Annual changes of ship emissions around China under gradually promoted control policies from 2016 to 2019

Ship emissions and coastal air pollutions around China are expected to be alleviated with the gradually implemented of domestic ship emission control (DECA) policy. However, there is so far a lack of a comprehensive post assessment on the ship emission response after the policy implementation. This study developed a series of high spatiotemporal ship emission 10 inventories of China’s inland rivers and the 200 Nm zone from 2016 to 2019 based on an updated Ship Emission Inventory Model (SEIM v2.0) and analysed the interannual changes of emissions under the influence of both ship activity increase and gradually promoted policy. The route restoration technology in SEIM v2.0 has greatly improved the spatial distribution of ship emissions and the river vessels (RVs) are better distinguished by using the spatial frequency distribution method. From 2016 to 2019, SO2 and PM emissions from ships decreased by 29.6% and 26.4%, respectively, while ship NOX emissions increased 15 by 13.0%. Although the DECA 1.0 policy has been implemented since 2017, it was not until 2019 with the DECA 2.0 that significant emission reduction was achieved, e.g., 33.3% regarding SO2. Considering the potential emissions brought by continuous growth of maritime trade, however, an even larger emission reduction effect of 39.8% was achieved in 2019 compared with the scenario without switching cleaner fuel. Although ocean-going vessels (OGVs) contributed to approximately 2/3 of ship emissions in Chinese waters, 2/3 of them came from ships registered in other countries. Containers 20 and bulk carriers are still the dominate contributors to ship emissions, and newly-built, large ships and ships using clean fuel oil are taking an increasingly large proportion in emission structure. The four-year consecutive daily ship emissions were presented for major ports, which timely reflects the response of step-by-step DECA policy on emissions and may provide useful references for port observation experiments and local policy making. In addition, the spatial distribution shows that a number of ships detoured outside the scope of DECA 2.0 in 2019 to save the cost on more expensive low sulphur oil, increasing 25 emissions in farther maritime areas. The multi-year ship emission inventory provide high-quality datasets for air quality and dispersion modellings, as well as verifications for in-situ observation experiments, which may also guide further ship emission control direction in China. https://doi.org/10.5194/acp-2021-212 Preprint. Discussion started: 6 April 2021 c © Author(s) 2021. CC BY 4.0 License.


Introduction
Shipping is an important anthropogenic source of air pollutants and greenhouse gases, which has come into the view of 30 scientists and public since the end of last century (Corbett and Fischbeck, 1997;Capaldo et al., 1999;Lawrence and Crutzen, 1999). Air pollutants emitted from ships can be further transported to inland areas by the onshore flow, along with the atmospheric chemical transformation, aggravating air pollution and endangering human health (Endresen et al., 2003;Eyring et al., 2007;Eyring et al., 2010;Corbett et al., 2007). In the past decades, despite the improvement of global fuel quality and engine post-treatment technology, shipping emissions are still increasing driven by ever-growing maritime trade (IMO, 2020;35 UNCTAD, 2019). Recent study shows that global shipping emission constitutes 3% of anthropogenic CO2 emissions in 2017 (IMO, 2020), and much more proportions for reactive gases, e.g., 20% to NOX and 12% to SO2 emission (McDuffie et al., 2020). China, as the largest maritime trading country, sitting on seven of the world's top ten ports with even more densely distributed coastal ports, is meeting a more tough challenge due to the lagging emission control measurements compared to European and American countries (Mao and Rutherford, 2018b). 40 In recent years, numerous researchers attempted to quantify ship emissions of China and evaluate their air quality impacts.
These studies suggest that ship emissions of SO2 in China are nearly 5 times of that from road transportation , and emissions within 12 nautical miles (Nm) accounts for ~ 40% of the total emissions from all ship emissions in coastal areas Li et al., 2018). Exhaust emissions from ships contributed significantly to air pollution in coastal 45 provinces, especially in the Yangtze River Delta (YRD), with the highest increase of annual PM2.5 concentrations reaching up to 4 ~ 5μg/m 3 (Chen et al., 2018;Liu et al., 2018;Lv et al., 2018;Feng et al., 2019;Wang et al., 2019). During ship-plumeinfluenced periods, ships can even contribute 20% ~ 30 % of the total PM2.5 concentrations (Fu and Chen, 2017;Chen et al., 2018). The adverse impact brought by ship emissions also lay huge burden on human health, causing 14,500 ~ 37,500 premature deaths in a large scale of East Asia and also hundreds of that in regional areas of Pearl River Delta (PRD) of China 50 (Liu et al., 2016;Chen et al., 2019a).
Theses previous evaluations have made great efforts to support the formulation of China's domestic emission control area (DECA) initially designed for Bohai Rim Area (BRA), YRD and PRD and later upgraded to the whole water areas of 12 Nm from the baseline of Mainland China, where ships entering the DECA were required to switch clean fuel oil with lower sulphur 55 content. However, theses assessments are mostly so-called "prior assessments", namely, evaluations of the cost and benefits of environmental and health improvement by assuming control scenarios based on the earlier ship activities before the implementation of the policy. Under the circumstance of increasing shipping demand, synchronized by the step-by-step implement of control measures, the "post evaluation" is of equally importance to seek the response of actual ship activities and emissions to the policies as well as to provide powerful foundations for in-situ observation experiments (Wu et al., 2021). 60 Although a number of studies have demonstrated the air quality benefit due to switching low sulphur oil in local port areas https://doi.org/10.5194/acp-2021-212 Preprint. Discussion started: 6 April 2021 c Author(s) 2021. CC BY 4.0 License. (Zhang et al., 2018a;Zhang et al., 2019b;Zou et al., 2020;Zhang et al., 2018b), there is so far a lack of a comprehensive national-scale evaluation to reflect the benefits of gradually promoted DECA policy, which is vital to guide further ship emission control direction in China.

