Exploring Air Quality in Western U.S. since 1971

Caleb Weis, Eva Wu, and Jimin Han

Introduction

For our project, we wanted to explore the levels of air pollution, and the Air Quality Index values, throughout the United States during the years after the passage of the Clean Air Act of 1970, which brought with it the first establishment of National Ambient Air Quality Standards, which have remained largely constant since then. In order to answer this question, we drew upon data from the US EPA. More details on this dataset can be found below, and in the “data” folder of our repo. For our visualizations, we decided to make an interactive shiny app with a choropleth map of the western US over time, colored by county-level pollution/AQI levels, and some line plots of state-level pollution/AQI levels over time. More details on our app can be found below, and the app itself is in the “shiny” folder of our repo.

Data

The data wrangling aspect of this project was challenging, and very time-intensive. Loading data from the API could be a very arduous process, and so we decided to load it in 10-year increments. Even then, we knew that we could not load daily data—that would have been county-level data on 6 different pollutants every day for an end-goal period of 50 years and for multiple states. We initially tried to load the next time increment of data, quarterly, but that too proved time-intensive, and so we ultimately decided to use yearly increments. In addition, due to the sheer amount of data that we still had to load, we decided to focus our analysis on the 11 states which make up the “Western” region of the United States, a decision primarily informed by the knowledge that this is the most wildfire-prone region in the United States (Supporting data from Statista).

After loading all of the data, we put it through numerous rounds of adjustment, filtering, merging, summarizing, pivoting, etc. in order to get to our final dataset.

First, we used different air quality measures to construct the Air Quality Category variable, a 6-level index used by government agencies to communicate to the public how polluted the air is. We utilized a US EPA-designed function called con2aqi (literally, concentration to air quality index) in order to transform rounded pollution levels into air quality index values. We then took the maximum index value produced by each pollutant in a given county and year, and this value was the final AQI value for that county in the given year. To graph this in the standard format, we then transformed this variable into a discrete variable with six levels, and color-coded these levels using the standard color scale (most of this information is from this document published by the US EPA). All of these transformations were saved into the dataset, enabling us to finally use this dataset to produce the final choropleth map in the way described below.

We loaded and cleaned California data first. In order to reuse the code for other states, we made one assumption - no state kept better air quality records than California. We made this assumption after some brief high-level googling, but it was not a fully backed-up assumption. Through making this assumption, we saved ourselves the time that it would have taken to have checked the availability of data on certain pollutants for each state over each 10-year period. Instead, we used the template created by loading the California data, in which we did exhaustively check for the availability of data on each of the six pollutants in our analysis, and assumed that this would translate to other states. We found that it overwhelmingly did translate, meaning that most states did not have data on fewer pollutants than did California during these periods. However, there is a possibility that some states had data on more pollutants than did California during these periods, and that data was left out of our project.

Here is a snippet of the data:

year	state	county	county_code	state_code	AQI	pollutant	units_of_measure	arithmetic_mean	air_quality_index	fips
1971	Arizona	Cochise	003	04	33	SO2	Parts per billion	23.268075	Good	04003
1971	Arizona	Coconino	005	04	2	SO2	Parts per billion	0.733333	Good	04005
1971	Arizona	Gila	007	04	115	SO2	Parts per billion	48.123255	Unhealthy for Sensitive Groups	04007
1971	Arizona	Greenlee	011	04	107	SO2	Parts per billion	88.596000	Unhealthy for Sensitive Groups	04011
1971	Arizona	Maricopa	013	04	247	CO	Parts per million	3.478998	Very Unhealthy	04013

The following is a high-level description of an example tibble generated by our function.

Variable names	Description
year	Year of recorded measurement.
state	State name.
state_code	Code for chosen state.
county	County name.
county_code	Code for county in that state.
AQI	Air Quality Index, used by government agencies to communicate to the public the air pollution level.
Air Quality Index	The levels of air pollution. Low AQI = “Good”, high AQI = “Hazardous”.
pollutant	Name of pollutant.
arithmetic_mean	Arithmetic mean of pollutant concentration.
units_of_measure	Units of measure of pollutant concentration.

