Satellite imagery reveals increasing volatility in human night-time activity

The illuminated Earth, viewed from space at night, is a powerful testament to human presence, revealing a ‘Black Marble’ increasingly delineated by the light of human settlements, industries and energy infrastructures. Artificial light at night (ALAN) extends visibility beyond daylight hours, enabling round-the-clock movement, gathering and continuation of daily life. Yet, ALAN is far more than a visual spectacle: it is a direct, measurable signal of human activity, reflecting how we build and power our settlements, the dynamics of our economies and our responses to both crises and opportunities⁶. The variability of ALAN mirrors the pace and nature of human activity, manifesting as either abrupt events, such as new constructions or disasters, or gradual trends driven by long-term economic or demographic forces. Understanding the direction, location and intensity of these changes is, therefore, important for assessing the full scope of global change and its impact on human infrastructure and energy transitions⁷. We define brightening and dimming as sustained increases and decreases in radiance (excluding transient noise), respectively, driven by either abrupt events or gradual trends, as shown in Extended Data Fig. 5.

For decades, global assessment of ALAN trends has centred on a narrative of continuous and widespread brightening, with dimming viewed as a localized exception. This perspective stems largely from temporally aggregated night-time light (NTL) satellite observations^8,9. Annual and multi-year composites, such as those produced from earlier Defense Meteorological Satellite Program Operational Linescan System (DMSP-OLS)¹⁰ and later from Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB) data^11,12, have been instrumental in mapping long-term ALAN trends², documenting the expansion of urban extent¹³ and estimating the light pollution effects¹. Monthly VIIRS DNB products subsequently improved temporal granularity, enabling analyses of intra-annual variability¹⁴, such as seasonal impacts¹⁵ and epidemiological dynamics¹⁶. However, despite these strengths, temporal aggregation inherently masks short-term fluctuations and the bidirectional changes arising from the coexistence of brightening and dimming.

In reality, human activities and the resulting changes in ALAN are not uniform, linear or steady processes. They operate across a spectrum of timescales, ranging from gradual shifts such as suburban expansions or LED adoptions to discrete, abrupt events such as industrial constructions, changes in municipal lighting codes, power outages triggered by disasters and the disruptions of infrastructure or destructions of buildings due to armed conflicts^17–20. Focusing only on the net change derived from temporal composites masks the specific timing and impacts of these incidents. This oversight limits our ability to truly link cause and effect, assess policy effectiveness with precision, and understand the full ecological implications of a rapidly changing nocturnal environment.

More recently, advances in NASA’s daily Atmospheric- and Lunar-BRDF (bidirectional reflectance distribution function)-corrected Black Marble NTL product³ have provided an unprecedented opportunity to quantitatively characterize the finer daily night-time light dynamics of Earth. Unlike earlier nightlight products that primarily served visualization or coarse trend analysis, Black Marble applies comprehensive corrections for atmospheric conditions, terrain and lunar illumination to improve radiometric stability, and at the same time provides information on the contaminated pixels, such as clouds and snow²¹. This enables consistent detection of subtle yet meaningful changes that reflect real-world dynamics, making it a trusted source of information for stakeholders. However, this abundance of daily data comes with its analytical hurdles. The signal is subject to substantial noise from a variety of sources, including atmospheric interference, variations in sensor viewing and local geometry, and ephemeral conditions such as snow cover²². The magnitude of this combined noise is highly variable and can often exceed the subtle, real-world changes we aim to detect, making most current algorithms ineffective at separating the true signal from the noise.

This study presents the first comprehensive global analysis of ALAN change dynamics derived from daily Black Marble NTL data. By adapting a continuous change detection algorithm^4,5 (Methods), we quantified the timing (day-of-year and year), intensity (area-averaged radiance change), type (abrupt or gradual), and direction (brightening or dimming) of ALAN changes for every 15-arc-second pixel (about 500 m at the equator) across the primary inhabited landmasses of Earth (70° N–60° S) from 2014 to 2022. By tracking each distinct change event, our analysis captures the full trajectory of development and decline over time, explicitly accounting for pixels experiencing multiple changes.

