Table  shows the averages and standard deviations of meteorological conditions during Day 1, Night, and Day 2 in the Ctrl runs for each case. As observed in subtropical marine stratocumulus , the LES runs show a strong diurnal cycle in cloud properties. Figures  and  show the cloud LWP across the model domain in each case during Day 1 and Night, respectively, for the Ctrl runs. In all cases, mesoscale roll convection develops, similar to that seen in the LES runs of a case study using aircraft-observed fields . In the Pristine6 runs, the roll structure is less coherent, especially during daytime, and the cloud structure more closely resembles open cellular convection. This change in the cloud field organization is likely caused by cloud-base precipitation. Solar absorption may also help to break the roll structure .

During daytime, the in-cloud LWP (LWPCLD) is lower than at night (Table ), as solar absorption weakens the net radiative cooling at cloud top, which is the primary source of turbulence in marine low clouds. It is notable that the LWPCLD in the Polluted3.5 and Polluted1.5 runs is much lower than 60 gm-2, contradicting the argument by that a LWP <60gm-2 is difficult to sustain in steady state. Our inclusion of the diurnal cycle (in contrast to ) may permit lower LWP values to be sustained in our cases, but this is unclear. The CF in Pristine6 also varies diurnally from ∼50% during the day to ∼80% at night. During the daytime, weak turbulence is unable to supply water vapor from the subcloud to the cloud layer so that the cloud is depleted by drizzle. During the night, the turbulence is intensified and cloud water is recovered. The Middle6 and Polluted6 cases remain overcast through the entire day, without stabilization of the boundary layer due to drizzle evaporation. As the free troposphere (FT) becomes drier, the CF decreases because the lifting condensation level (LCL) becomes closer to the inversion due to the incorporation of dry free-tropospheric air into the boundary layer.

For the Pristine6 run, the cloud-base precipitation rate Rcb varies diurnally from 0.40 mmd-1 during the day to 0.81 mmd-1 at night. Rcb in the Middle6 run is much weaker due to the low coalescence efficiency of the smaller cloud droplets, and this precipitation is too weak to reach the surface. Rcb in the three polluted cases is negligible (<0.01mmd-1). Similarly, the surface precipitation rate Rsfc in the Pristine6 run changes from 0.15mmd-1 during the daytime to 0.31mmd-1 at night. re is largest in Pristine6 and becomes smaller as Na increases. Among the three polluted cases, the cloud droplet effective radius re is largest for the case with the moist FT (Polluted6) due to thicker clouds and a comparable number of cloud droplets.

Figure  shows vertical profiles of qt and sl during day and nighttime. Generally, qt and sl profiles are more stratified during daytime than nighttime, as solar absorption at cloud top weakens the production of turbulence by longwave cooling. Evaporation of rain drops below cloud base stabilizes the MBL, so the Pristine6 case is the most stratified of the cases. The MBL is more stratified with a moister FT than a drier FT, as a drier FT is more transparent to outgoing longwave radiation so that cloud-top cooling is more effective.

The entrainment rate (we) also varies depending on meteorological conditions (Table , Fig. ). Factors controlling we include boundary-layer turbulence, the strength of the buoyancy jump across the inversion Δb, the boundary-layer depth zinv, and an efficiency term A that incorporates the microphysical (e.g., evaporation and sedimentation) impacts on entrainment. We use the parametric formula suggested in : 1we=Aw*3zinvΔb. Here, w* is the convective velocity scale, a measure of the buoyant production of turbulence, defined as the vertical integral of the buoyancy flux over the boundary layer: w*=2.5∫0zinvw′b′‾dz1/3. In this work, we define B‾, as the vertical integral of buoyancy production normalized by the boundary-layer depth: B‾=w*3/zinv. Therefore, Eq. () becomes 2we=AB‾Δb. Thus, 3A=weΔbB‾.

