Development of a moving point source model for shipping emission dispersion modelling in EPISODE-CityChem v1.3

. This paper demonstrates the development of a moving point source (MPS) model for simulating the atmospheric dispersion of pollutants emitted from ships under movement. The new model is integrated into the chemistry transport model EPISODE-CityChem v1.3. In the new model, ship parameters, especially speed and direction, are included to simulate the instantaneous ship positions and then the emission dispersion at different simulation time. The model was ﬁrst applied to shipping emission dispersion modelling under simpliﬁed conditions, and the instantaneous and hourly averaged emission 5 concentrations predicted by the MPS model and the commonly-used line source (LS) and ﬁxed point source (FPS) models were compared. The instantaneous calculations were quite different due to the different ways to treat the moving emission sources by different models. However, for the hourly averaged concentrations, the differences became smaller especially for a large number of ships. The new model was applied to a real conﬁguration from the seas around Singapore that included hundreds of ships and their dispersion was simulated over a period of a few hours. The simulated results were compared to 10 measured values at different locations, and it was found that reasonable emission concentrations were predicted by the moving point source model. model MAFOR with a 3D Eulerian chemistry transport model EPISODE-CityChem. In these studies, the chemistry transport models (mainly Eulerian models) have shown

To evaluate the contributions of ship emissions on air quality in coastal areas, atmospheric dispersion modelling of the pollutants, such as NO x , SO 2 and Particulate Matter (PM), in a regional or city scale by considering the local meteorological conditions, topographical information, turbulent diffusion, and chemical transformations is a useful approach. Different dispersion softwares such as a Gaussian model and a Eulerian model have been developed and widely applied in numerical simulations (Milazzo et al., 2017;Gibson et al., 2013;De Nicola et al., 2013;Mallet et al., 2018;Kukkonen et al., 2016). The 25 most common and simplest one is a Gaussian-based model that assumes the dispersion of air pollutant to follow a Gaussian distribution. Merico et al. (2019) applied a steady-state Gaussian-based model − ADMS-5 − to estimate the dispersion of emissions from ships which are mainly in the hoteling and maneuvering phase in the harbour area of an Italian port city Bari.
The same model was also used by Cesari et al. (2016) for a case study of ship emissions in Brindisi, Italy. Another popular steady-state Gaussian plume model, AERMOD, recommended by United States Environmental Protection Agency (EPA) was 30 also widely used by different groups (Gibson et al., 2013;Cohan et al., 2011). Abrutytė et al. (2014) employed AERMOD to simulate the dispersion of NO x from ships in Klaipeda port. AERMOD was also used by Fileni et al. (2019) and Cohan et al. (2011) to evaluate the contribution of ships on PM emissions in harbour cities. The Gaussian plume model is able to save a lot of computational cost, however, it suffers from several limitations, such as assuming a steady-state solution, a spatially uniform meteorology and straight line trajectories (Bluett et al., 2004), that make it not suitable under many conditions for air 35 quality modelling. In addition to the simple Gaussian plume models, some advanced, unsteady Gaussian puff models (such as CALPUFF), which can simulate the effects of time and space-varying meteorological conditions on pollutant transport, transformation and removal (Bluett et al., 2004), are developed as well. CALPUFF has been widely used for simulating the dispersion of ship emissions. Jahangiri et al. (2018) applied CALPUFF to predict the average values of the ship emissions on the port area of Brisbane, Australia for the whole of 2013. Poplawski et al. (2011) and Murena et al. (2018) also employed 40 CALPUFF to evaluate the effects of cruise ships on air quality in the harbours of Victoria, Canada and Naples, Italy respectively. Compared to the Gaussian plume models, the advanced unsteady Gaussian puff models overcome some limitations, for example, the causality effects can be simulated by CALPUFF. Furthermore, a Lagrangian or Eulerian chemistry transport model (CTM) that solves the advection-diffusion equation and the atmospheric chemistry to predict the transport and chemical reactions of emission species received an increasing attention the ability of predicting the pollutant concentrations at the locations of interest, and hence are a good approach for investigating the environmental impact of ship emissions in coastal cities.
