Introduction

Greenhouse gas emissions from human activities contribute to global warming (Guan et al. 2018; Han et al. 2024; Li et al. 2022). Greenhouse gases listed by the Kyoto Protocol include carbon dioxide (CO2), methane (CH4), nitrous oxide (N2O), hydrofluorocarbons (HFCs), perfluorocarbons (PFCs), and sulfur hexafluoride (SF6) (Telesetsky, 1999). CO2 is an important indicator in global climate change simulation research and international climate policy (Valjarević et al. 2022). The natural carbon cycle maintains a relative balance in carbon revenue and expenditure, but human activities (such as burning fossil fuels and land use changes) increase CO2 emissions, upsetting this equilibrium and influencing climate feedback mechanisms (Falkowski et al. 2000). According to measurements by the NOAA Global Monitoring Laboratory, CO2 concentrations has risen from 336.85 ppm in 1979 to 422.77 ppm in 2024 (Lan et al. 2025). Nowadays, anthropogenic CO2 is acknowledged as one of the primary sources of climate change (Hu et al. 2024; Rahman and Kashem, 2017). It is crucial to create national inventories to track anthropogenic CO2 emissions in order to lessen the adverse effects of climate change (Shi et al. 2021). Under the Paris Agreement, Parties are required to determine and report their anthropogenic CO2 emissions at the national level (Klein, 2017). Understanding the global carbon cycle and guaranteeing the effective execution of the United Nations Framework Convention on Climate Change (UNFCCC) depend on monitoring the spatiotemporal changes in anthropogenic CO2 emissions (Hegglin et al. 2022; Li et al. 2022). This is also crucial for informing international climate change policy initiatives for sustainable development (Figueres et al. 2018; Hua et al. 2023).

National inventories focus on detailed information on CO2 emissions from various human activities in peacetime (IPCC, 2006). In recent years, geopolitical conflicts have occurred frequently, and sociopolitical tensions have been rising in many parts of the world (Mortoja and Yigitcanlar, 2022; Söder, 2023). In this context, anthropogenic CO2 and other greenhouse gas emissions are highly sensitive to abrupt disruptions in human activities. Geopolitical conflicts, for example, often result in the mass displacement of populations from urban and industrial centers, leading to substantial reductions or redistributions of emissions associated with residential activities and transportation (Gao et al. 2021). Simultaneously, conflicts inflict severe damage on power plants, industrial facilities, and fuel supply chains, causing profound disturbances in energy production and consumption patterns, which in turn significantly alter greenhouse gas emissions (Sasmoko et al. 2023). Nevertheless, monitoring and quantifying these emission changes remain a considerable challenge. Geopolitical conflicts often cause countries and regions to fall into political, economic, and social chaos, and government resources and attention are forced to turn to emergency and security affairs, thus weakening the ability to monitor and collect CO2 emission data (Bun et al. 2023). In addition, during conflicts, infrastructure may be damaged, information transmission is interrupted or data management systems are paralyzed, resulting in the inability to determine and report anthropogenic CO2 emissions (Bun et al. 2024). The lack of timely and reliable data makes it difficult to formulate and implement effective emission reduction strategies, posing serious challenges to fulfilling the emission reduction obligations under the Paris Agreement. Near-real-time monitoring techniques enable the timely detection of abnormal fluctuations in CO2 and other greenhouse gas emissions during geopolitical conflicts, thereby providing robust data support for the periodic reporting obligations under the UNFCCC. Such datasets facilitate the quantitative assessment of emission variabilities during conflict periods, which is essential for ensuring the equitable allocation of carbon credits and the effective operation of compensation mechanisms within carbon trading schemes (Adediran and Swaray, 2023). Furthermore, these data lay the foundation for developing an accountability framework for conflict-related emissions (e.g., environmental damage compensation), thus informing decision-making processes for post-conflict carbon emission responsibility and advancing climate justice practices (Mubarik et al. 2024).