As approximated by Eqs. (5) and (6) in , entrainment efficiency (A) increases with entrainment-zone cloud liquid water amount, which is largely determined by sedimentation velocity and cloud thickness , and with the reduction in buoyancy due to evaporation by the dilution of cloud water with above-inversion air e.g.,. The expressions in Eqs. () and () are approximate. If, for example, entrainment is related to a different metric of boundary-layer turbulence than B‾ e.g.,, the computation of A from simulated values of we, B‾, and Δb will be affected. As a result, changes in A with aerosol perturbations may result from both changes in entrainment efficiency and errors in the approximations embedded in Eq. (). Figure  shows run-averaged (after the spin-up) A, B‾, Δb, cloud liquid water mixing ratio at zinv-50 m (qc,inv), and we in the Ctrl runs. Generally, A is proportional to qc,inv, consistent with the results of .

In the Pristine6 Ctrl run, we is significantly weaker than the other runs, mainly due to small A and B‾. The low A value may be attributed to low qc,inv caused by the high sedimentation velocity of large cloud droplets, whereas the low B‾ value is caused by the suppression of turbulence from rain formation, which warms the cloud layer, and precipitation evaporative cooling below cloud. Across the weakly precipitating and non-precipitating cases, we is quite similar, but the factors driving we are different. In the Middle6 and Polluted6 runs, where the FT is moist, A is high but B‾ is low. The high A value is mainly due to larger qc,inv, while the moist overlying FT leads to weaker cloud-top radiative cooling and low B‾ e.g.,. For drier FTs, A and qc,inv decrease as clouds become thinner, but B‾ increases as cloud-top cooling becomes more effective. For all cases, we is greater at night than during the day, due to stronger net radiative cooling at cloud top. In the Pristine6 run, we at night (2.72 mms-1) is 3 times greater than during the day (0.92 mms-1). This is partly attributed to the large variation in the CF, in addition to the stronger net radiative cooling at night.

Figure  shows Hovmöller plots of the boundary-layer mean 〈Na〉 (Fig. 5a–e; hereafter, angled brackets represent the boundary-layer mean of a variable) and LWP (Fig. 5f–j) averaged along the y axis for the Plume runs for each case. For all of the cases, the ship tracks are approximately parallel to the roll axis, which strongly affects the lateral dilution of the plume . Except in the Pristine6 run, the plume edge tends to align with the mesoscale cell boundaries, showing that the mixing across the width of a cell is rapid, whereas the mixing to adjacent cells is relatively slow.

During Day 1 in the Pristine6 run, there is a fringe of clear sky at the edges of the plume (Fig. f), consistent with and . These cloud-cleared regions are caused by a mesoscale circulation, generated by the gradient in precipitation rates between the region with a higher NCCLD, due to the injected plume, and the background. The widths of the cloud-cleared regions become broader up until the early afternoon, and they then narrow. At night, the roll structure develops within the plume, and the cloud-free fringes disappear. Clouds within the plume become overcast, while those in the background remain patchy but thick. On Day 2, the areas within the plume remain mostly overcast, while clouds in the background regions are more broken. Starting at sunrise on Day 2, 〈Na〉 in the background air starts to be depleted, likely because turbulence after sunrise is too weak to sustain the cloud number concentration against the loss by coalescence. In the other cases, there is no visible change in cloud morphology with aerosol injection.

In all of the cases, the cloud droplet effective radius averaged over cloudy grids (re) robustly decreases with aerosol injection (Table ), with stronger decreases when unperturbed re is large. The decrease in re is greatest in the Pristine6 case (-1.3, -1.7, and -1.6 µm for Day 1, Night, and Day 2, respectively), followed by the Middle6 case (-0.9, -0.9, and -1.6 µm) for Day 1, Night, and Day 2, respectively), despite lower aerosol injection rates in these runs. Among the three polluted cases, the re decrease is larger in the moister cases, as unperturbed re is larger; the increase in the Nc is very similar in all cases (Table ).

