Al Based PM 2.5 Retrieval and Spatiotemporal Downscaling Using Earth Observation Data

Authors

  • ALLISON LOU Aspiring Scientists' Summer Internship Program Intern
  • Seren Smith Aspiring Scientists' Summer Internship Program Co-mentor
  • Chaowei Yang Aspiring Scientists' Summer Internship Program Mentor
  • Hai Lan Aspiring Scientists' Summer Internship Program Co-mentor
  • Yun Li Aspiring Scientists' Summer Internship Program Co-mentor

DOI:

https://doi.org/10.13021/jssr2021.3274

Abstract

Sensors used to collect data on the physical world can sometimes unexpectedly lose their readings due to sensor or communication errors. The objective of this project was to develop an innovative methodology to retrieve particulate matter 2.5 (PM 2.5) data records on a global scale from Purple Air sensors. An additional objective was to further downscale the spatiotemporal resolution to 1 km and the hourly level in some key regions by using artificial intelligence (AI) models. Major tasks focused on using deep learning for PM 2.5 retrieval as well as prediction and downscaling with the end goal of having a PM 2.5 estimation covering on a global scale and an hourly 1 km * 1km granularity of PM 2.5 for the Los Angeles region. Accomplishments consisted of gathering all timeseries data for global Purple Air sensors from 2018 and for California Purple Air sensors from 2019, all of which was uploaded to the project's drive for further cleaning into usable models.

Published

2022-12-13

Issue

Section

College of Science: Department of Geography and Geoinformation Science

Categories