vegetation canopy27. Based on previous studies8, we used Ta during the preseason (a 40-day fixed period before vegetation phenology occurs) for the analysis. Specifically, we calculated the spring mean temperature (DOY from 51 to 90) and autumn mean temperature (DOY from 191 to 230), according to the long-term mean SOS and EOS in 428 cities. To ensure the choice of different temperature periods does not affect results, we further conducted three sensitivity analyses using mean temperature for the periods 31–90 DOY, 41–90 DOY and 51–110 DOY for SOS and 191–251 DOY, 201–250 DOY and 221–280 DOY for EOS.

Vegetation phenology

We acquired the VIIRS Global Land Surface Phenology Product (GLSP) Collection 1 VNP22Q2 from the NASA Land Processes Distributed Active Archive Center Distribution Server (https://lpdaac.usgs.gov/products/vnp22q2v001/) to extract the yearly vegetation phenology indices at a 500-m spatial resolution spanning from 2014 to 2020. The VNP22Q2 V001 was calculated via hybrid piecewise logistic modeling from a time series of the two-band enhanced vegetation index (EVI2) derived from the VIIRS Nadir Bidirectional Reflectance Distribution Function Adjusted Reflectance (NBAR). The product was validated and showed high agreement with Landsat phenological observations and in situ leaf bud break observations from USA National Phenology Network55. With the aim to investigate the overall spatial pattern of phenology across rural-to-urban gradients rather than individual trees in cities, the VNP22Q2 dataset offers the advantage of credible and long-term time series within aggregated buffer areas. In our study, we selected the ‘onset greenness increase’ as the SOS date in spring and the ‘onset greenness decrease’ as the EOS date in autumn, at which time canopy greenness begins to decrease.

We also utilized a 30-m urban phenology product developed by Li et al.56 to validate the spatial pattern of phenology in cities. It captured the phenology dynamic in 147 cities in the USA from 1982 to 2015, which aligns well with in situ observations from the PhenoCam network and Harvard Forest. We chose phenological data from 62 cities, each exceeding 500 km2 in size with a period of 2014‒2015 (Supplementary Fig. 16). These cities are included in both our MODIS analysis and the 30-m dataset, ensuring a consistent and focused examination of urban phenological patterns across multiple data sources.

Land use and land cover

Our work used yearly land use and land cover collection 6.0 MODIS yearly product following the International Geosphere-Biosphere Programme classification at a 500-m spatial resolution from 2014 to 2020 to select study land cover types. We focused on the phenology of forest and urban vegetation; therefore, we selected six land cover types of forest and urban, including evergreen needleleaf forests, evergreen broadleaf forests, deciduous needleleaf forests, deciduous broadleaf forests, mixed forests, and urban and built-up lands. Cropland was excluded from this study owing to the high intensity of human activities and irrigation in these areas. We removed other land use and land cover types in each yearly image and then masked the corresponding vegetation phenology, Ta and ALAN images based on the yearly land use and land cover types.

The National Land Cover Database provides spatially explicit and reliable information on land cover and land cover change across the USA at 30-m resolution57. It released nine epochs from 2001 to 2021 (2001, 2004, 2006, 2008, 2011, 2013, 2016, 2019 and 2021). In this study, we utilized the 2013 National Land Cover Database map to match the 30-m urban phenology dataset and selected six land cover types, including deciduous forest, evergreen forest, mixed forest, developed open space, developed low intensity and developed medium intensity, to represent forested and urban areas in each city.

City boundary dataset and impervious surface area

We defined urban areas for each study city on the basis of the global urban boundary (GUB)58 in 2018. The urban boundaries were developed by the kernel density estimation approach and cellular-automata-based urban growth modeling and improved by the morphological approach. Compared with other products, GUB has been demonstrated to capture the urban fringe areas well and show good agreement with the high-resolution Google Earth images58. We implemented a three-step city selection process outlined below to guarantee valid areas in both rural and urban regions. First, we selected Northern Hemisphere cities with areas larger than 100 km2, which ensured the sufficiently available areas for subsequent calculations. Second, we referred to the method developed by Zhang et al.59 to expand rural areas outward to the urban boundary to avoid a size imbalance between the rural and urban areas. The extended distance from the urban boundary was determined as follows:

$${D}_{{{\mathrm{ex}}}}=\left(\sqrt{2}-1\right)\sqrt{\frac{{A}_{{\mathrm{r}}}}{\uppi }},$$
(1)

where Ar is the urban areas and Dex is the distance of the extended rural areas. This function has an advantage over a fixed threshold by adopting different extended distances with different city sizes and can double the area in each city59. This step also prepared the divided buffers along rural–urban gradients (Supplementary Fig. 1). Third, we selected the city as the study city only if the area of valid phenology pixels within the Ar exceeded 50% of the city’s total area. In total, we selected 428 study cities in the Northern Hemisphere that met these conditions (Extended Data Fig. 1).

