Integrated processes analysis and systematic meteorological classification of ozone episodes in Hong Kong

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 111,, doi: /2005jd007012, 2006 Integrated processes analysis and systematic meteorological classification of ozone episodes in Hong Kong Jian-Ping Huang, 1,2 Jimmy C. H. Fung, 1 and Alexis K. H. Lau 1,3 Received 19 December 2005; revised 6 June 2006; accepted 5 July 2006; published 24 October [1] We have conducted a systematic and comprehensive survey of 54 O 3 episodes that occurred over Hong Kong during using the Pollutants in the Atmosphere and their Transport over Hong Kong (PATH) model system. We identify the relative contributions of regional transport and local chemical production for these O 3 episodes. Two regimes with different relative regional or local contributions are identified. Subsequently, meteorological analyses are used to help identify the synoptic patterns associated with the O 3 episodes of these two regimes. In particular, we find that the O 3 episodes are dominated by regional transport when a tropical cyclone/typhoon is located over the northwestern Pacific or the South China Sea to the east or southeast (a C pattern), when an anticyclone (i.e., a local high-pressure system) appears over mainland China to the north ( A ), or both ( AC ). On the other hand, local chemical production dominates when a trough is situated to the east over the South China Sea ( T ). Strongest regional transport is associated with the C or AC synoptic patterns. These results are likely related to the stronger horizontal advection of O 3 under the C, A, or AC patterns (as compared with T ), as well as the stronger production of O 3 in the Pearl River Delta region under clear sunny skies at the periphery of a tropical cyclone/ typhoon ( C ) and/or under a high-pressure system ( A ). These four characteristic synoptic patterns, namely, C, A, AC, and T, account for 33%, 17%, 28%, and 22% of the 54 episodes, respectively, between 2000 and Finally, back trajectory analyses also show that the lower regional contributions in the trough cases ( T ) are related to their airflow pathways not passing through the O 3 -rich Pearl River Estuary. This study also shows that a systematic and comprehensive application of integrated process analysis can be used to identify and summarize the salient features (weather patterns in this case) of specific type of air quality events. Citation: Huang, J.-P., J. C. H. Fung, and A. K. H. Lau (2006), Integrated processes analysis and systematic meteorological classification of ozone episodes in Hong Kong, J. Geophys. Res., 111,, doi: /2005jd Introduction [2] Hong Kong is situated at the southeastern coast of China and adjoins the rapidly developing Pearl River Delta (PRD) region. The rapid urbanization and industrial development of the PRD have resulted in a large increase in anthropogenic emissions over the past two decades. Between 1990 and 1995, NO x emissions increased by 21% over the PRD region, and they are predicted to increase by a further % between 1995 and 2020 [Streets and Waldhoff, 2000]. As a result, Hong Kong and the PRD have been suffering from frequent and severe air pollution events [e.g., CH2M Hill (China) Limited, 2002; Wang and Kwok, 1 Department of Mathematics, Hong Kong University of Science and Technology, Kowloon, Hong Kong. 2 Now at Department of Marine, Earth and Atmospheric Sciences, North Carolina State University, Raleigh, North Carolina, USA. 3 Also at Institute for the Environment, Hong Kong University of Science and Technology, Kowloon, Hong Kong. Copyright 2006 by the American Geophysical Union /06/2005JD ; Lee and Hills, 2003; Huang et al., 2005; Fung et al., 2005; Lam et al., 2006]. For instance, Figure 1 shows that the number of air pollution episode days increased significantly from 1999 to Figure 1 also shows that O 3 is the dominant contributor to the occurrence of air pollution episodes in Hong Kong, and that the number of O 3 episode days increased substantially between 1999 and An episode day is defined as one with an Air Pollution Index reading of (API) above 100, which is mainly associated with hourly O 3 concentrations over 120 ppb, or 24-hour NO 2 concentrations over 150 mg m 3, or 24-hour PM 10 concentrations over 180 mg m 3 at any of the ambient air quality monitoring stations set up by the Hong Kong Environmental Protection Department (HKEPD). The locations of these air quality monitoring stations are shown in Figure 2a, together with the complex terrain. A better understanding of what happen in an O 3 episode is one of the major challenges for improving the air quality in Hong Kong and the PRD region. [3] In the lower troposphere, O 3 is formed through a series of photochemical reactions involving oxides of nitro- 1of14

2 [5] However, a systematic and quantitative evaluation on how weather systems impact the occurrence of O 3 episodes in Hong Kong is still lacking. In particular, most of the earlier air quality studies are based on individual case studies, but it is often difficult to gauge the representativeness of individual cases. In this study, the PATH modeling system is used to Figure 1. Number of air pollution episode days (the Air Pollution Index, API > 100) and the contributing pollutants (O 3,NO 2, and PM 10 ) during gen (NO x =NO+NO 2 ) and volatile organic compounds (VOC) under strong solar radiation. Surface O 3 concentrations may reach episodic levels because of substantial contributions from strong photochemical production and atmospheric transport processes [e.g., Akimoto et al., 1996; Gaza, 1998; Chan and Chan, 2000; Pont and Fontan, 2000; Dayan and Levy, 2002]. Typically, regional transport is controlled by flow patterns, i.e., prevailing winds on which air pollutants can be transported from polluted regions upwind to downwind areas. O 3 production is associated with specific weather conditions such as high temperature, strong solar radiation, high humidity and so on. Those conditions follow different synoptic patterns. The climate of Hong Kong is governed by the East Asian monsoon. During late summer and early fall (i.e., August and September) when O 3 episodes are most common, frequent weather systems affecting Hong Kong include tropical cyclones or typhoons over the South China Sea or the northwestern Pacific, regional high-pressure systems (anticyclones) to the north over mainland China, and lowpressure systems (troughs) to