The automatic identification system (AIS) is transponder technology on board ships to enhance safety by broadcasting ship information via VHF (very high-frequency) channels. It works in conjunction with the Global Positioning System (GPS) to broadcast information such as ship position, speed and heading, together with static ship information such as ship name, ship length and ship type. Each AIS-equipped vessel is identified by a unique Maritime Mobile Service Identity (MMSI) number, which is also part of the AIS data.

We have set up an observation location in Nanjing, where an AIS receiver is located on the roof of the meteorological building in NUIST. Ships transmit AIS signals at intervals varying from every 3 s to a few minutes to provide information on their position. The black dots in Fig. 1c are an example of the locations of ships according to the AIS signals received in a time interval of 4 h. We see that the antenna can receive signals within at least a 50 km radius. A region with a longitude of less than 118.95∘ E and a latitude of more than 32.05∘ N was selected as the area where all ships, including those with weak AIS transmitters, can be tracked under all conditions. AIS data were collected for 365 d from September 2018 to August 2019. Based on the AIS information, ships were classified into seven types: cargo ships, tankers, passenger ships, tugboats, dredgers, patrol vessels and others. Cargo ships include container ships, bulk carriers and ro-ro (roll-on/roll-off) vessels (excluding passenger ships); tankers comprise liquid chemical tankers, liquefied gas tankers and oil tankers; and passenger ships consist of ferries and ro-ro passenger ships. Patrol vessels are classified separately because of their small size in combination with high speed.

The AIS data show that approximately 2000 boats pass through the observation area each day, with a drop in the number of ships on specific days. Figure 2 shows the number of unique ships per day over a year. The gray box in Fig. 2 shows the Chinese New Year period when the number of vessels dropped sharply. The decrease in the number of ships on other days is usually related to the weather conditions. On 26 November 2018, very thick fog occurred, and the travel of both ships and vehicles was disrupted. On 24 February 2019, the local meteorological bureau issued a fog warning, so the number of passing vessels decreased. On 10 August 2019, typhoon Lekima made landfall and brought catastrophic damage to southeastern China. The number of ships coming from the typhoon region was greatly reduced, as was the number of local vessels in Nanjing. On 5 January 2019, the Jiangsu Maritime Bureau issued a navigational warning for construction work in the Baguazhou branch of the Yangtze River in Nanjing.

To estimate shipping emissions from the Yangtze River, we adopt the AIS method to obtain high-resolution ship information and emission factors. Equation (1) is used for calculating the emissions of various pollutants such as NOx, SO2, PM2.5 and PM10 from a single ship based on its main power (Jalkanen et al., 2012). We only considered the emissions of the main engine of the ship while ignoring the relatively lower emissions from the auxiliary engines and boilers. 1E=P⋅L⋅fEF⋅fLLAM⋅fF⋅fC⋅T, where E is the emission from the main ship engine, P is the main engine power of the ship (See Sect. 2.2.1), L is the main engine load factor (See Sect. 2.2.2), fEF is the emission factor, fLLAM is the low-load adjustment multiplier of the main engine, fF is the fuel correction factor, fC is the control factors for implementation of greener technology, and T is the sailing time.

Below we will further discuss each element of this equation.

The emission factor (fEF) refers to the mass of pollutants emitted per unit of work done by a single vessel in 1 h (g kW-1 h-1). It is an essential parameter for calculating ship emissions and has not yet been studied for China. For example, the emission factors used by Wan et al. (2020) for Beijing–Tianjin–Hefei, the Yangtze River Delta and the Pearl River Delta were taken from literature referring to Europe and the United States (USEPA, 2000; Entec, 2002; EPA, 2009). Ship emission factors used by Fan et al. (2016) for the Yangtze River Delta and the East China Sea were based on an EPA report (2009) and on Goldsworthy and Goldsworthy (2015), the latter being a study for the Australian region. The emission factors used by Fu et al. (2012) for the Shanghai port were taken from the air pollutant emission inventory of the Port of Los Angeles (Archana et al., 2013). Lacking studies of ships of the Chinese fleet, our study also relies on emission factors for other regions, following the scheme in Table 1.