65
With the advent of the big data era, characterization of ship emissions has evolved from the earliest "top-down" estimation based on global fuel consumption Endresen et al., 2003) to the "bottom-up" model based on the big data of ship automatic identification system (AIS) (Jalkanen et al., 2009;Winther et al., 2014;Liu et al., 2016;Johansson et al., 2017;Nunes et al., 2017). The AIS-based ship emission inventories have great advantages in improving the spatial-temporal resolution for numerical simulation as well as providing possibilities of near-real-time emission estimation to meet regulatory 70 needs (Miola and Ciuffo, 2011;Nunes et al., 2017;. However, emission calculation methods based on big data greatly depend on the data quality, thus demanding tedious steps for data cleaning. As AIS signal loss occurs in many cases, dealing with long-time missing AIS signals has been one of the key technical problems both for scientific research and supervision (Zhang et al., 2019c;Peng et al., 2020;Zhang et al., 2020). Without targeted measures, the estimated ship emissions would be spatially and temporally misallocated, thus further raise uncertainties of environmental impact assessment. 75 In this study, we developed the ship emission inventory for inland rivers and the 200 Nm zone of China from 2016 to 2019 based on the continuous global AIS data and the updated version of Shipping Emission Inventory Model (SEIM v2.0). The global AIS database including annually ~30 billion AIS signals and Ship Technical Specifications Database (STSD) covering over 3.5 million individual vessel profiles were combined as fundamental data for emission calculation. The previous SEIM 80 model was upgraded to SEIM v2.0 through the following three improvements: 1) developing a route restoration module to restore the most likely trajectory for missing AIS signals; 2) identifying river vessel from AIS data based on spatial frequency distribution of ship trajectories; 3) incorporating step-by-step Chinese emission control policy with daily scale. The four-year consecutive daily ship emissions and structure were analyzed from the national to port level to track the variation at a fine time scale. The interannual spatial change of emissions from ocean-going vessels (OGVs), coastal vessels (CVs) and river vessels 85 (RVs) were presented and compared. In addition, another scenario without the DECA policy was performed to evaluate the effect of China's gradually implemented DECA policy, considering the actual change of interannual ship activities. Results of this study provide high-quality emission inventory data for the further numerical simulation of air quality and health benefit of ship emission reduction. The SEIM model has been established in our previous work to develop multi-scale ship emission inventory with high spatial and temporal resolution based on a combination of satellite-based and terrestrial-based AIS data (Liu et al., 2016;Fu et al., https://doi.org/10.5194/acp-2021-212 Preprint.  2017; Liu et al., 2018). In this model, emissions were calculated based on the instantaneous operating status and power changes for each individual ship between two successive AIS signals, usually ranging from a few seconds to a few minutes. Each active 95 ship in AIS data was dynamically matched with its technical profiles for classification and emission calculation. With highfrequency AIS signal transmit time and geographic locations, the total emissions could be ultimately aggregated by that from all ships of all time intervals in the whole year, resulting an inventory with high temporal and spatial resolution. Technical details including the data collection and cleaning, calculation formula, emission factor adoption as well as default parameter setting of the SEIM model has been introduced in our previous studies (Liu et al., 2016;. Currently, SEIM 100 considers ship emission for both air pollutants (e.g., SO2, PM, NOX, CO and HC) and greenhouse gases (e.g., CO2, CH4 and N2O), from main engines, auxiliary engines and boilers.
To reduce the uncertainties of emission calculation, we introduced several techniques in the previous version of SEIM: 1) a double-nested research domain was applied to reduce the boundary effects for regional inventories; 2) the Gradient Boosting 105 Regression Tree (GBRT) method was adopted to estimate the default values of missing ship properties; 3) the propeller law was used to calculate the instantaneous engine loads; 4) the 10-minute linear interval interpolation method was used to figure out long-distance AIS signal gaps (Liu et al., 2016). These all contributed to improve the reliability of ship emission inventories.
Here, we introduce a refined version of the SEIM v2.0 to describe the improvement of the model that applied for estimating annual ship emission inventory around China. The major improvements include: 1) developing a route restoration module to 110 restore the most likely trajectory for missing AIS signals; 2) distinguishing river vessel from AIS data based on spatial frequency distribution of ship trajectories; 3) incorporating step-by-step Chinese emission control policy with daily scale to timely reflect the actual emission level. Figure 1 shows the current structure and flow chart of the SEIM v2.0, which is composed of several key modules: data pre-115 processing, route restoration, emission calculation, policy-abutted modification and post-processing. First, the originally collected raw AIS data and ship profile data from multiple sources are combined to form a ship activity database and STSD, and the RVs will be identified based on ship trajectories. Second, a route restoration module is applied for long-time gaps in AIS data, in which the 10-minute linear interpolation will be applied on the shorted paths instead. Third, the instantaneous emission along with movement of ship's trajectory will be calculate based on a series of extra prepared parameters and factors. 120 Then, the policy-abutted modification will be applied for vessel entering the DECAs to switch low sulphur fuels. Finally, the ship emission inventory datasets will be established and used for visualization and multidimensional analysis. As most of the technical methods have been described in our previous work, such as GBRT methods, emission calculation algorithm, extra parameter preparations, we focus on the study area definition, the latest data evaluations and the improvements of the SEIM v2.0 to introduce the technical details for developing the ship emission inventory around China. 125

