Data Quality Assessment and Management applied to Community-Level Air Quality Monitoring Data

Abstract:

Centralized data management repositories must employ well-designed Data Quality Management Plans to ensure that publicly available air quality data products are assigned proper data confidence levels. Community-level air quality monitoring presents new challenges for quality assurance systems in centralized data repositories. Wide disparities in the grade of monitoring platforms and site-level quality assurance (QA) demand wider scope in quality control (QC) methods and tools at the central database level. With the introduction of commercially accessible, low-cost sensors (LCS) available to the public, the challenge of evaluating data quality at the centralized database level requires an expanded set of QC methods.

CARB’s Community Air Quality Viewer (AQview) is a centralized database repository and public data portal for storing, providing, and visualizing community-level air quality data collected to support California’s AB 617 legislation. In contrast to regulatory air quality monitoring, the monitoring data collected to support AB 617 is done using a variety of instruments ranging from regulatory and research grade monitors to networks consisting of many LCS devices. This presentation will describe elements of AQview’s Quality Assurance Data Management Plan for handling the quality assignment of data stored in AQview at both the single record and derived data product level. AQview incorporates a staged Level 1 and Level 2 QC pipeline consisting of sub-hourly and hourly data branches. QC methods applied to raw sub-hourly data records will be described and contrasted with methodologies applied to hourly averaged data sets. A discussion on the application of outlier detection tests based on both parametric statistical models and domain-based non-parametric models will also be given along with example test case results. This presentation will include commentaries on the use of a data visualization tool specifically designed for support in evaluating the performance and optimization of QC checks, an automated methodology for assessing changes in site-level data confidence over time (sensor degradation and drift), and innovative, predictor-based outlier detection methods currently being explored. A description of the computational environment and technologies for supporting AQview’s QC pipeline will also be presented.