Global information on human activity can be visualized through Earth observations using nighttime light (NTL) satellites (Elvidge et al. 2009; Li and Cao, 2024; Zou et al. 2024). Common data sources include the Defense Meteorological Satellite Program Operational Linescan System (DMSP-OLS) (Huang et al. 2014) and the Suomi National Polar-Orbiting Partnership Visible Infrared Imaging Radiometer Suite (NPP-VIIRS) (Elvidge et al. 2017). On the one hand, NTL data have been considered as an effective tool for detecting geopolitical conflicts, such as the Iranian, Syrian, and Yemeni crises, among others (Jiang et al. 2017; Li et al. 2015, 2021). For example, the research in (Li and Li, 2014) revealed the national and provincial losses during the Syrian Civil War using monthly composites of data from DMSP-OLS. On the other hand, since NTL data can accurately depict the fine-grained spatial patterns of human activities, it has also been used to quantify anthropogenic CO2 emissions at various administrative scales by spatially disaggregating statistical data (Gao et al. 2023; Guo and Wang, 2023). Some studies have put forward global CO2 emission datasets by combining NTL data with ground observations (Gaughan et al. 2019). For instance, Oda and Maksyutov disaggregated national fuel estimates and created the Open Source Data Inventory for Anthropogenic CO2 (ODIAC) using a point source database and NTL data (Oda et al. 2018). In a similar fashion, another widely used dataset that reports emissions as national totals and global gridmaps is the Emission Database for Global Atmospheric Research (EDGAR) (Crippa et al. 2020). There is typically a lag of more than a year or two in the generation of these datasets. Recently, the Global Gridded Daily CO2 Emission Dataset (GRACED) was developed for near-real-time monitoring of anthropogenic CO2 emissions by integrating several data streams, including point sources, national fossil fuel and cement, and country-level sectoral emissions (Dou et al. 2022; Liu et al. 2020a). Despite its ability to monitor changes in anthropogenic CO2 emissions during geopolitical conflicts in a timely manner, GRACED is still constrained by the availability of statistical data and ground observations in conflict zones. Estimates of anthropogenic CO2 emissions are obtained using data from other regions when relevant data in conflict zones is not available, which could lead to significant errors and uncertainty.

Based on the above research and potential deficiencies, this study integrates NTL data and GRACED to highlight spatiotemporal changes in anthropogenic CO2 emissions during geopolitical conflicts. This study not only provides unique insights into how geopolitical conflicts rapidly change anthropogenic CO2 emissions, but also helps to understand the broader environmental impacts of geopolitical conflicts. To achieve these goals, we choose Ukraine as the study domain, as it is experiencing drastic changes in anthropogenic CO2 emissions during the Russia-Ukraine war. The main contributions of this work are summarized as follows: (1) The monthly changes in anthropogenic CO2 of different sectors in Ukraine during the Russia-Ukraine war are estimated. (2) The spatial heterogeneity of the changes in anthropogenic CO2 of different sectors in Ukraine during the Russia-Ukraine war is also analyzed.

Materials and methods

This section introduces the study case, datasets, and our newly proposed model for monitoring changes in anthropogenic CO2 emissions.

Study case

Since February 2022, Russia has invaded Ukraine on several fronts, with attacks on both military and civilian targets seriously affecting Ukraine’s population displacement, national economy, and ecological environment (Rawtani et al. 2022). The impact of the Russia-Ukraine war goes far beyond the heavily industrialized area of fighting, as civilians in the hinterland and close to the front lines are constantly in danger from drone attacks, airstrikes, and indiscriminate shelling. The Armed Conflict Location & Event Data Project (ACLED) has recorded approximately 40,000 incidents of political violence in Ukraine, one year after the war began (Raleigh et al. 2010). Over 5 million Ukrainians have been internally displaced, and another 8 million have been compelled to apply for asylum outside. In this study, Ukraine (currently experiencing a Russian invasion) is chosen as a study domain to examine changes in anthropogenic CO2 emissions (Fig. 1A).

Fig. 1: Spatial distribution of sectoral CO₂ emissions.
figure 1

A Study domain and B–D CO2 emissions of residential consumption, ground transport, and industry sectors across Ukraine in January 2022.

Full size image

Located in the East European Plain, Ukraine is the second largest country in Europe, with an area of approximately 603,500 square kilometers. The country presents a landscape consisting of fertile farmland, vast green areas, and highly urbanized and industrialized built-up areas (Fig. S1). Industrial activities are mainly concentrated in the regions of Donetsk, Luhansk, Dnipro, and Zaporizhia, while large green areas, including nature reserves, forests, meadows, and farmlands, serve as important buffers to mitigate urban heat islands and improve ecological services (Morar et al. 2022).

Ukraine experiences a temperate continental climate characterized by cold winters and warm summers, with pronounced temporal and spatial variability. Utilizing the Terra Moderate Resolution Imaging Spectroradiometer (MODIS) Land Surface Temperature/Emissivity Daily (MOD11A1) product, we derive the spatial distribution of average annual land surface temperatures in Ukraine for the years 2000, 2010, and 2020 (Fig. S2A–C), along with the temperature changes observed in 2020 relative to 2000 (Fig. S2D). Higher temperatures and significant warming trends are most evident in the southern and eastern regions, where industrial activities are heavily concentrated. These temperature shifts are closely linked to changes in vegetation phenology, extended growing seasons, and alterations in regional carbon cycles (Piao et al. 2007).