Study area
Ships have strong spatial mobility, unlike the on-road mobile sources that mostly have fixed geographical range of activities.
Due to the complexity brought by the inconsistency of the ships' flag state, operating country and activity location, there is hardly unified standard to determine the attribution country of ship emissions. In this study, the target area for developing ship emission inventory is the navigable inland rivers and the coastal waters approximately within 200 Nm away from the Chinese 130 mainland's territorial sea baseline (hereinafter referred to as 200 Nm zone), as shown in Fig. 2. We defined the target area due to the following reasons. First, ship emissions occurred in this region are proved to have significant contribution to air pollution and human health of China (Lv et al., 2018), thus it is reasonable to regard China as a receptor and investigate the regional air quality impact from the surrounding ships. Second, as the current DECA is limited to 12 Nm to the baseline of territorial sea, far less than proposed area of the international ECA (200 Nm), thus it is possible to provide a scientific reference for further 135 scheme design by investigating the emission variation in the 200 Nm zone. In addition, the 200 Nm zone is the water region with the most intensive ship traffic and complex route, which is an appropriate demonstration area to test the effect of route restoration. The study area is also generally consistent to the research scope of other AIS-based ship emission inventory of China so as to make comparison of the corresponding results.

140
A double-nested domain is set to calculate ship emissions and reduce the boundary effect, in which the outer domain (D1) is 0°-90°N and 90°E-140°E and the inner domain (D2) is 14°-43°N and 104°E-130°E. The spatial distribution of emissions will be retained and presented with D2 as the boundary, and the statistical results for China will be finally made for the inland river and the 200 Nm zone. Figure 2 also shows the scope of the DECA 1.0, which include three areas, namely, BRA, YRD and PRD (often called DECA 1.0), and the coastal areas of DECA 2.0, which is approximately equal to the area from the coastline 145 to 12 Nm from the Chinese mainland's territorial sea baseline (hereinafter referred to as baseline). Meanwhile, ship emission within different coastal areas, i.e., from coastline to 12 Nm, 12-50 Nm, 50-100 Nm and 50-200 Nm from the baseline are also decomposed to investigate the spatial variation, which are also illustrated in Fig. 2.

Data pre-processing and evaluations
The global dynamic AIS data for the whole year of 2016-2019 (from January 1 st to December 31 st ) with averagely 30 billion 150 signals per year are collected to build a ship activity database, which provide high-frequency information including signal time, coordinate location, navigational speed and operating status, etc. The STSD has also been updated to 2019, which describes ship properties such as vessel type, dead weight tonnage (DWT) and engine power, designed speed, flag state, etc. Besides the ship data collected from Lloyd's Register and the Classification Societies of various countries, we have expanded the database to incorporate fishing ships from Global Fishing Watch (GFW) (Kroodsma et al., 2018). As the AIS data are composed of the 155 satellite AIS signals and the terrestrial-based AIS signals, same messages received from multiple base stations may lead to large quantities of duplicates, especially when ships are berthing. To deal with the redundant information and compress the https://doi.org/10.5194/acp-2021-212 Preprint. Discussion started: 6 April 2021 c Author(s) 2021. CC BY 4.0 License. data size, the time spans of continuous AIS signals with their instantaneous speeds both equal to 0 and displacements less than 0.01 degree were enlarged to 10 minutes. In this way, on the premise of keeping the total operation time unchanged, the volume of the raw AIS data was reduced. Table 1 shows the statistical results of the AIS messages and active ships for different years 160 in this study. The increasing trend of total vessel DWT and decreasing trend of the number of identified ships operating around China indicate the improvement of average loading capacity per ship. Detailed processing method of data collection, cleaning, matching and verification are described in our previous works (Liu et al., 2016). Most vessels show constant daily operating hours except a slightly decrease in the Spring Festival. However, fishing ships drops significantly in Summer due to the fishing-off season. Figure 3b shows the cargo fleet structure from the perspectives of vessel number, total DWT and total installed power of main engines. In terms of the vessel numbers, the fishing ship accounts 170 for the largest proportion of 42.5%, while general cargo also accounts for 29.8%, respectively. As for total DWT, the proportion of bulk carrier reaches 49.5%, and the oil tanker also occupies a considerable proportion (23.4%). For the total power of main engines, the proportion of container (35.4%) exceeds that of the bulk carrier (28.0%), indicating a higher engine power demand per unit volume for containers. Owning to the distinct technical specifications of different ship types, the number of vessels of each type would not be linear with their DWT, power, navigation time, and thus emissions. 175