the south and east over the South China Sea. Studies have shown that some of these systems can have significant impact on O 3 episodes in Hong Kong [Chan and Chan, 2000; Lee and Hills, 2003]. [4] In a previous study [Huang et al., 2005], a technique called integrated process rate (IPR) analysis was used to quantify the relative contributions of local ozone formation and regional transport to ozone in the Hong Kong region. This integrated process rate analysis embedded in an air quality model, the San Joaquin Valley Air Quality Study/ Atmospheric Utility Signature Predictions and Experiments (SJVAQS/AUSPEX) Regional Modeling Adaptation Project (SARMAP) air quality model (SAQM) [Chang et al., 1997], was applied to a typhoon-related O 3 episode [Huang et al., 2005] in Hong Kong. The results show that two thirds of total O 3 production within the lower atmospheric boundary layer attributed to regional transport from the PRD region during one typhoon-related synoptic pattern of O 3 episode. Figure 2. (a) Locations of the air quality monitoring stations referred to in this study. The topography of Hong Kong is also shown with contours at 200 m intervals [from Huang et al., 2005] (general air quality monitoring stations: Central/Western (CW), Kwai Chung (KC), Kwun Tong (KT), Sham Shui Po (SP), Tsuen Wan (TW), Sha Tin (ST), Tai Po (TP), Tung Chung (TC), Yuen Long (YL), and Tap Mun (TM); roadside air quality monitoring stations: Eastern (EN), Causeway Bay (CB), Central (CL), and Mong Kok (MK)). (b) Spatial distribution of weather stations of the Hong Kong Observatory (HKO) (x axis is longitude, y axis is latitude, and unit is degree). 2of14

3 Figure 3. Configuration of the four nested domains in the MM5 (solid lines) and in SAQM (dashed lines) models; resolutions from the outmost to the innermost domains are 40.5 km, 13.5 km, 4.5 km, and 1.5 km, respectively [from Huang et al., 2005]. simulate 54 O 3 episodes over Hong Kong from 2000 to (Here, an O 3 episode is defined as a continuous period when the hourly O 3 concentration is above 120ppb, and it may consist of multiple days). The IPR analysis embedded in the air quality model, SAQM, is used to study how different weather patterns affect O 3 episodes in Hong Kong. 2. Model Description and Configuration [6] The PATH model is a three-dimensional air quality modeling system. It consists of three submodels: the fifth generation Pennsylvania State University/National Center for Atmospheric Research (NCAR) Mesoscale Modeling System, MM5 [Grell et al., 1994]; an emissions processing module, the 1995 version of the United States (US) Emissions Modeling System (EMS-95) [Emigh and Wilkinson, 1995]; and an Eulerian photochemical and transport model, SAQM. The PATH model has been extensively tested and verified by the HKEPD for air pollution modeling in Hong Kong and the PRD region [Huang et al., 2005] MM5 and SAQM [7] MM5 is a community mesoscale meteorological model widely used for weather prediction, air quality, and hydrological studies [Warner et al., 1991]. MM5 provides meteorological inputs for air quality models like SAQM, the U.S. Environmental Protection Agency (EPA) Models-3 Community Multiscale Air Quality (CMAQ) modeling system [Byun and Ching, 1999], and others. [8] SAQM was originally developed by the State University of New York and was evaluated extensively by the California Air Resources Board as part of the SARMAP study in the San Joaquin Valley in central California [Chang et al., 1997]. It has been modified by HKEPD to study air quality in Hong Kong [CH2M Hill (China) Limited, 2002; Huang et al., 2005]. The Carbon Bond-IV chemical mechanism is used in this study. It contains 78 thermal-kinetic reactions, 11 photolytic reactions, and 56 reactive species. [9] Both MM5 and SAQM are used with a configuration of four nested domains, as shown in Figure 3. The horizontal resolutions from the outermost to the innermost domains for both models are 40.5 km (D1), 13.5 km (D2), 4.5 km (D3), and 1.5 km (D4), respectively. For SAQM simulations, each domain contains grid cells, whereas for MM5 and EMS-95 simulations, the numbers of grid cells along the west-east and north-south directions are , 85 73, 61 55, and grid cells for the four nested domains from the outermost to the innermost, respectively. In the vertical direction, 26 sigma layers are used in the MM5 simulation, defined unequally from the ground to the model top (100 hpa), while 15 sigma layers are used in SAQM simulations, with the lowest ten layers coinciding with those of MM5 and the 10th layer is about 1050 m above ground level. [10] Recent studies on the land use category in MM5 [Fung et al., 2005; Lo et al., 2006] have shown that correct simulation of the diurnal variations in the flow field is very important in air quality simulations. In these studies, a comparison of two separate MM5 land use data sets (i.e., US Geological Survey (USGS) and PATH, each with different parameter values and different spatial distributions) was performed to understand the importance of land surface processes and land-atmosphere interactions in the evolution of mesoscale weather phenomena during a highpollution episode in Hong Kong. Specifically, the relative importance of six land use parameters including the roughness length, thermal inertia, soil moisture availability, albedo, surface heat capacity and surface emissivity are studied. These results suggested that the availability of soil moisture is important and has a controlling influence on the flow patterns and the surface fluxes. The daytime wind field pattern depends more strongly on the availability of soil moisture than does that of the nighttime. These studies also point to the importance of coupled land surface modeling in air quality simulations, particularly for cities with complex terrains and urban characteristics similar to Hong Kong. Therefore, over Hong Kong and the PRD, an up-to-date land use data set at 30 m resolution, originally compiled by the Planning Department of the Hong Kong government in 2003, is reformatted to replace the default MM5 30 arcsecond land use data set based upon a 1993 data set from USGS. A comparison between two data sets is shown in Figure 4. Figure 4a depicts the land use category distribution over the PRD region based on the USGS data set of the default MM5 system (v3.6.6), and Figure 4b shows the corresponding land use categories from the updated data set. Urban areas are shown in red. Figure 4 shows much larger urban areas in 2003 compared with conditions in 1993; quantitatively, the area designated as urban increased by a factor of 24. With development and urbanization, anthropogenic emissions have increased significantly, and local meteorological conditions including urban heat island and coastal land-sea breeze effects, have changed remarkably. Several studies [Fung et al., 2005; Lo et al., 2006] have shown that urbanization strengthens sea breeze circulation. With a higher sensible heat flux, urban areas have higher temperatures in the lower parts of the atmosphere. These studies also suggested that the air pollution meteorology of the PRD region is strongly influenced by the local sea breeze circulation. The sea breeze convergence region is identified as the principal feature contributing to the re- 3of14