The low-load adjustment factor (fLLAM) is used when the main engine load factor is less than 0.2. When the load factor is below 0.2, there is an increase in emission intensity due to the inefficient engine use at low speed. Hence, the emissions from low-load vessels were multiplied by the low-load adjustment factor (EPA, 2009). The low-load adjustment multipliers for the main engine are shown in Table S1.

The fuel correction factor (fF) corrects the emission factors according to sulfur content of the fuel used by the ships. Ordinary inland vessels typically use general diesel oil (GDO), while larger inland ships and direct river and sea vessels use marine fuel oil (MFO). MFO includes residual oil (RO) and marine distillates (MD). The sulfur content requirements for various fuels are shown in Table S2. The emission factors mentioned above are obtained based on the assumption that marine fuel is heavy oil with a sulfur content of 2.7 % (EPA, 2009). In October 2018, the DECAs began responding to a new policy that the sulfur content of inland marine fuel cannot be higher than 0.001 % in 2019 (Wang et al. 2021). In our study, the actual inland ship emissions are calculated based on the sulfur content of 0.001 %, for which a fuel correction factor is needed. In addition, we will calculate ship emissions for an average fuel sulfur content of 1.5 % (Xu et al., 2019) for a scenario where the new policy would be absent. The fuel correction factor fF for each air pollutant is shown in Table 2.

The control factor fC is to potentially correct for higher-quality marine diesel engines or the availability of emission reduction facilities on the ships as a result of emission control policy. There is currently limited legislation to control ship emissions in inland waterway zones in China and insufficient monitoring infrastructure for enforcement; therefore the control factor is set to 1.

To evaluate the emissions derived using our method, we compared them to three other emission inventories in the YRD, including the satellite-derived NOx inventory DECSO (Daily Emission derived Constrained by Satellite Observations) and the following bottom-up inventories for East Asia: MEIC (Multi-resolution Emission Inventory for China) and SEIM (Shipping Emission Inventory Model).

We analyzed the collected AIS data from September 2018 to August 2019 and concluded that the number of ships for each defined type was stable throughout the time period. Detailed data on the number of vessels per month are listed in Table S3. Figure 4 shows the contribution of NOx, SO2, PM10 and PM2.5 from different ship types. Emissions from cargo ships are higher than for other types of ships, followed by tankers and tugs, with dredgers and patrol boats making the lowest contribution to the emissions. Cargo ships contribute more than 58 % of the ship emissions of all species in the Nanjing section of the Yangtze River. This is because cargo ships are the dominant vessel type in this region and their number accounts for about 81 % of the total.

Dividing the total emissions from each ship type by the number of ships of that type, we obtained and analyzed average ship emissions per ship type. For a single vessel, tugs emit more pollutants (Fig. S2). This is related to the higher power needed for the engine of a tugboat, which was also concluded by Xu and Bai (2017). Compared with cargo ships, tankers have a relatively higher contribution to SO2. The sulfur content of tanker fuel is slightly higher than the sulfur content of diesel fuel for cargo ships (Zhu et al., 2019). Because SO2 emissions are directly related to the sulfur content of fuel oil, tankers will emit more SO2.

We calculated the monthly emissions from the AIS data in the observational area, as is shown in Table 6. Since the same calculation method is adopted for all four pollutants, their spatial and temporal variations are consistent. Figure 5 shows the monthly variations of NOx emissions from ships from September 2018 to August 2019. We see that, except for the sharp decrease in the number of ships in February, the number in other months has basically stabilized at around 60 000. The monthly emissions are highly related to the number of ships and the speed of ships.