Route restoration
Even if the AIS data has high frequency to report ship activities, there are sometimes long periods of signals loss due to equipment failure or manual shutdown. This kind of signals only accounts for a minority of AIS data, but may lead to large deviation of the amount and distribution of ship emission especially in case of long operating hours. To solve this problem, a 180 route restoration module was developed in SEIM v2.0 to predict the most likely navigation trajectories of the lost signals and spatially reallocate ship emissions. This method has been previously experimented by Johansson et al., (2017) in a global scale.
Here, we referred to their method and applied to China with more refined resolution.
The ship route restoration method is based on the Dijkstra algorithm (Cherkassky et al., 1996), which interpolates the lost 185 signals evenly on the shortest shipping route connecting two endpoints, namely, the experiential routes. Thus, a comprehensive ship route network need to be established before applying the route restoration algorithm. As the global AIS data provide massive signals of ship locations, the historical navigation trajectories for all in-service vessels are clearly visible on map.
Based on the aggregated ship traffic distribution and the geographic domain of D1 in this study, the shipping route map was drawn and split into 870 arcs connected by 656 nodes, as depicted in Supplement Fig. S1. Regarding the shipping route map 190 as an undirected graph, by applying the Dijkstra shortest-path algorithm, the shortest route path between each node-pair can be calculated, as well as the geodesic distance aggregated by all arcs. In this way, the ship route network connected with nodes and arcs were established ahead and the shortest geodesic paths for all node-pairs were pre-stored as database to look up, so as to improve the operation efficiency.
195 Figure 4 illustrates the diagrammatic sketch of the ship route restoration algorithm, taking a segment of AIS positions as an example. The method can be summarized as following steps: 1) For each two consecutive AIS points and , judge the geographical relationship between line and the continent; 2) If line intersects with the continent and they are not contained in the continent, apply the route restoration algorithm by firstly finding the nearest start node ! and end node ! by traversing the pre-stored node library; 3) Look up the shortest path connecting nodes ! and ! (eg., ! " # $ % ⋯ ! ) from 200 pre-stored ship route network database and calculate average speed resulted from the geodesic distance For each arc , -, calculate ship emissions based on average speed, instantaneous power and emission factors; 6) Calculate emissions ∑ summed from each time span along the restored route.

Classification of OGV, CV and RV 205
In the SEIM v2.0, vessels are classified into OGVs, CVs, and RVs for emission estimation. In China, inland vessels are having an increasing number with AIS equipment installed these years. As the fuel standard for RVs are more stringent compared to OGVs, it is necessary to distinguish them from the AIS data to calculate emissions accurately. In methodology, since OGVs are mostly engaged in international trade following the management of International Maritime Organization (IMO), they are identified by meeting the condition that both valid IMO number and the Maritime Mobile Service Identify (MMSI) number 210 are available. CVs and RVs are both domestic vessels designed to operate in rivers and coastal areas, respectively. However, in some cases, they do cross each other's navigational waters when the inland waterway system borders the coastline (Mao and Rutherford, 2018a). Thus, we identified RVs by activity frequency distribution based on the navigation trajectories for each vessel. By defining the geographic domain of D2 in Fig. 2, vessels with more than 50% of the AIS signals throughout the whole year occurred on inland rivers are considered as RVs (Fig. 5a). This method allows the possibilities for CVs and OGVs 215 sometimes travelling into the estuaries. Finally, vessels that are not identified by OGVs and RVs are regarded as CVs. CVs operating around the coastal seas of China, seldom contacting with other countries. RVs mostly mainly sail on the Yangtze 220 River and Pearl River systems, with a small proportion wandering in coastal seas. These results of spatial distribution of OGVs, CVs, and RVs indicate that the identification method is basically satisfactory.