4 Figure 4. Comparison of (a) the 1993 USGS land use data set with (b) the updated land use data set in 2003 over the PRD region. gional haze problem. This recirculation of pollutants requires further investigation. [11] Four-dimensional data assimilation (FDDA) is widely used in meteorological modeling, especially in support of air quality studies. As suggested by Stauffer and Seaman [1994], analysis nudging and observational nudging have been applied to the outermost domain (D1) and the innermost domain (D4), respectively. Hourly surface wind observations at 36 automatic weather stations of Hong Kong Observatory (HKO) are used in the observational nudging for MM5 simulations in D4, while data at another 10 HKO stations selected randomly but uniformly distributed cross the HK territory, are used to evaluate the performance of MM5 on surface wind simulation (the locations of all HKO wind monitoring stations shown in Figure 2b). The SAQM default values are used for initial conditions (ICs) and boundary conditions (BCs) in the largest domain (40.5 km). The SAQM constant profiles for major chemical species (e.g., O 3, CO, NO, NO 2,SO 2 ) generated by the Hong Kong Environmental Protection Department for its air quality study in Hong Kong with PATH model system [CH2M Hill (China) Limited, 2002] are employed to this study. A spin-up period of 2 days is used to minimize the influence of initial conditions during this study. The ICs and BCs for an inner domain are provided by the simulations from its outer domain Emissions Inventory [12] The emissions inventory for the PATH system is compiled with EMS-95 on the basis of data collected for The inventory consists of five categories of emissions sources: point sources (e.g., power plants), mobile sources (e.g., vehicles, railways, aircrafts), area sources (e.g., domestic and commercial fuel combustion), shipping sources, and biogenic sources (consisting of emissions from soils and vegetation, which are based on the information on the total amount of landmass covered by different vegetation types in a given domain in conjunction with a specific meteorological data). Emissions in the first two domains (D1 and D2) are the same as report in a previous study [Huang et al., 2005], but those in the PRD (D3) and Hong Kong (D4) are updated with new data for Figure 5 shows emission rates of NO x and VOC over the PRD region and Hong Kong. Both NO x and VOC emissions are concentrated around urban areas, showing similar patterns as in the land use map (Figure 4). The total daily emissions of VOC, NO x, CO, and SO 2 over D3 (PRD) and D4 (HK) are presented in Tables 1a and 1b. The percentage of contributions from individual sources to total NO x and VOC emissions is shown in Figure 6. Point and mobile sources account for the majority of NO x emissions (75% in PRD and 77% in Hong Kong), while area and shipping sources account for the remaining parts (25% in the PRD and 23% in Hong Kong). In contrast, the contributors to VOC emissions are area, mobile, and biogenic sources (99% in PRD and 98% in Hong Kong), but area sources have a higher percentage of contribution in Hong Kong than in the PRD (51% in Hong Kong versus 28% in the PRD), while biogenic sources contribute three times more VOC in the PRD than in HK in terms of percentage contribution (34% in PRD versus 13% in HK). Overall, the contributions of individual emission sources are similar in Hong Kong and PRD for NO x emissions but different for VOC emissions. Finally, we also note that the ratio of VOC to NO x is 1.55 in PRD, approximately twice that in Hong Kong Process Analysis Method [13] Process analysis is a diagnostic tool, commonly used in air quality models, such as PATH [Huang et al., 2005], CMAQ, the Comprehensive Air Quality Model with extensions model (CAMx) [ENVIRON, 2002]. It quantifies the impact of individual physical processes and chemical reactions on the formation of O 3 and other species. Two process analysis methods are embedded in SAQM: integrated process rate (IPR) and integrated reaction rate (IRR) analyses. The former assesses the effects of all physical processes and 4of14

5 HUANG ET AL.: IPR ANALYSIS OF O3 EPISODE Figure 5. Spatial distributions of emission rates of (a) NOx and (b) VOC over PRD; (c) NOx and (d) VOC over Hong Kong (unit is g/s). the net effect of chemistry, while the latter tracks individual gas-phase chemical transformation pathways to identify the relative contributions of individual chemical reactions to the chemical species of interest (e.g., O3). The change in chemical species concentrations due to different physical and chemical processes is governed by the following partial ðuci Þ þ ðvci Þ þ ðwci Þ þ þ Pchem;i Lchem;i þ Ei þ dry ð1þ where Ci is the concentration of chemical species i; u, v, and w are the three components of the velocity vector in the x, y, and z directions, respectively; Ke is the eddy diffusivity used to parameterize the subscale turbulent fluxes of trace species; Table 1a. Summary of Total Emissions Over D4 (i.e., Hong Kong)a Total VOC NOx CO SO2 Area Biogenic 32.1 Mobile Point Shipping Total Pchem,i and Lchem,i are the chemical production and loss rates, respectively, due Ei is the source to i i and are the rates of change dry in concentration due to cloud processes and dry deposition, respectively. Each term on the right-hand side of equation (1) represents the contribution of individual processes to the total concentration of each chemical species. The processes include horizontal advection, vertical transport, chemical reactions, emissions, and dry and wet deposition. 3. Process Analysis of a Typhoon-Related O3 Episode 3.1. Observation of the 4 6 September 2002 O3 Episode [14] As an example of an O3 episode and the IPR analysis, the episode that occurred between 4 and 6 SepTable 1b. Summary of Total Emissions Over D3 (i.e., PRD Region)a VOC NOx CO SO2 a a Unit is tons per day. Area Biogenic Mobile Point Shipping Total Unit is tons per day. 5 of 14

