We used the open VIIRS Aerosol Optical Thickness (AOT) data, which permits a measure of average aerosols suspended in the air over each SEZ. AOT is suitable for ambient air quality applications because is related to particulate matter (PM) concentrations in the atmosphere; in this case, higher values of AOT will correspond to higher PM. VIIRS AOT products are typically used to study daily air quality forecasting, wildfires, dust storms, or haze episodes. VIIRS AOT data is provided at different wavelengths, as particle scattering varies by wavelength. For purposes of Phase I, we use the VIIRS AOT measurement at 550 nm, around the middle wavelength over which AOT is measured. Relative changes in AOT can be used to examine pollution changes over time. Unfortunately, VIIRS data is relatively low resolution (at 6 km/pixel), but due to this low resolution, we expect any algorithm to scale to multiple zones fairly easily.

Conclusions

We found that we could obtain a reasonable estimate of air pollution using the open VIIRS AOT data, which seems to show daily fluctuations. Like the daily nightlight data, however, we were forced to manually search and query datasets as we have no indication if there are clouds or other artifacts and then georeference the data obtain data from SEZ regions.

In the future, we would like to correlate the AOT data with sensors on the ground (e.g., the network of OpenAQ sensors) measuring particulate matter at a variety of sizes. This will allow us to ensure that we can accurately estimate air quality from such a small sample subset of the data. For example, the AOT measurements (possibly using multiple wavelengths) would be combined with ground training data to develop a model transforming observed AOT particulate matter concentrations. Without this additional data, it’s difficult to be confident in our estimate. We’d also want to further investigate how AOT data, which measures 550nm particles, will correlate with other accepted metrics like PM 2.5, which are of particular interest to some groups given their implication in causing health problems. Again, ground-based sensors would be very helpful for creating this type of model, but currently, these sensors are unlikely to be co-located within the zones of interest.