Artificial light at night widely affects ecosystem energy use

Artificial light pollution is accelerating across the globe^1,2 and has widespread consequences for people^3,4 and the planet^5–7. Shifts in the luminance and spectral composition of the nocturnal environment modify the physiology, behaviour and ecological interactions of organisms^7–11, which together play a fundamental role in ecosystem metabolism^12,13. Ecosystem metabolism, comprising gross primary production (GPP) and ecosystem respiration (R_e), directs the magnitude and direction of carbon–climate feedbacks via net ecosystem exchange (NEE)¹⁴. Around one quarter of global terrestrial ecosystems are exposed to artificial light at night (ALAN)¹⁵, but the effects on ecosystem metabolism are currently unknown.

Changing daily and seasonal cycles of light and dark¹⁰ could decouple the timing of biological processes across trophic networks¹⁶. Trophic groups are also exposed to ALAN at different intensities and have varying sensitivities to luminance and spectral composition¹⁷. Plant responses to photoperiod are influenced even at low ALAN intensities^18,19, and longer-term exposure influences seasonal phenology, growth form, resource allocation and, thus, potentially carbon fixation²⁰. High ALAN intensity exposure in urban areas disrupts the behavioural patterns of nocturnally migrating birds²¹ and plant diversity²² and restructures soil microbial communities, reducing the functional genes involved in nutrient regulation and plant health²³. Together, the observed effects of ALAN across levels of biological organization and diverse taxa suggest a potential cascading impact on ecosystem structure and function. Previous studies of ALAN effects, however, have focused on local or experimental manipulations, leaving uncertainty about whether ALAN effects persist at the ecosystem level and longer timescales.

GPP and R_e are fundamentally constrained by shortwave (solar) radiation (SW) and temperature (T), respectively^24–26. That is, SW determines the direction and duration of energy flow between the atmosphere and ecosystems, and T determines the rate of reactions¹². Although ALAN is not expected to influence SW or T directly, artificial light could disrupt the processing of energy according to these fundamental constraints via acclimation, compensation and adaptation strategies^27,28. A better understanding of the magnitude and direction of ALAN effects on ecosystem metabolism could help constrain carbon–climate processes in Earth system models (ESMs)²⁹. Specifically, largely uncertain ESM processes and their response to climatic factors could be compounded by the chronic effects of pervasive anthropogenic stressors, such as ALAN.

To detect the potential influence of ALAN on ecosystem metabolism, we investigate half-hourly and mean daily ecosystem carbon fluxes (F_c; F_c: NEE, GPP and R_e) measurements against their fundamental constraints, T and SW, according to the modified Arrhenius equation of Weyhenmeyer¹²:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{{\rm{c}}}={T}^{4}\sigma {{\rm{e}}}^{\frac{-\mathrm{SW}}{\sigma {T}^{4}}}{k}_{{F}_{{\rm{c}}}}-b,$$\end{document}Fc=T4σe−SWσT4kFc−b,where F_c is ecosystem C flux (NEE, GPP, R_e) (in µmol CO² m⁻² s⁻¹), T is temperature in Kelvin, σ is the Stefan–Boltzmann constant (in J m⁻² s⁻¹ K⁻⁴) (5.67 × 10⁻⁸), SW is incoming SW (in J m⁻² s⁻¹,) k_Fc is the slope of the linear relationship and b is the intercept. The function establishes a biophysically grounded baseline for different F_c by capturing their shared fundamental constraints (T and SW). The use of the modified Arrhenius function in this study primarily serves as a comparative baseline rather than a mechanistic model, enabling deviations attributable to chronic ALAN effects to be identified relative to fundamental energetic constraints.

The variance weighting substantially reduced residual heteroscedasticity across all F_c, with the scale estimate reduced by ~95% and adjusted R² (R²_adj) reduced by 0.06–0.11 in weighted compared with unweighted final GAMMs, indicating improved model stability through decreased overfitting to high-variance observations (Supplementary Table 4 and Extended Data Fig. 2). Figure 3, right panels, shows weighted GAMM estimates of relative changes in each F_c across gradients of ALAN intensity and ND. Partial effect surfaces illustrating nonlinear ALAN × ND interactions at half-hourly timescales are presented in Extended Data Fig. 3, along with residual diagnostics indicating no substantial autocorrelation after model fitting.