6 Figure 6. Contributions of individual sources to total emissions of (a) NO x and (b) VOC in PRD; (c) NO x and (d) VOC in Hong Kong. tember 2002 is discussed here. This O 3 episode is related to the presence of typhoon Sinlaku (see Figure 7). Typhoon Sinlaku formed on 29 August 2002 to the east of the Mariana Islands and then tracked westward. When the typhoon was located to the east of Taiwan, the large-scale subsidence effect in its periphery made the air over Hong Kong and the PRD region very stable, which brought in weather conditions conducive to O 3 formation and buildup (high temperature, clear sky with strong solar radiation, weak surface northerly winds and a stable lower atmospheric boundary layer (ABL)). The highest hourly O 3 concentration of 156 ppb was recorded by HKEPD at Tung Chung (TC, located in the southwestern part of Hong Kong; see Figure 2a) on 6 September High O 3 concentrations Figure 7. Weather chart at 1400 LST on 6 September 2002, as an example of the C synoptic pattern for O 3 episodes. 6of14

7 Table 2. Performance Statistics of MM5 on Modeling Surface Winds (Wind Speed (ws) and Wind Direction (wd)) During Four Different Synoptic Patterns of O 3 Episodes in Hong Kong a O 3 Episode, mean_obs mean_mm5 std_dev_obs std_dev_mm5 rmsd ws, Index of Agreement yymmdd_dd Pattern ws, m s 1 ws, m s 1 ws, m s 1 ws, m s 1 ms 1 ws wd _07 C _12 A _22 AC _08 T a std_dev, standard deviations; rmsd, root-mean-square differences. were also observed at other air quality monitoring stations, such as 109 ppb at Sham Shui Po (SP), 121 ppb at Tsuen Wan (TW), 116 ppb at Sha Tin (ST), 94 ppb at Central Western (CW), and 100 ppb at the remote background station at Tap Mun (TM). The high levels of O 3 throughout Hong Kong, with highest levels in the west suggest a regional-scale O 3 episode, with significant transport of O 3 from the PRD into Hong Kong [Fung et al., 2005; Lam et al., 2006] Evaluation of MM5 Simulations [15] A 3-day simulation from 0200 local standard time (LST) on 4 (1800 UTC on 3) to 0200 LST on 7 (1800 UTC on 6) September 2002 was conducted. The initial and boundary conditions for the outmost domain are provided by 6-hourly Global Final (FNL) analysis with a horizontal resolution of 1 1 degrees in latitude and longitude. Twoway nesting is applied to the four MM5 domains. In the simulations, the medium-range forecast (MRF) planetary boundary layer scheme is applied to all four domains [Hong and Pan, 1996], while the Grell cumulus parameterization scheme is used in D1 and D2 but not in D3 and D4 as recommended by Grell et al. [1994]. [16] The ability of MM5 to simulate the surface winds (both speed and direction) is evaluated with observational data from the monitoring network operated by the Hong Kong Observatory, and the performance statistics are summarized in Table 2. The statistical parameters examined include observed and predicted means, standard deviations (std_dev), root-mean-square differences (rmsd) for wind speed (ws), and indices of agreement for wind speed (ws) and wind direction (wd). The index of agreement is defined by I ¼ 1 P N i¼1 P N i¼1 ðp i o i Þ 2 ; ð2þ ðjp i ojþjo i ojþ 2 where p i and o i are the simulated and observed values, respectively, for measurement i. N represents the total number of measurements. o denotes the average value of observations. A model index value of 1 indicates perfect agreement between model and observation while an index of 0 indicates no agreement. [17] The observed and simulated surface winds show similar means (2.06 m s 1 for observed wind speed versus 2.11 m s 1 for simulated wind speed) and standard deviations (1.15 m s 1 for observed wind speed versus 0.98 m s 1 for simulated wind speed). We can see low values of the root-mean-square difference (0.87) for wind speed and high values of the index of agreement for both wind speed (0.85) and wind direction (0.82). In addition, Figure 8 shows comparisons of simulated and observed temperature profiles at Kings Park (Hong Kong) on 6 September Under the influence of the typhoon, the atmosphere was mostly stable below 850 hpa, as evidenced by the measures of 5.7 C/km in the morning (0800 LST) and 7.7 C/km in the evening (2000 LST) when an inversion layer with rate of 2.6 C/km occurred near the surface. These are well captured by the model, suggesting that the MM5 model is able to simulate the subsidence effect associated with the typhoon, which sustained a stable ABL Evaluation of Surface O 3 Simulations [18] Figure 9 shows the time series of observed and simulated O 3 mixing ratios at six air quality monitoring stations (their locations are shown in Figure 2a). It shows that both the spatial and temporal variations in surface O 3 are well captured by SAQM, including the good match in the timing of the event, and the higher concentrations in the western part (TC; Figure 9e) than in the eastern part (TM; Figure 9b). A similar performance by SAQM is also found at other HKEPD monitoring stations (not shown). [19] The performance of SAQM on surface O 3 simulations is further evaluated through statistical methods, Figure 8. Simulated (solid lines with circles, from domain D4) and radiosonde soundings reported (dashed lines with stars) temperature profiles at King s Park (Hong Kong) on 6 September 2002; radiosonde soundings are available at 0000 and 1200 UTC (0800 and 2000 LST). 7of14

8 Figure 9. Hourly time series of simulated and observed surface O 3 at (a) Tsuen Wan, (b) Tap Mun, (c) Sham Shui Po, (d) Sha Tin, (e) Tung Chung, and (f) Central/Western between 4 and 7 September including normalized bias (D) and normalized gross error (E d ) for concentrations above a prescribed threshold, to demonstrate the model s capability in simulating high levels of O 3 concentrations [Russell and Dennis, 2000]. [20] The normalized bias, D (for hourly observed values of O 3 40 ppb) is defined as D ¼ 1 N X N j¼1 C p x i ; t j Co x i ; t j : ð3þ C o x i ; t j [21] The gross error, E d (for hourly observed values of O 3 40 ppb) is given by E d ¼ 1 X N C p x i ; t j Co x i ; t j ; ð4þ N C o x i ; t j j¼1 where N is the number of samples at monitoring station x i, C o (x i, t j ) and C p (x i, t j ) represent observed and predicted values at monitoring station x i for hour t j, respectively. [22] The statistical results are presented in Table 3. The mean normalized bias (D) of all general air quality moni- 8of14

