Abstract View
Factors Influencing Ambient Particulate Matter in Delhi, India: Insights from a Machine Learning Model
KANAN PATEL, Lea Hildebrandt Ruiz, University of Texas at Austin
Abstract Number: 192
Working Group: Urban Aerosols
Abstract
Concentrations of ambient particulate matter (PM) depend on various factors including primary sources, meteorology, and chemical transformations. The concentrations/composition, sources and dynamics of PM can be estimated by combining ambient field measurements with machine learning and statistical analysis tools. New Delhi, India is the most polluted megacity in the world and routinely experiences extreme pollution episodes. As part of the Delhi Aerosol Supersite Study, we measured online continuous PM1 (Particulate matter of size less than 1µm) concentrations and composition using an Aerosol Chemical Speciation Monitor (ACSM) for 4+ years, starting January 2017. PM1 is typically composed of organics as well as inorganics such as chloride, ammonium, sulfate, and nitrate ion species. To understand the factors that influence PM1 variability, we built a machine learning model using random forest regression that estimates PM1 species concentrations by using ambient temperature, relative humidity, planetary boundary layer height, wind speed, wind direction, precipitation, agricultural burning fire counts, and solar radiation. We used hour of day, day of week and month of year to account for emissions specific to certain times (e.g., emissions from traffic may be more important during vehicular rush hours). We demonstrate the applicability of this model to 1) capture temporal variability of the PM1 species, 2) to understand the influence of individual factors/features via sensitivity analyses, which is otherwise difficult to interpret because of multicollinearity between the variables and 3) to predict the PM1 concentrations during the COVID-19 lockdowns and use the differences between predicted and actual concentrations to quantify the role of activity restrictions during the COVID-19 lockdowns on air quality. Overall, our model provides novel insights into factors influencing ambient PM1 in new Delhi, India, demonstrating the power of machine learning models in atmospheric science applications.