Recently, a new weather station has been created that runs a neural network that can predict air quality according to local ozone concentration. Although you’ve seen home weather stations before, you’ve never developed one to these specifications.
The initial idea was from self-taught full-stack developer Kutluhan Aktar, who set out to design a low-cost weather station that predicted air quality levels. In this sense, tropospheric ozone is an ideal measuring element as it is potentially dangerous, and can cause breathing difficulties and trigger asthma attacks.
Ozone occurs at the surface of the ground when emissions from vehicle leaks and industrial plants react in the presence of sunlight, so its presence is a useful marker that its precursors are also floating.
With a weather station and ozone detector on his balcony, Aktar was able to collect a data set with an Arduino Nano 33 board and send it via Bluetooth to a Raspberry Pi 4 installed inside. This was used along with the local air quality index to train an artificial neural network based on TensorFlow Kera H5 so that it could make predictions of outdoor air quality only from meteorological and ozone data.
This weather station displays all the data it collects, such as ozone concentration, wind speed, temperature, and atmospheric pressure. It also generates air quality estimation graphs according to three categories: good, moderate, and unhealthy.