Data

VIIRS NTL imagery

A global daily measuring system for Earth system science and applications is provided by the Day-Night Band (DNB) sensor included in the VIIRS. The atmospheric and Lunar bidirectional reflectance (BRDF)-corrected Black Marble NTL product (VNP46A2) provides daily DNB data at 500 m spatial resolution, with operational correction for surface reflected lunar radiance (Román et al. 2018). We acquired daily NTL imagery over the study domain between January 1, 2021, and February 28, 2023, from the VNP46A2, which is publicly available from the National Centers for Environmental Information (NCEI) of the National Oceanic and Atmospheric Administration (NOAA). Then, we down-sample the collected NTL imagery to a spatial resolution of 10 km by grid summing.

GRACED

GRACED has been tracking the spatiotemporal fluctuation features of global anthropogenic CO2 emissions from various sectors around the world since January 1, 2019, thus supporting adjustments of various climate policy measures. Specifically, GRACED is produced with a temporal resolution of one day and a global spatial resolution of 10 km. Here, we collect daily anthropogenic CO2 emission data of residential consumption, ground transport, and industry sectors between January 1, 2021, and February 28, 2023, from GRACED. Figure 1B–D present the spatial distribution of anthropogenic CO2 emissions from different sectors in Ukraine before the Russia-Ukraine war (i.e., January 2022).

Methods

Figure 2 presents a schematic overview of the proposed model for monitoring monthly changes in anthropogenic CO2 emissions of different sectors using VIIRS NTL data and GRACED. We first aggregate the daily NTL data for each month during the war period using a minimum composite technique, and then calculate the monthly NTL changes relative to a reference period. Meanwhile, the monthly mean composite NTL data and the monthly mean composite GRACED data from the year before the Russia-Ukraine war are used as baseline data for determining CO2 emissions per unit of NTL. On this basis, the monthly changes in anthropogenic CO2 emissions of different sectors during the war period are estimated and then validated by the monthly changes derived from the GRACED data during the war period and the reference period.

Fig. 2
figure 2

Workflow for monitoring monthly changes in anthropogenic CO2 emissions of different sectors using VIIRS NTL data and GRACED.

Full size image

Monthly changes in the NTL composite

The monthly changes in NTL caused by the Russia-Ukraine war are reflected based on the difference between the NTL during the war period and the NTL during the reference period. Let \({{\rm{NTL}}}_{{m}_{i},{y}_{j}}=\left\{{{\rm{NTL}}}_{{m}_{i},{y}_{j}}^{1},\ldots ,{{\rm{NTL}}}_{{m}_{i},{y}_{j}}^{n}\right\}\) be a set of daily NTL data, where n is the number of days for month i of year j. Given that NTL data are easily disturbed by abnormal factors such as fire and explosions during the war period (Ialongo et al. 2023; Wang et al. 2021), we generate the monthly minimum composite of NTL data (from January 2022 to February 2023) by using the minimum function of Google Earth Engine (GEE) to reduce data uncertainty as follows:

$${{\rm{NTL}}}_{{m}_{i},{y}_{j}}^{min}=\arg \min {{\rm{NTL}}}_{{m}_{i},{y}_{j}}.$$
(1)

Then, the monthly minimum composite of NTL data is imported into the ArcGIS 10.7 software for spatial analysis. Taking January 2022 as a reference period before the war, the monthly changes in the NTL minimum composite during the war period are calculated using the raster calculator tool:

$${{\rm{DNTL}}}_{{m}_{i},{y}_{j}}={{\rm{NTL}}}_{{m}_{i},{y}_{j}}^{min}-{{\rm{NTL}}}_{Jan,2022}^{min}.$$
(2)

Considering the extensive destruction of civilian infrastructure and the resulting population relocation during the Russia-Ukraine war (Rawtani et al. 2022), we assume that the war period will not have a higher NTL value than the non-war period. As a result, these observations will not be included for subsequent estimation of the changes in CO2 emissions when \({{\rm{DNTL}}}_{{m}_{i},{y}_{j}}\) is larger than zero.

Estimation of monthly CO2 emission changes

Here, we develop a spatial model to estimate the monthly changes in anthropogenic CO2 emissions across several sectors. Based on the daily NTL data and daily anthropogenic CO2 emission data from GRACED, the monthly mean composite NTL data and the monthly mean composite anthropogenic CO2 emission data during the baseline period (from January 2021 to December 2021) are first generated via the mean function of GEE:

$${{\rm{NTL}}}_{{m}_{i},{y}_{j}}^{mean}=\arg {\rm{mean}}\,{{\rm{NTL}}}_{{m}_{i},{y}_{j}},$$
(3)
$${{\rm{GRACED}}}_{{m}_{i},{y}_{j},{s}_{k}}^{mean}=\arg {\rm{mean}}\,{{\rm{GRACED}}}_{{m}_{i},{y}_{j},{s}_{k}},$$
(4)

\({{\rm{GRACED}}}_{{m}_{i},{y}_{j}}=\left\{{{\rm{GRACED}}}_{{m}_{i},{y}_{j},{s}_{k}}^{1},\ldots ,{{\rm{GRACED}}}_{{m}_{i},{y}_{j},{s}_{k}}^{n}\right\}\) is a set of daily GRACED data of sector k, and n is the number of days for month i of year j (i.e., 2021).