In pollutant dispersion modelling, it is necessary to include an appropriate assumption or model for treating the emission sources. For a typical setup of shipping emission dispersion simulation, the definition for an emission source usually depends on the ship status, namely that the ship is at berth (hoteling) or on cruise (maneuvering and cruising). Usually, the ship at hoteling 60 phase is treated as a fixed point emission source (Merico et al., 2019;Poplawski et al., 2011;Deniz and Kilic, 2010;Lucialli et al., 2011;Formentin, 2017), which is a reasonable assumption as the ship stops at the dock and generates emissions from its chimney as a single point. For the ships under movement, different models have been applied to treat the emission sources in different studies. Iodice et al. (2017) assumed the emissions from the moving ships were generated at multiple fixed points along the predefined navigation route. Saxe and Larsen (2004) used fixed points to represent the average positions of the ships in both maneuvering and hoteling modes. Murena et al. (2018) also treated the ships in maneuvering and navigation modes as fixed point emission sources. In another group of studies, a line emission source model was widely applied to simulate the moving ships in maneuvering or cruising mode (Poplawski et al., 2011;Kotrikla et al., 2013;Deniz and Kilic, 2010;Lucialli et al., 2011). In the line source (LS) model setup, the ship emission is assumed to be constantly emitted along the entire ship route which is assumed as a straight line. In some other cases, ships in hoteling or maneuvering modes were treated as area 70 sources (Kotrikla et al., 2013;Formentin, 2017;Abrutytė et al., 2014), however, this assumption is not commonly used to treat ship emissions.
From the above literature review, it is evident that either a (or multiple) fixed point(s) source model or a line source model is commonly used for ships under movement in the air pollution dispersion modelling. However, neither of these assumptions is realistic as the ship position is changing when it is moving. The ship movement is not explicitly included in current air pollution 75 dispersion modelling and leads to a research gap. In this paper, a moving point source (MPS) model was hence developed, that can update the ship positions at different time based on the ship speed and direction and then simulate the emission release from the moving ships in the dispersion modelling. The new developed MPS model was integrated into the 3D Eulerian chemistry transport model EPISODE-CityChem (Hamer et al., 2020;Karl et al., 2019) and applied to predict the dispersion of NO 2 species generated by the ships in cruising mode in a simplified simulation, and the simulated results were compared to those 80 obtained by using LS and fixed point source (FPS) models. In addition, the new MPS model was applied to a real case study to predict the concentrations of NO 2 and PM 2.5 species contributed by all ships around the Singapore city, and the simulated results were compared to the measured values in different observation stations. The MPS model introduces a new approach for treating the ships and other objects under movement in the atmospheric dispersion modelling, and will increase the knowledge of the atmospheric environment modelers. The model setups and important simulation results are presented in this paper. 85 of the dispersion model EPISODE, which is originally developed by Slørdal et al. (2003Slørdal et al. ( , 2008. In this section, the dispersion model, the MPS model, simulation setups and configurations of the case studies are introduced. 2.1 Dispersion model EPISODE-CityChem simulates the transport, chemical reactions and deposition of pollutant species in both a 3D Eulerian grid and a ground-level sub-grid (Hamer et al., 2020;Karl et al., 2019). A typical Eulerian grid has a horizontal resolution of 1 km by 1 km, while the vertical grid size varies from several meters (near ground) to several hundreds meters (higher layer) with a total height up to several kilometers. The sub-grid has a better resolution with a typical size of 100 m by 100 m horizontally.