We used the annual maps of global artificial impervious area (GAIA)60 to calculate the impervious surface area index (β, the percentage of impervious surface area in one pixel) in each city. The GAIA dataset provides a global urbanized area at a high resolution of 30 m from 1985 to 2018 detected from the entire archive of Landsat images60. It was widely agreed that the GAIA maps had more than 90% accuracy and fit well with other products in temporal trends at local and global scales59. Hence, we aggregated the 34-year maps as one image at a resolution of 1 km to calculate β (ranging from 0 to 1) (Supplementary Fig. 15). Then, we defined the buffers from rural to urban areas within the extended city boundary as above.

Köppen–Geiger climate map

We used the Köppen–Geiger climate global map28 at 1-km resolution to examine the varying effects of ALAN and Ta across climate zones. The Köppen–Geiger climate classification is a widely utilized system for categorizing global climates on the basis of air temperature and precipitation. It has been used in a broad range of topics, such as climate change research, physical geography, hydrology, agriculture and biology28. The climatic type for each city was determined on the basis of the central points of their respective areas, and then climate groups with more than ten cities were selected. As a result, 7 climate zones with 427 cities were used in the analysis, including desert (BW) and steppe (BS) in arid regions, dry summer (Cs), dry winter (Cw) and no dry season (Cf) in temperate regions, and dry winter (Dw) and no dry season (Df) in cold regions.

Definition of buffers along rural–urban gradients

We proposed a framework to define ten gradually impervious surface buffers representing the intensity of urbanization in each selected city. This framework aimed to consider the urban spatial heterogeneity and precisely quantify the spatial pattern of phenology, ALAN, and Ta and their interactions under the development of urbanized degree from rural to urban areas (Fig. S1). Firstly, we calculated the impervious surface area index (β, ranging from 0 to 1) in pixels at 1-km resolution using GAIA with a high resolution of 30 m within the city boundary60. Next, we utilized the kernel density estimation to spatially smooth the β to include the effect of the surrounding environment on the pixel. Then, we aggregated these pixels into ten β categories with an interval of Δβ = 0.1 and calculated the average for each buffer in each city. The analysis of β distribution in 428 Northern Hemisphere (NH) cities in ten buffer zones provided exact proportion of impervious areas along rural–urban gradients (Supplementary Fig. 15). In this way, we obtained a more representative distribution of built-up areas and reduced the impact of peculiar environments25. The numbers from 1 to 10 indicated the ten buffer zones from rural to urban areas. We focused on the mean SOS, EOS, ALAN and Ta in each buffer and assessed the spatial pattern of vegetation phenology, ALAN and UHI across the rural–urban gradients. Data preparations were processed on the Google Earth Engine platform, and data (the mean vegetation phenology, ALAN and Ta over the ten buffers for each of 428 cities between 2014 and 2020) were exported for the following analysis in MATLAB 2022a.

Analysis of the ALAN and UHI impacts on phenology

Curve fitting and urban–rural difference calculation

To quantify the nonlinearly spatial pattern of vegetation phenology, ALAN and Ta with increasing impervious fractions from rural to urban areas in each city, we compared five algorithms and finally adopted three mathematical models according to the highest coefficients of determination (R2), the lowest bias and the lowest root mean square error. The quadratic polynomial model was selected in fitting both the SOS/EOS–β and Ta–β relationship, while the logistical growth curve model was selected for the ALAN–β relationship. The parameters in these chosen models represent underlying biophysical processes. The equations were as follows:

$${y}_{{{\mathrm{sos}}}/{{\mathrm{eos}}}}={a}_{1}{\beta }^{2}+{a}_{2}\beta +{a}_{3}$$
(2)
$${y}_{{{\mathrm{ALAN}}}}=\frac{{b}_{3}}{1+{b}_{1}{{\mathrm{e}}}^{{b}_{2}\beta }}$$
(3)
$${y}_{{T{\mathrm{a}}}}={c}_{1}{\beta }^{2}+{c}_{2}\beta +{c}_{3},$$
(4)

where ysos/eos, yALAN and yTa indicate the multiyear average of SOS and EOS dates, ALAN and Ta in each buffer; β is the impervious surface area index; and the fit parameters are ai (i = 1, 2 and 3) for SOS or EOS curves, bj (j = 1, 2 and 3) for ALAN curves and ck (k = 1, 2 and 3) for Ta curves. To minimize the uncertainties from remotely sensed datasets, we used the multiyear average of each index during 2014‒2020 in every impervious surface buffer and focused on the spatial distribution of indices from rural to urban areas in each city. The standard error was calculated by the sample set including the number of cities for the entire or continental groups in each buffer for 7 years. The relationship between phenological indices and β was quantified by a quadratic polynomial model for both SOS and EOS. The interception of the fit curve (parameter a3 in equation (3)) in each city was approximately the baseline phenology date in the first buffer (Fig. 1a). For example, the SOS observed from remotely sensed products was 94.4 (DOY) in the first buffer, while the a3 was 94.9 (DOY) (Fig. 1a). The logistical growth curve model was applied to the ALAN–β curves. Moreover, we found that the parameter b3 in the logistics model for ALAN was about twice the maximum of ALAN in the tenth buffer. For the Ta–β curve fitting, we applied the quadratic function model (equation (4)); the constant c3 indicates the minimum temperature in rural areas, and c1 and c2 are fit parameters in the models. We calculated the determination coefficient (R2) and P value to measure the reliability of these mathematical models in capturing the pattern of these indices in each city. These statistical results demonstrated the effectiveness of our models in fitting these index changes (Supplementary Tables 2–4).

