Publications for this study were retrieved from the Web of Science (WOS) Core Collection and China National Knowledge Infrastructure (CNKI) databases, spanning the years 2008 to 2025. The keywords searched for were “Forest” and (“C storage” or “C sequestration” or “C sink”) and (“Measurement method” or “Technique” or “Technology”). The search period was from January 2008 to April 2025. Details of the Boolean search query are provided in Appendix A. During screening, iterative refinements were applied to exclude irrelevant records. In total, 2,427 publications related to forest C-storage estimation techniques were retained (Figure 1), including research articles and review papers. All records were saved and subjected to visual analyses to summarize journals, temporal trends, countries, institutions, authors, and keywords.

In the visual analysis, each network node represented a distinct entity, and the temporal dimension was indexed by publication year (Huang et al., 2023). Foundational literature on forest C-storage estimation in our knowledge network can be traced to 2008, when Professor George Burba introduced eddy-covariance (EC) corrections to reduce cold-season bias in ecosystem CO₂ uptake (Burba et al., 2008). Accordingly, 2008 was selected as the starting year.

International policy milestones align with subsequent methodological development. Under the Kyoto Protocol, Parties were required to meet quantified emission-reduction targets during 2008–2012 (Protocol, 1997; Rosenqvist et al., 2003). This period was designated as Stage 1 (Figure 2). Following the entry into force of the Paris Agreement on 4 November 2016, global climate governance entered a new phase (Paris Agreement, 2015; Grassi et al., 2017). Consequently, the years from 2013 to 2017 were defined as Stage 2. In 2018, the “REddyProc” package addressed key challenges in processing raw EC flux data (Wutzler et al., 2018), and network-level advances helped resolve several long-standing issues (Baldocchi et al., 2018), motivating Stage 3 (2018–2025).

Overall, the period 2008–2025 is divided into three stages: 2008–2012 (Stage 1), 2013–2017 (Stage 2), and 2018–2025 (Stage 3). This study systematically reviews publications within each stage to characterize distributional patterns and temporal evolution in forest C-storage estimation techniques.

The accuracy of forest C stock studies primarily relies on multi-source data fusion and model optimisation (Nguyen and Saha, 2024). By integrating ground plot measurements with RS technologies (such as Sentinel-2 and LiDAR), alongside ML models (e.g., RF, fully convolutional neural networks) and traditional regression methods (e.g., polynomial regression), researchers have made significant advancements in estimating forest C storage (Nasiri et al., 2022). Among these, regression models, especially those focusing on C storage estimation, have been refined over the years to increase their accuracy and applicability (Sun and Liu, 2019). In addition, C mapping techniques, including RS-based spatial analysis and visualization tools, have been developed to provide more detailed and geographically accurate assessments of forest C stocks (Lamichhane et al., 2019).

Over the past decade, these methods have been applied across a range of forest ecosystems, from tropical to temperate forests, and have shown a marked improvement in C stock predictions. The integration of these approaches, coupled with newer algorithms such as BO-RF (Bayesian Optimized RF), has significantly enhanced the reliability of C stock estimates by improving model precision (R²). While precision metrics like R² have been widely applied in previous studies, recent advancements in algorithms and optimisation of feature variables have further bolstered the accuracy and reliability of C stock estimates. Hence, in this study, a meticulous selection was made from 2,427 publications, ultimately identifying 487 papers as the most relevant to the present study. From these publications, key data such as study locations, latitude and longitude, research methods, and precision (R²) were extracted and recorded. Moreover, all study location coordinates were converted into decimal format and visually displayed in Figure 2a.

Furthermore, the research areas were categorized into three scales: large, medium, and small (Huang et al., 2022). Large-scale regions generally refer to global or continental areas, medium-scale regions typically refer to national or state levels, and small-scale regions usually refer to local or regional areas. It is important to note that ownership of large-scale studies often depends on the primary research institution producing the research article. The present study systematically reviewed the literature related to these three area scales and detailed the level of research into forest C storage estimation techniques across different spatial dimensions.

Bibliometric analysis was conducted using Cite Space (V6.2.R6) software package. This package is a metric instrument that displays knowledge flow and distribution of data, within a field from perspective parametres such as scientometrics, data analysis, and information visualization, and then generates knowledge graphs for different themes (Chen, 2006). The results can intuitively show the research trends and research hotspots in certain fields, and the size of each node in the graph represents the frequency of a particular item like keywords or authors (Wang G. et al., 2023).