95
The governing advection-diffusion and mass conservation equations for the averaged concentrations in the main Eulerian grid model are indicated as: where C i is the concentration of species i (i=1:N , where N is the total number of species), u, v and w are the three wind 100 velocity components, K (H) and K (z) are the horizontal and vertical eddy diffusivities, and R i and S i represent the source and sink terms. The estimation of eddy diffusivities is based on the mixing length theory (K-theory) (Slørdal et al., 2003).
In the simulation, the emission source term, wind field and other meteorological conditions are assumed hourly constant. The emission species will be simulated until it is fully diluted or outside of the simulation domain. The photochemistry simulated in the Eulerian grid has several options, such as EMEP45 , EmChem03mod and EmChem09mod (Simpson 105 et al., 2012), that are modified or updated from the European Monitoring and Evaluation Programme (EMEP) (Simpson, 1995).
In this study, the chemical scheme applied is EmChem09mod.
In the sub-grid receptor model, the pollutants generated by emission sources (either a point or a line source) are calculated by using simple Gaussian models. The LS model used in the EPISODE-CityChem package is a steady-state integrated Gaussian plume model (HIWAY-2) with a simplified street canyon model (SSCM), which will affect the concentrations on the receptor 110 points close to the line source (usually within an influence distance of 500 m). The emitted mass from line sources will be integrated into the 3D Eulerian model in each simulation time step. For the point source, a Gaussian segmented plume model, SEGPLU (Walker and Grønskei, 1992), with the use of a Weak-wind Open Road Model (WORM) (Walker, 2011) meteorological pre-processor (WMPP) is implemented to treat the pollutants released from an individual point source as discrete emissions of finite length plume segments, that emitted in each time interval (∆t). In the calculations, the plume 115 rise (due to buoyancy or momentum) is estimated based on Briggs's algorithms (Briggs, 1969(Briggs, , 1971(Briggs, , 1975, which consider the different atmospheric stability conditions (such as neutral-unstable and stable conditions). The effects of stack downwash (Briggs, 1973) and plume penetration (Weil and Brower, 1984) on plume height are considered as well. The transport, growth, chemical reaction and deposition of the plumes will be estimated based on the local meteorological conditions where the plumes stay, and the plume mass will be integrated into the Eulerian grids when the segmented plume grows to a predefined 120 size (usually when σ y /dy=4 or σ z /dz=4, where σ y , σ z are Gaussian dispersion length scales in cross wind direction and vertical direction for a plume, and dy, dz are the horizontal and vertical sizes of an Eulerian grid cell). The existing plumes will contribute to the concentrations in the sub-grid receptors. The emission concentration at the receptor points will be finally estimated as the sum of the Eulerian grid concentration and contributions from line and point sources, described as Eq. (3).
where, C t rec is the receptor point concentration at time t, C t−1 m is the Eulerian grid concentration at previous time step (estimated by Eq. 1), C t l and C t p are the concentrations contributed from line and point emission sources, L and P are the total numbers of line and point emission sources. In the sub-grid modelling, the stack downwash, dry and wet deposition, and plume rise and penetration are considered as well, and the photochemistry applied is EP-10 plume scheme (Karl et al., 2019). More details about the EPISODE-CityChem software can be found in the papers written by Hamer et al. (2020) and Karl et al. (2019).  Table 1.  Table 1. turning angle (θ) ship turning angle -360∼360 • (1) θ=0 • : moving straightly ( Fig. 1(a)); (2) θ>0 • : turning clockwisely ( Fig. 1(b)); (3) θ<0 • : turning counter-clockwisely ( Fig. 1(b)).
Note: In each simulation hour, all five variables are assumed constant and only need to be updated hourly.
Based on these five variables, the ship position is estimated and updated in each simulation time step (∆t). As shown in Fig.   2, ship emission is assumed to be emitted in a virtual point of the short ship line that represents an average ship position for each individual time step. The emitted plume is then treated by using the SEGPLU model in EPISODE. As mentioned above, the parameters (such as plume size and location) of each individual plume will be updated in each time step, and then their contributions to the 3D Eulerian cell and sub-grid receptors were calculated. In this model, the ship movement parameters, 145 mainly ship speed, direction and turning angle, are constant for each simulation hour, and hence, the virtual point is usually taken as the middle point of the short travel distance during each ∆t. Apparently, the ship and plume positions will be more realistic when time step is smaller. For the time step that ship starts or stops to move, the estimation of the virtual point takes time factor into account, as presented in Eqs. (4) and (5).
where, − → x i is the virtual position (x, y) of a ship during i th time step in each simulation hour, d − → ) is the actual ship travel distance in i th time step, t i =i∆t is the actual time difference from the start of each simulation hour for i th time interval, − → u s is the ship velocity. As mentioned above, the five parameters in the simulation are only updated 155 in each hour, assuming that ship is moving in a straight line or a curve with a constant speed during each hour. This is actually a limitation for the current version of MPS model compared to a real ship if its movement parameters change frequently, however, the model used in this study is to address the idea of a MPS model that has the potential of tracking the ship movement and then better simulates the dispersion details of ship emissions. The current version of MPS model provides an alternative option for simulating the dispersion of ship emissions in a harbour city. In addition, it should be highlighted that it is possible to define 160 an arbitrary ship movement by using the MPS model, once the real ship movement data collected at different time is added to the MPS input files in the simulation. In this study, the turning angle (θ) is set as 0 • for all moving ships, and the start time (t s1 ) and stop time (t s2 ) for all moving ships are assumed to be 0 s and 3600 s respectively, due to the lack of such information for each ship.

