Since 1 January 2020, International Maritime Organization (IMO) Marine Environment Protection Committee (MEPC) regulations have limited sulfur in marine fuels from 3.5 % by mass to 0.5 % or required exhaust gas cleaning systems (scrubbers) to achieve an equivalent reduction in sulfur oxide (SOx) pollution (IMO, 2019). These IMO 2020 fuel sulfur regulations and the resulting decrease in sulfate aerosol (airborne particulates) are expected to have large benefits to public health (Partanen et al., 2013; Sofiev et al., 2018; Zhang et al., 2021). They are also expected to have an undesired side effect, however: as sulfate aerosol cools the climate by reflecting sunlight directly and indirectly via changing cloud properties, the IMO 2020 SOx reductions may accelerate global warming.

Shipping effects on clouds were first identified in the mid-1960s in satellite imagery of ship tracks, or curvilinear cloud perturbations following individual ships (Conover, 1966; Twomey et al., 1968). For the same amount of liquid water within a cloud, increasing aerosol increases the cloud droplet number concentration (Nd) and decreases the cloud-top effective radius (re), brightening the clouds (Twomey, 1974, 1977). Cloud macrophysical adjustments to this aforementioned Twomey effect have been observed within ship tracks as well and can reinforce the microphysical brightening effect by suppressing drizzle (Albrecht, 1989; Goren and Rosenfeld, 2012) or counteract it by enhancing entrainment (Chen et al., 2012; Coakley and Walsh, 2002; Toll et al., 2019). Understanding how much greenhouse gas warming is masked by these aerosol–cloud interactions from shipping and other forms of pollution is the largest source of uncertainty in quantifying present-day anthropogenic radiative forcing (Forster et al., 2021).

Although ship tracks have long served as “natural experiments” for testing hypotheses about aerosol–cloud interactions in cases of clear causality (Christensen et al., 2022), until recently, attempts to observationally assess regional- to global-scale cloud perturbations and forcing from shipping have found negligible (Schreier et al., 2007) or null effects (Peters et al., 2011) due to the small fraction of ships that form easily identifiable tracks and the large background variability in cloud properties. New methods using machine learning have identified many times more ship tracks than has been possible with manual identification (Watson-Parris et al., 2022; Yuan et al., 2022, 2019), and analyses tracking air masses from ship locations have shown that cloud adjustments differ systematically between easily identifiable and “invisible” ship tracks (Manshausen et al., 2022). Using some of these newer methods, it has been shown that ship track occurrence decreased regionally after the introduction of emission control areas around North America and Europe and then globally after the IMO 2020 regulations went into effect (Gryspeerdt et al., 2019; Watson-Parris et al., 2022; Yuan et al., 2022). Large-scale changes in cloud microphysical and macrophysical properties have been more equivocal, however. Yuan et al. (2022) found smaller Nd increases within ship tracks after the IMO 2020 regulations, as expected, but, paradoxically, greater re decreases than before and no difference in cloud brightness. Watson-Parris et al. (2022) did not find evidence for a change in global or regional Nd after the IMO 2020 regulations despite the clear decrease in ship tracks, with a possible exception in the southeastern Atlantic.

In this work, we assess the detectability of large-scale cloud perturbations from the IMO 2020 regulations by revisiting an alternate solution to the limitations of “bottom-up” methods tracking individual ship tracks: a “top-down” statistical approach developed by Diamond et al. (2020), hereafter D20, to identify regional-scale cloud perturbations within a shipping corridor in the southeastern Atlantic Ocean basin. A unique meteorological setup makes that region ideal for estimating causal aerosol effects: near-surface winds blow parallel to the shipping corridor and closely constrain the pollution, which also happens to intersect a major stratocumulus cloud deck. D20 used a universal kriging method (see Zimmerman and Stein, 2010, and references therein) to estimate counterfactual fields of cloud properties and radiation in the absence of the shipping corridor based on the observed spatial statistics of nearby, non-shipping-affected grid boxes. They found significant increases in Nd and cloud albedo (a measure of cloud reflectivity) and decreases in re within the stratocumulus deck but estimated that several years' worth of data were needed to detect a clear signal. Thus, it is possible that the effect of the IMO 2020 regulations will have just become detectable using their method.

