This study examined the Daikoku Pier in the Port of Yokohama, Japan (Figure 1). At Daikoku Pier, T1, T2, C1–4, P3, and T3–8 terminals are currently used as seaport automobile terminals. The terminals have gradually been converted from container terminals to automobile terminals. A maximum of 11 Ro-Ro ships can dock at the same time (City of Yokohama, 2022). The terminals that have recently been renovated or reopened as automobile terminals include C3 on 4 August 2020; C4 on 1 April 2022; and P3 on 15 April 2022 (Yokohama Kawasaki International Port Corporation, 2020; Yokohama Port Corporation, 2022; City of Yokohama, 2022). P3 and T3–T6 have been used as both automobile terminals and cruise ship terminals (Keihin Port Office, Kanto Regional Development Bureau, Ministry of Land, Infrastructure, Transport and Tourism, 2022).

The flowchart of the analysis procedure is shown in Figure 2. Planet Dove/SuperDove images (Planet Team, 2017) are commercial data. Planet Explorer, a platform that subscribers can use to search and download images, was used to evaluate the daily coverage and to estimate the number of vessels.

First, we searched all images of the study site from 2018 to 2023 and classified the imagery via visual interpretation into three categories: (1) data available for analysis (available data), (2) data not available for analysis due to cloud (data with cloud), and (3) lack of data. The standard for visual interpretation was based on whether the automobile terminals, as well as vessels, were visible in the images. Even when the data were with cloud, if the automobile terminals and vessels were visible then the images were classified as available data. Our survey was conducted twice for confirmation. We evaluated the agreement rate between the two surveys; then any discrepancies were corrected.

Second, we counted the number of vessels at automobile terminals from corrected ‘available data’ using Planet Explorer. Although it was possible that vessels other than Ro-Ro ships (non-Ro-Ro ships) may call at the automobile terminals, this was not considered to be the majority, so no distinction was made. We counted the number of vessels two times, and then the discrepancies were corrected. This result was compared to the actual schedule information provided by the Yokohama Port Information System at the time of Planet Dove/SuperDove images observed to verify the visual interpretation accuracy. This schedule information was publicly available on the website and included the name of the vessel, type of vessel, and arrival and departure times by terminal in Japanese. However, the container vessels in C3–4 terminals, before being in service as automobile terminals, and the cruise ships in P3 and T3–T6, were excluded because it was not directly related to the automobile terminals targeted in this study. The length of the cruise ships that call at Daikoku Pier range from 290 to 315 m (e.g., Cunard, 2024; Princess Cruise, 2024; MSC, 2024); notably, these cruise ships are much longer than the standard-size Ro-Ro ship of ∼200 m (e.g., K-Line, 2024; NYK Line, 2024; Wallenius Wilhelmsen, 2024).

Finally, the number of vessels counted by Planet constellation were used to analyze the actual usage of the study site. R version 4.22 (R Development Core Team, 2019) was used to construct a heatmap, bar charts, and boxplots of our findings.

Planet Dove/SuperDove images were classified into three classes: available data (Figures 3A, B), data with cloud (Figures 3C, D), and lack of data (Figures 3E, F). As a supplement, Figure 3B was classified as available data due to the absence of cloud cover of the automobile terminals, while Figure 3C was classified as data with cloud due to cloud cover over T1, C3 and C4 terminals. The daily results are presented as a heatmap in Figure 4A and bar charts aggregated by year in Figure 4B. The agreement rate between the first and second visual interpretation was 90.5%. The number of days in a year for the three data classes were 105–155 for available data, 67–86 for data with cloud, and 124–194 for lack of data, from 2018 to 2023. The results over the 6-year period showed the following percentages for the class types: 37.1% available data, 21.2% data with cloud, and 41.8% lack of data. The overall available data ranged from 1 to 22 days per month. There was a period of up to 11 consecutive days with available data from February 3–13, 2018, and a period of up to 29 days without available data from June 27–25 July 2019. On average, the available data was available once every 2.7 days.

The daily number of vessels counted from available data is shown as a heatmap in Figure 5A. The agreement compared to the actual schedule information is shown as heatmap in Figure 5B, and the accuracy was 89.0%. The remaining 11.0% mismatch were due to misses in visual interpretation, as some general cargo vessels were not counted due to their small size. From the actual schedule information, non-Ro-Ro ships such as general cargo vessels, work vessels, cement carriers, tugboats, and container vessels were sometimes called at the Daikoku Pier’s automobile terminals. Hence, this visual interpretation result of the number of vessels include non-Ro-Ro ships as well as Ro-Ro ships. In a few other cases, the visual interpretation results were considered more accurate than the actual schedule information.

The number of vessels were counted from the available data of Planet Dove/SuperDove imagery from 2018 to 2023 at Daikoku Pier in the Port of Yokohama, Japan. While these numbers include Ro-Ro ships, non-Ro-Ro ships and a few mistakes, the analysis was generally good and accurate reflected the actual situation. We investigated how these results could be used for actual daily port usage of automobile terminals.