We analysed 1.16 million daily NASA Black Marble NTL images constrained to the analysed NTL areas (15.16 million km², 10% of global land; Supplementary Fig. 3 and Supplementary Information Section 1), excluding persistently dark regions and areas with ephemeral natural light events that did not meet our persistence criteria (Methods). Validation against independent stratified random sample units (n = 2,071 for abrupt changes and n = 1,902 for gradual changes) confirmed global reliability (Fig. 1 and Extended Data Table 1). To quantify these dynamics, we analysed the area of change, the radiance change (overall shift in light output) and the change intensity. Unless specified, all regional and global values were derived from the native 15-arc-second grid results projected to a ‘500-m’ nominal resolution equal-area sinusoidal grid.

Our data reveal a dynamic night-time environment in which changes are frequent rather than sporadic. Over 2014–2022, the ever-changed area, defined as the total area experiencing at least one ALAN change, was substantial (3.51 million km²; Supplementary Table 1). Globally, although nearly half (51%) of these altered areas saw only gradual changes, more than one-third (35%) experienced both abrupt change events and longer-term gradual changes, with the remainder (14%) marked solely by abrupt changes (Extended Data Table 2). Further analysis at the country or territory level reveals diverse national profiles in the typical frequency of abrupt and gradual changes (Supplementary Fig. 1).

More revealing of the true dynamics is the total cumulative ALAN change area (or gross area change) during 2014–2022 (Extended Data Table 3), which sums the area changed each year, thereby accounting for pixels that underwent several changes (for example, a brightening followed by a dimming, or multiple abrupt events). Unbiased area estimates indicate that between 2014 and 2022, 2.05 (95% confidence interval (CI) [1.79, 2.32]) million km² underwent abrupt ALAN changes, and 19.04 (95% CI [18.10, 19.98]) million km² experienced gradual changes. Asia, particularly China and India, accounted for the largest cumulative area of ALAN change. This cumulative change area over the 9 years is 5.5 times the global lit area of the 2014 baseline, with an average of 6.6 changes per ever-changed area, indicating that each parcel of lit land experienced several significant ALAN alterations during the study period (Fig. 2a,b, pie charts).

The high frequency of abrupt change underscores the transient nature of ALAN signals across the globe, with more than 20% of affected areas experiencing abrupt changes more than once (Fig. 2a). The spatial pattern, observed in many parts of the world, reflects ongoing cycles of construction and demolition (for example, in China²³ and India²⁴), energy instability (for example, grid failures in Venezuela²⁵ and energy crises in Lebanon²⁶), fossil fuel operations (for example, gas flaring in Texas, USA²⁷) or societal disruptions (for example, conflicts in the Middle East¹⁷) (Fig. 2a). By contrast, gradual ALAN changes reveal long-term trends, with 94% of affected areas experiencing changes that persisted for more than 1 year, particularly developed regions such as the USA and Europe (Fig. 2b). These gradual shifts are often tied to slow demographic-economic dynamics, dedicated dark-sky conservation efforts (for example, the UNESCO Starlight Reserves), or systematic infrastructure upgrades (for example, LED replacement programs of Europe²⁸). The only places in which lighting remains mostly unchanged are uninhabited regions and regions with limited development, such as protected natural reserves and remote deserts (Fig. 2b).