9 HUANG ET AL.: IPR ANALYSIS OF O3 EPISODE Table 3. Average Performance Statistics of SAMQ on Surface O3 Simulation at All General Air Quality Monitoring Stations During Four Different Synoptic Patterns of O3 Episodes in Hong Konga O3 Episode Pattern D Ed Cr _ _ _ _09 C 1.7% 22.1% 0.66 A 11.1% 24.1% 0.59 AC 15.2% 26.6% 0.58 T 8.6% 21.9% 0.59 a D, mean normalized bias; Ed, mean gross error; and Cr, mean correlation coefficient _07 represents the case on 4 7 September 2002, similarly for other three cases. toring stations (see Figure 2a) around Hong Kong is about 1.7%. The mean gross error (Ed) is around 22.1% and the mean correlation coefficient is around These results suggest reasonable performance by SAQM in modeling surface O3 according to the performance criteria proposed by the United States Environmental Protection Agency (USEPA), which stipulates a normalized bias ± 15% and a gross error ± 35% [Russell and Dennis, 2000]. [23] Figure 10 shows the evolution of the spatial distributions of simulated O3 together with the simulated winds at the surface in D3 (PRD) and D4 (Hong Kong) on 6 September At 1100 LST, the O3 distribution was rather uniform and concentrations were less than 80 ppb. The surface winds were dominated by northeasterlies over the whole PRD region and no sea breeze was observed. Later, a sea breeze developed along the coastal areas and strengthened gradually through midafternoon. As a result, convergence zones were created over Lantau Island and the western part of Hong Kong. In the meantime, an O3 plume with concentrations higher than 100 ppb was found to extend southward over the Pearl River Estuary. Under the influence of the sea breeze, transport of this O3 plume into the western part of Hong Kong is seen in Figure 10d. This matches well with the occurrence of high O3 levels in the later part of the afternoon at Tung Chung (Figure 9e) and in the western part of Hong Kong Integrated Process Rate Analysis of O3 Formation [24] We use IPR to investigate O3 formation over Hong Kong in the innermost domain (D4) from 0900 LST (initial) to 1500 LST (finial) on 6 September 2002, when high levels of O3 were observed. Figure 11 shows the change in the vertical profiles of O3, and the four main controlling processes including gas-phase chemical production (CHEM), horizontal transport (HADV), vertical transport (VTRA), and dry deposition (DEPO) during this period. The O3 mixing ratios in different layers increased significantly at 1500 LST compared with those at 0900 LST, especially within the lowest eight model layers (Figure 11a). On the ground, O3 was consumed by chemical reactions, and removed by deposition, but this was more than compensated by the vertical transport from upper level, and the horizontal advection from outside Hong Kong (Figure 11b). Within the lower ABL (surface to about 500 m), the increase in O3 can be attributed mainly to chemical production and horizontal advection. By integrating the effect of both processes over the lower ABL from the surface to 500 m above Figure 10. Spatial distributions of simulated surface O3 and wind at (a) 1100 LST and (b) 1400 LST in PRD; (c) 1100 LST and (d) 1400 LST in Hong Kong on 6 September of 14

10 ground level (AGL), we found that horizontal advection and photochemical production respectively contributed about 75% and 25% of total O 3 within the lower ABL. This further quantifies and confirms the significance of regional transport of O 3 from the PRD into Hong Kong during a typhoon-related O 3 episode. 4. Process Analyses of O 3 Production for All O 3 Episodes and Identification of Controlling Synoptic Patterns [25] The PATH model system is applied to study 54 O 3 episodes from 2000 to The configurations of the MM5 and SAQM simulations are identical to those described in section 3. The performance of both models is evaluated in the same way, including temporal variations, spatial distributions, and statistical calculations. The model performance statistics for four O 3 episodes shown in Tables 2 and 3 are used to demonstrate the performance of MM5 and SAQM in modeling surface winds and O 3, respectively. Figure 11. (a) Evolution of O 3 vertical profiles and (b) IPR analysis on O 3 formation during the 6-hour period from 0900 to 1500 LST on 6 September, The deposition term only occurs in the first layer with value about 50 ppb/6 hours. Figure 12. Percentage contributions of horizontal advection (HADV) versus chemical production (CHEM) for 54 O 3 episodes over the Hong Kong domain from 2000 to A color code is used to identify the synoptic pattern for each episode. Four episodes are cases _07 (an episode occurred on 4 7 September 2002, similarly for other cases), _12, _22, and _09, corresponding to typical examples of four synoptic patterns, C, A, AC, and T, respectively. In Table 2, we can see that the observed and simulated surface winds are close in terms of similar means and stand deviations, e.g., their differences range from 0.28 to 4 for means and from 4 to 0.19 for standard deviations. High values in the index of agreement (0.85 for wind speed and 0.82 for wind direction) are found. MM5 has a similar level of performance in the other cases that are not shown here. MM5 simulations provide necessary and reasonable meteorological inputs for the air quality model, SAQM. [26] As shown in Table 3, in the SAQM simulations, the mean normalized bias (D) of all air quality general monitoring stations around Hong Kong (see Figure 2a) is about 11.1% for case _12, 15.2% for case _22, and 8.6% for case _09 (case _04 was discussed in section 3.3). The mean gross error (E d ) is around 24.1% for case _12, 26.6% for case _22, and 21.9% for case _09. The mean correlation coefficient (Cr) is around 0.59 for case _12, 0.58 for case _22, and 0.59 for case _09. Those parameters suggest the reasonable performance of SAQM in modeling surface O 3 in reference to USEPA standards (see section 3.3). [27] To gain further insight into the formation mechanisms of O 3 episodes in Hong Kong, IPR analysis is used to quantify the different physical and chemical processes during the O 3 episodes. The percentage contributions from horizontal advection versus those from local chemical production for 54 O 3 episodes during 2000 to 2004 are shown in Figure 12 and their corresponding individual values are given in Table S1 1. The values shown represent the spatial average over the innermost domain (D4). Two regimes with different relative local or regional contributions to ozone episodes in HK are identified as follows: 1 Auxiliary materials are available at ftp://ftp.agu.org/apend/jd/ 2005jd of 14

