Abstract View
Multi-Modal Data Sharing and Synthesis for Low-Cost Sensors
ALIAKSEI HAURYLIUK, R. Subramanian, Carl Malings, Ashley Angulo, Randy Sargent, Ana Hoffman, Albert Presto, Carnegie Mellon University
Abstract Number: 314
Working Group: Translating Aerosol Research for Societal Impact: Science Communication and Public Outreach
Abstract
A network of Real-Time Affordable Multi-Pollutant (RAMP) sensors has been continuously operating in the Greater Pittsburgh Area since August 2016. In an effort to publicly disseminate the collected data, we developed or partnered in developing several methods of presenting the data. Each method addresses a different use case and target audience, with the consideration of the audience’s level of engagement and comfort with data analysis.
The methods include:
1. Sending location-specific air quality alerts based on data from the sensors
2. Displaying interpolated air quality data on a publicly-accessible map
3. Publishing aggregate data in weekly and monthly reports
4. Sharing processed data with OpenAQ and CREATE Lab’s Environmental Sensor Data Repository (ESDR) platforms
5. Making freely available the software tools for collecting and processing the raw data from the sensors, and providing documentation and tutorials for their use.
The alerts and map are targeted at the general public, while the reports are most suited to answer questions from sensor hosts and community groups. Processed data are intended for power users, while raw data and software tools have researchers in mind.
This approach utilizes multi-modality in an attempt to answer different questions regarding the current or past state of air quality in the region. It relies on aggregated historical data for providing context and analysis on near-real-time readings to address more immediate concerns about pollutant levels. Tools like recurring reports provide a line of communication with the sensor hosts and community groups, who are concerned with air pollution trends in their area. Furthermore, data aggregation can be used to identify undesirable sensor operation or conditions which result in poor corrective model performance. This is useful both for quality control internally and communicating uncertainties to the consumers of data from low-cost sensors.