Importantly, those changes are shifting in both directions, with brightening accounting for 65% of abrupt and 71% of gradual changes, whereas dimming accounts for 35% of abrupt and 29% of gradual changes (Fig. 2c). Using the Stokes and Seto framework²⁹ (Extended Data Table 4), we found more than half of the abrupt brightening was driven by non-residential development and electrification, highlighting the role of infrastructure expansion and rural electrification in shaping short-term ALAN spikes. By contrast, abrupt dimming was mainly attributed to reductions in gas flaring (46%), driven by government regulations, gas infrastructure upgrades and operational volatility³⁰. Gradual changes followed a similar brightening–dimming ratio but with distinct leading drivers. Gradual brightening was often tied to concurrent change, such as steady urban expansion, whereas gradual dimming was primarily caused by de-electrification, reflecting long-term declines in lighting infrastructure or energy access. Moreover, the global bidirectional patterns indicate that frequent brightening regions were primarily driven by concurrent change and non-residential development, typically reflecting rapid urbanization and development (for example, Eastern and South Asia³¹), or emerging economies expanding energy access (for example, West Africa³²) (Extended Data Fig. 1a). Conversely, regions with frequent dimming reflect energy conservation initiatives in high-income countries/territories (for example, Eastern USA, Western Europe³³) or systemic instability, such as the rolling blackouts caused by load shedding in South Africa³⁴ (Extended Data Fig. 1b).

The long-term global trajectory points towards a brighter planet, evidenced by a net 16% increase in total ALAN radiance from 2014 to 2022 (Fig. 3e), outpacing global population growth³⁵. Specifically, brightening contributed a radiance increase equivalent to 34% of the 2014 baseline, while dimming offset this by 18%. Yet, this aggregate figure conceals a widespread coexistence of brightening and dimming (Fig. 3). Mapping the spatial distribution of ALAN change area and change intensity between 2014 and 2022 reveals substantial regional heterogeneity (Extended Data Fig. 2 and Supplementary Fig. 2). Broad patterns of brightening are evident across the world (Supplementary Table 3), led by Asia, and reflecting continued urbanization, industrial expansion and rural electrification. Although China and India show the largest national-level increases (Fig. 3a, L1), our data show strong regional contrasts. In China, brightening is concentrated in the eastern and central regions, driven by urbanization and industrial activity, whereas western areas show lower levels of change and more spatially fragmented patterns. In India, southern regions experienced sustained brightening throughout the study period, reflecting higher levels of urbanization and economic development, whereas northern regions exhibited brightening primarily in the early years (Fig. 1a,b), driven by national rural electrification and street lighting programmes that expanded power access and lighting infrastructure. Much of Sub-Saharan Africa also shows a strong brightening signal, reflecting development that illuminates previously unlit regions³⁶.

Substantial dimming patterns are also observed in several regions (Fig. 3a). Europe presents a particularly clear and structured dimming pattern, with a 4% net decrease in ALAN radiance relative to its 2014 baseline. Notably, dimming alone accounts for an 18% reduction, and these changes align closely with national borders, highlighting the impact of country-specific lighting regulations. Extensive areas experienced these reductions, most notable in France (33% net drop), the United Kingdom (22%), and the Netherlands (21%) (Fig. 3a, L2, and Supplementary Fig. 2). This reflects a combination of widespread technological shifts from older, less efficient lighting to newer LED systems, measures to reduce light pollution and energy use, and broader national and EU-level energy efficiency mandates^18,37,38.

By contrast, the dimming observed in Venezuela (Fig. 3a, L3) is not driven by regulation or technology but stems from systemic collapse. Here, ALAN radiance declined by more than 26% relative to its 2014 baseline, reflecting economic downturns, widespread infrastructure decay and lack of investment³⁹. These cases highlight how reductions in night-time illumination can arise not only from deliberate energy-saving or pollution-mitigation strategies but also from infrastructural breakdown and economic instability.

The USA offers a microcosm of this complexity (Fig. 3a, L4). The West Coast brightened with ongoing population growth and vibrant economies in its main urban centres⁴⁰, whereas the East Coast and parts of the Midwest dimmed because of de-densification in some older urban cores⁴¹, the decline of certain manufacturing sectors and the adoption of energy-efficient lighting technologies. Central USA regions, particularly areas overlying the main oil and gas shale basins such as the Permian (Texas) and Bakken (North Dakota), exhibit highly volatile ALAN signals. These regions exhibit strong bidirectional changes: intense brightening during boom periods due to drilling activity and gas flaring, followed by equally sharp dimming as operations scale down or shift location²⁷. This reflects oil and gas extraction cycles, in which the timescale of lighting changes is often dictated by the specific methods and phasing of extraction activities, rather than broader oil price fluctuations alone. Similar dynamic signals are observed in oil-producing regions across the Middle East.

