Long-Term Observed Visibility in Eastern : Temporal Variation, Association with Air Pollutants and Meteorological Factors, and Trends (Supplementary Materials)

Sec. S1 Change point detection for changed or modified targets

A change point detection (CPD) technique was employed in order to determine the potential inhomogeneity in visibility data due to changed or modified targets in the past. CPD is a statistical method that essentially compares two sample distributions at any point of time in a time series, formed by the original data divided by the periods before and after that time, then performing a significance test against the null hypothesis of the same population distribution [1–2]. If the null hypothesis fails, that particular time is a change point (i.e., inhomogeneity). Here, CPD was implemented using the library CPM [2] on the statistical R software [3], with the option of a Mann– Whitney nonparametric test (at a significance level of 0.1). In addition to the changed or modified targets, any abrupt change due to climate or air pollution can also induce a change point, as well the given visibility dependence on both meteorology and air pollution, thus confounding the matter. To mitigate such effects to some degree, we assumed that such abrupt changes are less likely to occur over a relatively short-term period (here, 10 years or less). With this, in our CPD processing, given a monthly average visibility series spanning 1981–2015 (35 years), the CPD is to be individually run for the following six 10-year sub-periods (with five years overlapping): 1981–1990, 1986–1995, 1991–2000, 1996–2005, 2001–2010, and 2006–2015. The entire data is viewed as homogeneous if all of the sub- periods pass the CPD test. All of the visibility data at the selected TMD (Thai Meteorological Department) stations were found to pass the CPD test.

Table S1. Numbers of days before and after the met-screening by station and by season.

Total no. of days excluded Initial no. No. of days with a % Days finally Station (missing and by of days missing value remaining met-screening) Dry season Chon Buri (CB) 6343 1 (<1%) 1170 81.6 Laem Chabang (LC) 2718 1 (<1%) 539 80.2 Sattahip (ST) 6343 11 (<1%) 1410 77.8 Rayong (RY) 1631 62 (3.8%) 396 75.7 (CT) 6343 1 (<1%) 1020 83.9 Sa Kaeo (SK) 2718 1 (<1%) 121 95.5 Prachin Buri (PB) 6343 18 (<1%) 385 93.9 Wet season Chon Buri (CB) 6440 2 (<1%) 1524 76.3 Laem Chabang (LC) 2760 3 (<1%) 604 78.1 Sattahip (ST) 6440 14 (<1%) 2156 66.5 Rayong (RY) 1656 0 (0%) 664 59.9 Chanthaburi (CT) 6440 0 (0%) 3789 41.2 (SK) 2760 1 (<1%) 836 69.7 Prachin Buri (PB) 6440 13 (<1%) 2096 67.5

Atmosphere 2019, 10, 122; doi:10.3390/atmos10030122 www.mdpi.com/journal/atmosphere Atmosphere 2019, 10, 122 2 of 5

Table S2. Explanatory variables initially tried and finally used for the generalized linear models of

visibility (VIS) and PM10.

VIS PM10 Variable CB LC RY CB LC RY Temperature √ √ √ √ √ √ Relative humidity  √   √ √ Cloud cover    √   Mixing height       Wind speed   √   √ Recirculation factor √    √  Back-trajectory cluster       Persistence √ √ √ √ √ √ Adjusted R2 (as % of variance explained) 42.6 61.7 52.7 63.5 54.4 51.3 (using all of the variables above) Adjusted R2 (as % of variance explained) (using only the variables marked by the 41.2 59.7 51.9 62.7 53.9 49.7 “√” symbol)

Table S3. Visibility (VIS) and seasonality index (SI) by station and by season. SD—standard deviation.

Dry season Wet season Station SI (%) (mean  SD, km) (mean  SD, km) VIS before met-screening Chon Buri (CB) 8.3 ± 2.3 10.9 ± 2.9 31.3 Laem Chabang (LC) 6.7 ± 2.3 9.9 ± 3.1 47.8 Sattahip (ST) 8.9 ± 3.5 11.6 ± 3.4 30.3 Rayong (RY) 7.4 ± 2.7 9.6 ± 3.2 29.7 Chanthaburi (CT) 8.2 ± 1.8 8.1 ± 2.3 −1.2 Sa Kaeo (SK) 8.6 ± 1.6 9.4 ± 2.1 9.3 Prachin Buri (PB) 8.8 ± 1.5 10.1 ± 2.5 14.8 VIS after met-screening Chon Buri (CB) 8.6 ± 2.3 11.8 ± 2.3 37.2 Laem Chabang (LC) 6.9 ± 2.3 10.7 ± 2.7 55.1 Sattahip (ST) 9.2 ± 3.5 12.5 ± 2.9 35.9 Rayong (RY) 7.8 ± 2.9 11.1 ± 2.8 42.3 Chanthaburi (CT) 8.4 ± 1.8 10.0 ± 1.6 19.0 Sa Kaeo (SK) 8.7 ± 1.6 10.1 ± 1.7 16.1 Prachin Buri (PB) 8.9 ± 1.4 11.2 ± 1.6 25.8

Atmosphere 2019, 10, 122 3 of 5

a) b) c)

d) e) f)

g)

Figure S1. Diurnal variation of visibility and other meteorological variables at Chon Buri (CB), Laem Chabang (LC) and Rayong (RY): a) visibility (VIS) (before met-screening); b) visibility (VIS); c) relative humidity (RH); d) temperature (TEMP); e) mixing height (MH); f) wind speed (WS); and g) global radiation (GR) . All of the results are of after the met-screening. For CB and RY, a relatively lower visibility is seen during the night. Note that no nighttime monitoring at LC was made. The increased visibility in the afternoon is attributable to warmer and more unstable conditions, which promote a higher mixing height and more atmospheric dilution. Also, the increased wind speed (due to the more unstable condition) and reduced relative humidity (due to the warmer condition) in the afternoon give favorable conditions for visibility to improve. The relatively low visibility in the night and early morning is potentially caused by a higher relative humidity and low mixing height. The intensified air pollution from traffic during rush hours (e.g., at 0700 LT) is likely to be responsible for a slight drop in visibility at CB, whose background is urban.

Atmosphere 2019, 10, 122 4 of 5

Figure S2. Frequency (days) of each visibility class (I: Good, II: Moderate, and III: Poor) by station in the dry season. At each station, the left- and right-hand-side bars are of before and after the met- screening, respectively. Here, CB: Chon Buri, LC: Laem Chabang, ST: Sattahip, RY: Rayong, CT: Chanthaburi, SK: Sa Kaeo, PB: Prachin Buri.

a)

b)

Atmosphere 2019, 10, 122 5 of 5

c)

Figure S3. Chon Buri (CB), Laem Chabang (LC), and Rayong (RY) and their nearby PCD (Pollution Control Department) stations (Google Maps). Roads and highways are marked by yellow or white lines, but industrial estates are marked by red circles.

a) b) c)

d) e)

Figure S4. Diurnal variation of air pollutants at Chon Buri (CB), Laem Chabang ( LC), and Rayong

(RY): a) PM10; b) CO; c) SO2; d) NOx; and e) NMHC.

References 1. Reeves, J.; Chen, J.; Wang, X. L.; Lund, R.; Lu, Q.Q. A review and comparison of changepoint detection techniques for climate data. J. Appl. Meteorol. Climatol. 2007, 46, 900-915. 2. Ross, G. Parametric and nonparametric sequential change detection in R: The cpm package. J. Stat. Softw. 2015, 66, 3. 3. R Development Core Team R: A language and environment for statistical computing (version 3.1.0). Vienna R Foudation Stat. Comput. 2014.