11 Figure 13. Weather charts on (a) 7 October 2003, (b) 11 October 2004, and (c) 20 October 2003, given as examples of the T, A, and AC synoptic patterns, respectively. [28] 1. Regime I is associated with cases in which more than 50% of total O 3 production is from local chemical production. Analyses of weather maps show that these cases were all associated with the synoptic situation when a trough can be seen to the east of Hong Kong over the South China Sea (and sometimes extending over the northwestern Pacific). We define this type of synoptic pattern as T (trough), and a typical example is shown in Figure 13a. A total of twelve cases are found to belong to this T pattern, corresponding to about 22% of the O 3 episodes. The IPR analyses suggest that local chemical production of O 3 in D4 (the Hong Kong domain) is more important to the occurrence of O 3 episodes under this synoptic condition. [29] 2. Regime II is characterized by cases in which more than 50% of total O 3 production in D4 is from regional transport. A review of the synoptic weather maps shows that these cases are associated with the presence of (1) an anticyclone (i.e., a high surface pressure system) over mainland China to the north of Hong Kong (synoptic pattern A, example shown in Figure 13b), or (2) a tropical cyclone over the South China Sea or the northwestern Pacific to the east or southeast of Hong Kong (synoptic pattern C, example shown in Figure 7), or (3) both an anticyclone to the north, and also a tropical cyclone to the east or southeast (synoptic pattern AC, example shown in Figure 13c). In the past five years, twelve, nine, and twelve O 3 episodes were found to be associated with the C, A, and AC synoptic patterns, respectively, within this regime. [30] Note that these synoptic patterns are necessary but not sufficient conditions for ozone episodes, and that the two regimes differ in their episodic extent. We note also that there are a number of similarities in the weather conditions over Hong Kong during all the O 3 episodes, irrespective of the type of synoptic pattern. First, all these cases are associated with little or no cloud cover, strong solar radiation, and high temperatures, which are necessary conditions for photochemical production of O 3. Moreover, the background large-scale winds over Hong Kong are weak northerly winds (which also include northeasterlies or northwesterlies). These conditions help to create a relatively stable lower tropospheric boundary layer and strong sea breeze circulation, all of which are in favor of O 3 formation and accumulation in or around Hong Kong (see section 3.3). [31] However, there are also subtle differences between the four synoptic patterns, particularly with respect to the presence of large-scale subsidence. It is well known that large-scale subsidence is often found at about km away from the center of tropical cyclones [Huang et al., 2005]. Hence, under the C or AC synoptic patterns, when a tropical cyclone is about km from the Pearl River Delta, the region already feels the storm s impact with large-scale subsidence over much of southern China. Anticyclones (i.e., high-pressure systems) are also associated with weak winds, clear skies and wide subsidence [Xu et al., 2005]. Therefore, when the region to the north of Hong Kong is under the influence of an anticyclone ( A ), conditions favorable to O 3 formation can also be expected. In other words, under the C, A, or AC synoptic conditions, the large-scale conditions to the north of Hong Kong are favorable to the formation of O 3. On the other hand, under the T synoptic pattern when there is a trough to the east of Hong Kong (see Figure 13a), there is rarely often a subsidence over southern China, and the conditions may not be favorable for the formation of O 3 (although the local conditions in Hong Kong may be favorable). 11 of 14

12 Table 4. Types and Frequencies of O 3 Episode Days and Associated Weather in Hong Kong and Its Vicinity ( ) Number of O 3 Episode Days ( ) Type Surface Pressure Pattern Surface Wind Over Hong Kong C (cyclone) tropical cyclone over western Pacific or northerly/northeasterly/northwesterly South China Sea A (anticyclone) anticyclone over mainland of China northerly or northeasterly AC tropical cyclone over western Pacific or northerly/northeasterly/northwesterly (anticyclone/cyclone) South China Sea, and anticyclone over mainland of China T (trough) low-pressure trough in vicinity of Hong Kong variable cyclone flow [32] The general characteristics of the four categories of O 3 episodes are summarized in Table 4, including the surface pressure patterns, the dominant surface winds, and the frequency of O 3 episodes in Hong Kong over the past five years. The C, A, and AC synoptic patterns are found to be most conducive to the occurrence of O 3 episodes. About 33%, 17%, and 28% of O 3 episodes in the past five years (2000 to 2004) resulted from these three synoptic patterns, respectively. The remaining high O 3 episodes (22%) happened when a low-pressure trough was found to the east of Hong Kong over the South China Sea. [33] In summary, regional formation of O 3 is expected to be more prominent under the C, A, and AC synoptic conditions, whereas O 3 is preferentially formed under the T conditions, and this is likely the reason why we see much stronger horizontal advection effects during C, A, and AC, but not during condition T. Finally, we also note that the subsidence effect of a typhoon can be much stronger than that of a typical anticyclone, and that may explain the findings that the strongest advection cases (pattern C or AC ) are all associated with tropical cyclones. 