Tracking the global reduction of marine traffic during the COVID-19 pandemic

The COVID-19 pandemic has resulted in unparalleled global impacts on human mobility. In the ocean, ship-based activities are thought to have been impacted due to severe restrictions on human movements and changes in consumption. Here, we quantify and map global change in marine traffic during the first half of 2020. There were decreases in 70.2% of Exclusive Economic Zones but changes varied spatially and temporally in alignment with confinement measures. Global declines peaked in April, with a reduction in traffic occupancy of 1.4% and decreases found across 54.8% of the sampling units. Passenger vessels presented more marked and longer lasting decreases. A regional assessment in the Western Mediterranean Sea gave further insights regarding the pace of recovery and long-term changes. Our approach provides guidance for large-scale monitoring of the progress and potential effects of COVID-19 on vessel traffic that may subsequently influence the blue economy and ocean health.


Post-processing S-AIS data
In order to remove inland water bodies from the satellite AIS dataset (S-AIS), grid cells from the Caspian Sea and with ≥95% land coverage were removed from the analysis. Land area coverage estimates were estimated based on the GSHHS shoreline database (version 2.3.7) 1 , a highresolution land mask that represented narrow straits (e.g. Suez Canal or Panama Canal). Further quality control procedures included the removal of ocean grid cells with vessel average speed values above a given threshold (i.e. 99th percentile) and small patches of isolated cells. Overall, we detected a total number of 1,143 patches across all processed months (n = 12). Each monthly map contained one large patch of interconnected grid cells (patch size > 2.5×10 8 km 2 , n = 12).
From the remaining patches, we filtered out patches of size < 769 km 2 (i.e. area of one grid cell in the equator). Visual inspection indicated that small patches with no variability in either the average speed or total number of vessels (i.e. standard deviation equals zero) were likely to be the result of spurious detections (e.g. linear patterns across same longitudes or latitudes), and were also filtered out. Finally, marine traffic density maps were converted to the Mollweide projection with a WGS84 datum as it is a single global projection that preserves geographic area and allows data transfer and analysis among operating systems and software.

Comparison between T-AIS and S-AIS in the Western Mediterranean
Terrestrial AIS (T-AIS) coverage was not homogenous in the Western Mediterranean due to a non-uniformly distribution of antennas (i.e. few antennas in North Africa, see www.marinetraffic.com). In order to assess the potential spatial and temporal bias in T-AIS we compared density estimates with the S-AIS dataset. First, we aggregated the T-AIS data into the same spatial (0.25 x 0.25 degrees) and temporal (monthly) resolutions than the S-AIS data, and reclassified vessels types into five categories (i.e. combined "recreational" with "other"). Then, we calculated the absolute difference between equivalent months across the study period (January -June in 2019 and 2020). Results show that S-AIS provides more information in areas with poor T-AIS coverage (Fig S8). We found a marked underestimation of traffic density by T-AIS in (1) areas with a lower density of antennas (i.e. north Africa), and (2) in areas furthest from the coast (e.g. ocean area between Balearic Islands and Sardinia). Moreover, we found higher underestimates during winter months, when AIS detection range could be reduced by adverse metocean conditions. Overall, results align well with previous studies 2 and support our approach of restricting the analysis to EU coastal areas only.

Global estimates of the annual trends in shipping occupancy
To compare our estimates of changes in shipping occupancy (Fig 4b) with pre-COVID growth rates, we predicted occupancy for 2020 based on the annual trend from the previous 10 years.
Given that no similar AIS dataset was available with an equivalent coverage, we used vessel density estimates from altimetry sensors 3 . Ship density estimates from altimetry were biased towards large vessels, which have higher radar reflectivity, and were acquired at coarser spatial resolution (i.e., 1 x 2 degrees). However, altimetry estimates provide a longer term series that can be combined with AIS data 4 . We derived annual occupancy maps following the same procedure used with S-AIS data (i.e. the extension of all cells with presence of marine traffic). Then, we used an autoregressive integrated moving average (ARIMA) model to forecast the expected occupancy in 2020-2025 based on the 2000-2019 period. The ARIMA model was performed using the forecast package with the auto.arima function, which implements an automatic stepwise model selection algorithm 5 . The selected model corresponded to the ARIMA (0, 1, 0), a random walk with drift, which provided an increase estimate of the occupancy for 2020 of 2.98% (Fig S9). While our approach assumes the same increase across the globe and do not distinguished between sectors, our estimate is similar to previous growth rates for some merchant vessels (e.g. 5% for roll-on/roll-off cargo ships, 2% for bulk carriers) 6 .