Predicting Air Pollution in Ulaanbaatar, Mongolia — Part I, Introduction
This is part 1 of a 4 part series on predicting air pollution (specifically PM2.5 levels) in Ulaanbaatar, the capital city of Mongolia. Part 1 details the problem of air pollution, some solutions that have been proposed and attempted, and my idea to predict pollution levels using machine learning. Here you can find Part 2, Part 3, and Part 4.
For those living in Ulaanbaatar, the capital of Mongolia, air pollution is a fact of life during winter months. This choking smog blots out the sun and smells of burnt coal. The smell lingers in your hair and on your clothes. The source of the pollution? Raw coal burning stoves located in the outskirts of the city, called ger districts (named for Mongolia’s nomadic yurt dwellings). While coal fired power plants and vehicles contribute as well, the primary source is for cook and heating fires in homes around the city.
Those living in the ger district heat their gers or self made homes with stoves originally meant for wood or dried animal dung. These stoves don’t efficiently burn the coal, putting a large amount of particulate matter into the air. Using raw coal is an alternative chosen out of necessity. The harsh winters and price of wood makes these fuel sources impractical.
Solving air pollution is a huge challenge for Mongolia. Residents of the ger district often have no choice but to live in Ulaanbaatar, the largest city and main economic engine of the country, in order to support their families. The city was designed for about half a million people, but today 1.4 million people reside in the city. This is nearly half the population of the country.
The health impact of air pollution is quite severe. It is estimated that 29% of cardiopulmonary deaths and 40% of lung cancer deaths are attributable to air pollution (link). Mothers and young children are especially vulnerable. During winter months (when pollution is highest) fetal deaths increase 3.5 times and birth defects are more common.
A recent report from UNICEF found the cost of treating air pollution related illnesses and indirect costs of missed work to be 5.4 billion MNT ($2.2 million USD) per year.
“Children living in a highly polluted district of central Ulaanbaatar were found to have 40% lower lung function than children living in a rural area.” — UNICEF
One of the earliest solutions for solving the air pollution problem was to replace the inefficient stoves with new clean burning stoves. This project, starting in 2013 and finishing in 2018, is funded primarily from the World Bank through a $15 million USD loan. These stoves have been mostly panned by those who actually use them, while a report by the Millennium Challenge Corporation has praised their adoption.
These stoves greatly reduce the output of particulate matter compared to the traditional stoves. However they are slow to heat and offer a small cook surface. In addition they are only able to be filled from the top meaning cooking and refilling of fuel are not practical together.
Other solutions focus on educating the public. Project MASC is an organization that promotes the use of face masks that filter pollution and indoor air filtration. A recent Facebook event hosted by the US Embassy in Mongolia had experts discuss the health impacts of air pollution and the use of face masks and air filters.
The US Embassy is launching a public social network campaign on the promotion of public awareness on the effects of air pollution. The campaign broadcasts each Friday at 10:00 AM. www.facebook.com
Perhaps the only true way to truly reduce air pollution is to eliminate the problem itself, Ulaanbaatar’s ger districts. Yet this doesn’t seem feasible given the fact that apartment prices remain out of reach for most of Ulaanbaatar’s residents. As a result tens of thousands of apartments remain empty.
Discovering Pollution Myself
As a resident of Ulaanbaatar it is a daily ritual of mine to check the air pollution levels each morning before leaving for work. This information comes from two sources, the government air pollution monitoring stations (Agaar.mn) and the US Embassy PM2.5 station (stateair.mn). I use the Air Matters iOS app and it delivers push notifications throughout the day as air pollution changes around me (Agaar.mn has an app but it doesn’t give push notifications).
These measurements help my wife and I inform our day and especially our weekends. Do we go to the playground with our son? Should we wear a face mask? Can we open the windows at work or in our apartment?
For the past year I have observed pollution here in the center of the city where I live and I have noticed several things:
- While winter months have the worst air pollution, spring and fall have bad times as well.
- The air pollution levels change quite a lot over the course of the day. Morning and evenings are the worst.
- Air pollution stations frequently don’t work, especially when temperatures are very low.
- Different parts of the city have very different levels of pollution.
- The same pollution measures are not recorded at each station. Several stations do not have PM2.5 monitors.
Predicting Air Pollution
Making personal observations is all fine and good, but it doesn’t give a rigorous understanding of air pollution. I wanted to learn what environmental conditions have an impact on air pollution and how those conditions can be used to predict PM2.5 levels in Ulaanbaatar. Why PM2.5? First, it is one of the most dangerous forms of air pollution and is the most common in Ulaanbaatar. Second, it is the form of pollution tracked by the US State Department at the US Embassy here in Ulaanbaatar, and they make their data available to the public in a very easy to use format.
“When it’s cold, the pollution is bad.” — Typical person
When I first discussed this idea with expats and Mongolians, the majority of the response was a bit disheartening. Most seemed resigned to the fact that in the winter pollution is always bad and that predicting it would yield a predictable result, “When it’s cold, the pollution is bad.”
Many were quite surprised I wanted to predict pollution at all. The status quo wasn’t good enough for me. After all, I noticed that even during very cold days pollution levels in the middle of the day weren’t all that bad. So I started collecting data in the hopes of learning more, and hopefully I would be able to collect enough to actually predict PM2.5 levels.
Benefits of Predicting
As with any project, the benefits must be clear from the start. Here are what I believe can be some benefits of predicting PM2.5 levels:
- Knowing mask/air filtration needs for certain times of day. (Just like we ask ourselves whether we should bring an umbrella if it will rain.)
- Knowing whether it will be safe to open the windows today. Often people only look at the pollution levels once a day (if at all). Simply looking out the window isn’t a reliable indicator of pollution.
- Knowing if it is a good idea to go outside or to the park, and at what time. Also knowing what part of the city will have cleaner air at a certain time of day can be valuable.
Also, when we are able to predict pollution we can start to unlock further possibilities and benefits. When we can predict pollution we…
- …can predict pollution we may be able to forecast hospital admissions. Meaning hospitals could better know how to manage their scarce resources.
- …can give guidance to the public to keep their windows sealed during times of high pollution. Instead of giving general advice what about a push notification saying that pollution is going to be very bad for the next n hours?
- …can consider closing schools and public buildings during times of especially high pollution. If children and public servants will be exposed to dangerous levels of pollution, should they stay home?
These are just a few possibilities. I believe even more benefits and possibilities will be found over time.
As with anything, there will always be limitations and caveats. At the risk of underselling the potential here are a few issues:
- WHO guidelines and the US EPA give recommendations based on 24 hours of exposure. When is the last time a kindergarten student was outside for 24 hours? The AQI categories are based on these exposure estimates.
- With any predictive model there will be some error margin. Our goal is to predict the pollution as reliably as the weather can be predicted. At the risk of oversimplification we would like to be able to accurately predict the pollution category (based on the US EPA categories) 70% of the time.
- Asking residents to stay inside may not be better for their health. Many homes (especially gers) are not sealed properly and don’t have air filtration.
- Making predictions aren’t useful if the public doesn’t use them. In Mongolia the government is often expected to lead on issues such as these, so without buy-in from the city or national government predictions will be of limited value.
This was part 1 in a 4 part series on predicting air pollution in Ulaanbaatar. In Part 2 we will explore visualizations of pollution and weather in Ulaanbaatar to better understand it. Part 3 will focus on the machine learning model I have built. Part 4 will be focused on deploying the model to the public and how we will be tracking its accuracy.
Have a question? Send me an email at email@example.com.