Approach

In order to explore air quality, we created a shiny app containing choropleth maps and line plots. Our app contains 4 main tabs with different information on air pollution plus a tab for data and intro. We chose to make a shiny app because it allowed us to not only present, but also interact with the visualizations by selecting features such as pollutant type, year, and state.

We used the sidebar layout so that we could select pollutant type in the sidebar and click on tabs on the main panel to see different plots. We put pollutant on the side rather than under each tab because it was a common variable for both the map (1st tab) and the line plot (3rd tab). We included different types of pollutants because each pollutant might indicate a different reason for or consequence of the corresponding air pollution level. Including them will better inform expert users about air quality. We also included the AQI plots, because this allowed us to better communicate the overall air quality to laypeople without getting into too much details in pollutant types.

The 1st and 2nd tabs each contains a choropleth map of western US states and counties which allow users to compare air quality across space. The 1st map filled by selected pollutants, and the 2nd map filled by AQI levels. The slider allows users to select a specific year or animate across years. The maps show the county-level choropleth of pollution levels in the selected year accordingly.

The 3rd and 4th tabs each contains a line plot of air quality across years, which allow users to explore change of air quality across time and to compare air quality trends across states. The 3rd tab has pollutant levels on the y-axis as the air quality measure, and the 4th tab has AQI on the y-axis as the air quality measure. The selectize input allows users to select up to 8 out of the 11 states, while preventing them from mispelling or inputing states that were not in our data. The limit of 8 states is for clarity of graphing. We don’t want to cram the graph with too many lines, which hinders our ability of interpretation. The resulting line graphs show the trend of air pollution levels of the selected states across years.

The 5th tab contains a table showing our data. It has a search box on the top, allowing users to type in any year, county, state, pollutant, etc. values if they want to check on the data containing the typed value only.

The 6th tab contains a brief intro of our project and our app, so that users can better navigate it.

Code & Visualizations

The shiny app can be accessed using this link.

Below are some sample graphs of placeholder year 2010 and placeholder pollutant PM2.5.

Here is the map of PM2.5 data in 2010.

counties_air %>%
  filter(year == 2010 & pollutant == "PM2.5") %>%
  ggplot() +
  geom_sf(data = WEST.SF) +
  geom_sf(mapping = aes(fill = arithmetic_mean), color = "grey40") +
  coord_sf(datum = NA) +
  scale_fill_gradient(name = paste0("Pollution level \n(Micrograms/cubic meter (LC))"),
                      low = "lightyellow", high = "darkred",
                      na.value = "white") +
  theme_void() +
  labs(title = "Map showing county-level air quality in 2010",
       subtitle = "measured by PM 2.5") +
  theme(legend.position = "left")

Here is the AQI map in 2010.

# create a color scale
cols <- c("Good" = "green", "Moderate" = "yellow", "Unhealthy for Sensitive Groups" = "orange",
          "Unhealthy" = "red", "Very Unhealthy" = "purple", "Hazardous" = "maroon")

counties_aqi %>%
  filter(year == 2010) %>%
  ggplot() +
  geom_sf(data = WEST.SF) +
  geom_sf(mapping = aes(fill = air_quality_index), color = "grey40") +
  scale_fill_manual(values = cols) +
  coord_sf(datum = NA) +
  labs(title = "Map showing county-level air quality measured by AQI",
       fill = "AQI Levels") +
  theme_void() +
  theme(legend.position = "left")

Here is a sample line graph for PM2.5 levels of some selected states.

input_state <- c("California", "Arizona", "Washington")
air_quality_state %>%
  filter(state %in% input_state & pollutant == "PM2.5") %>%
  ggplot(aes(year, air_qual_year, color = state)) +
  geom_line() +
  theme_light() +
  labs(title = paste0("Air Quality Across Years in ", paste0(input_state, collapse = ", ")),
       subtitle = paste0("measured by PM2.5"),
       x = "Year", y = "Polution level (Micrograms/cubic meter (LC))",
       color = "State")