In the Pristine6 and Middle6 cases, the cloud-base precipitation rate Rcb decreases in response to aerosol injection. Reduction in droplet size (Table ) reduces collision–coalescence efficiency. dRcb in the Pristine6 case is -0.06, -0.26, and -0.04 mmd-1 for Day 1, Night, and Day 2, respectively; dRcb in the Middle6 case is -0.03, -0.07, and -0.06 mmd-1, respectively. The decrease is larger during nighttime and early morning because the background precipitation rate is higher. As there is negligible precipitation in the three polluted cases, dRcb is also negligible (Table ). Likewise, Rsfc also decreases in the Pristine6 case (-0.02, -0.09, and -0.01 mmd-1 for Day 1, Night, and Day 2, respectively). Precipitation is too weak to reach the surface in the Middle6 case, so dRsfc∼0.

As a result of the reduction in re, the cloud-top entrainment rate we is enhanced in all cases due to the sedimentation–entrainment feedback (Table ). Turbulence production through drizzle suppression is more effective than the sedimentation–entrainment feedback with respect to enhancing turbulence , so dwe is greatest in the Pristine6 case (Table ), where we in the Plume run is increased by 30 %–100 % over the Ctrl run. The increase in we in the Middle6 case is likely also aided by Rcb suppression. For the three polluted cases, dwe is weaker than in the two precipitating cases, but it generally increases with free-tropospheric moisture.

From Eq. (), the relative change in entrainment rate can be expressed as 4dwewe=dAA+dB‾B‾-dΔbΔb. Run-averaged values of dwe/we, dA/A, dΔb/Δb, and dB‾/B‾ (Fig. ) clearly show that entrainment efficiency changes (dA) dominate changes in entrainment. These changes are much greater in the Pristine6 case than in the Middle6 and three polluted cases (note the different axes), probably due to the greater reduction in re and, thus, also sedimentation velocity. Likewise, among the three polluted cases, dA/A is greater for a moister than a drier FT, which is also attributed to a greater reduction in re. Reduced turbulence (i.e., negative dB‾/B‾) partly mitigates the A-driven increase in dwe. This is largely a damping of the MBL turbulence by the increased entrainment itself , although dB‾>0 in the Pristine6 case due to turbulence invigoration from precipitation suppression. dB‾/B‾ is more negative in the Polluted6 case than in the Polluted3.5 case, likely because dA/A is greater, and the associated enhancement of entrainment results in increased stratification in the boundary layer.

In contrast to the other polluted cases, dB‾/B‾ in Polluted1.5 is positive. This is mainly attributed to greater cloud cover in the Plume run than in the Ctrl run (Table ), leading to a stronger longwave radiative cooling and, thus, an intensification of turbulence. An increase in cloud thickness with greater entrainment can happen in a mixed layer but typically requires a moist FT. Here, increased B‾ is occurring with an extremely dry FT, which one would expect to be detrimental to cloud amount. This behavior may be due to a response (increase) in surface fluxes, as discussed in the following section. Finally, it should be noted that dΔb/Δb contributes only weakly to dwe. In all cases, dΔb increases because stronger entrainment tends to sharpen the inversion.

These results imply that, in general, entrainment enhancement of weakly precipitating and non-precipitating MBLs is mainly caused by an increase in entrainment efficiency, which leads to stronger stratification of the boundary layer. In a precipitating MBL, although there is also a large increase in entrainment efficiency, buoyancy production by suppression of precipitation leads to a more turbulent boundary layer. The combination of the enhancement of A and B‾ drive a large enhancement of the entrainment rate.

As illustrated in Appendix , LWP adjustments are controlled by changes in zinv and zcb, which are determined by both the mean and coupling state of qt and sl. In this subsection, we show the changes in the fluxes of qt and sl into the MBL in response to aerosol injections, which affect the MBL mean state. To understand the different responses of qt and sl, perturbations in net fluxes of qt and sl into the boundary layer, averaged throughout the simulation (dFqt and dFsl, respectively), are given in Fig. . Bars represent dFqt and dFsl by entrainment (orange), by the rest of the processes (blue), and the sum of all terms (hatched). The tables below the histograms in Fig. 7 show the net fluxes due to different processes (i.e., entrainment – ENT, surface flux – SFX, sedimentation – SED, radiation – RAD, and sum of all terms – SUM). The methods to calculate the net fluxes are given in Appendix .