Simulation setup 165
The purpose of this study is to evaluate the new developed MPS model, and two simulations are conducted in Singapore area in this paper. One is a simplified dispersion simulation that only includes moving ships with simplified input conditions, and the results by using the MPS model are compared with the LS and FPS models. Another is a real case study that simulates all the ships around Singapore city by the new model and compared with the concentrations of emission species measured in different stations. To simplify the simulation, a constant meteorological condition taken from two weather stations (as shown in Fig. 3(b)), one in south of Singapore and another in north of Singapore, was applied to the entire simulation period. The details of the weather conditions are shown in Table 2, where the wind inputs in two weather stations were assumed as 2 m s −1 and 180 • (blown 180 from south to north). A build-in meteorological pre-processor, MCWIND, was used to first guess and estimate the local wind speed and direction in the simulation domain, based on the input values from the weather stations, and then they were adjusted to the given topography to obtain the 3D divergence-free and mass-consistent diagnostic wind field (Hamer et al., 2020). Other meteorological parameters (such as vertical temperature gradient) in the simulation area were also estimated by MCWIND. In this paper, the calculated wind field in the ground-level layer for the dispersion modelling is shown in Fig. 4. Table 2. Meteorological inputs applied in MCWIND pre-processing utility for the simplified simulation.  were constant as the ship operating condition was unchanged, and no background concentrations were used. In addition, the 200 chimney height is assumed to be 30 m for the large size ships (such as the liquid bulk ships) and 10 m for the small ones (such as the leisure ships), while exit gas is assumed to be at 20 m s −1 with 300 • C for all ships. The ship building height for each ship is set as 5 m below than the chimney height, and the width is assumed as 20 m for the large size ships and 5 m for the small ones. These choices have a very minor effect on the results (Appendix C). In this study, the ship emission sources were treated by using three different models, namely moving point, fixed point and line sources, and the simulated emission profiles 205 were compared. The simulation setups are summarized in Table 3. In addition, sensitivity studies were conducted by changing the mesh density, simulation time step and emission source setups (such as exit velocity, chimney height and ship building dimensions), and the simulated concentration profiles for different species (such as NO 2 , SO 2 and PM 2.5 ) were compared as well. The simulation results for the sensitivity studies are presented in Appendices A -D.
engine type (slow, medium and high speed diesel engine, gas turbine and steam turbine), k is fuel type (bunker fuel oil, marine diesel/gas oil, gasoline), m is ship operation mode (cruising, hoteling, maneuvering).  The MPS model was applied to a real case study in this paper as well. The hourly averaged emission values for several hours (11 am to 4 pm on April 23, 2020) in Singapore were simulated by using the MPS model, and the results at different observation stations were compared to the measured data. The model setups (such as the grid size) and numerical methods (such as MEET method for emission rate calculation) are same with those used in the simplified simulation, except those (such as the meteorology and background concentrations) introduced in this section. The configuration and setups of the simulation 220 is shown in Fig. 6.  Table 3.

Results and Discussion
In this section, the results for two studies are presented.  Figure 7(a) presents the instantaneous NO 2 concentration near ground simulated by the MPS model. Based on the 2D plots, it clearly shows that the species concentration inside the plume is gradually reduced in the opposed direction to the ship movement, which is reasonable. As the ship moves to west-south direction and keeps emitting emissions at different positions along its route, the early generated emission will be transported by wind to further north and then diluted, and hence, the emission plume is formed with minimum concentration at the east side and peak value at the west side. The simulated results

245
indicate that the MPS model gives a quite reasonable prediction for the distribution of emissions released by a moving ship. In comparison, the LS model gives quite different results, as shown in Fig. 7(b), that the simulated NO 2 species is distributed in a much wider area with a relatively smaller peak concentration. In the dispersion modelling, a line source is a very common model for treating a moving ship, assuming that the ship continuously generates emissions along the entire line in the simulation. As a result, more emissions appear near the entire ship route and then gradually diluted in the downwind side. Compared 250 to the real condition, it is unrealistic as the ship keeps moving and is not able to emit emissions from the entire ship route simultaneously. Furthermore, since the total emission rates (g s −1 ) generated by the ship are same for the MPS and LS models, the NO 2 emission rate at each point along the ship route (or saying emission rate intensity (g s −1 m −1 )) in the LS model is much smaller than the MPS model. Hence, the maximum NO 2 concentration generated by the LS model has a relatively smaller value than the MPS model.