Then, the monthly mean composite is imported into the ArcGIS 10.7 software. The subsequent formulas are all numerically calculated using the raster calculator tool of the software. The monthly CO2 emissions per unit of NTL of different sectors during the baseline period are calculated by:

$${{\rm{CEPN}}}_{{m}_{i},{y}_{j},{s}_{k}}=\frac{{{\rm{GRACED}}}_{{m}_{i},{y}_{j},{s}_{k}}^{mean}}{{{\rm{NTL}}}_{{m}_{i},{y}_{j}}^{mean}}.$$
(5)

Based on the monthly changes in the NTL minimum composite during the war period (Eq. (2)) and the monthly CO2 emissions per unit of NTL during the baseline period (Eq. (5)), we estimate the monthly changes in anthropogenic CO2 emissions of different sectors. Specifically, the monthly changes in CO2 emissions in the residential consumption sector (relative to the reference period) are estimated as follows:

$$\begin{array}{ll}{{\rm{MCCE}}}_{{m}_{i},{y}_{j},residential}={{\rm{TAF}}}_{{m}_{i}}\times {{\rm{CEPN}}}_{{m}_{i},{y}_{j},residential}\\\qquad\qquad\qquad\qquad\quad\times\,{{\rm{DNTL}}}_{{m}_{i},{y}_{j}},\end{array}$$
(6)

where \({{\rm{TAF}}}_{{m}_{i}}\) is a temperature adjustment factor for compensating the variations in air temperature that are considered in GRACED. The \({{\rm{TAF}}}_{{m}_{i}}\) is calculated according to the monthly mean composite of GRACED data during the baseline period:

$${{\rm{TAF}}}_{{m}_{i}}=\frac{{{\rm{GRACED}}}_{{m}_{i},2021}^{mean}}{{{\rm{GRACED}}}_{Jan,2021}^{mean}}.$$
(7)

Given that the frequent movement, displacement, and humanitarian assistance of Ukrainian civilians during the war will lead to intensive transport activities and CO2 emissions, the monthly changes in CO2 emissions ofthe ground transport sector relative to the reference period are estimated as follows:

$$\begin{array}{ll}{{\rm{MCCE}}}_{{m}_{i},{y}_{j},transport}=(-1)\times {{\rm{CEPN}}}_{{m}_{i},{y}_{j},transport}\\\qquad\qquad\qquad\qquad\;\;\times\,{{\rm{DNTL}}}_{{m}_{i},{y}_{j}}.\end{array}$$
(8)

Concerning the industry sector, the monthly changes in CO2 emissions relative to the reference period are estimated by multiplying the monthly changes in the NTL minimum composite during the war period and the monthly CO2 emissions per unit of NTL during the baseline period:

$${{\rm{MCCE}}}_{{m}_{i},{y}_{j},industry}={{\rm{CEPN}}}_{{m}_{i},{y}_{j},industry}\times {{\rm{DNTL}}}_{{m}_{i},{y}_{j}}.$$
(9)

Note that the geographic coordinate system of all the input and output data mentioned above is the World Geodetic System 1984 (WGS84). The spatial distribution of resultant CO2 emissions across different sectors is extracted according to the extent of Ukraine using the clip tool of ArcGIS 10.7 software and exported.

Validation of monthly CO2 emission changes

Here, we generate monthly mean composites of GRACED data by aggregating the daily GRACED data (Eq. (4)) during the war period (from February 2022 to February 2023) and the reference period (i.e., January 2022). Then, the monthly changes in the GRACED mean composite of different sectors relative to the reference period are calculated by:

$${{\rm{DGRACED}}}_{{m}_{i},{y}_{j},{s}_{k}}={{\rm{GRACED}}}_{{m}_{i},{y}_{j},{s}_{k}}^{mean}-{{\rm{GRACED}}}_{Jan,2022,{s}_{k}}^{mean}.$$
(10)

\({{\rm{DGRACED}}}_{{m}_{i},{y}_{j},{s}_{k}}\) is used as the ground truth for validating the monthly CO2 emission changes estimated from VIIRS NTL data. In particular, we use linear regression to compare the estimated and ground truth values. According to the R2 and slope fitting metrics, we examine the effectiveness of the proposed spatial model for estimating monthly changes in CO2 emissions of different sectors.

Results

In this section, we first analyze the NTL changes in Ukraine across various scales. Then, the estimation of changes in CO2 emissions of different sectors is performed.

