A comparative study of three deep learning models for PM2.5 interpolation

Lixin Li, Weitian Tong, Adolphe Some

Research output: Contribution to book or proceedingConference articlepeer-review

Abstract

PM2.5 is a pollutant particulate matter with diameter less than 2.5 micrometer. There exist many stations installed in the world to measure its concentration. Some areas without any proper equipment nor any station installation must rely on interpolation techniques to approximate its concentration. So, there is a need of interpolation technique to approximate the concentration of the pollutant in those areas. The faster and more accurate interpolation technique can help identify more polluted areas and thus efficiently take some measures to reduce PM2.5 harmful effects. We explored three different neural networks, i.e., Bidirectional-Long Short-Term Memory (Bi-LSTM), Gated Recurrent Unit (GRU), Temporal Convolutional Neural Networks (TCN), to interpolate the PM2.5 concentration over the southeast region of the U.S. We investigate different data preprocessing techniques and the effects of spatiotemporal correlation on the models. We finally compare these models and make a choice on the model that is more appropriate for PM2.5 interpolation.

Original languageEnglish
Title of host publication26th International Database Engineered Applications Symposium, IDEAS 2022
EditorsBipin C. Desai, Peter Z. Revesz
PublisherAssociation for Computing Machinery
Pages16-24
Number of pages9
ISBN (Electronic)9781450397094
DOIs
StatePublished - Aug 22 2022
Event26th International Database Engineered Applications Symposium, IDEAS 2022 - Budapest, Hungary
Duration: Aug 22 2022Aug 24 2022

Publication series

NameACM International Conference Proceeding Series

Conference

Conference26th International Database Engineered Applications Symposium, IDEAS 2022
Country/TerritoryHungary
CityBudapest
Period08/22/2208/24/22

Keywords

  • Bidirectional-Long Short-Term Memory
  • Fine particle matter PM2.5
  • Gated Recurrent Unit
  • Interpolation
  • Temporal Convolutional Neural Networks

Fingerprint

Dive into the research topics of 'A comparative study of three deep learning models for PM2.5 interpolation'. Together they form a unique fingerprint.

Cite this