china and the world discuss the environment

  • linkedin group
  • sini weibo
  • facebook
  • twitter

Sign up for email updates

文章 Articles

Seeing China’s pollution from space

Angel Hsu


US scientists have used satellite data to assess a decade’s worth of PM 2.5 levels. Angel Hsu, one of the team, explains what they found.

article image

The Great Wall may not, after all, be visible from space – but Chinese air pollution is.

A team of researchers at Battelle Memorial Institute and Columbia University, in collaboration with Yale University, recently used satellite readings to produce data on fine particulate concentrations in Chinese provinces. While these satellite measurements are not perfect, they provide the first estimates of ground-level annual average concentrations of the pollutant PM 2.5 for all of China over the last decade.

PM 2.5 is the term for particulate matter measuring 2.5 microns or less in diameter and has become a focus of public safety campaigners in China in recent months. Fine particulates have the ability to penetrate human lung and blood tissue and can lead to asthma, cardiovascular disease and cancer.

How are the satellite measures of PM 2.5 derived? In short, scientific instruments aboard the satellites assess something called Aerosol Optical Depth (AOD). This is a measure of the degree to which aerosol particles prevent the transmission of light either through absorption or scattering.

Several studies have developed algorithms and models to relate the AOD measures to ground-based measurements of particulate matter. Of course, relating these atmospheric column measurements to ground-level measurements is tricky and depends on the vertical structure, composition, size, distribution and water content of the atmospheric aerosol. Therefore, regional differences and climatology also play a role in the extrapolation of PM 2.5. (The methods used to extrapolate PM 2.5 measurements are described in full here.)

The PM 2.5 concentrations are expressed in terms of average exposure by province or municipality. The population-weighted exposure for a specific province is calculated by multiplying the satellite-estimated PM 2.5 concentration for each grid cell by the percentage of the province population that lives within that grid cell and producing an average for all of the grid cells within a province. 

This means that PM2.5 concentrations for more heavily populated areas within a province will count more towards the provincial average than the outlying or sparsely populated areas. This standard approach addresses the cases where, for example, there is relatively clean air over a large percentage of a province, but nobody lives there to be “exposed”, while a large proportion of the population lives in a small area with high PM 2.5 concentrations (or vice versa).

As a result, the population-weighted numbers are more telling of actual exposure to fine particulate matter. Simply put, these numbers represent an average air quality situation an average citizen in the Chinese province in question would face on any given day.

What do the measurements show?

The figures provided below reveal telling trends for PM 2.5 data in China. All but four provinces (excluding Taiwan) have average annual exposures to PM 2.5 above levels recommended by the World Health Organization (WHO). Figure 1, below, shows a map of population-weighted fine particulate matter concentrations in China’s 31 provinces in 2007. Most provinces exceed the WHO recommendation for PM 2.5 levels, which is set at an annual average of 10 micrograms per cubic metre.



Figure 1. Annual-average population-weighted fine particulate matter concentrations (PM 2.5) for Chinese provinces, including Taiwan, in 2007. 

The time series data provided in Figure 2 and Figure 3 offers an insight into PM 2.5 trends in different Chinese provinces. PM 2.5 concentrations are the highest in Shandong and Henan provinces. Beijing, Shanghai and Guangdong province have experienced slight decreases in annual average PM 2.5 levels over the last three years, although concentrations have remained fairly steady over the last nine years. Unsurprisingly, less developed western provinces such as Tibet and Inner Mongolia have the lowest fine particulate matter concentrations.

 Figure 2. Nine-year time trend of average annual PM 2.5 concentration data for selected provinces and municipalities in China.

Of course, as with any type of modeling, there is an associated uncertainty. In particular, satellites aren't as good at reading AOD over bright surfaces such as snow and deserts, and they also can't tell you about vertical distribution of particles in the atmosphere (for example, they can't distinguish particles high up or close to the surface). The uncertainty with the model we used is about +/- 25 percent, which translates into 6.7 micrograms per cubic metre.

Satellite measurements do not ultimately match up to data from earth: ground-based, in-situ measurements are ideal. However, satellite air-quality measurements can help to fill in spatial and information gaps where ground-based monitoring stations are not available. Furthermore, satellite measures provide consistent, repeated monitoring that allow for comparison over time and between areas. As Chinese policymakers face up to PM 2.5, data from space can help them understand what they are dealing with.

 Figure 3. Nine-year time trend of average annual PM 2.5 concentration data for provinces and municipalities in China, including Taiwan. (Click to enlarge the image)

Angel Hsu is a doctoral student at the Yale School of Forestry and Environmental Studies and project director for the 2012
Environmental Performance IndexMap produced with funding from the NASA Earth Science Division Applied Sciences Program, by scientists at Battelle Memorial Institute with the guidance of CIESIN at Columbia University and YCELP at Yale University.

Homepage image provided by SeaWiFS Project, NASA/Goddard Space Flight Center, and ORBIMAGE



评论 comments


评论 comments



嗨 Hi Guest user

退出 Logout /

发表评论 Post a comment

评论通过管理员审核后翻译成中文或英文 最大字符 1200

Comments are translated into either Chinese or English after being moderated. Maximum characters 1200

排序 Sort By:




A question from Sina Weibo

Ma Jun from the Institute of Public and Environmental Affairs:

Henan and Shandong's pollution levels are very high, whereas pollution levels in Shanxi seem to be relatively low. Would it be possible to explain more clearly how the annual population-weighted PM2.5 concentrations were calculated?




Sina Weibo Question 2

In 2007, out of China's 31 provinces, autonomous regions, and municipalities, 27 have population-weighted fine particulate matter pollution concentrations in excess of 10 micrograms per cubic meter (/m3), the WHO's annual air quality standard. Shandong, Henan, and Hebei were the highest ranking, with Beijing coming in at number 10. I don't know how the figures for each province were calculated.