NTL changes in Ukraine across various scales

Figure 3 A–C present the monthly NTL minimum composite in January 2022 (one month before the war began), August 2022 (six months after the war began), and February 2023 (one year after the war began). With the war going on, Ukraine has suffered a precipitous decrease in NTL, with many illuminated areas becoming dark. For quantitative analysis, we first assess NTL changes at the oblast (administrative division or region) scale (Fig. 3D). Compared with the western region in the hinterland, the central and eastern regions close to the front lines exhibit relatively more NTL reduction. Six months on from the invasion, 63% of oblasts show a reduction of more than 50% in NTL. One year on from the invasion, the NTL in oblasts of Lviv, Kiev, Poltava, Dnipropetrovsk, and Kharkiv decreased by 41%, 78%, 67%, 75%, and 69%, respectively.

Fig. 3: Monthly NTL minimum composite and changes.
figure 3

Monthly NTL minimum composite in Ukraine in A January 2022, B August 2022, and C February 2023, and NTL changes at D oblast scale and E city scale.

Full size image

In addition, we select seven cities severely affected by the war for further analysis, including Lviv City, Kiev City, Kherson City, Dnipropetrovsk City, Zaporizhia City, Kharkiv City, and Donetsk City (Fig. 3E). The NTL in the cities of Kiev, Dnipropetrovsk, Zaporizhia, and Kharkiv has a reduction of more than 60% six months after the war began. And the cities of Kiev, Kherson, Dnipropetrovsk, Zaporizhia, and Kharkiv show a reduction in NTL of more than 80% one year after the war began. In contrast, while being threatened and attacked by drone attacks and airstrikes, Lviv City in eastern Ukraine exhibits a relatively lower drop in NTL. It can also be found that the NTL of Donetsk City has not changed significantly. On the one hand, this may be due to the fact that this city has been in conflict since 2014, resulting in a small difference between the NTL during the war and during the reference period. On the other hand, as the main conflict area between Russia and Ukraine, the frequent explosions and flames in Donetsk City may also lead to abnormalities and uncertainty in the NTL.

Estimation of changes in CO2 emissions of different sectors

Based on the aforementioned NTL changes in Ukraine, we estimate the changes in CO2 emissions of residential consumption, ground transport, and industry sectors, respectively (Figs. 4–6). The spatial distribution of residential consumption CO2 emission declines in Ukraine is presented in the Fig. 4A and D. Relative to the reference period, a total of 413 kt and 136 kt declines in residential consumption CO2 emissions are observed six months and one year after the war began, respectively. The CO2 emissions in the residential consumption sector are counted at the oblast scale (Fig. 4C and F). Six months after the war began, the top five oblasts with the largest decline in monthly CO2 emissions are Kiev (59 kt), Dnipropetrovsk (47 kt), Kharkiv (43 kt), Lviv (33 kt), and Donetsk (32 kt). One year after the war began, the top five oblasts with the largest decline in monthly CO2 emissions are Kiev (19 kt), Kharkiv (17 kt), Dnipropetrovsk (14 kt), Donetsk (14 kt), and Odessa (7 kt). As for the ground transport sector, the spatial distribution of CO2 emission increases in Ukraine is presented in Fig. 5A and D. In total, 106 kt and 33 kt increases in ground transport CO2 emissions are observed six months and one year after the war began, respectively. Figure 5C and F present the oblast-scale CO2 emissions of the ground transport sector. Six months after the war began, the top five oblasts with the largest decline in monthly CO2 emissions are Dnipropetrovsk (13 kt), Kiev (11 kt), Kharkiv (9 kt), Donetsk (8 kt), and Lviv (6 kt). One year after the war began, the top five oblasts with the largest decline in monthly CO2 emissions are Dnipropetrovsk (5 kt), Donetsk (4 kt), Kharkiv (4 kt), Kiev (4 kt), and Poltava (2 kt). Regarding the industry sector, the spatial distribution of CO2 emission declines in Ukraine is presented in Fig. 6A and D. Six months and one year after the war began, 324 kt and 139 kt declines in ground transport CO2 emissions are observed, respectively. Figure 6C and F present the oblast-scale CO2 emissions of industry sector. Six months after the war began, the top five oblasts with the largest decline in monthly CO2 emissions are Dnipropetrovsk (126 kt), Zaporizhia (47 kt), Donetsk (22 kt), Lviv (17 kt), and Luhansk (12 kt). One year after the war began, the top five oblasts with the largest decline in monthly CO2 emissions are Dnipropetrovsk (57 kt), Zaporizhia (24 kt), Donetsk (13 kt), Luhansk (12 kt), and Kharkiv (5 kt).

Fig. 4: Residential consumption CO₂ emission changes.
figure 4

A National spatial distribution, B linear regression, and C oblast-level estimation of residential consumption CO2 emission decline six months after the invasion. D National spatial distribution, E linear regression, and F oblast-level estimation of residential consumption CO2 emission decline one year after the invasion.