The ALAN change intensity quantifies the average strength of those changes, whereas change area captures how broadly illumination shifts spread (Extended Data Fig. 2). This distinction differentiates widespread but subtle changes from more localized but intense transformations. Globally, a complex picture emerges when examining these intensities relative to the 2014 baseline (Supplementary Table 4). Areas experiencing brightening saw their average intensity increase by 9%, whereas those experiencing dimming saw a decrease of 10%.

Regionally (Extended Data Fig. 2f–j), Asia and Africa exhibit the most intense bidirectional change intensities (brightening: 10–11% relative to their initial radiances; dimming: −11% to −15%), underscoring them as hotspots of strong development and retreat. By contrast, the USA showed widespread but moderate change intensities (brightening: 8% of initial radiance; dimming: −6%). Apart from high-intensity oil extraction regions, shifts elsewhere reflect incremental, low-density suburban sprawl⁴². Unlike high-density redevelopment, this horizontal expansion is additive, distributing artificial light across vast peri-urban areas without requiring the disruption of existing infrastructure, resulting in widespread but gradual trends⁴³. India presents a similar profile of widespread cumulative change area accompanied by moderate change intensities. This contrasts sharply with China, where high-intensity signals are consistent with a strategy of vertical urbanization and high-density land conversion⁴⁴. This specific evolution of land use creates a unique NTL signature with periods of dimming (demolition) followed by explosive brightening (vertical reconstruction). Consequently, higher urban density acts as a multiplier for NTL volatility, whereas lower-density sprawl acts as a dampener.

Temporal trends in radiance changes (2014–2022) show a critical evolution of global ALAN dynamics (Fig. 4). Although the global total net radiance change trended upward (P < 0.05), this conceals an intensification (P < 0.05) of both brightening and dimming over years (Fig. 4a), indicating that the global nightscape is becoming more dynamic and volatile overall. Countries and territories undergoing net brightening are often not doing so in a uniform, one-sided way (Supplementary Table 5). In most cases, both brightening and dimming radiance changes are increasing simultaneously (Fig. 4b). This bidirectional dynamism is pronounced in East Asia and Africa, where expanding light footprints coexist with intensifying pockets of dimming. By contrast, the USA and Australia show significant (P < 0.05) intensification only in brightening. Meanwhile, most of Europe, South Africa and New Zealand exhibit intensifying bidirectional trends despite their net dimming trajectories.

This intensification of bidirectional change is further corroborated by disaggregated global trends in change area and change intensity. The global annual area experiencing dimming grew significantly (expanding by 12,875 km² yr⁻¹, P < 0.05) (Extended Data Fig. 3a, right), driven by gradual shifts in Africa, Asia, Europe and South America (Extended Data Fig. 3b and Extended Data Fig. 4c). Conversely, the brightening area showed a slight, non-significant decrease (Extended Data Fig. 3a, middle). In terms of change intensity, the global brightening trend increased significantly (increasing by 0.04 nW cm⁻² sr⁻¹ yr⁻¹, P < 0.05) (Extended Data Fig. 3c, middle). Meanwhile, the intensity of dimming events, led by abrupt events (Extended Data Fig. 4b), also showed a negative trend of −0.03 nW cm⁻² sr⁻¹ yr⁻¹ (Extended Data Fig. 3c, right).

