The data is sourced from the global eddy covariance flux observation network FLUXNET2015 dataset. This dataset integrates observations of land–atmosphere interface fluxes from different regional flux networks using consistent processing procedures. It includes carbon fluxes, water fluxes, energy fluxes, and auxiliary meteorological variables. FLUXNET2015 is generated through the coordination of multiple regional networks and undergoes unified data quality control, gap-filling, and variable derivation processes, resulting in standardized products at various temporal resolutions (e.g., half-hourly, daily, monthly, and annual) for scientific research. Among these, the monthly products are aggregated from high-frequency raw data following a standardized workflow and undergo quality checks, preserving key flux variables and quality flag information to facilitate cross-site comparisons and integrated analyses. Based on the FLUXNET2015 database, a monthly flux dataset for Arctic flux observation sites was extracted, encompassing key ecosystem process variables at the site level, such as CO2, evapotranspiration, and energy exchange. All sample data were uniformly organized according to the SUBSET/FULLSET specifications of FLUXNET2015, ensuring consistency in variable definitions and quality control with the original data, while preserving monthly temporal resolution information. This makes the dataset suitable for seasonal analyses of carbon–water–energy cycles in the Arctic region and for integrated studies with other multi-source data.

 The processing of the dataset is based on the FLUXNET2015 data processing framework. Starting from the half-hourly or hourly data provided by the original flux sites, monthly-level products are generated through unified quality control, gap-filling, and temporal aggregation procedures. First, rigorous quality control is applied to the original flux and environmental variables, including preprocessing steps such as timestamp consistency checks, inter-variable consistency validation, and removal of unreasonable values. This ensures the consistency and comparability of data across different sites. Subsequently, the commonly used Marginal Distribution Sampling (MDS) method is employed to fill gaps in missing observational data. This method fills missing values by identifying observational records from periods with similar meteorological conditions, thereby preserving the characteristics of the original data to the greatest extent. Additionally, reanalysis data (e.g., ERA-Interim) can be optionally incorporated to fill longer-term gaps, enhancing temporal continuity and spatial consistency.

 Data quality control is a core step in ensuring the scientific validity and comparability of this Arctic monthly flux site dataset. This dataset adopts the unified data quality control framework of FLUXNET2015, which applies rigorous QA/QC procedures to conduct consistency checks and processing on the original high-frequency flux and meteorological observations from each site. The objective is to minimize errors arising from data measurement, missing values, and processing steps. During the quality control process, FLUXNET2015 provides quality flags (QC) and uncertainty quantification metrics for each variable. These are used to distinguish between original measured values and gap-filled values, as well as to indicate gap-filling results of different quality levels (e.g., high-quality filling, moderate-quality filling, or lower-quality filling). These flags are aggregated and summarized during the generation of the final monthly products, enabling the identification of high-quality observations or data segments with potentially higher uncertainty. At coarser temporal scales such as monthly, the quality flags indicate the proportion of original measurements versus high-quality gap-filled data within the given period, providing essential reference information for subsequent analysis and interpretation.The FLUXNET2015 processing pipeline also incorporates techniques such as gap-filling with reanalysis data for longer missing periods, energy balance adjustment, and multi-method filtering. These approaches enhance data completeness and consistency across multiple dimensions. Uncertainty estimates, quality flags, and processing notes for different sites and variables can be accessed in the FULLSET or SUBSET products. This information offers critical guidance for data users in evaluating the reliability and applicable conditions of the data.