A program to pilot different methods of combining local data with global scientific products and models to develop city-scale air quality analytics.
Cities around the world often struggle to find the financial and human capital resources to understand the levels, sources and impacts of air pollution. Many of the world’s most polluted cities have the least information about their air quality, preventing them from managing it effectively. However, there are a growing number of globally available, publicly funded, open resources for tracking, forecasting and attributing air pollution. These resources can be used to improve local access to air quality analytics, thus enabling action to reduce air pollution, improve lives and protect the environment.
CityAQ began as a partnership with WRI, NASA Global Modeling and Assimilation Office (NASA-GMAO), and eight cities to develop a globally scalable approach to local air quality forecasting for city-regions.
The project originally focused on generating air quality forecasts for eight cities and metropolitan areas, including Addis Ababa, Ethiopia; Bogota, Colombia; Jakarta, Indonesia; Kigali, Rwanda; Leon-Salamanca-Celaya Metro, Mexico; Monterrey Metro, Mexico; Guadalajara Metro, Mexico; and Sao Paulo, Brazil. CityAQ also advised Quito, Ecuador on how to best visualize and share near real-time air quality data.
CityAQ is creating scalable models for combining locally available air quality monitoring information with global modeling outputs, satellite products and other open analyses to develop customized tools for cities. Visit TheCityFix Learn to access a catalog of clean air learning products designed for city officials, civil society, practitioners and stakeholders.
Our first pilot combines local monitoring data with the outputs of NASA’s global GEOS Composition Forecast model (GEOS-CF) to develop optimized air quality forecasts for subnational air quality managers. By developing this product, we aim to:
1. Provide participating cities with useful air quality forecasts.
City officials and air quality managers can use this forecast to anticipate air quality events, communicate with stakeholders and more effectively manage local interventions.
2. Refine a methodology for combining locally held information with globally consistent analytics to offer new tools for city and regional decision-makers.
WRI supports cities in developing tools to share relevant data with NASA and other platforms such as OpenAQ. NASA applies a machine learning algorithm to local monitoring data and GEOS-CF model outputs to generate more accurate local air quality forecasts.
3. Develop the data infrastructure to make localized forecasts available and usable by all cities around the world.
The programming workflow was designed to ingest local monitoring data, combine it with the GEOS-CF model outputs and return the combined forecasts to an Application Programming Interface (API) developed by Development Seed. The API ensures that the forecasts are accessible to all users, including city participants and platforms such as Resource Watch. Access the API documentation for more information as well as a script developed by Resource Watch to download the data from the API.
4. Design a scalable approach for engaging with users to co-create air quality tools that leverage and extend the existing scientific analysis.
WRI engages with participating cities to conduct a qualitative assessment of user needs and use cases for locally corrected forecasts. Watch our AQ Tech Talk to learn more about how we developed city-level air quality forecasting in Mexico City.
*Disclosures: Factors affecting air quality can change over the course of a day, which can lead to differences between the forecast and the actual levels observed. The forecasts were adjusted using validated monitoring data where available. Where validated data was not available, the raw data observations were used to adjust the forecast model. A concentration level of zero indicates that the forecast is not available for that station. This is the case for those stations that didn't have validated or raw data of enough quality to train the machine learning algorithm. The performance of the forecast model for each monitoring site has not been determined.