Full size image
Fig. 5: Ground transport CO₂ emission changes.
figure 5

A National spatial distribution, B linear regression, and C oblast-level estimation of ground transport CO2 emission decline six months after the invasion. D National spatial distribution, E linear regression, and F oblast-level estimation of ground transport CO2 emission decline one year after the invasion.

Full size image
Fig. 6: Industry CO₂ emission changes.
figure 6

A National spatial distribution, B linear regression, and C oblast-level estimation of industry CO2 emission decline six months after the invasion. D National spatial distribution, E linear regression, and F oblast-level estimation of industry CO2 emission decline one year after the invasion.

Full size image

Linear regression analyses are conducted to assess the degree of agreement between the estimated and reference CO2 emissions for each sector throughout the war period. For the residential consumption sector, there is a significant degree of congruence between the estimated and reference values in both August 2022 (R2 = 0.76, ρ < 0.05) and February 2023 (R2 = 0.87, ρ < 0.05) (Fig. 4B and E). For the ground transport sector, the estimated and reference values exhibit a substantial degree of consistency in both August 2022 (R2 = 0.64, ρ < 0.05) and February 2023 (R2 = 0.74, ρ < 0.05) (Fig. 5B and E). For the industry sector, linear regression results demonstrate significant consistency between the estimated and reference values in both August 2022 (R2 = 0.91, ρ < 0.05) and February 2023 (R2 = 0.89, ρ < 0.05) (Fig. 6B and E). Additionally, Tables S1–S3 provide the statistical findings of linear regression on the monthly variations in the estimated and reference CO2 emissions of different sectors. There is a notable degree of agreement between the estimated and reference values for each month during the war. Specifically, the R2 ranges from 0.61-0.87, 0.51-0.74, and 0.69-0.93 for the residential consumption, ground transport, and industry sectors, respectively.

Furthermore, the monthly changes in CO2 emissions of different sectors relative to the reference period are shown for the whole Ukraine and also for seven cities severely affected by the war, i.e., Dnipropetrovsk City, Donetsk City, Zaporizhzhia City, Kiev City, Kharkiv City, Kherson City, and Lviv City (Fig. 7). Since the invasion in late February 2022, the ongoing war in the first half of the year has caused significant changes in CO2 emissions of different sectors relative to January 2022. Beginning in mid-September, the Russia-Ukraine war was reduced from a full-scale war to a local war due to the declaration of defeat on Russia’s fronts in the Kiev and Kharkov directions. Then, Russia began regular air and missile attacks on Ukraine’s energy facilities and civilian infrastructure from October to December 2022, resulting in more pronounced changes in CO2 emissions of different sectors during this period.

Fig. 7: Monthly sectoral CO₂ emission trends.
figure 7

Monthly changes in CO2 emissions of different sectors for A the whole Ukraine, B Dnipropetrovsk City, C Donetsk City, D Zaporizhzhia City, E Kiev City, F Kharkiv City, G Kherson City, and H Lviv City.

Full size image

Discussion

In the context of Russia-Ukraine war, this research estimates the monthly changes in CO2 emissions of different sectors using NTL imagery. The spatial heterogeneity of the anthropogenic CO2 changes is also uncovered across various scales. According to the results of linear regression tests, the estimated changes exhibit good consistency with the reference changes from the GRACED data during the war (Tables S1–S3). However, the difference between the estimated and reference values of different sectors needs to be investigated. Regarding the residential consumption sector, GRACED calculated CO2 emissions by assuming that the annual totals remain unchanged (Dou et al. 2022; Liu et al. 2020a). The war-induced changes in residential consumption CO2 emissions from the GRACED data could be ignored or underestimated. The estimated changes derived from NTL data are greater than those derived from GRACED, and the slopes of linear regression are larger than 1. As for the ground transport sector, GRACED calculated CO2 emissions by makingthe assumption that there would be little change in the spatial distribution of ground transport within a nation (Dou et al. 2022; Liu et al. 2020a). But the massive population movement and displacement at the beginning of the war, as well as the return of civilians and the regularization of air and missile attacks after September 2022, are likely to exacerbate CO2 emissions related to ground transport. Correspondingly, the estimated values are larger than the reference values in those months, with the slopes of linear regression being larger than one. Concerning the industry sector, the slopes of linear regression are not consistently positive or negative. It may be explained by the fact that statistical information on industrial production and electrical generation in Ukraine during the conflict was unavailable (Dou et al. 2022; Liu et al. 2020a). In this case, datasets from other countries or regions are adopted for calculating industry CO2 emissions in GRACED. Therefore, it is reasonable that the estimated changes in industry CO2 emissions are much larger than the reference changes from GRACED. In addition, it can be observed from the results in Tables S4 that the fitting effects of the three sectors are different. The fitting results of residential consumption and industry sectors are better than those of the ground transport sector, which could be related to the difference in activity intensity reflected by NTL in various sectors and NTL sensor limitations. On the one hand, various human activities produce different lighting intensities and durations at night, which determines the different abilities of light data to capture changes in these activities (Shi et al. 2021). The industry and residential consumption sectors usually produce consistent and stable NTL signals, while the activities of the ground transport sector are mainly manifested in road lighting and moving vehicle lights, which are more dispersed in space and highly time-varying, making it difficult to stably reflect their overall activity level through a single indicator. On the other hand, NTL data have issues with sensor saturation or insufficient resolution in different brightness ranges (Sun et al. 2024; Zheng et al. 2023). NTL in concentrated industrial and residential areas is relatively less susceptible, while scattered and weaker road and vehicle lights are more susceptible to sensor limitations and atmospheric interference, which could statistically increase data uncertainty and regression model errors.