Ship emission control policy
In recent years, a series of policy documents have been issued to control the air pollution from ships, among which the most effective measure is the establishment and implement of DECA (MOT, 2015(MOT, , 2018. China's DECA policy were put into effect 225 step by step from 2016 to 2019. Figure 6 summarizes the evolutionary of DECA including the control area and fuel standards, as well as their comparison with international ECA. Before the global sulphur cap taking effect in 2020, the heavy fuel oil (HFO) with sulphur content as high as 3.5% has long been used in ships all over the world. In 2015, China initially established three DECAs along the coastline (DECA 1.0), covering the most densely distributed area of ports, with gradual mandates for ships to use low sulphur fuel (LSF) with sulphur content <0.5% m/m from core ports to the whole regions and from berthing 230 to all operating modes, in order to reduce SO2 and PM emissions. In 2018, an upgraded DECA 2.0 was proposed to expand the region to cover the entire coastline (within 12 Nm from the Chinese mainland's territorial sea baseline, Fig. 2) in which ships are required to use LSF regardless of the operating status. In addition to fuel requirement, the DECA 2.0 policy also defined the control requirement of NOX emissions from ships that diesel engine above 130 kW built or modified on or after  Table S1). Before the 240 mandatory date of January 1 st 2017, core ports in YRD and Shenzhen port pioneered the DECA 1.0 policy nine months and three months earlier, respectively. Core ports in YRD are supposed to implement the DECA 2.0 policy three months before fully coming into effect in January 1 st 2019. Meanwhile, RVs are required to use the general diesel fuel (GDO) with much lower sulphur content, gradually iterating from 350 ppm to 10 ppm, finally keeping pace with the China V standard of on-road diesel fuel in 2018. 245 Based on the above investigation, a policy-abutted modification module was developed in SEIM 2.0 to incorporate the actual implement of ship emission control policies with daily scale in China. At each AIS signal point, according to the geographic location, signal time and the operating mode, the vessel will be dynamically judged whether it enters the scope of DECAs at that time and select the required fuel type and sulphur content. Then, a fuel correction factor (FCF), resulted from the quotient 250 of the emission factors of the switched fuel and original fuel, will be further multiplied in the emission calculation formula. In this way, a high spatial and temporal resolution ship emission inventory in line with actual implementation condition of control policy will be finally developed.

Simulation scenario setting
To comprehensively investigate the effects of gradually implemented DECA polices under the condition of growing waterway 255 transport demand, we designed another scenario (No-DECA scenario) in SEIM v2.0, as listed in Table 2. Compared to the base scenario embedded with actual DECA policy described in section 2.4.3, the No-DECA scenario was designed to assume vessels observed in AIS data of target year not to implement the DECA policy, namely, to simulate the ship emission of China's inland waters and the 200 Nm zone supposing all active vessels continued to use fuels with sulphur content at pre-DECA level. By comparing the emission result from Base scenario and No-DECA scenario, the objective emission reduction 260 effect of gradually implemented DECA policies could be vividly illustrated.

Annual ship activities and emissions
With the development of China's waterway transport, seaborne trade has been increasing through 2016-2019. As illustrated in 265 Fig. 7a, Chinese ports' total passenger turnover, cargo turnover and cargo throughput remained stable rise and added by 10.9%, 6.8% and 17.4% in 2019 compared to 2016, respectively. Growing water transport demand stimulated the increase of ship activities and improvement of fleet loading capacities (Table 1) Table S2) Li et al., 2018;Huang et al., 2019). On the one hand, our study established a larger ship activity database based on global AIS data (~30 billion 275 signals per year), and the incorporation of GFW database also improved the recognition of ships, especially CVs and RVs in China. On the other hand, the annual increase of ship activity driven by maritime trade could also contributed to ship emission growth. Among all vessels, OGVs composed the largest part in ship emission, with a proportion of 70.4% regarding SO2 and 59.7% regarding NOX in 2016. CVs ranked after OGVs, with 29.4% contribution to SO2 emission and 27.1% to NOX emission; while RVs' composition was relatively small, accounting for 13.2% for NOX and <1% for SO2. The contribution of RVs to 280 SO2 emissions was much lower than NOX, as RVs were considered to use GDOs with significantly lower sulphur content than HFOs. In addition, as we identified RVs based on spatial frequency distribution of ship trajectories in AIS, which allows RVs sometimes operating in coastal waters, the identified vessels of RVs as well as emissions might be higher than that in Li et al., (2018).  Table S2). In terms of NOX, however, emissions continuously increased year by year, 290 with a total increase of 13.0% from 2016 to 2019; while emissions of other pollutants also showed a gradual increase trend (Supplementary Table S2). Therefore, the ship DECA policy has a significant impact on reducing SO2 and PM emission but current vessel engine emission standard only have limited influence on controlling NOX emission. In addition, although the DECA 1.0 policy has been implemented since 2017, it was not until 2019 that significant emission reduction was achieved.