Furthermore, in this paper, the “ggplot2” package in R 4.5.0 software was used to draw the bar chart, the” metafor” package was used to draw the global distribution map, and it was drawn in combination with ArcGis10.8.1. Data analysis was conducted using IBM SPSS 27 (Feng et al., 2024).

The literature discussing forest C storage estimation techniques came from 121 countries located around the world. These countries were primarily distributed across the Americas, Europe, Asia, and Oceania (Figures 2a,b). The number of publications relating to forest C storage estimation techniques and the cumulative number of publications over the period ranging from 2008 to 2025 showed a steady increase in growth trend, as seen Figures 2c,d. Stage 1 in the development of forest C storage estimation techniques ranged from 2008 to 2012. During this period the average annual global number of publications was 78, with an average annual growth rate of 6%. In China, the average annual number of publications was 23, with an average annual growth rate of 39%. During Stage 2 (i.e., 2013–2017) and Stage 3 (i.e., 2018 to 2025), the average annual global publication growth rates were 1 and 3%, respectively (Figure 2c). The cumulative number of publications globally showed a continuous linear upward trend, reaching a total of 2,427 publications, with China contributing 25.5% to the publication total (Figure 2d).

The literature containing forest C storage estimation techniques covered 78 disciplinary categories and involved 823 published journals. Among the top ten disciplines was Environmental Science, which had the highest number of publications (718) (Figure 3a) and accounted for 18% of the total number of publications (Supplementary Table S1). This was followed by Forestry and Ecology, with 456 and 383 publications respectively, representing 12 and 10% of the total number of publications. The journal “Global Change Biology,” a journal closely allied to the Environmental Science discipline, published 1,306 articles, which equated to 3% of the global total. The total number of publications in “Forest Ecology and Management,” “Science,” and “Nature” was 3,072, and accounted for 7% of the total number of publications (Figure 3b; Supplementary Table S2).

Since 2008, the citation frequency of “Global Change Biology,” “Tree Physiology,” and “Journal of Geophysical Research—Atmospheres” significantly increased during Stage 1, with sustained popularity lasting 4 to 5 years (Figure 3c). The journals “RF” and “Science of the Total Environment” maintained their popularity from 2022 to 2025, with citation levels of 21 and 18, respectively (Supplementary Table S3).

The present study found the United States had the highest number of publications, with 691 papers, and accounting for 29% of the total number of publications (Figure 4a). China followed with 468 publications, making up 20% of the total number of publications. Other countries with high publication numbers included Germany (240 papers, 10%), Australia (163 papers, 7%), and France (142 papers, 6%) (Supplementary Table S4, Figure 4a). The collaboration network existing between the United States and China produced the most publications, indicating that they hold a leading global position in forest C storage estimation techniques research (Figure 4b). Among the 234 publishing institutions worldwide, the Chinese Academy of Sciences ranked first with 208 publications, accounting for 22% of the total number of publications (Supplementary Table S5). This was followed by the United States Department of Agriculture (121 papers, 13%), the United States Forest Service (98 papers, 11%), and the University of California (87 papers, 9%) (Figure 4c). The University of the Chinese Academy of Sciences ranked seventh, and the remaining institutions were located in Europe (Supplementary Table S5). These findings demonstrate that the United States is the most research active country in the field of forest C storage estimation technology, followed by China (Figure 4d).

The study identified 613 authors contributing to forest C storage estimation (2008–2025). Among the most cited, M. Reichstein (231 citations, H-index 114) introduced the model–data fusion framework, integrating soil respiration and environmental drivers to construct cross-scale models. This approach significantly improved the accuracy of dynamic simulations of forest C fluxes and advanced the use of deep learning in Earth system science (Reichstein and Beer, 2008). Eric A. Davidson (220 citations, H-index 102) combined experiments and modeling to reveal the temperature sensitivity of soil C decomposition (Davidson and Janssens, 2006). He further developed the DAMM model, linking enzyme kinetics with temperature response to simulate soil organic matter decomposition, representing a process-based modeling approach (Davidson et al., 2012).

Dennis D. Baldocchi (215 citations, H-index 133) promoted EC as a core method for ecosystem-scale CO₂ fluxes, establishing its role in long-term monitoring networks such as FLUXNET (Baldocchi, 2003). Richard A. Houghton (204 citations, H-index 100) revised C flux estimates from land-use change since 1850, providing one of the most widely used approaches for global C budgets (Houghton, 2003; Houghton et al., 1999). Rattan Lal (200 citations, H-index 136) highlighted the role of soil C sequestration in climate mitigation and developed frameworks to assess soil organic C, particularly valuable in the context of C neutrality (Lal, 2004, 2013).

