Air pollution poses significant environmental and public health challenges, necessitating efective air quality data management. However, current approaches face limitations in ensuring data quality, interoperability, and compliance with FAIR (Findable, Accessible, Interoperable, and Reusable) principles. In this work-in-progress, we present a prototype of a knowledge graph-based system designed to enhance air quality data management across its entire lifecycle, from collection and validation to publication. Our approach integrates semantic web technologies to explicitly represent data provenance, quality dimensions, and interoperability requirements. We apply our system to a case study in Uruguay, where air quality data is collected from multiple organizations, highlighting the challenges of cross-institutional data integration and validation. Preliminary results demonstrate improvements in data consistency, traceability, and usability. Future work will focus on refining scalability, enhancing data quality inference mechanisms, and integrating additional environmental datasets.
Air pollution is a critical global challenge with significant environmental and public health implications.
Monitoring and managing air quality data are essential for assessing pollution levels, identifying trends,
and supporting policy decisions to mitigate harmful efects [
In addition to the above mentioned intrinsic complexities related to the management of air quality data, the processes involved in data management are generally built from a wide range of subprocesses that include policies and actors from many diferent organizations. Among others, these organizations are typically governments, private companies, nongovernmental organizations and independent laboratories. The roles of each of the organizations involved are often described in applicable laws, commercial contracts, international agreements between counties, etc. Sometimes, roles arise from ad hoc exchange between organizations with no clear responsibilities or accountability for the quality of the data.
In this work, we present a prototype of a data management system designed to address the entire lifecycle of air quality data, from collection and validation to publication, where multiple organizations are involved. Our approach leverages knowledge graphs and semantic web technologies to enhance data integration, representation, and usability. A key novelty of our system is the explicit representation The 3rd International Workshop on Knowledge Graphs for Sustainability (KG4S2024) – Colocated with the 22nd Extended Semantic https://www.fing.edu.uy/~lorenae/ (L. Etcheverry)
CEUR
ceur-ws.org of air quality data quality dimensions using RDF, which enables structured, machine-readable metadata about the reliability and accuracy of air pollution measurements. This approach ensures transparency in data provenance and facilitates traceability across the data lifecycle, enabling users to assess the trustworthiness of air quality information efectively.
As this is an ongoing research efort, we present a prototype implementation along with preliminary results that showcase the feasibility of our approach. The current system provides an initial framework for integrating and managing air quality data, and future work will focus on refining its scalability, improving inference mechanisms for data quality assessment, and expanding interoperability with external environmental datasets.
The remainder of this paper is structured as follows: Section 2 discusses related work on air quality data management and semantic technologies. Section 3 presents our proposed system, describing its architecture and data processing pipeline. Section 4 describes the application of our approach to a case study involving Uruguayan air quality monitoring, and Section 5 concludes with insights on further work and future research directions.
Several studies focus on air quality management from a methodological perspective; however, they do
not specifically emphasize the information systems required for efective implementation [
Previous work already explored using knowledge graphs and semantic web technologies for air
quality data management. Still, none focuses on data provenance and/or modeling data quality and
validation processes. In [
Our approach integrates existing semantic web vocabularies to comprehensively represent the diferent aspects of the lifecycle of environmental data, specifically air pollutant data. By leveraging wellestablished ontologies, we ensure interoperability, provenance tracking, and structured data quality representation, aligning with FAIR principles. This section outlines the key components of our system, including the semantic vocabularies used, the system architecture, and the mechanisms for ensuring traceability across the data lifecycle. In particular we adopt concepts from the following ontologies and volcabularies
To efectively model and manage air quality data, we adopt a combination of semantic web vocabularies,
each addressing specific aspects of the data lifecycle:
1Semantic Sensor Network Ontology (SSN) https://www.w3.org/TR/vocab-ssn/
• SSN The Semantic Sensor Network (SSN) 2ontology is an ontology for describing sensors and
their observations, the involved procedures, the studied features of interest, the samples used to
do so, and the observed properties, as well as actuators. Specifically, the SOSA Module (Sensors,
Observations, Samples and Actuations) defines the core classes and properties, and provides
textual definitions and other annotations. This ensures consistency in representing air pollutant
concentrations, measurement units, and sensor specifications.
• PROV-O: The PROV Ontology3 is employed to model provenance information, including dataset
relationships and data generation processes. This enables traceability and accountability
throughout the data lifecycle.
• DQV (Data Quality Vocabulary)4: Integrated to represent data quality dimensions, metrics, and
values explicitly. These vocabularies allow us to encode quality metadata in a machine-readable
format, facilitating automated quality assessment.