5. Identification of Airflow Pathways During Different Synoptic Patterns of O 3 Episodes [34] As demonstrated previously, horizontal advection and local chemical production make different contributions during the four synoptic patterns of O 3 episodes. For example, cases _07, _12, _22, and _09 are four typical examples for synoptic patterns C, A, AC, and T, respectively. IPR results show that the ratios of regional transport versus chemical production are respectively 75% versus 25%, 58% versus 42%, 60% versus 40%, and 45% versus 55% during these four episodes. To obtain a better understanding of the role of regional transport during different synoptic patterns of O 3 episodes, the National Oceanic Atmospheric Administration s (NOAA) hybrid single-particle Lagrangian integrated trajectories (HYSPLIT) model is used to identify the airflow pathways reaching Hong Kong during these O 3 episodes. [35] The HYSPLIT model is a complete system for computing simple air parcel trajectories for complex dispersion and deposition simulations [Draxler, 1998]. The HYSPLIT model has been widely used to air quality studies in Hong Kong to identify the sources of air pollutants and air masses [e.g., Chan et al., 2003; Lee and Hills, 2003; Louie et al., 2005]. The backward trajectories in this study are calculated for domain 3 (a resolution of 4.5 km) using MM5 wind simulations with an interval of 1 hour. Figure 14 shows 24-hour backward trajectories of air masses reaching the TC station ending at 1500 LST (i.e., 0700 UTC) on the episode days. The backward trajectory at TC is used to show the impact of regional transport from the PRD on the air pollutant levels in Hong Kong. [36] The trajectories in Figure 14 show different pathways at four levels in the lower part of the ABL (20 m, 200 m, 500 m and 1000 m AGL) during the four different O 3 episodes. For the first case (020904_07, synoptic pattern C ), at 20 m AGL, air parcels originating from the inner PRD region in the vicinity of Zhongshan followed a local looping pattern between Zhongshan, Dongguan, Zhuhai, and Shenzhen before reaching Hong Kong. The trajectories at the 200 and 500 m AGL, followed a very similar pattern, passing through the inner Pearl River Estuary. This region has been shown to be a much more polluted area with higher levels of O 3 (see Figure 10b). Thus, after traveling through this O 3 -rich region, air parcels can carry more of the O 3 -rich air into Hong Kong. However, the trajectory at 1000 m AGL (close to the top layer of the ABL) did not show a reversed wind pattern. Thus regional transport from PRD was mainly concentrated the lower part of the ABL, where a breezes were developed. As analyzed earlier, about 75% of the total O 3 production was attributed to regional transport from outside Hong Kong in this case. [37] In the second case (041009_12, synoptic pattern A ), the trajectories at 20 m, 200 m, and 500 m AGL, also showed a very similar pathway pattern. A sea breeze was observed within the lower part of the ABL. Air parcels traversed the inner Pearl River Estuary, an O 3 -rich region, carrying some of the O 3 -rich air from the PRD into Hong Kong. In this case, about 55% of the O 3 production came from regional transport. This is less than that in the first case, which may partly be related to the trajectories spending less time over the Pearl River Estuary. [38] The third case (031019_22, synoptic pattern AC ) follows a similar pathway as the second case, with about 60% of the O 3 production from regional transport. [39] In the last case (031006_09, synoptic pattern T ), the airflow trajectories were different from those of the first three cases. The reversed wind pattern was not clear as in the previous cases. Air parcels did not pass through much of the Pearl River Estuary, and only 45% of total O 3 production came from regional transport. [40] In summary, the analyses suggest that the airflow pathways have a significant impact on the percentage of contribution from regional O 3 transport during O 3 episodes; the longer the time (during the day) the parcels stay over the Pearl River Estuary before arriving in Hong Kong, the 12 of 14

13 HUANG ET AL.: IPR ANALYSIS OF O3 EPISODE Figure 14. The 24-hour backward trajectories at 20 m, 200 m, 500 m, and 1000 m above ground level on (a) 6 September 2002, (b) 11 October 2004, (c) 20 October 2003, and (d) 7 October 2003 for C, A, AC, and T synoptic patterns, respectively. higher the percentage of contribution from regional O3 transport. 6. Summary and Conclusions [41] The PATH model system is used to investigate 54 O3 episodes during 2000 to A more up-to-date and higherresolution land use data set is used to replace that in the standard MM5 system over Hong Kong and PRD. FDDA is used in the MM5 simulations, including analysis nudging in the outmost domain (D1) and observation nudging in the innermost domain (D4). The MM5 simulations of surface winds and temperatures are evaluated in terms of temporal variations and statistical methods, and the results show high agreement between simulations and observational data. New emissions inventories, compiled for year 2001, provide emissions inputs to the air quality model. The surface O3 simulations are evaluated through direct observation-simulation comparisons and also by statistical methods. The results show reasonable performance of the surface O3 simulations, with respect to the criteria set by USEPA, such as the normalized bias and gross error. [42] Integrated process analysis embedded in SAQM is used to quantify the relative contributions of regional 13 of 14

14 transport versus local chemical production in the O 3 episodes. Two different regimes with varying importance of regional transport versus local chemical production are identified. Meteorological analyses are then used to help classify the synoptic patterns associated with these two regimes. Four recurring types of synoptic patterns are noted, namely, T (weak trough to the east of Hong Kong over the South China Sea, 12 cases), A (anticyclone or highpressure system over the continent to the north, 9 cases), C (tropical cyclone in the northwestern Pacific or the South China Sea, 18 cases), and AC (a mixture of A and C, 15 cases). [43] Under regime I, when most of the O 3 is formed through local chemical production and less than 50% is brought in through regional transport, all twelve cases are found to be associated with the T synoptic pattern. Regime II, characterized by 5075% of the O 3 over Hong Kong from regional transport, has forty-two cases and is identified with A, C, or AC synoptic patterns. [44] C, A, or AC synoptic patterns are associated with clear and sunny skies over southern China. The lower troposphere would be more stable when a tropical cyclone is in the vicinity ( C or AC ). Hence regional