To further highlight the contribution and methodological advantages of this study, the following discussion will focus on three aspects by comparing with related research: overcoming the limitations of conventional carbon satellites in detecting weak changes in anthropogenic CO2 emissions, resolving the issue of ground data scarcity during geopolitical conflicts, and offering data support for climate accountability and policy making. So far, many existing studies have demonstrated the effectiveness of using carbon satellites to monitor atmospheric CO2 concentrations (Pan et al. 2021; Wang et al. 2024; Wilmot et al. 2024). But in reality, changes in anthropogenic CO2 emissions are typically far less significant than variations in concentrations caused by transportation and interannual variability in the atmosphere (Liu et al. 2023a). It is challenging to precisely record such small changes using carbon satellite data. By combining global CO2 emission datasets prior to the conflict (e.g., GRACED) with NTL data during the conflict, this study offers a novel model to monitor changes in anthropogenic CO2 emissions in the context of geopolitical conflicts. To a certain degree, the proposed model could compensate for the limitations of carbon satellite data. In addition, the current global CO2 emissions datasets can well reflect the spatiotemporal characteristics of CO2 emissions from different human activities (Crippa et al. 2020; Dou et al. 2022; Oda et al. 2018). However, the statistical reports and ground observation data for dataset construction are often difficult to obtain or updated with lags during geopolitical conflicts (Levin et al. 2018; Ratnayake et al. 2022). This study uses NTL data as an alternative indicator, which not only compensates for the lack of data but also captures the immediate impact of conflict on CO2 emissions by comparing pre-war data with near-real-time changes. Furthermore, some studies have made significant progress in the development of policy frameworks and the adjustment of policy measures in recent years, such as the carbon market mechanism (Asadnabizadeh and Moe, 2024; Redmond and Convery, 2015; Schneider and La Hoz Theuer, 2019) and the climate accountability framework (Atapattu, 2020; Mees and Driessen, 2019; Williams, 2020). And policy design is gradually shifting towards data-driven, transparent, and cross-departmental coordination (Ali and Kamraju, 2025; van Deursen and Gupta, 2024; Hughes et al. 2020). However, it is challenging to satisfy the timely and reliable emission data demands of international regulators and policymakers, particularly in unpredictable contexts like geopolitical conflicts, because of inadequate measurement precision or a lack of frequent data updates (Ali and Thakkar, 2023; Feng et al. 2024). At the policy level, the proposed model can support spatial data on CO2 emissions in a dynamic manner. This will help improve the implementation of the climate accountability framework, in addition to providing decision makers with a new tool to assess CO2 emission changes in the context of conflict.

In the following, the limitations of this research and corresponding potential future research lines are discussed. First of all, Despite NTL imagery demonstrates great potential for anthropogenic CO2 monitoring during geopolitical conflicts, numerous research has demonstrated that the accuracy of NTL for CO2 emission estimation depends on a number of variables, including the population, economy, and natural environment (Liu et al. 2018; Shi et al. 2019; Sun et al. 2024). In particular, NTL tends to exhibit higher uncertainty for less developed or rural areas of the world with poor lighting (Pandey et al. 2017). In order to overcome such limitations (and the lack of NTL data) for anthropogenic CO2 emission estimation under geopolitical conflicts, it is of great importance to introduce point source data to enhance the understanding of spatiotemporal relationships between NTL and CO2 emissions. Specifically, social media data and data on political violence events can be considered (Liu et al. 2020b, 2024b; Raleigh et al. 2010). For example, some studies have modeled population displacements in Ukraine during the Russia-Ukraine war using social media data from sources including Facebook’s advertising platform and Twitter (Leasure et al. 2023; Liu et al. 2024). The near real-time data on the ongoing geopolitical conflicts from ACLED also provides spatial information on different types of political violence events that can also be used to reveal conflict-related CO2 emission processes and their environmental impact, including the use of petroleum products in military activities, the decomposition of war waste and fires in infrastructure, forests, and petroleum storage depots (Bun et al. 2024; Rawtani et al. 2022). Second, the current models may have oversimplified assumptions in complex and dynamically changing situations, and may not be able to fully eliminate the impact of factors such as temporary controls and damaged infrastructure. In future research, machine learning and deep learning can be introduced into the current model to capture nonlinear relationships and the impact of sudden events, thereby improving the model’s prediction accuracy and robustness for CO2 emission changes in complex environments (Ali et al. 2025; Bianchi and Putro, 2024; Guth and Sapsis, 2019; Qi and Majda, 2020). Third, this study is limited to analyzing the spatiotemporal changes in CO2 emissions under geopolitical conflict scenarios. A key direction for future research is to migrate the model to other emergency scenarios on a global scale, such as natural disasters (Mu et al. 2024), public health crises (Lan et al. 2021), energy crises (Liu et al. 2023), and others. In these different scenarios, energy consumption and human activity patterns may change significantly, which in turn affects the dynamics of CO2 emissions (Anser, 2019; Liu et al. 2022; Yu et al. 2022). By comparing cross-domain applications and verifications, the universality of the current model can be further validated. On this basis, a unified and flexible framework can be constructed to provide data support and a more comprehensive scientific basis for global climate accountability and environmental governance.

