Loading...

Developing Models to Estimate Ground Level PM2.5 Concentrations Using Satellite Measurements

Sotoudeheian, Saeed | 2018

811 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 51142 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Arhami, Mohammad
  7. Abstract:
  8. In this study different prediction models including linear mixed effect (LME), multi-variable linear regression (MLR), gaussian process model (GPM), artificial neural network (ANN) and support vector regression (SVR) were developed using satellite AOD product - with spatial resolution of 3 km – coupled with various auxiliary parameters to estimate ground-level PM2.5 over Tehran. The influence of site effect term on performance of LME models was evaluated using random intercept for monitoring sites. Results showed LME models without this term were able to explain variabilities of PM2.5 in ranges of 60 – 66% and 35 – 41% during model fitting and cross-validation (CV), respectively. By considering site effect term, the performance of LME models during calibrations and validations improved by 20% and 50% on average respectively. The best fit of LME models had a reliable performance during both model fitting (R2 of 0.76) and cross-validation (R2 of 0.6). However, LME models had a significant weakness in estimating high level of PM2.5 during CV. Among all other types of models, GPM with R2 value of 0.59 had the best performance during CV. While the best fit of LME and GPM had similar and reliable performances during CV, GPM was able to better predict extreme values of PM2.5¬. Therefore, GPM could be a better alternative for LME models in high levels of PM2.5¬. In the other part of study, the relationship between Elemental Carbon and AOD was evaluated using data from Terra and Aqua satellite. data were extracted for 3, 5, 10 and 15 km distance representing different scenarios, around 4 monitoring stations for 6-days intervals during 2015 in Tehran. The best performance was obtained for 5 km extraction radius scenario in which the data were integrated from two satellites. In this scenario, temperature as an auxiliary parameter coupled with AOD variable have improved model performance
  9. Keywords:
  10. Machine Learning ; Air Pollution ; Aerosol Optical Depth ; Particulate Matter Less than 2.5 mm ; Moderate Resolution Imaging Spectroradiometer (MODIS)Sensor ; Statistical Model ; Meteorological Conditions

 Digital Object List

 Bookmark

No TOC