The Russia-Ukraine war has caused significant population displacement and disrupted anthropogenic CO₂ emissions. Traditional data collection methods are often slow and insufficient during conflicts. This study addresses the need for rapid, fine-grained monitoring of these changes by utilizing social media big data, specifically geo-tagged tweets from Twitter. The research questions center on quantifying the spatiotemporal patterns of war-induced population displacement and the resulting changes in CO₂ emissions across different sectors. The importance of this study lies in its ability to provide near real-time information crucial for humanitarian aid and climate change mitigation strategies. Existing datasets like Facebook's advertising platform data provide valuable information but have limitations in spatial coverage and temporal resolution. Traditional CO₂ emission inventories are often delayed by a year or more, rendering them unsuitable for assessing immediate impacts. Satellite remote sensing can monitor CO₂ emissions but struggles with separating anthropogenic emissions from natural variations. This study aims to overcome these limitations by combining the strengths of social media data, specifically Twitter’s geo-tagged tweets, with existing CO₂ emission and population datasets. The integration of these data sources allows for a more comprehensive and timely understanding of the dynamic interplay between conflict, population movement, and environmental impact.

Several studies have used social media data to predict migration and population displacement, primarily relying on platforms like Twitter, Facebook, and Instagram. The use of Facebook's advertising platform data to estimate population changes in Ukraine during the war is a recent advancement. However, these studies often lack the fine-grained spatial resolution required to understand the localized impacts of conflict. Furthermore, limitations in data availability, especially for certain regions in Ukraine (like Luhanska and Donetska Oblasts before March 11th, 2022, the Autonomous Republic of Crimea, and Sevastopol), highlight the need for alternative approaches. The impact of geopolitical conflicts on anthropogenic CO₂ emissions remains largely unclear due to the lack of timely and high-resolution data. Existing global anthropogenic CO₂ emission datasets, such as GRACED, ODIAC, and EDGAR, provide valuable information but are often lagged, limiting their use in near-real-time monitoring of conflict impacts. This research builds on these previous efforts by incorporating geo-tagged Twitter data to improve the spatiotemporal resolution and timeliness of population and CO₂ emission estimations during the Ukraine conflict.

This study employs a three-pronged approach to estimate monthly changes in population and CO₂ emissions in Ukraine from January 1, 2022, to February 28, 2023. First, geo-tagged tweets from Twitter were collected and pre-processed to remove duplicates and account for user bias. The quantity of geo-tagged tweets in each 0.1° × 0.1° grid cell per month served as a proxy for the number of active Twitter users. The change ratio (CR) in the quantity of geo-tagged tweets was calculated relative to January 2022, representing the baseline period before the full-scale invasion. Second, baseline population data from the 2021 LandScan dataset were downsampled to a 0.1° × 0.1° resolution. The monthly population change (PC) was then estimated by multiplying the CR by the baseline population for each grid cell. This deterministic model assumes that the change in the number of active Twitter users directly reflects the change in the population. Third, anthropogenic CO₂ emission data from GRACED were used. Data for residential consumption, ground transport, and industry sectors were aggregated to monthly averages. Baseline CO₂ emissions per capita for each sector in 2021 were calculated and subsequently multiplied by the monthly population change (PC) to estimate monthly CO₂ emission changes. A temperature adjustment parameter (TA) was applied to the residential consumption sector to account for seasonal variations in heat emissions. The study hypothesizes that population displacement reduces residential and industrial CO₂ emissions while increasing ground transport emissions due to increased traffic congestion. The estimated CO₂ emission changes were validated against those calculated directly from the GRACED data for the same period (February 2022 – February 2023), using linear regression analysis to assess the strength of the correlation (R²) and the slope of the relationship between the estimated and reference values. Three representative Oblast-level administrative regions (Lviv, Kiev, and Donetsk) were selected for detailed analysis to illustrate the spatial variation in population and CO₂ emission changes.

The analysis revealed substantial population displacement in Ukraine in the year following the invasion. Over 11 million Ukrainians were displaced from their baseline 0.1° × 0.1° gridded regions by February 2023. Significant population declines were observed in Kiev Oblast (including Kiev City), Donetsk Oblast, and Lviv Oblast, with percentage declines of 44%, 24%, and 27%, respectively. The study found that the spatial distribution of population decline closely correlated with the intensity and location of geopolitical conflict events. Regarding CO₂ emissions, the study predicted a 336.21 kt decrease in residential consumption CO₂ emissions, a 58.83 kt increase in ground transport CO₂ emissions, and a 483.3 kt decrease in industrial CO₂ emissions. The strongest decreases in industrial CO₂ emissions were observed in Donetsk Oblast, reflecting the significant impact of conflict on industrial activities in that region. Linear regression analysis showed a strong positive correlation between the estimated and reference CO₂ emission changes for all three sectors, with R² values ranging from 0.57 to 0.93 for residential consumption, 0.41 to 0.9 for ground transport, and 0.74 to 0.99 for industry. The monthly analysis indicated that population displacement and changes in CO₂ emissions were most pronounced during the initial invasion (February-April 2022) and the period of increased missile strikes (October 2022-February 2023). The findings highlight the usefulness of combining social media data and existing emission datasets for near real-time monitoring of the impacts of geopolitical conflicts.

The strong correlations between estimated and reference CO₂ emission changes validate the proposed methodology's effectiveness in monitoring war-induced changes. The results demonstrate the potential of using geo-tagged Twitter data to supplement and improve the accuracy of existing CO₂ emission datasets, especially during periods of rapid change. The study successfully addresses the research question by providing a novel method for near-real-time monitoring of population displacement and its environmental consequences. The findings are highly relevant to humanitarian efforts and climate change mitigation by providing crucial information for targeted aid distribution and policy formulation. The spatial variation in population and CO₂ emission changes across different regions of Ukraine highlights the importance of considering the localized impacts of conflict. The higher-than-expected increase in ground transport emissions underscores the need to account for potential indirect effects of conflict, such as increased traffic congestion due to displacement.

This study introduces a novel method for near-real-time monitoring of war-induced population displacement and anthropogenic CO₂ emission changes using geo-tagged Twitter data, LandScan population data, and the GRACED CO₂ emission dataset. The findings demonstrate the reliability of this approach and its potential for improving the accuracy and timeliness of humanitarian and environmental assessments during conflicts. Future research could explore the integration of data from other social media platforms, incorporate demographic information, and use remote sensing data (e.g., nighttime light data) to enhance the accuracy and spatial resolution of the estimations. Applying this method to other conflict zones and exploring the long-term consequences of displacement on public health, ecology, and economic development are also important future avenues.

The study acknowledges several limitations. The reliance on geo-tagged tweets introduces potential biases due to the non-random sampling of Twitter users and the limited percentage of geo-tagged tweets available. The accuracy of population estimates depends on the assumption that changes in Twitter activity accurately reflect population changes. The model for estimating CO₂ emission changes relies on several assumptions regarding the relationship between population displacement and emissions in different sectors, which might not perfectly capture the complexities of real-world situations. Furthermore, the availability of data from GRACED is not uniform across all sectors and regions, potentially impacting the validation process.