Conclusion

We develop a new spatial model based on VIIRS NTL and GRACED data to monitor monthly changes in anthropogenic CO2 emissions of different sectors during geopolitical conflicts. Taking the Russia-Ukraine war as a case study, we estimate the monthly changes in CO2 emissions of different sectors and reveal the spatial heterogeneity of CO2 emission changes across various scales. Relative to January 2022, residential consumption, ground transport, and industry sectors are respectively observed to have CO2 emission changes of 413 kt, 106 kt, and 324 kt (six months after the war began), and of 136 kt, 33 kt, and 139 kt (one year after the war began). There is significant consistency between the estimated and reference CO2 emission changes for each month during the war. The R2 ranges from 0.61-0.87, 0.51-0.74, and 0.69-0.93 for the residential consumption, ground transport, and industry sector, respectively. In conclusion, this study provides a new perspective to improve the understanding of CO2 emission changes under geopolitical conflicts, as well as the potential use of applying the proposed spatial model to ongoing geopolitical conflicts around the world. Future perspectives include not only deepening the technical level of existing methods, but also expanding to multi-modal data fusion and interdisciplinary collaborative research, so as to build a more flexible and efficient system for global environmental monitoring and CO2 emission management. Specifically, the integration of multi-source remote sensing data and ground monitoring data can be explored to improve the spatiotemporal resolution of CO2 emission monitoring in geopolitical conflicts and emergency situations. In addition, advanced technologies such as machine learning and deep learning can be considered to further optimize data correction and model prediction capabilities, so as to provide more accurate and real-time data support for policy making, climate accountability, and environmental emergency management.

Policy recommendations

This study reveals the spatiotemporal impact of geopolitical conflicts on anthropogenic CO2 emissions from a remote sensing perspective. Relevant policies should make full use of this technology and data advantages to optimize the environmental monitoring system and enhance CO2 emission management capabilities. First, it is recommended that governments and international organizations increase investment to build a monitoring system that integrates multiple remote sensing data, such as nighttime lights, thermal infrared, aerosols, and high-resolution optical images. This system can not only complement the advantages of different data and improve the accuracy of capturing emission changes, but also achieve all-weather and dynamic monitoring (Jiang et al. 2023; Tian et al. 2024). Meanwhile, the system should have automated data processing and real-time early warning functions to ensure that abnormal fluctuations in environmental indicators can be quickly reflected in emergency situations, providing a scientific basis for government decision-making and emergency response.

Second, geopolitical conflicts and emergencies often go beyond the jurisdiction of a single country or department. Therefore, monitoring and assessing CO2 emission change requires cross-departmental and cross-regional collaboration (Xu et al. 2025). It is recommended that governments, scientific research institutions, international organizations and private enterprises build data sharing and joint monitoring mechanisms, break down information silos and establish a multilateral cooperation framework. By regularly holding international seminars and joint experimental projects, all parties can discuss issues such as data integration, model improvement, and emergency plans, thereby improving the accuracy and credibility of monitoring results.

Third, environmental governance and climate policy formulation cannot be separated from public supervision and extensive participation. It is recommended to build an open and transparent environmental data sharing platform to regularly release remote sensing monitoring data, emission estimation results and related analysis reports to the public. Transparent information disclosure will not only help to improve the accountability of governments and enterprises in carbon emission reduction, but also create conditions for public participation in environmental decision-making (Hahn et al. 2015; Li et al. 2017). This will help promote the formation of a green development model with the participation of the whole society and provide real and timely data information support for the international climate accountability framework.