Daily NTL observations offer an unprecedented window into societal dynamics, capturing short-term events often lost in coarser temporal composites (Fig. 5). This high-frequency perspective allows us to monitor the dynamics of night—the rapid and often abrupt fluctuations of human activity and societal responses as reflected in ALAN. Although annual and monthly data track broad long-term trends in energy use or ecological impact⁸, our daily-derived ALAN changes show crucial short-term, event-driven dynamics that shape those net long-term outcomes. By resolving the precise timing of these changes, we can move beyond broad correlations to link shifts in illumination directly to specific real-world shocks (for example, military actions, lockdowns or policy implementations), while capturing the full magnitude of transient events (for example, Fig. 1h) that would otherwise be smoothed out in monthly averages.

This granular analysis exposes societal dynamics and volatility that are often masked by net trends. A central insight from this study is that regions with comparable annual outcomes can behave very differently at finer timescales. For instance, areas with high fluctuation often signal instability (for example, armed conflicts in Syria), reflecting energy insecurity, economic precarity or sporadic conflict, even if their net trend appears stable. Conversely, low fluctuation suggests steady, uninterrupted development or intentional dimming (for example, Western Europe). Acting as a societal electrocardiogram, these dynamics are important for distinguishing resilient systems and those under acute stress.

Continent-level analysis shows a notable increase in ALAN volatility after 2020, evident in both brightening and dimming events, with dimming showing the steepest downturn (Fig. 5a). This growing volatility reflects a convergence of main global and regional disruptions, including COVID-19 lockdowns, accelerated LED transitions, light pollution policies and the energy crisis triggered by the Russia–Ukraine war. A prominent example is the global dip in ALAN radiance, primarily driven by the abrupt dimming events in Asia, which experienced the earliest and most extensive lockdowns during COVID-19 (ref. ⁴⁵) (Fig. 5b). This sudden decrease, visible with varying degrees across multiple continents in early 2020, aligns precisely with widespread lockdowns and curtailments of economic and social activity during the initial wave of the pandemic (Fig. 5d). Another compelling signal emerges in Europe during 2022 (Fig. 5c), in which a sharp and sustained decrease in ALAN radiance change is observed, contrasting with trends in most other continents during the same period. The timing aligns with the Russia–Ukraine conflict and subsequent European energy crisis, leading to widespread energy-saving measures across many European nations, notably France and Belgium⁴⁶ (Fig. 5e).

This global, high-resolution analysis of ALAN dynamics, derived from daily NASA Black Marble satellite observations, refines and expands our understanding of how humanity is altering the night environment³. Our findings show that the human light footprint is not a universally expanding entity but a dynamic system, characterized by the pervasive coexistence of brightening and dimming. The concurrent rising trends in the change area of dimming, the radiance change and the change intensity point to an intensification of this bidirectional dynamic. This signals an acceleration of the processes that modify the night-time environment, driven by increasingly potent and often counteracting forces, rendering the nocturnal world more volatile.

Consequently, our findings demand a re-evaluation of ALAN as a socioeconomic proxy. The inherent complexity and bidirectional nature of ALAN change are especially evident in regions undergoing rapid technological transitions, strong policy interventions or economic instability. Thus, simple correlations between net ALAN radiance change and indicators such as the gross domestic product may be misleading. These events act as exogenous shocks to the socio-technical system, creating dynamic, time-lagged interdependencies that require robust empirical modeling⁴⁷. Methodologies such as the simultaneous analysis of multivariate time series are essential here. Specifically, advanced econometric tools analysing impulse response functions offer the necessary framework to disentangle these complex interactions⁴⁸.

Moreover, our daily-derived, disaggregated dataset enables more sophisticated and accurate modelling. Understanding socioeconomic responses to ALAN reductions or fluctuations, not just increases, is an important, largely unexplored frontier, which requires such spatially and temporally explicit information. Furthermore, the demonstrated sensitivity to events such as pandemics and conflicts highlights the potential for near-real-time monitoring of societal disruptions, disaster impacts and infrastructure resilience. This ability, potentially extending to early warnings for economic cycles⁴⁹, positions daily ALAN analysis as a vital tool for informing policy, humanitarian aid and ecological response⁵⁰, ultimately strengthening societal resilience and guide resource allocation.

