Plan your trip: Forecasting seasonal pollutants

The main idea behind this project was to build an app/tool that would forecast when is the best time of the year to visit a given city based on low levels of pollution.

In a nutshell

• Data was gathered from all the stations that monitor Particulate Matter-10 (PM10) levels for each European capital and that are reported to the European Environment Agency. The records from a total of 318 stations was used for the analysis.
• The aim was to summarize all this information into an application that would advise the user when is the best time to visit a given city so that dangerous levels of PM10 are avoided. It was divided into two modules: PM10 Viewer and PM10 Forecaster.

With PM10 Viewer you can:

1. Visualize PM10 historical data for European capitals.
2. Inspect the levels of PM10 in a given day or period of time. For instance, you could analyze how the lockdowns due to Covid-19 influenced the emissions of PM10 in each city.

With PM10 Forecaster you can:

1. Visualize the forecast of PM10 levels by city using the Prophet algorithm (from Facebook) or a built-in generalized SARIMA model.
2. Interactively change the parameters of SARIMA to fine-tune and optimize the model for each specific city.
3. Forecast the best months (ranked) to visit a certain city according to the model of your choice (Prophet or SARIMA).

Continue reading to see how the analysis was performed and how the app would retrieve the requested information (with gifs examples).

1- Context

• Air pollution is responsible for many respiratory and cardiovascular diseases.
• Particulate Matter (PM) are one of the causal agents of such diseases. These particles are not composed of a single pollutant but rather a mixture of many chemical species (solids and aerosols).
• Some are emitted directly from a source, such as construction sites, unpaved roads, fields, smokestacks or fires.
• Most PMs form in the atmosphere as a result of complex reactions of chemicals such as sulfur dioxide and nitrogen oxides, which are pollutants emitted from power plants, industries and automobiles.
• PM are defined by their diameter (d) for air regulatory purposes:
• Those with a d < of 10 microns (PM10) are inhalable into the lungs and can induce adverse health effects, such as Chronic Obstructive Pulmonary Diseases (COPD).

In Europe, althoguh PM levels have slowly decreased over the years, COPD cases are steadily increasing

Click here to see an interactive map of the PM levels evolution by country!

Graphs and map plotted from data derived from the Global burden of disease 2019 (GBD 2019).

One way to mitigate the deaths attributed to pollution would be by reducing as much as possible our exposure time to this kind of pollutants.

2- Main objectives

• To build an App to visualize PM10 historical data.

If PMs follow a seasonal behavior, we could potentially avoid them by not visiting certain cities during specific periods of time.

• To investigate the seasonality of PM10 in Europe.

Assuming PMs are seasonal, is it the same for all cities in Europe?

• To forecast levels of PM10 in European cities.

This way we could predict which is the best month of the year to visit a given city based on low levels of pollutant.

3- Data processing

• PM data from all capitals of Europe was obtained from the European Environment Agency.
• For each city, the data recorded by the different sensors (dispersed within the city) was averaged based on the time of the recording.
• Streamlit was used as a tool to build the potential App.

4- PM10 Viewer

With PM10 Viewer you can:

1- Visualize PM10 historical data for European capitals:

> Thresholds for Air Quality Index are indicated in dashed lines.
> Seasons are colored for an easier inspection of seasonality.

2- Inspect the levels of PM10 in a given day or period of time:

> By specifying the date(s) in the 'From' and 'To' widgets.

5- Analysis of the seasonal effect

In order to analyze seasonality, a Seasonality Index metric was calculated as follows:

• Data was grouped by month.
• Seasonality Index (SI) = max(conc.) - min(conc.)
• Thus, the higher the SI, the higher seasonality is expected.

The Seasonality Index for each city was plotted in a map using the respective country geocoordinates. It was found that the cities with the highest seasonality index were geographically related and corresponded to the Balkans. Click here to see an interactive version of the map!

The Balkans accumulate most of Europe's power plants and households do depend on carbon, which seems to be reflected in the seasonality index.

This data correlates with the increased number of premature deaths attributed to PM2.5 in these regions as recently reported by the EEA and FAO:

6- PM10 Forecaster

With PM10 Forecaster you can:

1- Visualize the forecast of PM10 levels by city using Facebook Prophet or a built-in generalized SARIMA model.

> Prophet or generalized SARIMA model: SARIMA(0,1,1)(0,1,2)12
> In black: the actual data.
> In blue: the model.
> In green: the tested data used to calculate the error of the model (RMSE).
> Compare the two models: Prophet vs SARIMA

2- Change the parameters of SARIMA to fine-tune the model for the city.

> Fine-tune the trend and seasonal order parameters of SARIMA(p,d,q)(P,D,Q)s to find the best fit for the city.
> Compare the RMSE of your model to the built-in one.

3- Forecast the best months (ranked) to visit a certain city according to the selected model.

> Select the number of months to show in the ranking.
> For each month, the corresponding air quality index is returned.
> Switch between models to get the forecast.