255
In addition, the simulated emission profiles by using a FPS model are illustrated in Fig. 7(c). The FPS model is another commonly used assumption for treating the moving ship in the literature. In this study, the moving ship is assumed to stay in the middle point of the ship routine in each hour. As shown in Fig. 7(c), the NO 2 emission is blown to north by wind from the ship point and then diluted. Since the ship position is assumed unchanged during each hour in the simulation, the emission is distributed in a much smaller area with a much larger concentration compared to the other two models. Clearly, the FPS model 260 cannot reveal the effects of ship movement on emission dispersion.

Simplified simulation − results for case studies with more ships
After comparing the three emission source models for only one ship simulation, the three models were applied to a simplified study (cases 1-3) with 44 ships involved, in order to further evaluate the performance of different models for predicting the effects of moving ships on air quality in coastal cities. Both of the instantaneous and average results are presented in this section 265 to fully compare the different emission models. The meteorological conditions and simulation setups are same as presented in Tables 2 and 3.

Simulated results by using the moving point source model
In the case study for more moving ships, the simulation was first conducted by using the MPS model (case 1), and the instantaneous ground-level NO 2 concentrations at different time around Singapore area are plotted in Fig. 8. Based on the 2D 270 plots, it shows that the NO 2 emission moves to north from the ship positions and forms the higher concentration at t = 60 min compared to other simulation time, as most of ships are passing the same area during the first 60 minutes (Fig. 3(c)). The gas species then moves to west and east directions as the two groups of ships move towards their destinations, and the gas concentration is continuously diluted in the following simulations as the ships keep moving out of Singapore area. shows that less NO 2 species arrives at the ground when the plumes are closer to ships ( Fig. 9(a)), and then the gas species will be transported vertically to the ground as the plumes move to the downwind direction ( Fig. 9(b)). This is mainly attributed to the plume rise effects that the gas species exits the ship chimney with a certain velocity (in the simplified simulation, the exit velocity is assumed as 20 m s −1 for all 44 ships), and then the gas species will be blown by the wind (south to north) and only reaches the ground at a certain distance in the downwind direction. As a result, the peak NO 2 concentrations at ground level 280 appear on the locations that are far away from the ship routes but not near the ships, as shown in Fig. 8. As the emissions move further in the downwind direction, the plumes will be diluted vertically until fully disappeared, as shown in Fig. 9(c). In addition, the time history of NO 2 concentration recorded at the data point (shown in Fig. 3(b)) is also plotted as shown in Fig. 10, where it indicates that there are two peaks for NO 2 concentration when using the MPS model. The time history is reasonable. Based on the NO 2 curves, it indicates that the emission species generated by the ships take around 30 minutes to 285 reach the data point, and hence, the two peaks should be induced by the transport and accumulation of emissions generated during the first 60 minutes. As shown in Fig. 11, a large group of ships pass by or are close to the data point during the first 30 minutes and lead to a continuous emission accumulation to form the first peak concentration, and another group of ships pass by the data point later (from 40 to 60 minutes) to generate the second peak value. After 60 minutes, most of the ships have passed the data point (Fig. 12), and hence the NO 2 concentration is continuously decreased. The time series of 2D plots in Fig.   290 8 and the NO 2 concentration curve in Fig. 10 reveal that the effects of ship movements on emission distributions can be well captured by using the MPS model.

Comparison of three emission source models − instantaneous value
In the simplified case study, the simulation was then conducted by treating each moving ship as a line source (case 2). The instantaneous NO 2 concentrations contributed from the two groups of ships are plotted in Fig. 13 for different simulation time. Compared to the MPS results (Fig. 8), it clearly shows a much wider NO 2 distribution in Singapore area when using the LS model to simulate the moving ships, due to the continuous emission generation along the entire ship routes. For the LS model, the generated emissions have the continuous impact on a specific area, while the emissions emitted by the MPS model only have transient impact on the same area. As a result, when ships are concentrated in a small region (as shown in Fig. 3), the integration of simulated NO 2 emission generated by line sources induces a higher peak concentration than the MPS model 300 ( Fig. 13(a)), although the emission rate intensity for each line source is smaller as mentioned in section 3.1. As expected, when the ships are separated, the maximum NO 2 concentration for the line source becomes smaller than the MPS model, as shown in Figs. 13(b) and 13(c). The NO 2 time history curve for the LS simulation is also obtained as shown in Fig. 10. Compared to the MPS model, this simulated NO 2 concentration reaches to its peak at around t = 65 min and then is kept for around 15 minutes before it drops.