The change in fluxes from entrainment, dFqt,ENT and dFsl,ENT, in the Pristine6 case are -5.9 and 5.9 Wm2, respectively, which are about 5–6 times greater than in the Middle6 case and about 1 order of magnitude greater than the in the polluted cases. Among the three polluted cases, dFqt,ENT is greatest for Polluted1.5, as Δqt is more negative for the drier FT. Because Δsl is comparable across all of these cases (Table ), dFsl,ENT is mostly explained by dwe.

All of the other processes, such as the changes in the surface flux, sedimentation, and radiation, buffer the drying and warming of the boundary layer by entrainment enhancement. This buffering is most effective in the Pristine6 case, with surface moisture fluxes dFqt,SFX and precipitation suppression dFqt,SED together offsetting almost two-thirds of the entrainment drying. Similarly, dFsl,SFX, dFsl,SED, and dFsl,RAD together offset 84 % of the entrainment warming. The surface flux responses and the negative dFsl,SFX response are attributable both to changes in the MBL mean state and to changes in the degree of MBL coupling. The suppression of precipitation also induces moistening and cooling of the boundary layer (i.e., positive dFqt,SED and negative dFsl,SED, respectively). For the Pristine6 case, greater MBL cooling through the radiative flux response (i.e., negative dFsl,RAD) is caused by an enhanced CF at night in the Plume run.

The buffering is less effective in the Middle6 case, where only 20 % of the entrainment drying is offset by a reduced precipitation flux, and the contribution from surface evaporation is negligible. Approximately 60 % of entrainment warming is buffered by a reduction in surface sensible heat flux, reduced precipitation warming, and increased radiative cooling. Although the MBL becomes drier, decoupling suppresses heat and momentum transport, inducing only weak damping from the surface fluxes. Moistening and cooling by precipitation suppression is less effective in the Middle6 case than in the Pristine6 case, due to weaker suppression of Rsfc. As dCF is negligible in the Middle6 case, the increased radiative flux divergence in this case is driven by reduced solar absorption from a reduced cloud LWP (Table ).

The amount of buffering of entrainment drying by other responses depends on the free-tropospheric moisture. Of the polluted cases, dFqt,ENT is most negative for Polluted1.5 due to the drier FT. However, this drying is largely offset by dFqt,SFX so that dFqt,SUM is comparable to that in the Polluted3.5 case but weaker than that in the moister Polluted6 case. This can also be explained by the response of the coupling state of the MBL. The greater enhancement of entrainment in the moister FT induces greater stratification, making the damping by surface fluxes less effective.

dFsl,SFX, on the other hand, is more sensitive to sl, rather than to stratification. dFsl,RAD is more negative in the Polluted6 case than in the Polluted3.5 case, probably because the greater decrease in the LWPCLD (e.g., Table ) leads to a greater reduction in solar absorption. dFsl,RAD in the Polluted1.5 case is, however, much greater than in the Polluted6 and Polluted3.5 cases. This occurs because radiative cooling strengthens with a greater nighttime CF in the Plume than in the Ctrl run (Table ).

These results imply that drying and warming of the MBL by entrainment enhancement is controlled not only by Δqt and Δsl but also by the response of the entrainment rate to aerosol perturbation. In addition, the surface fluxes, precipitation, and radiation respond to aerosol-induced changes in the MBL state and play important roles in the MBL qt and sl budgets. This suggests that studies of aerosol–cloud interactions of a day or longer should include interactive surface fluxes so that the buffering mechanisms seen here are represented.

