Notebook: OpenAQ data overview
/notebooks/01_openaq-data-overview.ipynb/notebooks/01_openaq-data-overview.py
OpenAQ provides access to thousands of real-time air pollution measurement stations via a free API. For our target list of cities around the world, we want to explore the following questions:
- How many cities are covered by the OpneAQ database for individual pollutants?
- How homogeneous is the data across different stations for the same city?
- Will spurious measurements be a problem? How can we detect outliers?
- fetch all available data for the last 7 days for each city from the OpenAQ API
- save all data for stations providing any of ["o3", "no2", "so2", "pm10", "pm25"] within a radius of 25 km (i.e. API max)
- plot global maps of cities with available data and time series for all stations for the last 7 days grouped by city
- no data for any pollutant at any point over the last week for 50 out of the 153 cities (and some others with single/dodgy stations)
- best coverage for Europe, North America and Eastern Asia
- average of
$\sim$ 9 stations per location - most measurements for
$PM_{2.5}$ , fewest for$SO_{2}$ and$O_{3}$
- consistency between stations within the same city/region highly variable (e.g. see below for dominant diurnal ozone cycle in Warsaw; large spread in Athens, non-stationary):
- air quality stations likely clustered around pollution hot spots, e.g. see
$NO_{2}$ peaks around large streets (Hallein A10, Hallein B159) in Salzburg:
- large spread across stations for all pollutants in Lima (see below)
- e.g. for
$PM_{10}$ , choice of station/averaging makes difference between AQI level 1 (<20) or 6 (>150)
- some outliers and missing values which are not masked are easy to identify, as they are either 0 or -1000:
- but sometimes spikes of 2-3 data points are difficult to judge/interpret:
- no useful data for any pollutant for
$\sim$ half of all cities in OpenAQ data -> might need to consider additional data sources - spurious data/outliers for most cities -> robust quality screening will be necessary to detect and remove these data
- large spread between sensors within some cities -> develop methods to generate city-level estimates