305
In EPISODE-CityChem, hourly based simulations are conducted, and all the conditions such as meteorological parameters and emission setups are constant for every 60 minutes' simulation. As shown in Figs. 3 and 12, more ships pass the data point during the first 60 minutes and less ships pass by during the 2 nd simulation period (t = 60 − 120 min). When the LS model is used, a constant total NO 2 emission rate is generated during the 1 st simulation period (t = 0 − 60 min) and continuously affects the data point, leading to a concentration rise in the NO 2 curve to the peak value at around t = 65 min (Fig. 10). Then 310 the emission generation and dilution reach an equilibrium condition to maintain a constant peak concentration for a while, until the emissions generated by the ships in the 2 nd simulation period arrive at the data point. A smaller total emission rate is generated by the smaller amount of ships (during t = 60−120 min) near the data point area , and hence, the local concentration at the data point is reduced, shown as the NO 2 curve in Fig. 10. Clearly, the NO 2 concentration history obtained by the LS model cannot reveal the effects of real ship movements on emission dispersion, and hence, it is not an appropriate assumption 315 for simulating the instantaneous emission dispersion for ships in cruising mode compared to the MPS model.
In addition, the simulation was conducted by using the FPS model as well, assuming that the ships are staying at the middle points of the ship routes in each hour. The ground-level NO 2 distribution profiles are presented as Fig. 14. As expected, the emissions are distributed as separated plume segments, which are clearly not accurate. In Fig. 10, the NO 2 history curve for the fixed point assumption has the smallest peak value, as less NO 2 emission can be blown to the location of the data point.

320
The individual plume segments and the smallest single-peak NO 2 time history indicate that the FPS model is an inaccurate approach for simulating the emission release and dispersion from the moving ships. Based on the comparisons of the simulation results by using three different emission models, it suggests that the new developed MPS model can simulate more realistic ship movement and then instantaneous emission concentrations generated by the moving ships.

Comparison of three emission source models − average value 325
The simulation results in previous sections are the instantaneous NO 2 concentrations. In emission dispersion modelling, the average results (usually hourly based) are also important as they can be used for policy decision and for evaluating the longterm environmental impact. In this section, the hourly averaged results by using three different emission source models are compared as well, and the average NO 2 concentrations near ground at different simulation time are presented in Fig. 15. Based on the 2D plots in Fig. 15, it shows that the average NO 2 profiles by using the FPS model are much different from the other 330 two setups. As discussed above, the FPS setup is clearly inappropriate for modelling the moving ships.
The hourly averaged 2D plots in Fig. 15 also indicate that the simulated NO 2 emissions by using the MPS and LS models are distributed in similar area. This is because that the emissions for each ship are emitted along the same ship route for two models, although the location of NO 2 species generated by a MPS model changes along the ship route while the LS model emits emission along the entire route continuously. As a result, the accumulated NO 2 emissions will cover the similar area model gives a much different result. The simulation results suggest that the MPS model should be able to provide an alternative option to predict the hourly averaged emission concentrations and distributions in the air pollution dispersion modelling. Figure 16. Hourly averaged NO2 concentrations at a fixed point ("data point" in Fig. 3(b)) simulated by using different emission source setups.