The previous sections showed how the mean value and MBL coupling state of qt and sl respond to aerosol injection as well as which processes contribute to the response. In this section, we analyze how changes in cloud and MBL properties affect cloud LWP adjustments. Here, we assume that changes in the domain-mean cloud liquid water are determined by changes in the cloud fraction (dLWPCF), cloud thickness (dLWPh), and cloud adiabaticity (LWPfad). dLWPh is further decomposed into contributions to dLWP via changes in cloud thickness resulting from a response in the MBL mean state (dLWPMEAN) and in the degree of MBL coupling (dLWPCPL). Based on this, cloud LWP adjustments can be decomposed as follows: 5dLWP=dLWPMEAN+dLWPCPL+dLWPCF+dLWPfad. Details on how each term is calculated are given in Appendix . Bars in Fig.  show dLWP caused by changes in the cloud thickness from responses in the MBL mean state and the coupling state, changes in the CF, and changes in adiabaticity as well as the sum of all of the terms during (Fig. a) Day 1, (Fig. b) Night, and (Fig. c) Day 2.

On Day 1, dLWPMEAN is negative in all cases and is most negative for the Pristine6 case, followed by the Middle6, Polluted6, Polluted3.5, and Polluted1.5 cases. This is consistent with the degree of entrainment enhancement across the cases, indicating that MBL warming and drying by aerosol injection are mainly driven by entrainment enhancement. At Night, dLWPMEAN becomes more negative, except in the Polluted1.5 case. This implies that the system continues to move toward a drier and warmer steady state, which is consistent with the results of the simulations, which showed that the adjustment equilibrium timescale of cloud macrophysics (about 1 d) is much longer than that of the cloud microphysics (5–10 min).

On Day 2, dLWPMEAN in the Pristine6 case becomes positive (4.66 gm-2). This sign change in dLWPMEAN is mainly attributed to greater nocturnal drizzle suppression. Drizzle suppression is ineffective during daytime, as background Rcb is weak (e.g., Table ). Therefore, entrainment drying and warming are stronger than the damping effects discussed above. At night, drizzle starts to be strongly suppressed (e.g., Table ), leading to significant liquid water retention and turbulence generation. Thus, dLWP is driven more strongly by changes in sedimentation (i.e., surface precipitation), surface fluxes, and radiation during night. As a result, by the end of the night period, entrainment drying and warming is more than offset by these responses, leading to more positive dLWP. Accumulated drizzle suppression during nighttime leads to significant positive dLWPMEAN on Day 2. This leads not only to thicker clouds but also to a greater CF. During Day 1 and at night, dLWPCF is negligible (0.26 and 0.29 gm-2, respectively). On Day 2, dLWPCF becomes 8.44 gm-2, and this term dominates dLWP. Without significant suppression of precipitation, the weakly precipitating and non-precipitating cases do not have this sign change in dLWPMEAN.

Changes in the MBL coupling state in response to aerosol injection also have significant impacts on dLWP. Because turbulence in the MBL is intensified by drizzle suppression, dLWPCPL is positive in the Pristine6 case. More precipitation is suppressed at night, making dLWPCPL more positive at night than during daytime. During daytime in the Middle6, Polluted6, and Polluted3.5 cases, dLWPCPL is negative due to increased stratification of the boundary layer. dLWPCPL tends to be more negative during daytime when the background clouds are thick (Table ), due to a significant enhancement of the entrainment efficiency (i.e., Fig. ) leading to greater stratification. At night, however, dLWPCPL in these cases becomes negligible, as background drizzle is intensified. In the Polluted1.5 case, where the background LWP is very low, dLWPCPL is negligible during daytime. At night it becomes positive, and this term dominates total cloud LWP changes.

In the Pristine6 and Middle6 cases, where Rcb is not negligible, aerosol-induced Rcb suppression reduces the loss of cloud liquid water by drizzle, leading to an increase in the adiabaticity of MBL clouds . However, as there is no precipitation in the polluted cases, there is, of course, no reduction in Rcb (Table ), and dLWPfad is insignificant.