Contribution by flag state 295 Compared to the global ship emissions estimated by Forth International Maritime Organization (IMO) greenhouse gas (GHG)
Study (IMO, 2020), it is striking that OGVs in the 200 Nm zone of China contributed to 9.7 ~ 14.3% of global OGV emission (Supplement Table S3), despite only <1% of the world's sea area. However, we found that a considerable proportion of OGV emissions occurred in the 200 Nm zone was derived from vessels registered in other countries. Figure  of ship SO2 and NOX emissions, respectively, in the target region. Vessels registered in Panama were the second largest OGV emission contributor besides Mainland China, holding a proportion of ~18.3%. Other major contributors of ship emissions also included Liberia (~10.0%), Marshall Islands (8.0%) and Singapore (7.7%). From the perspective of interannual change, the contribution of Mainland China was raising over the four years, especially for NOX, which increased from 16.3% in 2016 305 to 21.0% in 2019; while the second largest contributor, Panama, had declined from 19.3% to 14.5% in the same period. With the gradual effectiveness of the DECA policy in China, it is equally important to pay attention to emissions from foreignregistered ships.

Emission composition variation 310
On a more refined time scale, we investigated the 5-day moving average ship SO2 and NOX emissions on a daily basis for the inland rivers and 200 Nm zone of China from 2016 to 2019, as shown in Fig. 9. It is evident that ship emission of SO2 was seasonally growing in 2016-2018, until a sharp drop on1 st January, 2019 due to the implementation of stringent control DECA 2.0 policy. The maximum daily ship emission intensity of SO2 reached 6.4×10 3 Mg/day on September 22 nd , 2018, 2.9 times of the lowest point, 2.2×10 3 Mg/day on January 1 st , 2019; while the daily discrepancy of ship NOX emission intensity also reached 315 3.0 times throughout the four years. The monthly variation of ship emissions for most vessel types was generally constant except a temporary decrease during Spring Festival in February (Fig. 9a). However, fishing ships showed significant seasonal variations, which declined annually in summer and return in autumn due to fishing ban in China. This has also been demonstrated by other studies .  Table S4. Containers had been accounted for the largest part and the contribution had been increasing through the four years, e.g., from 31.7% in 2016 to 42.9% in 2019 for SO2 (Fig. 9a). Although containers accounted for only 3.5% of vessel number and 4.6% of operating hour in Chinese waters, their relatively higher engine power contributed to significant 325 emission intensities compared to other ships of the same size, such as bulk carriers (Fig. 3). The HFO contributed to the majority of ship SO2 emissions due to its high content of sulphur, part of which, however, was gradually being substituted by marine gas oil (MGO) with the implement of DECA policy (Fig. 9b). In 2019, the MGO had accounted for 15.4% of the ship SO2 emission and 38.9% of NOX emission (Supplement Table S4). In terms of vessel build year, ships built after 2016 made an increasing contribution in annual NOx emission, reaching 10.6% in 2019 (Fig. 9c). Even though Tier II engine standard 330 had been applied to domestic ships built after 2016, ship NOX emissions were not found to decrease as the emission standard of Tier II only has minor improvement compared to Tier I. In addition, we also found that ships with larger DWT have a growing proportion in vessel fleet as well as emission contribution (Table 1 and Fig. 9d), indicating the developing trend of ship upsizing in the past few years. However, even though the newly-built, large-scale ships as well as ships using clean fuel oil are all taking an increasingly large part in emission structure, the updating iteration speed of fleet is not enough to reverse 335 the rising trend of NOX emission.

Emission variation of major ports
As is was step by step that the DECA policy was implemented in different ports in China, we extracted the 5-day moving average ship SO2 emission of major ports in BRA, YRD and PRD to track the consecutive emission changes throughout the four years, as shown in Fig. 10. In the initial stage, restriction on fuels with no more that 0.5% sulphur content was only 340 imposed on ships at berth for core ports in these three crucial port clusters ( Fig. 2 and Table S1). Before the mandatory date of January 1 st 2017, core ports in YRD and Shenzhen port pioneered the implement nine months and three months earlier, respectively, which significantly showed a decrease in ship SO2 emissions beginning from April 1 st and October 1 st in 2016, respectively. For other core ports in BRA and PRD, a noticeable decline could be observed on schedule on January 1 st 2017.
However, emission of ships at berth took a relatively smaller percentage (7.5% ~ 13.7%) in the 200 Nm zone according to our 345 results (Supplement Table S4), thus the emission reduction was rather conservative inside the DECA 1.0 region in 2018, even though the requirement was popularized to all ports. On the contrary, owing to intensified ship activities, ship SO2 emissions for some ports even largely increased, such as Ningbo-Zhoushan Port and Shenzhen Port, increased by 19. 4% and 11.4% in https://doi.org/10.5194/acp-2021-212 Preprint.  2018 compared to 2017. Fortunately, in 2019, when most rigorous DECA 2.0 policy was implemented, it is clearly illustrated in Fig. 10 that all ports' SO2 emissions were dramatically reduced. Core ports in YRD were supposed to implement the DECA 350 2.0 policy three months before fully coming into effect. Notably, those pilots witnessed an earlier decline in SO2 emission, which also proved the timely and flexible response of SEIM 2.0 model to the changeable DECA policy.
In addition to policy-driven emission changes, different ports showed distinct monthly emission variations highly related to their geographical location and ocean resources. For example, ship emissions in YRD region had a low point in July as their 355 activities were influenced by typhoon particularly in YRD (Weng et al., 2020); ship emissions in PRD region were higher in in spring and summer since wind direction were more advantageous for ship activity in spring and summer .
Besides, ship emissions in Ningbo-Zhoushan Port, Tianjin Port and Shenzhen appeared to be larger in spring and autumn, probably owing to the large-scale fishing ship operation (Chen et al., 2016;Yin et al., 2017). The above port-based emissions fully presented the daily ship emission variations for a long period from 2016 to 2019, which may also provide useful data 360 references for port observation experiments.

