Title Page
ABSTRACT
Contents
Chapter 1. Backgrounds 13
1.1. Introduction 14
1.2. Objectives and structure of the thesis 16
References 18
Chapter 2. Missing values estimation of PM₂.₅ spatial and constituent concentrations 29
ABSTRACT 30
2.1. Introduction 31
2.2. Materials and methods 34
2.2.1. Measurement and analysis 34
2.2.2. Missing value estimation methods 36
2.3. Results and Discussion 43
2.3.1. Missing values estimation of spatial PM₂.₅ mass concentrations 43
2.3.2. Missing value estimation of PM₂.₅ constituent concentrations 50
2.4. Conclusions 57
References 59
Chapter 3. Improving mortality risk assessment by estimating missing values 71
ABSTRACT 72
3.1. Introduction 73
3.2. Materials and methods 75
3.2.1. Mortality and meteorological data 75
3.2.2. Statistical analysis 75
3.3. Results and Discussion 78
3.3.1. The effects of missing value estimation on mortality risk assessment 78
3.4. Conclusions 85
References 86
Chapter 4. Improving source apportionment by estimating missing values 97
ABSTARCT 98
4.1. Introduction 99
4.2. Materials and methods 101
4.2.1. Receptor modeling 101
4.2.2. PMF input data QA/QC 103
4.3. Results and Discussion 106
4.3.1. Source apportionment of non-missing PM₂.₅ data 106
4.3.2. The effect of missing value estimation on source apportionment 114
4.4. Conclusions 135
References 137
Chapter 5. Conclusion 145
5.1. Summary 146
5.2. Conclusion 148
국문 초록 150
Table 2-1. Actual missing rates (%) for PM₂.₅ monitoring data 42
Table 3-1. Summary of mortality data in this study 81
Fig. 1-1. Structure of the thesis. 17
Fig. 2-1. Ambient air monitoring network in 25 districts (circle, 1~25) and one supersite (star, 10) in Seoul. 36
Fig. 2-2. Distributions of spatial PM₂.₅ mass concentrations in 25 districts of Seoul (red line: mean). 46
Fig. 2-3. Accuracy comparison for missing value estimation of spatial PM₂.₅ mass concentrations (RMSE: root mean squared error, time interval: hourly). 47
Fig. 2-4. Factor profiles of spatial PM₂.₅ mass concentrations in 25 districts of Seoul. 48
Fig. 2-5. The result of factorizing sites from 25 districts into five areas. 49
Fig. 2-6. Correlation of PM₂.₅ mass concentrations measured in district 25, Seoul, Korea (number: Pearson's correlation coefficient, clustered using dendrogram). 50
Fig. 2-7. Distributions of PM₂.₅ constituent concentrations (red line: mean). 53
Fig. 2-8. Accuracy comparison for missing value estimation of PM₂.₅ constituents concentrations (RMSE: root mean squared error, error bars: standard deviation of... 54
Fig. 2-9. Accuracy of missing values estimation for PM₂.₅ constituent concentrations (top: R², bottom: RMSE). 55
Fig. 2-10. Accuracy of missing value estimation for two or more types of PM₂.₅ constituent concentrations (top: R², bottom: RMSE). 56
Fig. 3-1. Schematic comparison of mortality risk assessments by preprocessing method for missing values of spatial PM₂.₅ mass concentrations. 82
Fig. 3-2. Association between daily PM₂.₅ mass concentration and relative risk of total mortality (gray: 95% CI) 83
Fig. 3-3. Comparison of relative risks for the same target date. 84
Fig. 4-1. Source profile of non-missing data. 113
Fig. 4-2. Schematic for chapter 4. 118
Fig. 4-3. Comparison of source profiles by preprocessing method for missing values of carbon species concentrations (error bar: DISP interval). 119
Fig. 4-4. Comparison of source contributions by preprocessing method for missing values of carbon species concentrations. 120
Fig. 4-5. Comparison of source profiles by preprocessing method for missing values of ionic species concentrations (error bar: DISP interval). 121
Fig. 4-6. Comparison of source contributions by preprocessing method for missing values of ionic species concentrations. 122
Fig. 4-7. Comparison of source profiles by preprocessing method for missing values of trace elements concentrations (error bar: DISP interval). 123
Fig. 4-8. Comparison of source contributions by preprocessing method for missing values of trace elements concentrations. 124
Fig. 4-9. Comparison of source contributions by preprocessing method for missing values of carbon and ionic species concentrations (error bar: DISP interval). 125
Fig. 4-10. Comparison of source contributions by preprocessing method for missing values of carbon and ionic species concentrations. 126
Fig. 4-11. Comparison of source contributions by preprocessing method for missing values of carbon species and trace elements concentrations (error bar:... 127
Fig. 4-12. Comparison of source contributions by preprocessing method for missing values of carbon species and trace elements concentrations. 128
Fig. 4-13. Comparison of source contributions by preprocessing method for missing values of ionic species and trace elements concentrations (error bar:... 129
Fig. 4-14. Comparison of source contributions by preprocessing method for missing values of ionic species and trace elements concentrations. 130
Fig. 4-15. Comparison of source contributions by preprocessing method for missing values of all PM₂.₅ constituents concentrations (error bar: DISP interval). 131
Fig. 4-16. Comparison of source contributions by preprocessing method for missing values of all PM₂.₅ constituents concentrations. 132
Fig. 4-17. Scatterplots of observed/reconstructed OC concentrations and distributions of observed concentrations by preprocessing method for missing values. 133
Fig. 4-18. Scatterplots of observed/reconstructed EC concentrations and distributions of observed concentrations by preprocessing method for missing values. 134