Abstract:

[Objective] With the increasing precision of environmental monitoring and the rapid advancement of multi-source observation systems, in situ observational data have become increasingly essential for atmospheric sciences, Earth system modeling, and environmental assessment. However, during data acquisition, transmission, and processing, in situ datasets are prone to missing observations, abnormal values, and non-standard formatting issues, which may degrade their reliability and limit their scientific utility. [Methods] This study utilizes the Level 1.5 datasets from the Atmospheric Composition Monitoring Network and the Aerosol Robotic Network, both released by the National Earth Observation Science Data Center of China, to construct a multi-dimensional quality evaluation framework for in situ datasets. The evaluation system is organized along four major criteria—data compliance, data missing conditions, data mis-measurement conditions, and potential data falsification—and is further operationalized into 15 refined quality indicators. The conceptual connotations of each quality indicator are systematically defined, and quantitative assessment is conducted by applying the MSCRED* unsupervised anomaly detection model as well as the K-nearest neighbor method for indicator-specific quantitative evaluation. [Results] Empirical evaluations were conducted on 1,468 observational stations included in the target dataset. The results show that the proposed evaluation system is capable of effectively revealing the quality heterogeneity among stations, and is able to characterize both the global and regional distribution patterns of in situ data quality levels. [Conclusions] This study proposes a multidimensional quality assessment framework and a quantitative integration approach tailored for in situ datasets. The framework can provide methodological support for diagnosing the current status of data quality, optimizing quality control procedures, and facilitating subsequent scientific applications.

Key words: in situ datasets, quality assessment, multidimensional indicator system, unsupervised anomaly detection, data authenticity