Global Air Quality Forecast

Despite the substantial negative impacts of air pollution, many countries lack reliable air quality information due to the prohibitive expense of establishing and maintaining the necessary infrastructure. A global air quality model forecasting system can provide air quality information to everyone.The NASA Global Modeling and Assimilation Office (GMAO) develops and maintains the GEOS system of models, which has a suite of capabilities including simulating weather, climate, chemistry-climate interactions, and now air pollution. They assimilate satellite and in situ data of some variables that affect weather and air quality into the GEOS model. The assimilation of these data improves the simulation of weather and air quality by providing the best representation of the atmosphere at the start of the forecast period.

⇒ Visualize the NASA composition forecasts for your city using FLUID.

Forecast System Description

• The GEOS Forward Processing (GEOS-FP) numerical weather prediction (NWP) model produces weather and aerosol analyses and forecasts in near-real time for research purposes.The GEOS-FP atmospheric composition forecasts are publicly available.  Once on the forecasts website, please click on "Chem Maps" or "Datagrams" under "Navigation". For instance, under "Datagrams" you can view  model output for various measurement stations and the world's megacities.
• The GEOS Composition Forecast (GEOS-CF) model produces high-resolution global air quality forecasts in near real-time. The innovation of the GEOS-CF system stems from the inclusion of Harvard University GEOS-Chem reactive gas-phase chemistry, which allows for the simulation of reactive gases (e.g., O₃, NO₂, VOCs) along with aerosol species already included in the GEOS-FP model. The GEOS-CF model produces 5-day forecasts of PM_2.5, O₃, NO₂ and many other air pollutants on a global scale with unprecedented spatial resolution of about 25x25 km². While this spatial resolution is not as fine as the resolutions of current regional air quality models, the model forecasts are for the entire globe. The near-term (< 2 year) goal is to move to about 12x12 km² resolution forecasts, which will be made possible by continuing advances in computing resources and efficacy improvements in the chemistry model. For instance, five years ago it was only possible to run GEOS-Chem globally at 200x250 km² horizontal resolution. Looking forward, the long-term goal for the GEOS-CF model is to continue to move to finer spatial resolutions -- both horizontal and vertical -- equivalent to the regional air quality forecasts. 

Forecasting and Risk Communication: Air Pollution in Your City

Forecasted air pollution values of PM_2.5, NO₂, O₃, and SO₂ are available for your city up to five days in advance. A health-based air quality index is also available to help with local risk communication. These values can be specifically tailored to your city by sharing local air pollution and daily health data. [more details]

Potential Applications: an Air Quality Index (AQI)

The GEOS-CF output could be used to generate an AQI for cities around the world. For example, the animation below shows the Canadian HAQI as applied to the output.

Health Air Quality Index (HAQI) calculated using the GEOS-CF model output for the time period October 10-23, 2017. The HAQI comprises of surface concentrations of NO₂, PM_2.5, and O₃, using the formulation by Stieb et al. (2008). Values range from 0 (white, good) to 10 (hazardous air quality conditions). The time frame captures extensive wild fires over western Canada, California and Portugal, as well as tropical storm Ophelia (October 15-16) transporting highly polluted air masses to the British Isles. Animation credit: Dr. Emma Knowland, NASA GMAO. 

Evaluating the Performance of the Air Quality Forecasts

The GEOS-FP and GEOS-CF models tend to perform well as compared to surface observations where air pollutant emissions from automobiles, industry, and power generation are characterized well, such as in places like the U.S. and Western Europe.

The models also tend to perform well at simulating wildfire and agricultural smoke events that affect whole regions, in large part to the assimilation of satellite data of aerosol optical depth and weather-related variables. For instance, the animation shows the transport of smoke from seasonal agricultural fires over India and PM_2.5 levels in Delhi. The surface PM_2.5 data from an air quality monitor at the U.S. Embassy. Karambelas et al. (2018) estimate that in northern India 463, 200 adults die prematurely each year from PM_2.5 and that 37,800 adults die prematurely each year from O₃. 

Modeled surface concentrations of particulate matter with diameter of less than 2.5 microns (PM_2.5), expressed in units of microgram per m³, for Delhi, India between February 1-28, 2018 (top panel). GEOS-CF model 1-day hindcast and five day-forecasts are performed once a day. These concentration time series are shown in the bottom panel. Modeled concentrations are shown in dark red (hindcast) to yellow (5-day forecast). Corresponding observations (from the OpenAQ database) are shown in black. Animation credit: Dr. Christoph Keller, NASA GMAO. 

However, for most of the world's cities, urban emissions are not well characterized and there are little or no publicly-available air quality data (e.g., OpenAQ) with which to evaluate the air quality forecasts. The animation shows that the model tends to over-predict observed PM_2.5 from the U.S. Embassy monitor in Jakarta.  Therefore, we are beginning to work with Indonesian scientists to benefit from their local expertise to, for instance, improve the pollutant emissions used in the model, to understand the unique meteorological variations that may occur there, and to access their air quality data, which are not publicly-available. We are also working with Brazilian scientists to improve the forecasts for Rio de Janeiro.

Modeled surface concentrations of particulate matter with diameter of less than 2.5 microns (PM_2.5), expressed in units of microgram per m³, for Jakarta, Indonesia between Oct 26 and Nov 10, 2017 (top panel). GEOS-CF model 1-day hindcast and five day-forecasts are performed once a day. These concentration time series are shown in the bottom panel. Modeled concentrations are shown in dark red (hindcast) to yellow (5-day forecast). Corresponding observations (from the OpenAQ database) are shown in black. Animation credit: Dr. Christoph Keller, NASA GMAO. 

Leads: Dr. Christoph Keller (christoph.a.keller@nasa.gov; USRA/GESTAR, NASA GMAO; not HAQAST funded), Dr. K. Emma Knowland (k.e.knowland@nasa.gov; USRA/GESTAR, NASA GMAO; not HAQAST funded)