To ensure our data analysis was robust to site bias across ALAN intensities (17, 22 and 47 sites at high, medium and low ALAN intensities, respectively, and 34 sites in North America and 52 sites in Europe), we repeated all GAMM and SEM analyses using a balanced dataset with an equal representation of low, medium and high ALAN intensity sites per continent (Extended Data Figs. 6–9). The models fitted to the balanced dataset showed consistently significant nonlinear interactions between ALAN and ND across temporal scales (Extended Data Fig. 6), and the annual SEM retained the core structure of ALAN indirectly influencing NEE through increased R_e (Extended Data Fig. 9). Whereas several weaker interactions (for example R_e ~ T and NEE ~ GS length) were no longer significant owing to reduced sample size, the SEM retained dominant pathways, and the overall explanatory power was comparable (R²_m = 0.64, R²_c = 0.70 for NEE). Notably, the standardized coefficients strengthened between ALAN ~ R_e and R_e ~ NEE when the SEM was fitted to the balanced dataset. Our observed ALAN effects on ecosystem metabolism are therefore robust to spatial imbalances in ALAN intensity across FLUXNET2015 site distribution.

Our study provides cross-continental evidence of ALAN's influence on ecosystem metabolism across timescales. We demonstrate that ALAN consistently modifies the relationship between F_c and their fundamental energetic constraints (Figs. 2–4). The R_e response to fundamental constraints was particularly sensitive to ALAN intensity at short (half-hourly and daily) timescales (Fig. 2 and Supplementary Tables 2 and 6). Alongside R_e, GPP and NEE exhibited significant nonlinear interactions between ND and ALAN intensity, revealing the importance of ALAN magnitude and timing in modulating ecosystem metabolism across scales (Figs. 3 and 4). At annual timescales, the influence of ALAN on NEE was primarily mediated through increased R_e rather than the direct suppression of GPP (Fig. 5). Taken together, our findings demonstrate the role of ALAN as a pervasive stressor capable of disrupting carbon balance across spatial and temporal scales.

The nonlinear influence of ALAN on ecosystem metabolism was strongly modulated by diel cycles and seasonality, demonstrating the importance of phenological dynamics³⁴ and biogeochemical feedbacks in shaping long-term carbon balance³⁵. The temporal aggregation led to notable shifts in the strength and direction of ALAN effects on R_e, whereas GPP and NEE displayed more consistent nonlinear responses to ND across timescales (Figs. 3 and 4). At the half-hourly resolution, short nights showed the strongest ALAN-induced increases in R_e (Fig. 3h,i), reflecting immediate physiological and microbial responses such as prolonged stomatal opening³⁶, sustained leaf dark respiration³⁷ and elevated microbial decomposition under disrupted circadian regulation³⁸. By contrast, daily mean nighttime R_e estimates indicated larger ALAN-related increases during longer nights (Fig. 4f), demonstrating how aggregation dampens short-term variability while revealing broader shifts in R_e across longer nights. GPP exhibited consistent positive or nonlinear ALAN effects across timescales (Figs. 3 and 4), probably driven by nocturnal illumination extending photosynthetic activity at medium ALAN intensities^39,40. Temporal scale and ND thus collectively shape ALAN's ecological impact, whereas diel averaging can obscure short-lived physiological responses while reflecting cumulative nighttime effects.

R_e exhibited greater sensitivity to ALAN than GPP across timescales (Figs. 2 and 3 and Supplementary Tables 2–11), and the SEM confirmed that ALAN primarily influences NEE indirectly via increased R_e at annual timescales (Fig. 5). The destabilizing effect of ALAN on production-respiration coupling will arise from shifts in multiple autotrophic and heterotrophic processes controlling carbon allocation and use efficiency^17,38. The greater R_e sensitivity may reflect a higher capacity of autotrophs to acclimate to ALAN through conservative growth strategies such as increased shoot-to-root ratios^20,41. In ecosystems dominated by C₃ plants, for instance, prolonged ALAN exposure can disrupt circadian regulation and prolong stomatal opening, reducing carbon uptake efficiency, increasing mortality and senescence, and leading to reduced GPP over time^42,43. Such trophic mismatches and shifts in carbon allocation are likely to accumulate across levels of biological organization, space and time^44,45, leading to progressive declines in NEE in illuminated ecosystems.