Real case study − comparison with measurement
After comparing with the LS and FPS models in a simplified study, the new developed MPS model was applied to the real 345 case by predicting the emission results generated by all ships (including those under cruise and at berth) around Singapore area during a couple of hours. The predicted hourly averaged NO 2 and PM 2.5 concentrations are compared to the observed results obtained from the Singapore National Environment Agency online data resource. The simulation results by using the LS model are also presented in Fig.17. Compared to the measured data, the NO 2 and PM 2.5 concentrations predicted by the MPS and LS models are generally similar, while the MPS model shows slightly better NO 2 results at the observation station A, which is closer to the ships than other stations. Fig. 18 also presents the overall averaged emission concentrations during the entire simulation period predicted by the MPS and LS models, and it clearly shows that the emission concentrations predicted by the two models are quite different especially at the locations close to the 375 ships. The different results for the two models are mainly attributed to the different treatments for the moving ships. However, as the emissions are transported to the locations far away from the ships (such as at the stations A-D), the differences of the emission concentrations predicted by the two models become smaller, due to the emission deposition and dilution. Although it may be expected that with a large number of ships and for large distances from the sources the LS model and the new MPS model will give similar results, the MPS model is a more realistic representation of the source and allows for greater granulation 380 of the emissions, swift response of the pollution dispersion model to any changes in the ship movement, and is expected to be equally accurate across all scales. Based on the results in this section and the simplified study, it is found that differences between the LS and the MPS models are small when a large number of ships are moving in a constant manner in the location of interest, however, the new MPS model could capture the impact of changes in the ship's course on the emission dispersion and hence provide more options to the modeller.

405
In addition, a real case study was conducted as well to further evaluate the MPS model by simulating all ships around Singapore area. Compared to the measured data, the MPS model was found that it can reasonably predict the emission concentrations at different observation stations located in Singapore, although gaps still exist due to the different setups and configurations between simulations and measurements. The LS model was compared in the study as well. The predicted emission concentrations by the MPS and LS models are quite different at the locations close to the ships, while these differences become smaller 410 at the locations far away from the ships as the emission is diluted and deposited. Compared to the measured data, the MPS and LS models perform similar while a slightly better NO 2 result was found for the MPS model at observation stations A, which is closer to the ships. The real case study together with the simplified study suggest that the MPS model is a more realistic representation of the emission source, and it allows for greater granulation of the emissions and swift response of the pollution dispersion model to any changes in the ship movement, compared to the LS and FPS models. The MPS also has a great 415 potential for a real-time simulation of the shipping emission dispersion, when using together with the automatic identification system (AIS) ship position data. Therefore, the MPS model should be a valuable alternative for the environmental society to evaluate the pollutant dispersion contributed from the moving ships.  (Karl and Ramacher, 2019) Data availability. The following datasets are available upon request from the authors. 1, input and output data of EPISODE-CityChem simulations for simplified cases with 1 ship in Singapore (∼0.6 GB); 2, input and output data of EPISODE-CityChem simulations for simplified cases with 44 ships in Singapore (∼1.1 GB); 3. input and output data of EPISODE-CityChem simulations for the real case study in Singapore (∼0.5 GB).

425
Appendix A: Parameter study − time step To further evaluate the MPS model, time step is investigated. In the reference case (case 1), the calculated time step is 15.8 s, and in this parameter study, time step is adjusted to two different values of 10 s (case S1) and 30 s (case S2) as shown in Table   A1. All other conditions and model setups for all three cases are same. The instantaneous NO 2 profiles at ground level for two additional time step simulations are plotted in Fig. A1. Compared 430 to the reference case (Fig. 8), it indicates that the NO 2 profiles at different simulation time are almost same for all three cases, although the local emission distributions and concentrations are slightly different. ical conditions (mainly temperature, wind speed and direction), in which Eulerian grid cell that the plume start point stays. In next time step, the plume position is updated and then its size and movement parameters are re-calculated based on the new meteorological conditions of the main grid cell, where the plume segment is transported. In addition, when the length scale of the segmented plume (σ y or σ z , which is highly affected by the meteorological conditions such as wind speed and temperature) 440 reaches to a pre-defined value (usually 1/4 of the Eulerian grid size), the plume mass will be integrated into the main Eulerian grid cell that the segmented plume locates and then deleted from the sub-grid model.
As the wind field (speed and direction) estimated by EPISODE-CityChem is spatially different (as shown in Fig. 4), the mass and number of plume segments and the values of other parameters (position, size, speed and direction) for each plume segment estimated in the dispersion modelling are different when using different time steps, and hence, the plume prediction 445 in the sub-grid modelling will be different to result in different emission concentrations. However, as the time step reduces to a relatively small value, the impacts of time step on simulation results are negligible as shown in this paper (Figs. 8 and A1).