• DataCube5 and QB4OLAP[
By combining these vocabularies, our system provides a unified framework for managing air quality data while adhering to FAIR principles. In order to distinguish raw data from validated data we will use the following terminology: i) Registers represent data points in the data staging area, and represent the output of a sensor at a certain date and time, while ii) Measures represent data points in the validating area, which are derived from the information contained in a Register.
We organize our approach based on three ontologies that reuse and extended concepts from existent ontologies and vocabularies: • AIRQorg - That allows to represent all relevant institutional information, including monitoring stations, agreements between institutions, and personnel involved in related tasks. • AIRQreg - That models all the registers as they are generated from the information sent from the
Sensors, including geographical information, pollutant, etc. • AIRQmed - That models all the validated measures generated from the registers. It also models the quality requirements (e.g. the value is within the sensor operating range) and provides information on the validation process, in particular on which agents participated in it.
Our system follows a structured data management pipeline that ensures the traceability, quality, and FAIR publication of air quality data. It is designed around three main components: Data Staging Area, Data Validation Area, and Data Publication, each playing a critical role in the data lifecycle. Figure 2 depicts the data flow through these components, showing how the proposed ontologies take part in the diferent stages. In the following subsections we outline the main responsibilities of each component.
The Data Staging Area is the initial entry point for raw air quality data collected from measuring stations and third-party environmental agencies. This component handles: 2Semantic Sensor Network Ontology (SSN) https://www.w3.org/TR/vocab-ssn/ 3PROV-O: The PROV Ontology https://www.w3.org/TR/prov-o/ 4Data on the Web Best Practices: Data Quality Vocabulary https://www.w3.org/TR/vocab-dqv/ 5The RDF Data Cube Vocabulary https://www.w3.org/TR/vocab-data-cube/
• Data Ingestion: Automated pipelines retrieve sensor data in various formats (CSV, JSON, XML, RDF). These pipelines are designed to handle heterogeneous data sources, ensuring seamless integration into the system. • Preprocessing and Harmonization: Raw data is standardized to ensure consistency in units, timestamps, and formats. The AIRQorg and AIRQreg vocabularies map diverse data representations into a unified model. • Initial Provenance Tracking: Metadata about data sources, timestamps, and ingestion processes is recorded using features from PROV-O included in AIRQreg. This ensures traceability from the moment data enters the system, providing a foundation for end-to-end provenance tracking.
Once ingested, data is transferred to the Data Validation Area, where quality evaluation is performed based on predefined criteria. This component includes: • Data Quality Assessment: Quality dimensions such as accuracy, completeness, consistency, and timeliness are evaluated using DQV and BigOWL4DQ. These vocabularies enable the explicit representation of quality metrics and their values, ensuring transparency in quality assessment. • Anomaly Detection & Correction: Missing values, outliers, and sensor errors are identified using statistical and rule-based methods. Automated correction mechanisms are applied where applicable, while unresolved anomalies are flagged for manual review. • Provenance and Traceability: All validation steps, including quality assessments and anomaly corrections, are logged using PROV-O. This ensures that transformations, quality evaluations, and modifications remain traceable throughout the data lifecycle.
Validated and enriched data is published in a structured and queryable format, supporting multidimensional analysis for decision-making. This component incorporates: • Multidimensional Modeling: Using DataCube and QB4OLAP, air quality data is organized into multidimensional structures. This allows users to aggregate and analyze data by dimensions such as time, pollutant type, and geographical region. • Linked Data Integration: Air quality data can be connected with external datasets (e.g., weather conditions, industrial emissions) to enable enriched analysis. This integration is facilitated by RDF-based linking mechanisms, ensuring interoperability across datasets. • FAIR Data Accessibility: The system ensures Findability through metadata indexing, Accessibility via standard SPARQL endpoints, Interoperability through RDF-based vocabularies, and Reusability with well-defined licensing and quality metadata. These features make the data accessible to a wide range of stakeholders, including researchers, policymakers, and the public. Finally, to ensure end-to-end traceability, we implement persistent identifiers by assigning unique URIs to datasets, observations, and validation results, ensuring long-term referenceability. Versioning and lineage tracking are managed using PROV-O to capture dataset versions, modifications, and transformations, allowing users to track changes and their impact. Additionally, provenance metadata is accessible through SPARQL queries, enabling stakeholders to verify data history and reliability, while access control mechanisms protect sensitive information while maintaining transparency for authorized users.
