Helen Coffman, Audrey Cabrera
Ryan Ohlinger, Arnav Gangal, Debbie Ryu, Hayden Souza, Ishaan Shah, Josh Chan, Renzo Tanaka-Wong, Vince Front, Grace Panos
In March 2020, multiple counties in California issued a lockdown in response to Covid-19. People halted their everyday activities, industries and travel paused their productions, and businesses shut down due to loss of income. As a result, the air quality improved, consumption of fossil fuel and nonrenewable energy sources was reduced, and recovery of the ecosystems. These positive effects on the environment demonstrated a solution for society to reduce the rate of climate change. Our project aimed to explore the impact of the pandemic on the environment, more specifically, how the pandemic impacted air pollution in California. We chose to investigate the change of three air pollutants - Sulfur Dioxide (SO2), Carbon Monoxide (CO), and Nitrogen Dioxide (NO2). We gathered past records of air pollutant levels from 2015 to 2020 to create prediction models that estimate future air pollutant levels under pandemic conditions and applied exploratory analysis to demonstrate a change in air pollutant levels. In addition, we built an interactive dashboard model of all the contributors to climate change and the impact the lockdown measures had across California.
Our exploratory data analysis results provided insight that the pandemic did not cause a significant change in pollution levels. We believe that the pandemic lockdown period between March 2020 to December 2020 was too short to reduce the pollution levels. However, our data analysis showed dramatic changes in travel and transportation (i.e., flights, jet, and vehicle fuel consumption) after March 2020. Even though the fuel consumption was reduced, it was not enough to lower the air pollution levels. Nevertheless, we investigated if the current fuel and energy consumptions would be enough to lower air pollutant levels in the future. When implementing our machine learning models (ARIMA and SVR), we discovered there was little to no correlation between the contributors and air pollutants. As a consequence, we could not accurately predict future pollutant levels due to fluctuating values and lack of data and predictors. Despite this problem, we used the weather data SVR model to predict future air pollutant levels and received more accurate results. In the end, we were able to explore the air pollutants’ concentration levels using machine learning models. The remainder of our dataset, consisting of air pollutant contributors, proved to be poor and insufficient for predicting air pollutant levels. In the future, when tackling a similar project, we believe we would need more data that is stronger and able to support our efforts in predicting air pollutant levels with SVR and ARIMA machine learning models.