Appendix B: Parameter study − grid resolution
Another parameter study was conducted by changing the horizontal (case S3 and S4) and vertical (case S5) grid size. In this section, the dispersion of ship emissions was first simulated by using two different main grid resolutions, where the horizontal 450 grid sizes (dx=dy) are changed to 700 m (case S3) and 1400 m (case S4), compared to the reference case (case 1 where dx=dy=1 km). To keep the same simulation domain, the main grid has 100×100 cells in horizontal direction for case S3 and 50×50 cells for case S4. All of other model setups (including vertical grid size and number) and initial conditions are same for these cases.
The simulated NO 2 profiles for two additional grid resolutions are plotted in Fig. B1, which shows similar results as the 455 reference case (Fig. 8) but with slightly different details. In one hand, the space-varying wind fields for the different gridresolution simulations are slightly different from the value in the reference case (case 1). As mentioned in last section, the parameters of the plume segments, such as location, size and speed, will be affected for the cases with different horizontal grid resolutions. In another hand, the plume mass will be added into the main grid cell that the plume locates, and then it will induce a relatively higher local concentration for the finer grid and lower concentration for the coarse grid, compared to the 460 reference case. Therefore, the integration of the sub-grid plume model with the main Eulerian model will be affected to result in different NO 2 concentrations for the different grid setups. However, compared to the coarse grid (case S4), when the grid size is reduced, the simulation results for cases 1 and S3 are much closer to show mesh independence, as presented in Figs. 8 and B1.
Similar study was conducted by changing the vertical grid size (dz) as well. In this study, a smaller dz (especially in lower 465 height) was applied in the simulation (case S5: 20 vertical layers), compared to the reference case (case1: 13 vertical layers).
All other setups and conditions are same for the two cases. Similar conclusions are obtained that the simulated NO 2 profiles are very similar with only slightly different details for the two different setups, as shown in Fig. 8 and B1. The parameter study suggests that the MPS model works well for the shipping emission dispersion modelling.
Appendix C: Parameter study − emission source setup 470 In this study, the ship exit gas velocity was assumed as 20 m s −1 for all ships, and chimney height was set as 30 m for large size ships and 10 m for small size ships (case 1). To investigate the effects of different setups for the MPS model on the simulation results, two additional cases were conducted by changing the exit gas velocity to 30 m s −1 (case S6) and by increasing the chimney height to 40 m for all ships (case S7), as shown in Table C1. The comparisons of the simulated results obtained by using different setups are presented in Figs. 8(a) and C1. Based on the simulated NO 2 profiles, it clearly indicates that the 475 predicted results using different setups for the emission sources are almost identical, suggesting that the impact of different emission source parameters (such as chimney height and exit gas velocity) on the simulation results is quite small.
In addition, the ship building height was assumed to be 5 m below the chimney height, and the building width was set as 20 m and 5 m for the large size and the small size ships respectively (case 1), as shown in Table C1. Another sensitivity study was conducted to evaluate the influences of different ship building dimensions on the simulation results, by assuming 20 m as the 480 building height and 15 m as the building width for all ships (case S8). As shown in Fig. C1, the predicted NO 2 concentrations by using different ship building setups are very similar, compared to the reference results (case 1) in Fig. 8(a). Therefore, it suggests that the impact of using these two different ship building setups on the simulations is negligible. Figure B1. Instantaneous NO2 concentrations near ground by using different grid resolutions. 1 st row: dx=dy=700m (case S3); 2 nd row: dx=dy=1400m (case S4); 3 rd row: small dz (case S5).  Figure C1. Instantaneous NO2 concentrations near ground by using different emission setups.

Appendix D: Results for different emission species
In this study, different species such as NO 2 , SO 2 and PM 2.5 were simulated. However, the distributions of different species 485 by using the same emission model (such as MPS model or LS model) are quite similar and only the concentration values are different, as presented in Fig. D1. As the purpose of this paper is to demonstrate the development of the new MPS model and to show the differences of the simulated results by using the MPS model and other two common models (LS and FPS models), the paper only presents the predicted NO 2 profiles in the simplified study. In the real case study, the predicted NO 2 and PM 2.5 concentrations were presented to compare with the measured data in the observation stations, due to the data availability. Acknowledgements. This study is funded by the National Research Foundation, Prime Minister's Office, Singapore under its CREATE programme. The authors appreciate the support for EPISODE installation from Dr. Matthias Karl in Helmholtz-Zentrum Geesthacht, Department

495
of Chemistry Transport Modelling, Geesthacht, Germany. KP also thanks Yichen Zong and Arkadiusz Chadzynski for their suggestions and help on obtaining input data for simulations.