Finally, it is crucial to recognize that current operational satellite NTL observations, such as those from VIIRS DNB, primarily capture post-midnight light emitted upwards from the surface, whereas ground-based monitors similar to those used for the World Atlas measure downward skyglow¹. These represent distinct physical quantities that are not directly comparable but provide complementary information regarding the night environment. Moreover, VIIRS DNB is most sensitive to a specific spectral range (around 500–900 nm), leaving a blind spot for trends occurring earlier in the evening or in different spectral bands^1,6. Yet, beyond these observational nuances, the overarching signal is unmistakable: the Black Marble of Earth is not merely growing brighter; it is pulsing with intensifying volatility, echoing the amplifying heartbeat of human activity.

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/s41586-026-10260-w.

This work was supported by the Terra, Aqua, Suomi-NPP and NOAA-20 programs of NASA (grant 80NSSC22K0199) and the Remote Sensing Theory for Earth Science program of NASA (grant 80NSSC20K1748). C.C.M.K. was supported by the New Earth Observation Mission Ideas program of ESA (contract 4000139244/22/NL) and by the DFG (grant 545098235). Y.Y. was partially supported by the NASA ACRES project (grant 80NSSC23M0034). The computational work for this project was conducted using resources provided by the Storrs High-Performance Computing (HPC) cluster. We extend our gratitude to the UConn Storrs HPC and its team for their resources and support, which aided in achieving these results. We acknowledge the use of AI-based language tools to improve the clarity and readability of the manuscript, and all content was reviewed and approved by the authors. Any use of trade, firm or product names is for descriptive purposes only and does not imply endorsement by the US government.

T.L., Z.Z. and Z.W. conceptualized the study; T.L., Z.Z. and Z.W. devised the methodology; T.L. was involved in production; T.L., S.Q., M.C., F.H., A.G., K. Song, J.W.S. and X.Y. performed the product validation; T.L., Z.Z., Z.W., C.C.M.K., M.O.R. and K.C. Seto conducted the formal analysis; T.L. and Z.Z. wrote the original draft; T.L., Z.Z., Z.W., C.C.M.K., M.O.R., K.C. Seto, Y.Y., S.Q., T.K., M.F., X.C., T.H.M., C.D.R., X.T., M.C., F.H., A.G., K. Song, J.W.S., X.Y., V.L.K. and C.D. reviewed and edited the paper; Z.Z. and Z.W. helped with funding acquisition.

The open-source data include the following: the VIIRS/NPP Gap-Filled Lunar BRDF-Adjusted Nighttime Lights Daily L3 Global 500 m Linear Latitude/Longitude Grid (VNP46A2) at https://ladsweb.modaps.eosdis.nasa.gov↗, the NASA Black Marble Collection 1 VIIRS/NPP Daily Gridded Day Night Band 500 m Linear Lat Lon Grid Night (VNP46A1) at https://ladsweb.modaps.eosdis.nasa.gov↗, the 2014 MODIS/Terra Land Water Mask derived from MODIS and SRTM L3 Global 250 m SIN Grid Collection 6.1 (MOD44W) at https://www.earthdata.nasa.gov↗, the World Bank-approved administrative boundaries at https://datacatalog.worldbank.org↗. The change dataset generated in this study is publicly available at 10.5281/zenodo.18264642.

The global ALAN change dataset and analyses were produced with custom code using MATLAB 2022b and Python 3.10, which are available at Zenodo⁶¹ (10.5281/zenodo.18264642).

The authors declare no competing interests.

Tian Li, Email: tianli@uconn.edu.

Zhe Zhu, Email: zhe@uconn.edu.

is available for this paper at 10.1038/s41586-026-10260-w.

The online version contains supplementary material available at 10.1038/s41586-026-10260-w.