In addition, Jerome Chave (191 citations, H-index 91) advanced allometric equations for AGB estimation based on tree diameter, height, and wood density. These models, validated with large datasets, have become the most widely applied tools for tropical forest C estimation (Chave et al., 2005; Chave et al., 2014). Marc Aubinet (177 citations, H-index 54) standardized EC data acquisition, processing, and error control through EUROFLUX, making it the international reference for ecosystem-scale C exchange (Aubinet et al., 2012).

These contributions established a multi-scale C estimation framework, spanning tree-level (allometric models), ecosystem-level (EC and soil respiration), and global-level (land-use and C cycle models). They remain highly applicable today and, when integrated with RS and AI, are expected to further enhance the accuracy and timeliness of forest C storage assessments (Supplementary Figure S1; Table 1).

From 2008 to 2012 (Stage 1), EC method, modeling, and RS technologies were widely applied to estimate forest C storage. Airborne LiDAR and allometric equations became the most applied technologies in Stage 2 (2013–2017). In Stage 3 (2018–2025), RF became the most frequently mentioned technology, followed by ML (Figure 5a). Notably, during this stage, forest inventory techniques, allometric equations, airborne LiDAR, digital soil mapping, and ML methods became active (Figure 5b). Among the group, both “ML” and “RF” methods registered the highest interest levels in recent years, with intensities of 13 and 9, respectively. Digital soil mapping and deep learning (DL) technologies first appeared in 2021 and 2022, respectively, and their popularity continues to the present day (Supplementary Table S6).

The keyword category was divided into four main clustering groups, with each cluster containing 103–140 keywords (Figure 5c). These clusters were ranked by their co-citation frequency, which included: (1). A cluster related to EC, covering 140 keyword nodes such as ML, spatial heterogeneity, and C budget, and with an S value of 0.8; (2). The soil organic C cluster, including digital soil mapping, RF, and other key technologies; (3). A cluster centered on radar technology, encompassing RS and airborne LiDAR, etc.; (4). A group representing ecosystem services, including forest logging and forest management (Figure 5c; Supplementary Table S7). Overall, “C sequestration” appeared as the most frequently mentioned keyword, occurring 299 times. “Forest” and “Climate change” appeared 276 and 264 times, respectively. In addition, “Storage” (210 times), “biomass” (176 times), and “EC” (162 times) also appeared as significant keywords (Figure 5d; Supplementary Table S8).

Modeling systematically extracted and analyzed data from selected articles. The trend in modeling accuracy over the three stages is presented in Figure 6a. In Stage 1 of the development of forest C storage estimation techniques (2008–2012), the average accuracy (R²) of the models was calculated to be 0.7. And by Stage 3 (2018–2025), the modeling accuracy had significantly improved, with an average R² reaching a value of 0.77. In particular, the accuracy of the TRIPLEX-Flux C exchange model significantly increasing over time, with R² initially being estimated to be 0.62 (Wang et al., 2012) and increasing to 0.97 (Chen et al., 2018) by the end of the final stage. Similarly, the R² of the RF model increased from 0.63 (Wu et al., 2016) to 0.97 (Zhu et al., 2023), while the R² of the 3-PG model was capable of reaching a value of 0.99. From a regional perspective, methods like Unmanned Aerial Vehicles (Zhang and Liang, 2020) and Global Forest AGB maps (Cunliffe et al., 2022) can be used for estimating large-scale C storage, with R² of 0.61 and 0.82, respectively (Figure 6b).

Between 2008 and 2025, 8, 16, and 67 countries conducted research closely related to C storage estimation at large, medium, and small scales, respectively (Figures 7a,c,e). During this period, China had an excellent performance in research publication volume at all scales, accounting for 30, 34, and 22% of the total publications in large, medium, and small scales, respectively. In comparison, the United States also showed high levels of activity in this research field and was ranked second overall. It contributed 10, 14, and 12% to publication volumes at the large, medium, and small scales, respectively (Figures 7b,d,f).

Before 2008, C estimation techniques primarily relied on ground plot surveys and biomass methods, which involved directly measuring plant dry weight or estimating C stocks using EC (e.g., NPP and respiration). The establishment of frameworks such as the ISO 14040/14044 and PAS 2050 further promoted the standardization of C accounting. In terms of biomass estimation models, the Brown model predicts forest biomass based on climatic parameters (e.g., temperature and precipitation), making it suitable for large-scale studies. The IPCC default model and biomass expansion factor (BEF) model provide general estimation methods based on vegetation type and parameters like tree height and diameter, offering essential tools for early C stock research.

