RESUMEN

In this study, six individual machine learning (ML) models and a stacked ensemble model (SEM) were used for daytime visibility estimation at Bangkok airport during the dry season (November-April) for 2017-2022. The individual ML models are random forest, adaptive boosting, gradient boosting, extreme gradient boosting, light gradient boosting machine, and cat boosting. The SEM was developed by the combination of outputs from the individual models. Furthermore, the impact of factors affecting visibility was examined using the Shapley Additive exPlanation (SHAP) method, an interpretable ML technique inspired by the game theory-based approach. The predictor variables include different air pollutants, meteorological variables, and time-related variables. The light gradient boosting machine model is identified as the most effective individual ML model. On an hourly time scale, it showed the best performance across three out of four metrics with the ρ = 0.86, MB = 0, ME = 0.48 km (second lowest), and RMSE = 0.8 km. On a daily time scale, the model performed the best for all evaluation metrics with ρ = 0.92, MB = 0.0 km, ME = 0.3 km, and RMSE = 0.43 km. The SEM outperformed all the individual models across three out of four metrics on an hourly time scale with ρ = 0.88, MB = 0.0 km, (second lowest), and RMSE = 0.75 km. On the daily scale, it performed the best with ρ = 0.93, MB = 0.02 km, ME = 0.27 km, and RMSE = 0.4 km. The seasonal average original (VISorig) and meteorologically normalized visibility (VISnorm) decrease from 2017 to 2021 but increase in 2022. The rate of decrease in VISorig is double than rate of decrease in VISnorm which suggests the effect of meteorology visibility degradation. The SHAP analysis identified relative humidity (RH), PM2.5, PM10, day of the season year (i.e., Julian day) (JD), and O3 as the most important variables affecting visibility. At low RH, visibility is not sensitive to changes in RH. However, beyond a threshold, a negative correlation between RH and visibility is found potentially due to the hygroscopic growth of aerosols. The dependence of the Shapley values of PM2.5 and PM10 on RH and the change in average visibilities under different RH intervals also suggest the effect of hygroscopic growth of aerosol on visibility. A negative relationship has been identified between visibility and both PM2.5 and PM10. Visibility is positively correlated with O3 at lower to moderate concentrations, with diminishing impact at very high concentrations. The JD is strongly negatively related to visibility during winter while weakly associated positively later in summer. Findings from this research suggest the feasibility of employing machine learning techniques for predicting visibility and understanding the factors influencing its fluctuations.

RESUMEN

This study deals with haze characteristics under the influence of the cold surge and sea breeze for Greater Bangkok (GBK) in 2017-2022, including haze intensity and duration, meteorological classification for haze, and the potential effects of secondary aerosols and biomass burning. A total of 38 haze episodes and 159 haze days were identified. The episode duration varies from a single day to up to 14 days, suggesting different pathways of its formation and evolution. Short-duration episodes of 1-2 days are the most frequent with 18 episodes, and the frequency of haze episodes decreases as the haze duration increases. The increase in complexity in the formation of relatively longer episodes is suggested by a relatively higher coefficient of variation for PM2.5. Four meteorology-based types of haze episodes were classified. Type I is caused by the arrival of the cold surge in GBK, which leads to the development of stagnant conditions favorable for haze formation. Type II is induced by sea breeze, which leads to the accumulation of air pollutants due to its local recirculation and development of the thermal internal boundary layer. Type III consists of the haze episodes caused by the synergetic effect of the cold surge and sea breeze while Type IV consists of short haze episodes that are not affected by either the cold surge or sea breeze. Type II is the most frequent (15 episodes), while Type III is the most persistent and most polluted haze type. The spread of haze or region of relatively higher aerosol optical depth outside GBK in Type III is potentially due to advection and dispersion, while that in Type IV is due to short 1-day episodes potentially affected by biomass burning. Due to cold surge, the coolest and driest weather condition is found under Type I, while Type II has the most humid condition and highest recirculation factor due to the highest average sea breeze duration and penetration. The precursor ratio method suggests the potential effect of secondary aerosols on 34% of the total haze episodes. Additionally, biomass burning is found to potentially affect half of the total episodes as suggested by the examination of back trajectories and fire hotspots. Based on these results, some policy implications and future work are also suggested.