Here, we apply an updated version of the D20 universal kriging algorithm to satellite retrievals of re and overcast albedo (Acld; top-of-atmosphere albedo when clouds are present) from the Clouds and the Earth's Radiant Energy System (CERES) Single Scanner Footprint (SSF) product for the Terra satellite (Loeb et al., 2018; Minnis et al., 2011). The reader is referred to Methods in Appendix A for further details about the data, universal kriging algorithm, and significance tests. Although D20 found a substantial decrease in cloud liquid water path within the corridor during the afternoon, no significant cloud macrophysical adjustments were found in the morning. We therefore interpret any changes in re and Acld using the Terra record (observations at ∼ 10:30 local time) as being dominated by the Twomey effect. We focus on both the austral spring season (SON; September–October–November), which features the strongest shipping signal (likely due to a combination of favorable meteorology and lower background Nd; Grosvenor et al., 2018), and the annual mean (ANN), which averages a greater number of observations and thus should minimize noise. For a variable X, the “factual” or observed value in the presence of the shipping corridor is referred to as “Ship” (XShip), the counterfactual value in the absence of shipping obtained via kriging is referred to as “NoShip” (XNoShip), and the Ship–NoShip difference is signified as ΔX and is interpreted as the effect due to the presence of the shipping corridor.

An unambiguous decrease in the magnitude of the re perturbation within the shipping corridor is evident in the post-regulation (2020–2022) data compared to the pre-2020 climatology (2002–2019) and the immediately preceding 3-year period (2017–2019) during austral spring (Fig. 1). Although several significant grid boxes (observations falling outside the 95 % confidence interval of the counterfactual) remain in the south of the domain, and thus some level of continued shipping influence is detected (as indicated by field significance at the ≪ 0.05 level), the microphysical changes are smaller and less clearly tied to the corridor; the signal is completely lost further north. Similar results are found for the annual mean values (Supplement Fig. S1), albeit with a clearer continued effect of shipping in the 2020–2022 data.

The shipping perturbation in overcast albedo is less well defined than that in the cloud microphysics, but there is still a clear perturbation in the 2002–2019 climatology and 2017–2019 data that is diminished in the 2020–2022 data in austral spring (Fig. 2). Similar results are found in the annual mean, although the 2020–2022 change is more ambiguous from visual inspection alone (Fig. S2). Lower background Acld values in 2020–2022, particularly in the annual mean (Fig. S2g), may be related to unusually warm sea surface temperatures (Figs. S3–S4); as dimmer clouds are relatively more susceptible to aerosol perturbations, this effect may partially obscure the decrease in cloud brightening from the IMO 2020 regulations.

To assess how anomalous the post-regulation 2020–2022 shipping perturbation values are, we compare them to those from prior 3-year periods by averaging over a core shipping corridor region (see Methods in Appendix A) and, to minimize effects from changing background conditions, also calculate perturbations as relative differences (100 % ⋅ ΔX/XShip). Full results are reported in Table S1 and summarized in Fig. 3. For the austral spring re perturbations, 2020–2022 is unprecedentedly weak (Fig. 3a) and does not overlap any prior period's value within their 95 % confidence intervals (Table S1 in the Supplement). The separation between 2020–2022 and any other period's values is not as clear for austral spring Acld (Fig. 3b), although the 2020–2022 perturbation values are the lowest on record and are the only period for which the effect is not distinguishable from zero at the 95 % confidence level (Table S1). For the annual mean re perturbations, the 2020–2022 values are lower than any other period and the difference with the climatological value is much larger than for any other period, although the separation is not as clear as for austral spring (Fig. 3c). While the annual mean Acld perturbations for 2020–2022 are also the lowest on record, the difference from climatology is not extreme compared to other periods (Fig. 3d).

To assess whether a reduction in the shipping effect after the IMO 2020 regulations went into effect is detected at various possible levels of confidence, Table S2 reports different percentiles of the ratio of the 2020–2022 relative differences over the climatology. A decrease in the re perturbation is detected at greater than 99 % confidence in the austral spring and at greater than 95 % confidence in the annual mean, whereas decreases in the Acld perturbation are only significant at the 90 % confidence level in the austral spring and within the interquartile range in the annual mean. We thus conclude that the effect of the IMO 2020 regulations has been clearly detected in the large-scale cloud microphysics and that there is strong evidence for a decrease in cloud brightness, although more years of data may be required for unequivocal detection of changes in overcast albedo.

There is a detectable change in large-scale cloud microphysical properties and evidence supporting a decrease in cloud brightening within the major southeastern Atlantic shipping corridor after implementation of the IMO 2020 fuel sulfur regulations, resulting in a positive IRFACI within the corridor of O(1 W m-2). Although this study did not address potential changes in cloud adjustments from the IMO 2020 regulations, this will be an important area of future work, especially as the fuel regulations are expected not only to decrease overall aerosol numbers but also shift them toward smaller sizes and sootier composition (Ault et al., 2010; Lack et al., 2011; Seppälä et al., 2021).