Beijing and IBM Using Satellite and Social Media Feeds to Forecast Air Pollution

While denying it in public until recently, the authorities have been aware of the problem for a long time: when Beijing was chosen to host the 2008 Olympic Games, they took major steps to reduce pollution levels in the city, halting constructions works, shutting down factories and power plants, imposing alternate-day driving rules. Read the full report @ Forbes

Those measures sure had a positive effect on the situation, leading the International Olympic Committee chief to praise the efforts, saying China had done “everything humanly possible” to reduce PM2.5 levels (which measure particulate emissions of carbon, nitrogen, sulfur, and heavy metals).

But such high-impact provisions are not easy to implement and come with a huge economic cost.Despite claims to the contrary, they ended up being more one-shot measures than a real solution, and the ‘blue skies’ didn’t last: just a few years later, things seemed to be just as bad as before, forcing the local government, in December 2015, to issue its first ever ‘red alert’.

While the problem is still far from be solved, then, a glimmer of hope for residents might come from a ten-year collaboration between the Beijing’s Environmental Protection Bureau (EPB) and IBM IBM +1.60%, a joint effort which goes under the name of “Green Horizons“. The project, which started in 2014, is focused on using IoT and cognitive computing to improve air quality management and forecasting.

“Using the machine learning technology developed for Watson, we take huge amounts of data from weather stations, satellites social media, in Beijing, and we can not only identify exact pollutants and their sources, but also create amazing correlations in the data,” IBM’s general manager of the Watson project, Harriet Green tells me. The forecasting window, with this new technology, has been extended from 2 to 10 days and the accuracy of the estimations jumped from 60% to 80%; this improvement was obtained, partly, combining traditional ground sensors information with data gathered from social media.

“Ground sensors actually are not enough to capture the whole picture of the pollution event, where does it come from, and where will it impact in the next few days. We use also data coming from the Chinese version of Twitter and Instagram to do some cross checking, and get a better understanding of the situation,” Zhang explains. A post on Sina Weibo or Tencent, a picture, a camera feed: everything is processed and analyzed.

Extra - Air Quality Forecasting in Delhi

Besides accuracy, the advantage of this cognitive, machine learning approach is that gives authorities some leverage to plan in advance which measures to take to reduce pollution and allows more targeted, tailored interventions than just shutting down more than 100 factories for weeks, like it happened in 2008.