The boxplots of the number of vessels by month and by day are shown in Figures 6A, B, respectively. The number of vessels ranged from 0 to 9. The median number of vessels was three–four per month. The interquartile range was smaller in August than in other months. The average number of vessels by month was low in August, January, and May (in the order of 2.8, 2.9, and 3.3, respectively) and high in July, June, and September (4.1, 4.0, and 4.0, respectively). In terms of the number of vessels per day, the average tended to be higher from the 25th to the 31st of the month, compared to other days. The average number was particularly high from the 28th onwards. The average for the 1st–24th was 1.9–4.1; for the 25th–27th, 4.4–4.6; and for the 28th–31st, 4.7–5.3. These results suggest that many vessels were docked in automobile terminals at the same time toward the end of the month, regardless of the actual month.

The Planet constellation was not able to observe the same location once a day. The number given for the ‘lack of data’ class increased from 2018 to 2020, whereas it dropped in 2021 and remained at the same level until 2023, despite the increase in the number of launched satellites every year since 2021 (Planet, 2021; Planet, 2022; Planet, 2023). However, 41.8% of the days lacked data over the 6-year survey period. Thus, improving the daily coverage remains a challenge for the satellite constellation. The ‘data with cloud’ ranged from 67 to 86 days in 2018–2023. The effect of cloud is unavoidable due to the characteristics of the optical sensors onboard. When searching images in Planet Explorer, the cloud cover filter can be set to any percentage. However, even if some part of the image is covered by cloud, the region of interest (in this study, automobile terminals) may not be covered by cloud. Filtering by percentage may exclude data available for analysis. In this study, we examined each image individually, so we considered we have evaluated the available data to the maximum extent possible. However, since the analysis was done by visual interpretation, the results may vary slightly depending on the analyst. Especially when the terminal is covered with thin cloud, the decision will be highly dependent on the analyst. It is desirable to develop a method to check this objectively in order to maximize the use of available data.

More than one Planet satellite image existed per month; however, data were missing for up to a maximum of 29 consecutive days in some cases, as mentioned earlier. Thus, it is necessary to understand the limitations of optical satellite constellations affected by cloud. For applications requiring more stable temporal frequency observations, the synthetic aperture radar (SAR) constellation (i.e., Synspective; iQPS) that provides high-resolution, weather-independent images day and night (Moreira et al., 2013) may be more appropriate and will become more readily available in the near future. There have been many vessel detection studies (Li et al., 2024; Yasir et al., 2023) and several maritime logistics studies using SAR data (e.g., Hamamoto et al., 2021; Johnsen, 2011; Li et al., 2022).

We counted the number of vessels by visual interpretation from Planet constellation’s imagery. The results showed good accuracy with the actual schedule information. The discrepancies were mostly a visual misinterpretation. The result of visual interpretation may differ among analysts. On the other hand, nowadays, satellite images are used not only for researchers who analyze satellite images, but also on a daily basis such as in the press (e.g., BBC, 2020). Therefore, it is also important to evaluate how much estimation can be done by visual interpretation. We evaluated the accuracy of the visual interpretation in this study was good. However, it was not a perfect match. Sometimes general cargo vessels were not counted because of their small size. Also, achieving perfect match may be difficult, as in some cases the visual interpretation results seem more accurate than the actual schedule information. It seems that some vessels were not properly registered in the Yokohama Port Information System. Hence, visual interpretation might be more in line with the actual usage in cases.

It is labor-intensive to conduct additional surveys over a wider area. Many methods have been proposed to detect vessels using optical satellite imagery (Kanjir et al., 2018) that may allow for a more efficient count. Alternatively, a vessel’s automatic identification system (AIS) that transmits the position information of vessels may be sufficient for counting the number of vessels in ports and harbors. The real-time and historical AIS data provide insights into shipping-derived impacts and conservation planning at multiple spatial and temporal scales (March et al., 2021). However, satellite imagery offers the advantage of detecting cargo such as automobiles stored in terminals with vessels.

The number of Ro-Ro ships was not constant and, in general, depended on both the day of the month and the month of the year. Satellite images were taken at certain times of the day, so not all vessels docked in Daikoku Pier automobile terminals were estimated; however, our results had some notable findings. The interquartile range of the number of vessels in August was smaller than that of other months. Notably, August is the Obon holiday period in Japan; companies including maritime logistics offices close temporarily during this time. The average number of vessels was slightly higher in June, July, and September due to the transport of automobiles before and after the terminals closed in August. At Daikoku Pier, the number of vessels increased at the end of month, regardless of the month. Automobiles are transported by car carrier from car factories located inland to Daikoku Pier; increasing traffic on the inland roads may be an issue. Additionally, leveling the number of Ro-Ro ships to call in port at the same time may help to mitigate the shortage of workers and storage space. Moreover, the 6-year period was affected by the 2019 Coronavirus (COVID-19) pandemic. The cruise ship, the Diamond Princess, reached the Port of Yokohama on the evening of 3 February 2020 (Jimi and Hashimoto, 2020; Yamagishi et al., 2020) and then anchored at Daikoku Pier P3 to T3. COVID-19 caused an unprecedented drop in maritime mobility, across all categories of commercial shipping worldwide (Millefiori et al., 2021).