Evaluation of route restoration
Since the shipping route restoration module was developed in SEIM v2.0 to solve the problem of AIS discontinuity, the spatial distribution of ship emission after route restoration was evaluated, as shown in Fig. 11. Direct interpolations for AIS signals 365 along the loxodrome would lead to part of emissions distributing on unrealistic routes, e.g., crossing the land areas, which could be even as long as connecting the South China Sea and the Bohai Sea (Fig. 11a). By using the route restoration method, the ship's navigation trajectory and emissions can be restored to more realistic shipping routes, thus reducing the deviation of the spatial distribution of emissions (Fig. 11b). Statistically, 15.3% of NOX emissions and 7.5% of SO2 emissions were spatially corrected in the study area. More improvements were obtained around Taiwan island, the Korean Peninsula and the Philippine 370 Islands, probably due to worse accessibility of high quality shore-based AIS signal. The misallocation of emission in China's land areas resulted in NOX underestimate of up to 2 ~ 4 Mg/grid in the downstream of Yangtze River and Pearl River, and the misallocation of emissions in water regions is more notable on shipping routes farther from the coast. This spatial improvement of ship emissions with the route restoration method would be expected to reduce uncertainties in the air quality model. Figure 12 presents the spatial change of SO2 and NOX emission in 2019 compared to 2016 form ships within different coastal region defined in Fig. 2. Remarkably, within 12 Nm, which approximately equates to the scope of DECA 2.0 in 2019, SO2 emission decreased by 78.8% (7.2×10 5 Mg/year) compared to 2016. Despite the year-by-year growth of seaborne trade, DECA policy effectively reduced ship-emitted SO2 overall and especially beneficial to coastal cities. On the other hand, however, we discovered that SO2 emission increased by 41.5% (1.3×10 5 Mg/year) in areas between 12-50 Nm from the baseline, especially 380 along the 12 Nm boundary. The proportions of ship SO2 emission from 12-50 Nm rose from 17.5% in 2016 to 35.3% in 2019, becoming the major spatial contributor in 2019. Emission of PM exhibited the similar pattern (Supplement Fig. S2a). This peculiar phenomenon implies the fact that some ships possibly made a detour to evade switching clean fuel oil, which could also be demonstrated by the larger growth rate in cargo turnover than throughput (Fig. 6a).

Spatial change of ship emissions 375
385 Figure 12b shows that NOX emission from ships occurred within 12 Nm of the baseline were continuously increasing from 2016 to 2018, until it declined by 5.0% (6.4×10 4 Mg/year) in 2019 compared to the last year. Meanwhile, NOX emissions occurred in areas between 12-50 Nm also turned to show a higher annual increase rate in 2019 (21.4%) than previous two years (7.4% ~ 8.2%). Such phenomenon once again proves the possibility of ship detour. Other species generally showed the similar emission pattern as NOX (eg., HC in Supplement Fig. S2b). In sum, DECA 2.0 policy has a positive effect on ships' 390 SO2 and PM emissions control as a whole and especially for coastal areas. However, a number of ships detoured outside the scope of DECA 2.0 to save the cost on more expensive clean fuel oil, which further elongated the sailing distance and thus aggrandized emissions in farther maritime areas. This reminds us to pay attention on additional environmental effect brought by detouring ships during the continuous implementation of DECA 2.0 policy.

Spatial changes of OGVs, CVs and RVs emissions 395
Interannual spatial change of OGVs, CVs and RVs were further compared for ship emissions of NOX and SO2, as shown in Fig. 13. Emission intensity of identified OGVs was apparently higher than CVs and RVs, demonstrating certain routes. The most intensive near-sea routes included China-Korea, China Mainland-Taiwan, the North Atlantic Route, Asia-Europe Route and routes between busy ports of China, such as main ports in BRA, YRD and PRD (Fig. 13a). Since the main shipping routes are rather close to the land, OGVs within 12 Nm of the baseline make up for approximately 38% and 32% of total OGV 400 emissions for NOX and SO2. From 2016 to 2019, OGV emissions were generally increasing in all regions, except SO2 emission in 0-12 Nm showing a significant drop down due to the DECA 2.0 policy.
As for CVs, approximately 80% of NOX emission and 70% of SO2 emission were annually distributed mainly within 12 Nm of the baseline, and the proportions occurred outside 12 Nm were greatly reduced compared to OGVs. Despite intensive 405 emission routes between coastal ports, notable emissions from CVs occurred more evenly distributed off the major routes (Fig.   13b), which were attributed to large quantities of fishing ships operating (Kroodsma et al., 2018). In the region of 0-12 Nm to the baseline, the annual SO2 emission reduction ratio of CVs (81.0%) in 2019 was even higher than that of OGVs (76.9%), indicating that CVs were more affected by the DECA 2.0 policy.