Ship emissions peaked in July followed by June and August. In July, the emissions are the highest due to the high river speed, and therefore more engine power (proportional to v3) is needed for upstream ships. From November to February, the emissions were lower, possibly because the water flow of the Yangtze River is lower at this period and ship activities during Spring Festival are reduced. In February, the number of ships dropped sharply, but the emissions from ships in February were similar to that from ships in January. This is due to the higher activity of tugs in February, which have larger engines and emit higher levels of emissions (Fig. S3). From February to March, we see that pollutant emissions increased due to the increase in the number of ships. This is closely related to the resumption of factory work and human activities after the holiday.

In this section we will discuss the uncertainty in our emission inventory. Our calculations have been based on the main engine only. However, during the navigation of a ship, the main engine and auxiliary engine of the ship are working at the same time. For a moored ship, the main engine stops working. On average, 17 % of the ships in the observational area are in dock every day, and this part of the ship emissions has been not taken into account, but the auxiliary engines are still working. Based on the study of Weng et al. (2020) in the Yangtze estuary, we estimate that our emissions show an underestimation of about 12 % because of ignoring the auxiliary engine and boiler emissions at berth and underway.

However, the locations of high ship emissions are consistent with previous studies. Zhu et al. (2019) pointed out that the distribution of ship emissions in the Jiangsu section of the Yangtze River in 2017 was uneven, with the emission rates in the Nanjing section of the Yangtze River and the Jiangyin section of the Yangtze River being relatively high. Xu et al. (2019) noted that for ports along the river, Nanjing port had the highest rate of ship emissions.

As the Yangtze River becomes wider when getting closer to the sea, the speed of the river will be reduced, and thus the emissions from ships can be lower. In the extreme cases of stagnant water, ship emissions can be reduced by a maximum of 3 %–33 % depending on the month. We estimate that this may lead to an overestimation of about 10 % in the ship emissions outside our study area around Nanjing.

Currently, the AIS-based approach is considered the best practice for ship inventories. However, there is still a lack of reliable local emission factors, auxiliary engine power ratings and fuel correction factor in the YRD region, which contributes largely to the uncertainties in this study. The selection of accurate emission factors is critical to the calculation of the ship emission inventory and the uncertainty that comes with it. The emission factors are closely related to the age and rotation of the ship's engine and engine load, and the fuel correction factor depends on the sulfur content of the marine fuel. Earlier heavy oil was a fuel of low quality with a sulfur content of about 2.7 %. In contrast, the fuel sulfur content in this study is only 0.001 %, while at the beginning of time period the fuel sulfur content may be as high as 1.5 %. For the scenario that the sulfur content is not regulated, we have calculated that the SO2 and PM emissions would be about a factor of 700–1000 higher.

In conclusion, our derived emissions have an underestimation of 12 % due to ignoring the auxiliary engines and boilers and an overestimation in some regions of about 10 % due to the slower river flow. Adding this to the uncertainties in emission factors, we estimate the total uncertainty to be 5 %–15 %.

To verify the calculated emissions of JSEI, we compared them with the ship emissions of SEIM for the overlapping region. Figure 8 shows the spatial distribution of the ship emissions of NOx, SO2 and PM2.5 for JSEI and SEIM and the ratio of JSEI to SEIM. JSEI and SEIM only calculate ship emissions, and they overlap for less than half of the domain. When comparing the overlapping grid cells, JSEI accounts for on average about 99 %, 0.05 % and 0.06 % of the SEIM emissions for NOx, SO2 and PM2.5, respectively. The average emissions from inland ships over rivers (JSEI) compare well with average emissions of sea-going ships (SEIM) for NOx. SEIM has higher values than JSEI for SO2 and PM2.5 because SEIM calculated the emissions for 2017, when only major ports needed to strictly control the sulfur content of marine fuel. The sulfur content of marine fuel was 0.001 % in our study. In comparison, the sulfur content of ocean-going marine fuel in 2017 was about 2.7 %, much higher than that of inland river ship fuel. Ship pollutants that are greatly affected by the sulfur content of marine fuel, such as SO2 and PM, will be reduced with the reduction of sulfur content. This shows that from 2017 to 2019, the policy was of great significance for ship emissions, effectively reducing the emissions of SO2 and PM. For NOx emissions, both inventories compare remarkably well.