请解释下为什么你的数据与以上的测量值不符? ?我很困惑,你说“卫星测量值与从地面上测得的数值不匹配”,但在同一个句子里,你又说卫星测量仍然是一个有用的工具。在对北京和陕西两地的测量中,卫星数据和地面数据之间的差异似乎过大,大于你的模式中所说的25%的不确定性。

另外,我很好奇,你能否详细解释一下这句话:该模式“不能测得大气中粒子的垂直分布情况” 。微粒的垂直分布对暴露在其中的人体健康影响有多大?

Pollution levels appear greatly understated

Greetings Angel, the numbers in your study, in particular the population-weighted exposure numbers, seem unreasonably low. For example, according to the BJEPB average PM2.5 concentrations in Beijing have been between 70-110 ug/m3 over the past decade. (
In Xi'an independent measurements of PM2.5 between 2004-2008 were at 182.2 ug/m3.

Would you please comment on why your data appears to be inconsistent with these measurements? I am confused that you state that "[s]atellite measurements do not ultimately match up to data from earth" but then in the same sentence say that it is still a useful tool. The differences in Beijing and Shaanxi between your satellite data and ground-based measurements seem to vary significantly more than the 25% uncertainty that you use to describe your model.

Also, I am curious if you could explain this statement in more detail that the model “can't tell you about vertical distribution of particles in the atmosphere”
. How much of an impact does vertical distribution have on human exposure to particulate?

人口加权PM2.5 年均浓度的计算方法

从文中看,人口加权PM2.5 年均浓度的计算方法应该是:A网格区域的污染浓度*A网格区域人口占全省人口百分比+B网格区域浓度*B网格区域人口占全省百分比+.........(其实就是加权平均数的计算方法而已)

Calculating the PM2.5 Concentration of a Province

As seen from the text, the method of calculating the population-weighted annual average of PM2.5 (fine particle pollution) concentration should be: [The pollution concentration of Grid A * The percentage of the entire provincial population living in Grid A] + [The pollution concentration of Grid B * The percentage of the entire provincial population living in Grid B] + .... (Actually this is the method for calculating any weighted average.)


空气质量指数可以用两种方式来表示——人口加权和地区加权。正如我解释的那样,特定省份的人口加权值是这样计算的:各网格内根据卫星数据得出的PM2.5浓度乘以该网格内居民占全省人口总数的百分比,即计算所有网格区域的大气污染暴露浓度的人口加权平均值后便得到人口加权大气污染暴露水平。我们在巴特尔的合作伙伴使用van donkelaar et al 2010 (的方法推算气溶胶光学厚度数据,然后用国际地球科学信息网络中心(CIESIN)的网格人口数据库把得到的数字转换成平均暴露水平数字。这些数据更多的反映的是长期曝露在PM2.5污染物中的情况,而不是短期情况。它实际上反映了某省某人某天感受到的PM2.5颗粒物污染情况。希望这些信息对你有所帮助。

Response to Ma Jun

There are two ways to develop these air quality indicators - population-weighted or area-weighted. As I explained, the population-weighted exposure for a specific province is calculated by multiplying the satellite-estimated PM 2.5 concentration for each grid cell by the percentage of the province population that lives within that grid cell and producing an average for all of the grid cells within a province. Our partners at Battelle used van donkelaar et al 2010 ( methods to extrapolate Aerosol Optical Depth (AOD) data and then used CIESIN's Gridded population dataset ( to translate these numbers into the average exposure numbers. These data more reflect long-term exposure to PM 2.5 rather than short-term exposures, so a PM 2.5 situation that an average person in x province would experience on a given day. I hope this helps.



Reply to Steven Andrews

Steven - the data are what they are. We're not making any claims of comparing them to official numbers. The satellite numbers provide "wall to wall" coverage so to speak of pollution across an entire province or municipality. You also know the uncertainty level associated with the satellite data, whereas you are not given any indication as to 1) how many ground-based monitors are included in those numbers; 2) whether they are regularly maintained, calibrated, etc.; 3) where they are located; 4) what uncertainty is associated in the links you provided. These numbers are for research and reference purposes only, to generate discussion and to see the potential for satellite data to help better understand air quality. We produced the same estimates for 132 countries in the 2012 EPI (, many for countries that do not measure PM 2.5. Having these numbers, knowing exactly how they were calculated and what the uncertainty is is already generating much discussion amongst policymakers. As I mentioned before, these are long-term exposure numbers and don't capture spikes in data. But health effects from air pollution are based on both long and short-term exposure.


很明显,这种违反基本科学常识的“分布图”是为了突出不同省份的PM2.5,特别是要突出北京地区的PM2.5是最高。但是,决定PM2.5 值的因素并非local的因素,而是global 的因素,Euroasia 和非洲的尘暴可以直接影响到美洲的PM2.5, 用这种特出邻近两个省的差异的手法去描述PM2.5,对了解和治理空气污染毫无帮助。

Distributive graphs seems like political tool

The most commonly-used scientific instruments in describing atmospheric indicators like temperatures and pressure are isotherm, isobar, etc. How come the distribution of PM2.5 fine particulate matter concentrations are divided according to administrative boundaries of provinces?
Apparently, such graphs do not only violate basic scientific knowledge, but also intend to specify the PM2.5 concentrations of different provinces, especially Beijing where concentration ranked the first. However, PM2.5 is not determined by local factors, but global reasons. For example, dust storm in Euro-Asia and Africa can directly affect PM2.5 of America. Measurement highlights the difference between two provinces is absolutely useless in governing air pollution.
Should there be any implications from the distributive graphs except political tool?

合作伙伴 Partners

项目 Projects