Is Beijing’s air quality improving?

Media outlets and researchers alike have noticed Beijing’s improving air quality. For example, NYtimes reported on July 19 that the air quality in Beijing showed signs of improvement for the first half of 2016 compared to other years. Looking at data all the way back to March 2010, it appears that there may be a break in the annual cycle of air pollution that peaks in winter. However, these cycles make it difficult to determine how much the air quality is actually improving.

It is useful to note at the outset that AQI scores between 0-50 are considered good, between 50-100 are considered moderate, and over 100 are considered unhealthy. For more information, visit the State Dept explanation. Prior to the recent dip, the previous monthly low was in January 2011, when the mean air quality for the month hit 100.26, just above the unhealthy threshold. Looking at the historical data, dips in AQI score have always been followed by spikes, most notably in February 2013, when the monthly high neared 250.

These spikes have been decreasing in magnitude, and the mean AQI scores for February, March, April, and May 2016 have all been below 100. This offers promising future for Beijing’s air quality. The annual spike in AQI occurs during the colder months, when the temperature inversion traps harmful pollutants close to the ground. Currently, the monthly mean AQI is tracking toward the first good classification in our data set. Data on months prior to March 2010 are spotty, making it hard to conduct meaningful analysis.

The change in trends is believed to have resulted in part from China’s aggressive targets published in the 12th five-year plan, covering 2010-2015. This plan included specific targets to curb emissions of CO2 as a proportion of GDP by 17%. While these targets are aggressive, China is starting from a low baseline and change is needed. Public health experts from University of California, Peking University, and Princeton University claim that there is still work to be done on reducing the contribution of household contributions to ambient air quality. To the extent that the air quality is improving, the health of Beijing’s residents depends on that trend continuing, because the effects of years of high PM2.5 levels may have lasting effects on resiratory health, according to the linked study linked.

To understand more about the future of air quality in Beijing, we analyzed the data going back to early 2010 and constructed a forecast of air quality based on time series data.

About the data

The raw data was collected from the US Embassy website, stateair.net, and it includes hourly averages for (almost) every hour from March 1, 2010 through May 30, 2016. The data are reported in PM2.5 levels (corresponding to the AQI score based on that variable) and concentration (the raw concentration reading). We cleaned the data and uploaded it to our CKAN library. We only report the AQI scores for simplicity, since they are calculated as a linear function of the concentration.The clean data are here.

There were 54,794 observations, with 2,365 missing (0.00%). To allow for some analyses that require no NA values, these missing values were imputed using a simple linear imputation (imputsTS()) from the nearest data points.

It can be difficult to see the patterns when looking at all the data at once. The calendar heat map below shows the AQI score for each day in the observed period.

To understand the cycles in the data, it has to be aggregated a bit.

Understanding the cycles

PM2.5 levels shift seasonally and over the course of each day. The decompose() function in R was used to analyze the seasonal variation, which appears below in the “seasonal” window.

Seasonal cycles

There is an annual high some time in late winter or spring followed by a decline over the summer months to an annual low period during the Fall. There was a peak in overall trend in January 2013, when the mean monthly AQI score was just over 243. After another peak in the overall trend in early 2014, there has been a downward trend in the AQI scores, a trend that appears to be accelerating with the latest release of historical data including April 2016. Controlling for this seasonal variation, the overall trend appears to be downward since the January 2014 peak.

plot(decompose(aqi_ts))

Daily cycles

Looking at the daily cycles compared with the mean plus/minus one standard deviation paints a more complicated picture. The variation in mean daily AQI scores is wide enough that the all scores fall within one standard deviation of the mean. The figures below illustrate the observed variation in the AQI throughout the day. In the figure on the right below, daily variation can be seen in the royal blue line, which shows the hourly mean AQI score. The highest hourly mean AQI scores occur between 8PM and 1AM, possibly corresponding with the evening rush hour. The lowest scores occur between 2PM and 4PM. Notably, the daily readings from May 2016, the outer red ring on the graphic to the right below, does not follow this daily trend. It shows a spike during the normal daily low, and a relative low point overnight. This could be due to a number of factors, and this post will not get into the possible causes. However, there is a mesmerizing visualization of the effect of wind on PM2.5, here.

The graph on the left below shows the mean AQI score for each hour throughout the day for each month between March 2010 and May 2016. The dark line on the graph to the right corresponds to the January 2013 spike mentioned above. This outlier does follow the mean daily cycle, with an exaggerated highs and lows.

Conclusions and Forecasting

Because the latest data seem to break from the historical trends, it is difficult to model future scenarios. We used a simple projection based on the historical data that mor heavily weights recent data compared with older data. The forecast, shown below, projects the recent downward trend into the future including seasonal effects, estimating that he AQI in Beijing will be consistently within the “good” range of below 50 starting in 2017. We will be interested to see the winter readings from Beijing as new data become available over the next few months.

In conclusion, despite the challenges of modeling the seasonal and daily cycles of air quality in Beijing pose challenges to interpreting the data and forecasting future scores, there does seem to be a break in the trend for Beijing’s air quality. We will monitor this data into the fall and winter and update this post with new analysis as the AQI score would typically reach its annual high during that period.