Here is a sample line graph for AQI of some selected states.

aqi_state %>%
  filter(state %in% input_state) %>%
  ggplot(aes(year, mean_aqi, color = state)) +
  geom_line() +
  theme_light() +
  labs(title = paste0("Air Quality Across Years in ", paste(input_state, collapse = ", ")),
       subtitle = paste0("measured by AQI (Air Quality Index)"),
       x = "Year", y = "AQI",
       color = "State")

Discussion

Map & Line Plot

In the 1st tab, we study the choropleth map of the US west coast states for 1971-2021, on various air pollutants. One caveat is that we lack a great deal of data from the 70s, especially for more rural counties. NO2 and Ozone are almost restrained to data from California in the 70s, while the PM measures were not collected in this dataset. But looking at the maps in general, we can see that the reported counties generally have high levels of pollutant concentration.

We can get a more holistic view of the trend if we look at the 2nd tab containing the line plot. Most notably, we can see that SO2, NO2, and CO have a similar trend in which the values for the 70s are extremely high and they converge to a lower level as we approach the 21st century. We hypothesize that this trend may be caused from the sheer lack of volume of data in the 70s (Most likely), and also that the decreasing trend represents an adapting mechanism for the counties after the policy enactment in 1970. PM10, although collected from the mid-1980s, have a similar downward trend. On the other hand, PM2.5 generally stays constant throughout the years except for an outlier in California. Ozone had a very unique trend in which all states experience a dip in ozone levels in the 1980s. More specifically, the levels fall from around 0.06 Parts per Million to less than 0.04, going back up to 0.06 at the beginning of the 90s. After the 90s, we note a rather constant level of Ozone despite fluctuations in year-by-year level.

AQI Map & AQI Line Plot

In the 3rd and 4th tabs respectively, we provide a similar set of maps and line graphs for AQI, which is a more general and intuitive measure of pollution derived from the pollutants we presented before. In the 1970s, where many of the values were fluctuating, we see a wide range of colors in the map. We see that amid mostly green (good) AQI levels, there are a mix of moderate and unhealthy levels near urban areas. However, beginning from 1980, the map turns almost completely yellow (moderate) and stays that way until 2021.

If we translate this trend to the AQI line plot, we can see that similarly yet much less extremely than our previous plots, the AQI level dips to the 50-60 range in the 1980s. One possible explanation is that this is due to the influence of the Ozone plot, because Ozone is more weighted when computing AQI. Reflecting how most pollutants converge to a lower level after the 1990s, we can rationalize the fact that AQI stabilizes to the moderate range (51-100).

Conclusion

Although the variance in the AQI levels are very small for all of the states, we can also note that the “larger” and “more urban” states tend to have slightly higher values of AQI. Those states are California, Arizona, Colorado, and Utah. We cannot generalize these results to any normative statements about air pollution or climate change. Also, we have some constraints due to a lack of data in certain regions. Thus, it will be reasonable to use this website as a supplementary source providing a convenient mapping of how pollutant levels have changed across time. Future scholars could first expand our computing power to obtain data that covers missing counties and the rest of the states. Then, they could also try to link this data with possible causes or results affected by the data, such as individuals’ health data, forest fire data, or the correlation with how “urban” some counties are.

Repo organization

• proposal folder. This folder includes the RMarkDown and knitted MarkDown documents which contain the information in this proposal.

• data folder. This folder contains RMarkDown documents used to load, examine, and wrangle data from the RAQSAPI and to load data from other sources, the csv files of data loaded from other sources, as well as the README file consisting of the codebook.

• Presentation folder. This folder contains code we used to create our xaringen slides.

• shiny folder. This folder contains an app.R document consisting of the code used to build the app and a data folder consisting of the data we used to make visualizations in the app.

• README_files folder. This folder contains select visualizations of air quality trends.

• This is the link to our shiny app website.

• Each folder has a README.Rmd and README.md file that describe the content of the folder.