410
Compared to OGVs and CVs, RVs have specific routes that were constrained by inland waterways, with the most intensive emission located on the Yangtze River and the Pearl River (Fig. 13c). Meanwhile, RVs also operate along Chinese coastal and produce a considerable proportion of emissions within 12 Nm of the baseline. With the increasingly stringent national fuel oil standards for RVs (MEE, 2018), i.e., sulphur content from 350 ppm before June 30 th , 2017 to current 10ppm beginning from January 1 st , 2018, SO2 emissions from RVs had been reduced to a rather low level, both for inland rivers and coastal areas. 415 However, other pollutants such as NOX emissions from RVs were still going uphill. In addition, although China has required certain categories of ships to install AIS equipment since 2010 step by step, a large part of small RVs in China have not been equipped with AIS . The lack of ship activity level and highly reliable local emission factors all brings uncertainties to the emission estimation of RVs. However, air quality and human health of inland cities near waterways could be impacted severely by RVs emissions . Therefore, RV emissions need to be stressed and worth further 420 investigation.

Monthly effect evaluation
Since the shipping activity increase and emission control collectively resulted in the past emission trend, we designed another scenario without DECA policy to evaluate the emission reduction effect considering the annual change of ship activities. Figure  Based on a four-year consecutive daily emission analysis, it is noticeable that the ship emission structure had been gradually changing, i.e., newly built, large ships and ships using clean fuel oil were taking an increasingly large proportion in emission structure. Containers and bulk carriers were still the dominant vessel type in ship emission composition. On a local scale, ship emissions in various ports exhibited different patterns in terms of daily variation. For example, ports in YRD were likely to 465 encounter typhoon in July and fishing ships were particularly abundant in BRA. Relevant findings may help provide useful data references for port observation experiments and local policy making.
The interannual spatial change of ship emissions also showed new characteristics. Through contrasting ship emissions within different distance from Chinese coastal baseline, we discovered that in 2019, a number of ships detoured outside the scope of 470 DECA 2.0 probably to save the cost on more expensive low sulphur oil, which further elongated the sailing distance and thus aggrandized emissions in farther maritime areas. This reminds us to pay attention on additional environmental effect brought by detouring ships during the continuous implementation of DECA 2.0 policy. In addition, the route restoration method developed in SEIM v2.0 effectively restored ship's navigation trajectory and emissions to more realistic shipping routes, thus reducing the deviation of the spatial distribution of emissions and could be expected to reduce uncertainties in the air quality 475 model.

Policy implication
Compared to the increasingly strict emission control policies of land-based sources and improving air quality in China, policies and regulations for the prevention and control of ship emissions would be more urgent to facilitate China's air quality to achieve the annual PM2.5 concentration standard of World Health Organization (WHO) Air Quality Guidelines (Wang et al., 480 2020;Zhang et al., 2019a). Although the current emission policy has achieved significant control effect on SO2 and PM emission, under the global low sulphur oil demand, China still needs to further apply for international ECA to enlarge the control area and strengthen the requirements for fuel quality. In order to make a comprehensive evaluation and in-depth improvement of the policy, attention is also needed during the design process of ECA scheme, such as the corresponding impact of ship detour and further expand DECA 2.0 so as to enlarge the reduction effects within 200 Nm zone. Meanwhile, 485 the international cooperation is also urgently called for to jointly control ship emissions due to ships' strong spatial mobility and the complexity of registration and operation. With the gradual cleaning of marine fuel and the obsolescence of HFO, ship emissions of SO2 and PM will be effectively mitigated in the near future. However, ship NOX emissions are still expected to increase until the gradual elimination of old ships and iteration of more stringent Tier III standard for newly built ships. Other related factors, such as engine type, NOX post-treatment technology etc. should be taken into consideration in the future. For 490 local decision makers, it is also important to make clear the local ship emission structure and meteorological conditions in order to conduct effective measures.

Data availability
The AIS data and STSD are restricted to the third party and used under license for the current study. 495

Code availability
Python codes used during the current study are available from the corresponding author on reasonable request.

Acknowledgements 500
This work is supported by the National Natural Science Foundation of China (grant nos. 42061130213 and 41822505). H.L.
is supported by the Royal Society of UK through Newton Advanced Fellowship (NAF\R1\201166).

Author contributions
XW and WY contributed equally. XW and WY designed the research and wrote the manuscript. LZ and DF provided multiple 505 analytical perspective on this research. ZS and XH helped collect and clean the ship data. LH provided guidance on the research and revised the paper. All authors contribute to the discussion and revision.