The ecological impacts of ALAN have primarily been examined at local scales^6,17, but landscape-scale factors will confound or amplify these localized effects². Urban proximity influenced GPP in our analysis, whereas ALAN directly influenced R_e (Fig. 5), suggesting distinct pathways through which nighttime lighting and urban characteristics modify ecosystem metabolism. Balancing sites across low, medium and high ALAN intensities further indicates potentially stronger mediating effects of ALAN on NEE via R_e (Extended Data Fig. 9). Despite the pervasive nature of light pollution, ALAN remains overlooked in ESM carbon–climate projections that otherwise account for climate and land use changes. Current observational data, however, do not enable the disentangling of the contribution of ALAN relative to sunlight in shaping F_c, and future targeted experimental studies will be needed to resolve these relationships.

Global eddy covariance networks such as FLUXNET are vital for monitoring ecosystem metabolism across diverse climates and land use types, but they are typically biased towards temperate regions, seminatural landscapes and dark skies (Extended Data Fig. 1). Urban flux towers are particularly scarce, and although networks such as Urban PLUMBER have been established, they do not measure F_c (ref. ⁴⁶). Similarly, available nighttime light satellite products used here (Visible Infrared Imaging Radiometer Suite (VIIRS) and Defence Meteorological Satellite Programme (DMSP)) are coarse in spatial resolution, are insensitive to blue light emitted by white light-emitting diode (LED) lighting⁴⁷ and cannot fully capture local heterogeneity in ALAN exposure at flux tower sites.

Enhanced satellite sensors with improved spectral and spatial resolution would advance ALAN monitoring², but ground-based measurements are also needed to capture how cloud cover exacerbates or reduces skyglow (brightening of the night sky) in high or low ALAN intensity areas, respectively⁴⁸. The coordinated expansion of eddy covariance flux tower networks along with complementary measurements, such as chamber-based respiration estimates and isotopic tracers, will be critical to disentangle the mechanisms by which ALAN alters ecosystem metabolism. Expanding ecosystem-level F_c measurements into urbanized, tropical, arid and high-latitude regions is vital to evaluate the global relevance of ALAN impacts on carbon cycling. While monitoring is essential, mitigation is also readily achievable.

Artificial light is ubiquitous and often beneficial, but the negative ecological effects of light pollution can be reduced while balancing societal benefits. Retrofitting LED lighting can reduce light pollution⁵, but it often results in over-illumination due to their higher efficiency⁴⁹. Given that lighting accounts for 20% of global electricity consumption and 6% of CO₂ emissions⁵⁰ and can exacerbate degraded air quality^51,52, mitigation interventions such as directional, dimmable and adaptive lighting designs² offer wider cobenefits. Unlike climate and land use change, the effects of light pollution could be mitigated overnight⁵³. Our study demonstrates the pervasive influence of light pollution on ecosystem metabolism across scales and highlights the urgent need to integrate ALAN into global change research, assessments of carbon–climate feedbacks and mitigation strategies. Developing a higher resolution understanding of species, community and ecosystem sensitivity to ALAN will be central to designing interventions that both safeguard biodiversity and preserve the land carbon sink.

We combined satellite-derived ALAN intensity metrics with eddy covariance flux measurements from 86 FLUXNET2015 sites in North America and Europe. The analyses were conducted at half-hourly, daily and annual timescales to capture short-term physiological responses, aggregated diel patterns and long-term ecosystem dynamics. All F_c (NEE, GPP and R_e) were first evaluated against their fundamental energetic constraints of T and SW according to the modified Arrhenius function (equation (1)) using LMMs. To explore nonlinear interactions, we then applied GAMMs which allow the flexible estimation of smooth terms. At annual timescales, we used piecewise SEM to account for collinearity among multiple drivers and to partition direct and indirect effects of ALAN, GS length and climatic variables on F_c. This hierarchical modelling framework enabled the consistent evaluation of ALAN effects across temporal scales.

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41558-025-02481-0.

Supplementary Information Supplementary Tables 1–11. Reporting summary