Furthermore, we evaluated the differences in phenology, ALAN and Ta between urban and rural areas worldwide. The differences were calculated as follows:

$$\Delta I={I}_{{{\mathrm{urban}}}}-{I}_{{{\mathrm{rural}}}},$$
(5)

where ∆I indicates the urban–rural difference for the phenological indicators (SOS and EOS), ALAN and Ta, and Iurban and Irural represent values in urban and rural areas. In the present study, Iurban was calculated as the mean of the sixth to the tenth buffer while Irural was calculated as the mean of the first to the fifth buffer in each city. To ensure the robustness of our results, we also calculated the differences by testing the mean of the first–second (or first–third) buffers in rural areas and the ninth–tenth (or eighth–tenth) buffers in urban areas and obtained similar results. We reported the results using the mean differences of these indices between the first to the fifth and the sixth to the tenth buffers (Supplementary Fig. 10). Furthermore, we evaluated regional differences among three continents: Asia, Europe and North America.

Partial correlation analysis

To determine the contribution of ALAN and Ta to phenology change along the rural–urban gradients, we calculated the partial correlation coefficients of ALAN and Ta for SOS and EOS along ten buffers using all 7-year series in each city. We normalized the variables to avoid uncertainty caused by the magnitude of datasets. The partial correlation analysis calculated the coefficient of each factor while controlling for the other factor. This method effectively excluded the interactive effects of other variables in attribution analysis34,61 and allowed us to determine the significance of influential factors. Given that the diel variance of ALAN and Ta mainly affect vegetation physiological processes at night, we also performed parallel analysis using Tmax and Tmin and found consistent results. When determining the dominant factor between Ta and ALAN, we selected the variable with the greater absolute partial coefficient value in every city. For each city, we applied the Fisher Z transformation to access whether the effects of ALAN and Ta on phenology changes were statistically significantly different along rural–urban gradients, using a significant level of P < 0.05. In addition, we performed the spatial analysis to investigate the effects of ALAN and Ta on SOS and EOS based on latitude and Köppen–Geiger climate zones. We binned the partial coefficients with a 3° interval based on the latitude of each city and calculated the mean partial coefficient in each bin. Similarly, we calculated the mean of partial coefficients of cities grouped by Köppen–Geiger climate zones. In the group scale, the two-tailed Student’s t-test was then used to assess the significance of the differences in the effects of ALAN and Ta on phenology across all cities, as well as across climate zones or latitude intervals. All the analyses were processed on MATLAB 2022a.

Validation of spatial phenology pattern

Our study validated the spatial patterns of phenology shifts using two independent phenology datasets: the 30-m urban phenology product and the 500-m VIIRS dataset during the overlapping period of 2014‒2015 in 62 overlapping cities in the continental USA (Supplementary Fig. 16). We applied the framework to define buffer zones from rural to urban areas. The validation confirmed a strong agreement in the observed phenology shifts, particularly in advancing SOS and delaying EOS from rural to urban areas. To address the influence of land cover differences on the spatial pattern of phenology, we conducted two experiments using 30-m urban phenology product: (1) including developed open spaces and (2) excluding developed open spaces, which are a mixture of some constructed materials and vegetation, primarily in the form of lawn and grasses.

We found that 40 (64.5%) of 62 selected cities had developed open space area covering less than 10% of the total urban area (Supplementary Fig. 17). The overall spatial patterns of both SOS and EOS remained consistent in both experiments—both scenarios exhibited an advancement in spring phenology and a delay in autumn phenology along the rural-to-urban gradient (Extended Data Fig. 3). Furthermore, no significant relationship was observed between the phenology difference and the percentage of developed open space in 62 cities, indicating that the amount of developed open space is not the reason for the phenological difference between urban and rural areas (Supplementary Fig. 17). These findings suggested that the presence of developed open spaces does not alter the broader spatial pattern of urban phenology shifts. Finally, we conducted the partial correlation analysis of phenology shifts in response to ALAN and Ta using the 30-m phenology dataset. The assessment with this higher-resolution phenology dataset further validated our main conclusion that ALAN plays a more dominant role than Ta in extending the urban growing season, particularly by delaying EOS (Supplementary Figs. 12 and 13).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.