The overall LWP adjustment dLWPSUM in the Pristine6 case is positive, despite significant enhancement of we, due to responses in the coupling state of the MBL by precipitation suppression and the increase in the CF and fad offsetting entrainment drying. dLWPSUM increases with time, as drizzle suppression is accumulated. dLWPSUM during daytime in the Middle6 case, where the background LWP is greatest, is the most negative of all of the cases and becomes less negative as the background LWP decreases. dLWPSUM is more negative on Day 2 than on Day 1 in the weakly precipitating and non-precipitating MBLs (i.e., all cases other than Pristine6). The Middle6 case stands out as having a relatively smaller negative dLWPSUM at night than during either of the days, compared with the other cases. This is because the background Rcb in this case is larger so that positive values of dLWPCPL and dLWPfad compensate for the negative dLWPMEAN. In the Polluted1.5 case, where the background clouds are quite thin, the LWP adjustment is much weaker.

The lifetime of aerosol perturbations is an important factor for determining the full extent of cloud radiative responses to an injection event. The temporal evolution of the aerosol perturbations is analyzed by separating the changes in the total aerosol within the boundary layer (d〈Na〉) into contributions from different processes: entrainment (ENT), autoconversion (AUT), accretion (ACC), scavenging (SCV), sedimentation (SED), and surface flux (SFX). Tendencies of 〈Na〉 for AUT, ACC, SCV, SED, and SFX are calculated online (as illustrated in Sect. ), whereas the tendency of 〈Na〉 for ENT is estimated using we[Na(zinv+50m)-Na(zinv-50m)], where Na(z) is the total aerosol number concentration at level z. The contributions to d〈Na〉 by the end of the simulation are calculated as the difference in each tendency term between the Plume and Ctrl runs, integrated over the time of the simulation. Figure  shows d〈Na〉^, i.e., the contribution to d〈Na〉/dNainit from the different processes. For example, a unit of -1 cm-3cm3 means that the initial perturbation has dissipated completely over the time of the simulation (39.5 h), whereas +1 cm-3cm3 means the initial aerosol perturbation has doubled. Figure S5 in Supplement shows that the budget calculations agree well with the evolution of 〈Na〉.

Because the aerosol injections increase 〈Na〉 considerably, dilution by cleaner free-tropospheric air acts as a sink and negative d〈Na〉. d〈Na〉^ENT is smaller in the Pristine6 case than in the other cases (about -0.1 cm-3cm3). This is mainly because the FT is more polluted than the boundary layer, and the initial perturbation is not significant compared with the jump in Na across the inversion (see Table ). In the other cases, where the boundary layer is more polluted than the FT, d〈Na〉^ENT acts as a major sink, ranging from -0.45 to -0.47 cm-3cm3.

Another primary cause of differences in d〈Na〉^ across cases is collision–coalescence. For precipitating clouds, aerosol injections reduce coalescence scavenging, so that the sink of aerosol to accretion and autoconversion decreases. In the Pristine6 case, where precipitation is most strongly suppressed, accretion and autoconversion decreases together induce positive d〈Na〉^ (contributing 0.70 and 0.07 cm-3cm3, respectively). This implies that, in a strongly precipitating MBL, aerosol injection may induce a transition from open to closed cells. In a weakly precipitating case (Middle6), d〈Na〉^ due to change in accretion and autoconversion is positive but smaller (0.07 and 0.12 cm-3cm3, respectively), extending the lifetime of the aerosol perturbation. In non-precipitating cases, these effects are negligible.

Scavenging of aerosol by coagulation onto cloud droplets induces universally negative d〈Na〉^, as the higher concentration of Na produced by aerosol injection leads to a higher scavenging rate. However, these terms do not have a large impact in the overall d〈Na〉^ in any of the cases examined. Surface aerosol fluxes and sedimentation of rain and cloud droplets do not respond significantly to aerosol injection, so this does not contribute to d〈Na〉^.