Here JSEI is added to the MEIC to represent the value of total air pollutants, as the MEIC does not include emissions from ships. Figure 9 shows the spatial distribution of the contribution of the river emissions (JSEI) over the total emissions (MEIC+JSEI) for NOx, SO2, PM2.5 and PM10 along the river. SO2 ship emissions along the river account for an average of 0.14 % of total SO2 emissions and only 0.82 % in the highest areas. PM emissions from ships along the river contribute on average 0.01 % of the total PM emissions.

Even in a scenario where there is no policy regulation, PM emissions from ships account for only a relatively low proportion of total PM, with a maximum of 20 % and an average of 5 % in riverine areas, with PM2.5 accounting for a slightly higher proportion than PM10. In this scenario, SO2 emissions from ships contribute around 40 % of the total SO2 pollution on average, while it can reach 83 % in some areas. Under the regulation of the new policy, the SO2 and PM emitted from ships are greatly reduced. Therefore, it is important to set up control zones for sulfur emissions from ships.

We integrated JSEI to 0.1∘ resolution and 0.25∘ resolution to compare with DECSO and MEIC, respectively. Because DECSO only covers NOx emissions, we focus on the contribution of NOx emissions from ships to the total emissions. Figure 10 shows the spatial distribution of NOx emissions derived from JSEI, MEIC and DECSO in the Jiangsu section of the Yangtze River. MEIC does not include ship emissions and has a coarser resolution than the other inventories.

When compared to DECSO, NOx emissions from ships account for 6.1 %–74.5 % of the total emissions (Fig. 11). From Fig. 11, we see that the width of the Yangtze River varies from region to region. The width of the Yangtze River increases significantly in Taizhou, Changzhou and the sections of the river close to the estuary of the Yangtze River (red box), where the river flow slows down. Because of the lower river speed, ship emissions in these areas might be overestimated.

The JSEI NOx emissions contribute about 28 % of total NOx emissions of MEIC in a region of 10–25 km around the river (Table 7). Compared to DECSO, the JSEI accounts for 42.9 % of total NOx emissions in the region of 5–10 km from the river. The MEIC grid cells are larger than the DECSO grid cells and therefore include more emissions than for DECSO, so the share of ship emissions for MEIC is nearly half of that for DECSO. Even on the coarser grid of MEIC, the comparison shows that ship emissions from riverine areas account for at least 25 % of the total emissions, so ship emissions should not be ignored in NOx emission inventories. When comparing the total ship emissions with the total non-vessel anthropogenic emissions in Jiangsu province in 2017, NOx emissions from ships still account for 5 % of the total non-vessel anthropogenic NOx in Jiangsu province given by the MEIC (Table S4).

Because SEIM is an annual emission inventory, we only compare the monthly variation of JSEI with MEIC and DECSO. Since the ship emissions for all species show similar monthly variability, we focus here only on NOx. We see that the share of NOx emissions from ships in total NOx regardless of the inventory is highest in the summer (Fig. 12). For the MEIC, NOx emissions from ships can account for 17 %–28 % of the NOx total emissions, and for DECSO, NOx emissions from ships account for 29 %–57 % of the NOx total emissions. The JSEI / (MEIC + JSEI) ratio tends to coincide with the monthly variation in ship emissions, as there is no significant monthly variation in emissions of the MEIC. For DECSO, ship emissions accounted for more than 40 % of the total emissions in February, which shows that the pollution caused by ship activities during the Spring Festival is quite significant in a time period of lower emissions in general.