formation of O 3 over the PRD is expected to be more prominent under these conditions, but not under T. This is likely the reason why we see much stronger horizontal advection with the C, AC, and A synoptic patterns, but not with T. The NOAA HYSPLIT trajectory model is used to identify the air masses reaching Hong Kong during O 3 episodes. The result shows that airflow pathways have a significant impact on the percentage of contribution from regional transport under different synoptic patterns. Cases with tropical cyclones to the east ( C ) and/or anticyclones to the north ( A ) are associated with back trajectories through the inner PRD region (an O 3 -rich region in the simulations) and higher percentages of contribution from regional transport. In contrast, cases with troughs to the east or southeast ( T ) are associated with back trajectories not passing through the O 3 -rich inner PRD region, resulting in weaker regional transport of O 3. [45] Most of the earlier air quality studies were based on individual case(s), and it is difficult to gauge the representativeness of just a few cases. This study shows that a systematic and comprehensive application of integrated process analysis can be used to identify and summarize the salient features (weather patterns in this case) of specific type of air quality events. This approach is much more powerful than individual case studies for gaining a generic understanding of air quality events. [46] Acknowledgments. This work was supported by the RGC grant under NSFC/HKUST36 and , the UGC under grants HIA02/ 03.SC04 and N_HKUST630/04, and NSFC grant The authors thank the Hong Kong Environment Protection Department for the provision of the PATH model and air quality data and the Hong Kong Observatory for provision of the meteorological data. References Akimoto, H., et al. (1996), Long-range transport of ozone in the East Asian Pacific rim region, J. Geophys. Res., 101, Byun, D. W., and J. K. S. Ching (1999), Science algorithms of the EPA Models-3 Community multiscale air quality (CMAQ) modeling system, EPA/600/R-99/030, U.S. Environ. Prot. Agency, Washington, D. C. CH2M Hill (China) Limited (2002), Study of air quality in the Pearl River Delta region, Hong Kong Environ. Prot. Dep., Hong Kong Spec. Admin. Reg. Govt., Apr. (Available at environmentinhk/air/studyrpts/files/final_rept.pdf) Chan, C. Y., and L. Y. Chan (2000), Effect of meteorology and air pollutant transport on ozone episodes at a subtropical coastal Asian city, Hong Kong, J. Geophys. Res., 105, 20,707 20,724. Chan, C. Y., L. Y. Chan, J. M. Harris, S. J. Oltmans, D. R. Blake, Y. Qin, Y. G. Zheng, and X. D. Zheng (2003), Characteristics of biomass burning emission sources, transport, and chemical speciation in enhanced springtime tropospheric ozone profile over Hong Kong, J. Geophys. Res., 108(D1), 4015, doi: /2001jd Chang, J. S., S. X. Jin, Y. Li, M. Beauharnois, C. H. Lu, H. C. Huang, S. Tanrikulu, and J. DaMassa (1997), The SARMAP Air Quality Model, final report, Calif. Air Resour. Board, Sacramento. Dayan, U., and I. Levy (2002), Relationship between synoptic-scale atmospheric circulation and ozone concentrations over Israel, J. Geophys. Res., 107(D24), 4813, doi: /2002jd Draxler, R. R. (1998), Hybrid single-particle Lagrangian integrated trajectories (HY-SPLIT): Model description, NOAA Tech. Memo. ERL ARL0166, Silver Spring, Md. Emigh, R. A., and J. G. Wilkinson (1995), The Emissions Modelling System (EMS-95), user s guide geophysics report, Alp. Geophys., Boulder, Colo. ENVIRON (2002), User s guide, Comprehensive Air Quality Model with extensions (CAMx), version 4.00, ENVIRON Int. Corp., Novata, Calif. Fung, J. C. H., A. K. H. Lau, J. S. L. Lam, and Z. Yuan (2005), Observational and modeling analysis of a severe air pollution episode in western Hong Kong, J. Geophys. Res., 110, D09105, doi: / 2004JD Gaza, R. S. (1998), Mesoscale meteorology and high ozone in the northeast United States, J. Appl. Meteorol., 27, Grell, G. A., J. Dudia, and D. R. Stauffer (1994), A description of the fifth generation Penn State/NCAR Mesoscale Model (MM5), NCAR Tech. Note NCAR/TN-398+STR, Natl. Cent. for Atmos. Res., Boulder, Colo. Hong, S. Y., and H. L. Pan (1996), Nonlocal boundary layer vertical diffusion in a medium-range forecast model, Mon. Weather Rev., 124, Huang, J.-P., J. C. H. Fung, A. K. H. Lau, and Y. Qin (2005), Numerical simulation and process analysis of typhoon-related episodes in Hong Kong, J. Geophys. Res., 110, D05301, doi: /2004jd Lam, J. S. L., A. K. H. Lau, and J. C. H. Fung (2006), Application of refined land-use categories for high resolution mesoscale atmospheric modeling, Boundary Layer Meteorol., 119(2), , doi: / s Lee, Y. C., and P. R. Hills (2003), Cool season pollution episodes in Hong Kong, , Atmos. Environ., 37, Lo, J. C. F., A. K. H. Lau, F. Chen, and J. C. H. Fung (2006), Urban modification in a mesoscale model and the effects on the local circulation in the Pearl River Delta region, J. Appl. Meteorol., in press. Louie, P. K. K., J. G. Watson, J. C. Chow, A. Chen, D. W. M. Sin, and A. K. H. Lau (2005), Seasonal characteristics and regional transport of PM2.5 in Hong Kong, Atmos. Environ., 39, Pont, V., and J. Fontan (2000), Correlation between continental air mass and ozone concentrations, J. Geophys. Res., 105(D14), 17,699 17,708. Russell, A., and R. Dennis (2000), NARSTO critical review of photochemical models and modeling, Atmos. Environ., 34, Stauffer, D. R., and N. L. Seaman (1994), Multiscale four-dimensional data assimilation, J. Appl. Meteorol., 33, Streets, D. G., and S. T. Waldhoff (2000), Present and future emissions of air pollutants in China: SO 2,NO x, and CO, Atmos. Environ., 34, Wang, T., and Y. H. J. Kwok (2003), Measurement and analysis of a multiday photochemical smog episode in Pearl River Delta of China, J. Appl., Meteorol., 42, Warner, T. T., D. F. Kibler, and R. L. Steinhart (1991), Separate and coupled testing of meteorological and hydrological forecasting models for the Susquehanna River Basin in Pennsylvania, J. Appl. Meteorol., 30, Xu, H. M., S. P. Xie, and Y. Q. Wang (2005), Subseasonal variability of the southeast Pacific stratus cloud deck, J. Clim., 18, J. C. H. Fung and A. K. H. Lau, Department of Mathematics, Hong Kong University of Science and Technology, Kowloon, Hong Kong. J.-P. Huang, Department of Marine, Earth and Atmospheric Sciences, North Carolina State University, Raleigh, NC 27695, USA. (jhuang5@ ncsu.edu) 14 of 14