Since 2008, driven by international agreements such as the Kyoto Protocol and the Paris Agreement, RS and process-based models have rapidly advanced (Falkner, 2016; Grassi et al., 2017) (Figure 5). EC systems (Longdoz et al., 2008) and C cycle models (Metsaranta and Kurz, 2012) have enhanced understanding at the process level, while RS has expanded coverage to larger areas (Wicaksono et al., 2011) (Figure 6a). However, the application of this technology is not balanced: developed countries have integrated these technologies into national monitoring systems, while many developing countries face challenges such as weak infrastructure, limited funding, and land tenure disputes (Gizachew and Duguma, 2016) (Figures 4a,b). These disparities are not only significant at the scientific level but also directly limit countries’ ability to fulfill reporting obligations under the United Nations Framework Convention on Climate Change (UNFCCC) (Köhl et al., 2015; Lewis et al., 2019) (Figures 2a,b, 3). Hence, the development of forest C sequestration estimation technologies reflects both changes in scientific capacity and policy frameworks. Early reliance on EC and simple C cycle models reflected technical feasibility as well as the urgency of reporting requirements under the Kyoto Protocol. While scientifically reliable, these methods were limited to site-level observations, thus unable to provide the scale required for national or global reports.

In the next stage, LiDAR (Antonarakis et al., 2017; Assefa et al., 2017), RS (Galidaki et al., 2017), and allometric equations (Neigh et al., 2016) (Figures 3b, 5a) became the main tools. The main reasons were economic and operational: LiDAR could provide detailed canopy structure information at increasingly lower costs (Simonson et al., 2016), while allometric equations offered a practical choose between accuracy and simplicity (Huy et al., 2022; Leite et al., 2020), making them suitable for national forest inventories (Johnson et al., 2022). However, these methods still have inherent limitations. The high cost of LiDAR restricts its application in developing countries, while allometric equations face regional applicability issues (Bossy et al., 2025; Nandal et al., 2023), often underestimating biomass in diverse tropical systems. From an applied perspective, allometric equations (Johnson et al., 2024) and EC techniques are more suitable for small-scale and detailed studies, while RS technology is indispensable for medium- and large-scale regional studies (Fu et al., 2015; Wang J. et al., 2023). This discrepancy highlights the ongoing tension between technical complexity and practical feasibility. The continued reliance on simpler methods is not due to inertia but reflects the socio-economic realities of national forest inventories and C offset markets.

In recent years (2018–2025), the shift towards ML, RF, and dynamic ecosystem models (Sun et al., 2025) (Figure 5d) has been driven by improvements in computational power, the rise of open-access satellite data (e.g., Landsat, Sentinel), and the growing demand for higher accuracy under the monitoring and reporting frameworks of the Paris Agreement and UNFCCC. These methods have overcome previous limitations, enabling the capture of temporal dynamics and handling increasingly complex RS datasets (Chen Z. et al., 2023). Unlike earlier static methods, dynamic models and ML integrate heterogeneous datasets, enhancing scalability (Rodríguez-Veiga et al., 2017) (Figure 7). The advent of ML further enhances the modeling techniques’ ability to handle complex data, challenging transparency and reproducibility, which are critical for international C reporting (Bharadiya, 2023; Yu and Wang, 2025).

Additionally, soil, as an integral component of forest ecosystems (Chen et al., 2024), also requires precise C stock estimation for comprehensive forest C assessments (Coops et al., 2025). Similarly, advances in digital soil mapping have improved the estimation accuracy of underground C stocks, but fragmented soil datasets still introduce uncertainty (Assefa et al., 2017; Avitabile et al., 2016) (Figure 6b). The challenges are not only technical but also institutional: balancing method innovation with the transparency and comparability required by measurement, reporting, and verification MRV (Measurement, Reporting, and Verification) systems. However, their effectiveness depends on data availability and transparency, and the lack of open-access high-resolution RS data (particularly in tropical regions) limits their applicability, exacerbating global inequalities in C monitoring capacity.

Overall, the temporal evolution of forest C sequestration technologies reflects a clear trajectory: from descriptive, labor-intensive approaches to dynamic, data-intensive modeling. Yet each transition has left unresolved issues—unequal access for developing countries, trade-offs between precision and cost, and the slow policy uptake of advanced algorithms. Addressing these challenges will be critical for ensuring that technical progress translates into more effective global climate governance.