To evaluate our approach, we apply it in a case study based on the national agency responsible for air quality monitoring in Uruguay, focusing on data collected near industrial regions. This setting presents unique challenges regarding data validation, provenance, and cross-institutional data sharing, further highlighting the need for robust, FAIR-aligned data management solutions. As this is an ongoing research efort, we present a prototype implementation along with preliminary results that showcase the feasibility of our approach. The current system provides an initial framework for integrating and managing air quality data, and future work will focus on refining its scalability, improving inference mechanisms for data quality assessment, and expanding interoperability with external environmental datasets.
Air quality monitoring in Uruguay is managed by national and local agencies, with the Ministry of
Environment (MA) overseeing regulations and compliance. Several initiatives provide public access to
air quality data but face challenges in standardization and data quality. The Observatorio Nacional
Ambiental (OAN)[
Air quality data collection in Uruguay is fragmented, with institutions using heterogeneous formats and varying management criteria. The lack of a shared ontological framework hinders interoperability, while sensor drift, missing values, and inconsistencies afect data reliability. No unified methodology ensures traceability, versioning, or comprehensive quality assessment. Existing platforms prioritize data publication but ofer limited analytical tools for environmental professionals. Assessing pollution trends, detecting anomalies, and correlating industrial activity with air quality remain challenging. These issues not only impact decision-making but also overburden scarce human resources, leading to ineficiencies and compounding operational challenges.
To implement a Proof of Concept, we utilized data provided by the MA for the years 2021 to 2023, sourced from twelve air quality measuring stations. The MA dataset includes information for 11.7M records on 10 diferent pollutants, collected from 12 diferent stations, all located on Uruguayan territory. It is relevant to notice that the dataset had been built on a few basic guidelines as was meant to be used internally in the MA in a particular ad hoc scenario and was not intended to be shared with third parties. These characteristics of the source dataset should make this implementation a relevant Proof of Concept. Much of the mandatory information needed to apply our approach is missing from the dataset, notably the nature of agreements between involved organizations, identification of agents (personnel, software, or other) who reported the register information, and the review processes the registers underwent. Some of this information was intentionally left blank for later inclusion, while default entities were created for other fields, such as personnel inserting data in the landing or validation stages.
We have implemented a prototype of our system, including: a data ingestion pipeline capable of processing real-world air quality datasets, a validation module that applies data quality checks to detect anomalies, and a semantic data model ensuring compliance with FAIR principles. Data Quality measures were implemented. Validity on the values inserted, basically verifying values are within sensor’s range. Accuracy for valid measurements, checking outliers and completeness for sets of measurements, dividing the time window into a number of slots and verifying existence of measurements for each of the slots.
Preliminary results demonstrate improved data consistency and completeness compared to raw datasets, enhanced traceability of data sources and transformations that increase trust in the published data, and the ability to generate multidimensional reports for analyzing pollution trends over time. Upon loading the first subsets of the source dataset into the staging area, we issued SPARQL queries to identify registers that did not comply with basic requirements, such as values reported outside the sensor’s operating range and multiple data points for the same sensor and time. For example, a data point representing the current valid reading for the pollutant PM10 (particulate matter under 10 ) at the station DU_PDT2 (Paso de los Toros) with a reading time 2021-01-01T00:50:00 had previously been reported nine times over ten months. Morevoer, we can provide SPARQL queries to retrieve all the previous registers and the reasons why they where overuled by subsequent registers, a feature that is not currently available in the data management systems employed by the MA. Each register linked to the readings is also linked to an Agent (person, software, etc.), allowing for further analysis of the reasons behind these repeated observations. Our system is still in early-stage development, with ongoing eforts to scale the solution for real-time data streams, integrate external environmental datasets, such as meteorological information, and to improve visualization tools for technical staf and policymakers.
This work presents a knowledge graph-based approach to enhance air quality data management by integrating semantic web technologies. Our prototype demonstrates improvements in data consistency, traceability, and interoperability, addressing key challenges in standardization and quality assessment. Preliminary results highlight the feasibility of this approach in the Uruguayan context, showcasing its potential to facilitate FAIR-compliant data publication and enable better-informed decision-making. Future work will focus on scalability, refining quality assessment mechanisms, and expanding interoperability with additional environmental datasets.
During the preparation of this work, the authors used Mistral, chatGPT-4 and Grammarly to do Grammar and spelling checks. After using these tool(s)/service(s), the author(s) reviewed and edited the content as needed and take(s) full responsibility for the publication’s content.