Based on the budget approach presented here, we can infer the lifetime of aerosol perturbations. The sum of all of the terms (hatched bars in Fig. , d〈Na〉^SUM) represents the change in d〈Na〉^ over the duration of the experiment after injection (39.5 h). Thus, the lifetime can be roughly estimated by dividing 39.5 h by d〈Na〉^SUM. As d〈Na〉^ increases with time for the Pristine6 case, it is not possible to estimate a lifetime for injected aerosol in this case. The injected aerosol lifetime in the Middle6 case is ∼90 h; for the polluted cases, it is ∼65 h. These timescales are considerably larger than the typical age (∼7 h) of ship tracks detected using satellite observations . Coupled with the fact that the magnitude of the cloud LWP adjustments becomes stronger on Day 2 than on Day 1 (Fig. ), this points to the need to track cloud responses to aerosol injections over multiple days in order to provide an assessment of their overall radiative effect. After 2 d, most ship tracks will have lost their identity, and it may be difficult to distinguish regions containing injected aerosol from marine background conditions.

Previous sections have illustrated how the cloud LWP and cloud droplet number concentrations respond to aerosol injections for the different background meteorological and aerosol cases. This section analyzes injection-induced changes in the cloud radiative effect (dCRE) by decomposition into contributions from changes in the Nc (dCRENc), i.e., the Twomey effect, and from those due to adjustments in the cloud LWP (dCRELWP) and CF (dCRECF). The decomposition approach is described in Appendix . In all of the cases, the residuals (RES) between the actual dCRE and the sum of all components are much smaller than individual components, indicating that the decomposition works quite well. The values of dCRE given here are an average over 24 h, so the zero values during nighttime are accounted for. When comparing dCRE across cases, it should be noted that the domain size in the Pristine6 and Middle6 cases is twice that of the polluted cases and that the amount of aerosol injection is different among the cases (see Table ). The bar plot in Fig.  summarizes dCRENc, dCRELWP, and dCRECF on Day 1 and Day 2 for all of the cases.

For the Pristine6 case, dCRENc, dCRELWP, and dCRECF on Day 1 are -5.8 and -0.8, and -3.3 Wm-2, respectively. Positive adjustments in both the LWPCLD and CF (Table ) approximately double the brightening induced by the Twomey effect alone. On Day 2, contributions from all three cloud properties increase in magnitude (Fig. ), with dCRECF dominating the brightening.

For all other cases, the cloud fraction responses dCRECF contribute only minimally or not at all to dCRE, and CRE changes largely comprise Twomey effects and LWP adjustments only. In all cases other than the Pristine6 case, LWP adjustments are negative (dCRELWP>0). In the Middle6 case, LWP reductions offset almost 50 % of Twomey brightening on Day 1. On Day 2, there is a small amount of cloud brightening in the morning, but there is darkening during the afternoon that offsets the cloud brightening earlier in the morning (Fig. S7b). The actual dCRE during Day 2 of Middle6 indicates brightening, even though the sum of the Nc, LWP, and CF contributions to dCRE indicates darkening. This discrepancy is attributed to the near cancellation of the Nc and LWP contributions, which allows errors in predicting the individual contributions to dominate the total. For the three polluted cases, dCRENc is similar across cases, and it is slightly smaller on Day 2 than on Day 1 in all cases. Although a more evenly distributed plume would result in greater Twomey brightening on Day 2, this is offset by the fact that the magnitude of the aerosol perturbation is decreasing with time (Fig. ). The negative LWP adjustments in the polluted cases offset 10 %–30 % of the Twomey effect on Day 1; this increased to 20 %–50 % on Day 2. It is noteworthy that dCRELWP is more positive under a moist FT (Fig. ), a response that appears to differ from , wherein the strongest negative LWP adjustments occur with a very dry FT. This is due to the greater entrainment enhancement under a moist FT, leading to more significant stratification and a weaker surface flux response than under a dry FT, as described above in earlier sections.