At the spatial scale, the evolution of forest C stock estimation methods reflects a shift from localized plot studies to global monitoring networks (Figure 7). At the local scale, C stock estimates for forests mainly relied on manual data collection and simple models (Liang et al., 2022). Accurate ground measurements and ecological models allowed for highly precise estimates (Chave et al., 2005) (Figure 6), but these lacked scalability. While still crucial for validating RS data, these methods were costly and time-consuming. The advantage of this type of research lies in its ability to provide detailed ecological insights, especially revealing the impacts of climate change on forest C stocks (Cusack et al., 2016). However, relying solely on local studies can lead to fragmented knowledge, making it insufficient for supporting national-level reports and limiting its broader application in policy.

At the national scale, forest inventory data, containing detailed information about tree diameter, height, species, crown cover, and other parameters, combined with RS technologies, has made C stock assessment more systematic. Countries with established forest inventory systems, such as the United States and China, have used this data to design C sink policies and fulfill international reporting obligations. However, significant differences in governance capacities among countries exist. In countries with weak institutions, forest inventory data is often incomplete or outdated, reducing the accuracy of C estimates and impacting their credibility in international C markets. This disparity raises an important policy issue: Should global climate financing mechanisms invest more in building MRV capacity in developing countries rather than solely funding technological upgrades?

At the global scale, satellite-based Earth observation systems (e.g., MODIS, Landsat) and collaborative initiatives like GFOI and the Global C Project (Avitabile et al., 2016) have expanded monitoring coverage (Liu W. et al., 2024; Zhang et al., 2025), with integrating data from different countries being key for C stock estimation (Abbas et al., 2020; Tigabu and Gessesse, 2025). However, technical limitations still exist, such as the varying resolution of RS products, cloud interference affecting continuity, and data-sharing barriers hindering progress. Moreover, while global datasets are effective in monitoring long-term C dynamics, they often lack the socio-ecological details required for regional policymaking. Therefore, a key challenge is how to balance the broad scope of global monitoring with the depth of local social-ecological contexts.

Overall, the development trajectory of forest C stock estimation technologies is clear: from ground-based surveys to dynamic, data-intensive modeling, and from local studies to global monitoring systems. The expansion of forest monitoring at the spatial scale has brought both opportunities and constraints. Each technological evolution has improved accuracy and scalability but has also introduced new challenges: unequal access to data across regions, trade-offs between technical sophistication and economic feasibility, and the slow uptake of advanced algorithms in policy. Future progress should not rely solely on technological innovations but on addressing these systemic issues, including strengthening capacity-building in developing countries, improving data-sharing mechanisms, and ensuring transparency and comparability in advanced models. Only with integration at the technical, institutional, and socio-economic levels can forest C monitoring truly support effective climate governance and promote fair participation in global C markets.

Technological adoption in forest C estimation cannot be understood in isolation from broader drivers. Policy incentives, such as REDD+ and national C markets, have been instrumental in promoting MRV-related methods (Redd and Ecuador, 2012). In developed economies, significant public investment has accelerated uptake of costly RS systems, while in many African and Southeast Asian countries, weak infrastructure and limited financial resources restrict the deployment of advanced technologies (Ali, 2022; George, 2000). This imbalance raises concerns for the equity and credibility of global C accounting, as regions with the largest forest C stocks are often least equipped to measure them accurately (Gifford, 2020).

Socio-economic constraints also influence methodological choices. High-resolution RS and LiDAR demand substantial data storage, processing power, and technical expertise, which create barriers for under-resourced institutions (Di and Yu, 2023; von Essen et al., 2025). Data-sharing issues exacerbate this divide; proprietary restrictions on satellite datasets or national forest inventories impede collective progress (Gessler et al., 2024). In contrast, community-based ground measurements and low-cost allometric methods, while limited in scalability, continue to play a crucial role in local contexts where high-tech methods remain inaccessible (Brofeldt et al., 2014; Venter et al., 2015).

From an ecological perspective, methodological evolution reflects the need to capture increasing complexity. Static biomass equations overlook dynamic processes such as mortality, disturbance, and soil C fluxes. Recent digital soil mapping and deep learning approaches attempt to address these gaps, but they also introduce new uncertainties related to algorithm transparency and overfitting. This suggests that methodological progress has been less about linear “improvement” and more about navigating trade-offs between cost, complexity, and ecological realism (Ball et al., 2017).

Although this bibliometric synthesis offers a broad overview of advances in forest C storage estimation, several limitations warrant consideration (Huang et al., 2020).

Despite significant advances, several unresolved challenges limit the operationalization of forest C estimation techniques in policy contexts.

In summary, the evolution of forest C estimation technologies has been shaped as much by socio-economic and policy forces as by scientific innovation. To ensure their future feasibility, research must embrace interdisciplinary integration, global equity in data access, and stronger alignment with international climate reporting requirements.