Asunto(s)

RESUMEN

Visibility and aerosol optical depth (AOD) characterization, and their relationship with PM10 and local and synoptic meteorology, were studied for January-March in 2014 and 2015 over Bangkok. Visibility degradation intensifies in the dry season as compared to the wet season due to increase in PM10 and unfavorable meteorological conditions. The average visibility is lower in January and February as compared to the other months. Relatively higher AOD in March despite lower PM10 is attributed to the synergetic effect of moderate relative humidity, secondary aerosols, elevated aerosol layer due to summertime convection, and biomass burning. Larger variability in visibility and PM10 in winter months is due to more synoptic weather fluctuations while AOD shows similar variability for all months attributed partly to fires. Higher PM10 and moderate-to-high relative humidity cause lower visibility in the morning while it improves in afternoon as PM10 and relative humidity decrease. AOD is higher in the afternoon as compared to that in the morning and evening as it is less sensitive to diurnal change in aerosols and meteorology at the surface level. Visibility and AOD relationships with PM10 are dependent on relative humidity. Weaker winds lead to lower visibility, higher PM10, and higher AOD irrespective of wind direction. Stronger winds improve visibility and decrease PM10 for all directions while AOD is higher for all directions except eastern and northeastern. The back-trajectory results show that the transport of pollutant and moist air is coupled with the synoptic weather and influence visibility and AOD. Two low-visibility events were investigated. The first event is potentially caused by the combined effect of local emissions and their accumulation due to stagnant weather conditions, secondary aerosols, and forest fires in the nearby regions. The second event can be attributed to the local emission and fires in the nearby area with hygroscopic growth of aerosols due to moist air from the Gulf of Thailand. Based on these findings, some policy implications have also been given.

Asunto(s)

RESUMEN

This present work investigates several local and synoptic meteorological aspects associated with two wintertime haze episodes in Greater Bangkok using observational data, covering synoptic patterns evolution, day-to-day and diurnal variation, dynamic stability, temperature inversion, and back-trajectories. The episodes include an elevated haze event of 16 days (14-29 January 2015) for the first episode and 8 days (19-26 December 2017) for the second episode, together with some days before and after the haze event. Daily PM2.5 was found to be 50 µg m-3 or higher over most of the days during both haze events. These haze events commonly have cold surges as the background synoptic feature to initiate or trigger haze evolution. A cold surge reached the study area before the start of each haze event, causing temperature and relative humidity to drop abruptly initially but then gradually increased as the cold surge weakened or dissipated. Wind speed was relatively high when the cold surge was active. Global radiation was generally modulated by cloud cover, which turns relatively high during each haze event because cold surge induces less cloud. Daytime dynamic stability was generally unstable along the course of each haze event, except being stable at the ending of the second haze event due to a tropical depression. In each haze event, low-level temperature inversion existed, with multiple layers seen in the beginning, effectively suppressing atmospheric dilution. Large-scale subsidence inversion aloft was also persistently present. In both episodes, PM2.5 showed stronger diurnality during the time of elevated haze, as compared to the pre- and post-haze periods. During the first episode, an apparent contrast of PM2.5 diurnality was seen between the first and second parts of the haze event with relatively low afternoon PM2.5 over its first part, but relatively high afternoon PM2.5 over its second part, possibly due to the role of secondary aerosols. PM2.5/PM10 ratio was relatively lower in the first episode because of more impact of biomass burning, which was in general agreement with back-trajectories and active fire hotspots. The second haze event, with little biomass burning in the region, was likely to be caused mainly by local anthropogenic emissions. These findings suggest a need for haze-related policymaking with an integrated approach that accounts for all important emission sectors for both particulate and gaseous precursors of secondary aerosols. Given that cold surges induce an abrupt change in local meteorology, the time window to apply control measures for haze is limited, emphasizing the need for readiness in mitigation responses and early public warning.

Asunto(s)