Air pollution modeling is a complex problem involving high dimensional data and sparsity of high quality sensor data. Due to its high cost, high quality sensors are deployed primarily by the government in very few locations, often only a few per city. This has opened up the space for companies like PurpleAir to deploy low cost air quality sensors to monitor air pollution throughout cities and rural areas. These low cost air sensors, however, are unable to effectively measure ultra fine particles (UFP) (smaller than 1 micron), which are currently unregulated but crucial to monitoring air quality since these small particles are more likely to make it into the human circulatory system and cause health problems. This raises the question, how can we combine intermittent high quality sensors with pervasive low quality sensor to estimate the air quality at any place? And can we use this to inform us where to deploy the next sensor to reduce uncertainty?
Students use the Fixed Rank Filter (FRF) to create a data-driven statistics-based model to estimate air pollution. In the spring, we will tackle two fronts: (1) build a mobile experimental setup to be deployed in a vehicle and collect data in Claremont and (2) estimate air pollution using existing data sets from a collaborator using the FRF. In spring, the work will be for academic credit.
In the summer, we will develop algorithms to optimally select the best location to deploy a sensor to reduce overall uncertainty in the air pollution estimate and deploy the sensor.
Essay Prompt - What interests you about this research and what do you hope to get out of the research experience? Please also comment on whether you would like to and are available to do research in the summer as a paid position on campus.
You are interested in applying techniques in engineering and robotics to study air pollution and take a research project from the beginning to the end.