Phytoplankton Impact on Marine Cloud Microphysical Properties Over the Northeast Atlantic Ocean

Abstract The current understanding of the impact of natural cloud condensation nuclei (CCN) variability on cloud properties in marine air is low, thus contributing to climate prediction uncertainty. By analyzing cloud remote sensing observations (2009–2015) at Mace Head (west coast of Ireland), we show the oceanic biota impact on the microphysical properties of stratiform clouds over the Northeast Atlantic Ocean. During spring to summer (seasons of enhanced oceanic biological activity), clouds typically host a higher number of smaller droplets resulting from increased aerosol number concentration in the CCN relevant‐size range. The induced increase in cloud droplet number concentration (+100%) and decrease in their radius (−14%) are comparable in magnitude to that generated by the advection of anthropogenically influenced air masses over the background marine boundary layer. Cloud water content and albedo respond to marine CCN perturbations with positive adjustments, making clouds brighter as the number of droplets increases. Cloud susceptibility to marine aerosols overlaps with a large variability of cloud macrophysical and optical properties primarily affected by the meteorological conditions. The above findings suggest the existence of a potential feedback mechanism between marine biota and the marine cloud‐climate system.


Introduction
This file contains 2 supplementary texts explaining the detailed method to compare SYRSOC and MODIS cloud measurements at Mace Head as well as the detailed method of concentration weighted trajectory model. In addition, 16 graphs and 2 tables support the main text.

Text S1. Comparison of SYRSOC and MODIS cloud data at MHD
The Moderate-Resolution Imaging Spectroradiometer (MODIS) level-2 cloud product from both Terra (MOD06) and Aqua (MYD06) platforms (Platnick et al., 2017) were used to compare with SYRSOC output. MODIS has several spectral channels of which near-infrared band, including window channels centred near 1.6 and 2.1 μm, in addition to an AVHRR heritage 3.7 µm channel. These provide cloud microphysical information at 1-km horizontal resolution retrievals and temporal period of collection (every 5 minutes along the orbit track). One 5 min file, referred to as data granule, contains data from roughly 1354 1-km pixels across-track to 2030 km along-track of Earth located data. Thus, a data granule is comprised of approximately 2.7 M 1-km pixels.
We analyzed the granules from both Terra and Aqua which were coincident with the SYRSOC clean marine cases at MHD. This study is based on the 3.7 μm channel, which is less affected by retrieval biases than others (Grosvenor et al., 2018). Out of the 52 clean marine datasets distributed over 47 days (47×2×288 = 27,072 granules), 443 granules have an overpass within 6 hours around the midtime of SYRSOC measurements. We selected a time difference threshold of 6 hours to maintain conditions similar to SYRSOC measurement, without losing too many data points. In each granule, the two external edge scenes were excluded for data quality (Rosenfeld et al., 2019). To minimize the uncertainty, pixels within each granule that have the following conditions are considered: i. Liquid water-phase clouds.
ii. One cloud layer.
iii. Solar zenith angle (SZA) less than 65°, due to the reliability of retrieved cloud properties starts deteriorating with increasing SZA (Grosvenor & Wood, 2014).
iv. Highest 50% of cloud optical depth (COD) within the granule.
MODIS could not retrieve CDNC directly. To obtain CDNC, the input obtained from MODIS cloud products of Reff and COD was utilized to solve the following equation under the abovementioned conditions. where is a constant factor that relates to the mean volume radius and equal to 1.08 (Freud & Rosenfeld, 2012). is the liquid water density. is the rate of increase of with height for a moist adiabatic ascent and is referred to as the condensation rate (Brenguier et al., 2000) or the water content lapse rate (Painemal & Zuidema, 2011). is a weak function of temperature and pressure, varying between 1.8 and 2.25 ×10 -3 g m -3 m -1 in the temperature range 280-290 K at 980 hPa (Szczodrak, Austin, & Krummel, 2001). In our study, we assumed is a constant equal to 2×10 -3 gm m -3 m -1 . Grosvenor & Wood (2014) estimates a 2% of CDNC underestimation if assumed to be constant throughout the cloud instead of taking into account the temperature and pressure variation and smaller errors for shallower clouds.
MODIS data were averaged over an area from 52.33 to 54.33 °N and from 9.90 to 11.90 °W (2° lat/long from MHD toward the Ocean). Out of the 52 clean marine datasets, 27 cases were coincident with overpasses of the Aqua or Terra satellite over MHD and within 6 hours before and after the midtime of SYRSOC cases.
The comparisons are shown in Figure S2. The Reff by SYRSOC is slightly lower than that by MODIS, consistently with Preissler et al. (2016). CDNC is more scattered, consistently with the fact that it is still a very uncertain parameter (Grosvenor et al., 2018), nevertheless, the data fall around the 1:1 line.

Text S2. Three-dimensional Concentration Weighted Trajectory (3D-CWT)
The Concentration Weighted Trajectory ( ) model is a tool for evaluating possible impacts of the long-range aerosol transport, by combining the residence time (trajectory points) of air masses over geographic regions with particle concentrations at a specific point (receptor site). Here, the threedimensional (3 − ) model was used to identify the NEA sea region (1) acting as an emission source of marine aerosols and (2) Where is the index of the trajectory corresponding to each cloud case, is the total number of trajectories which is equal to 52 and is the Reff or CDNC value for each cloud case observed at the sampling location (MHD) on the arrival of trajectory . is the number of trajectory points (air mass residence time) in the grid cell computed by excluding those exceeding the height value , which is the observed cloud base height for the specific cloud case. Similarly, the previously mentioned equation was applied to in situ Na (36 data points because Na is not available for all cloud cases).
In general, the back trajectory arrival points (corresponding to MHD position) and the closest 3 points for each trajectory (3h before arrival time) were excluded from the analysis since they bias the statistics of both and . Since the grid cells containing a low number of endpoints, extremely long distant trajectory points from MHD, can affect the robustness of values, we removed each cell having < median of (less than 5 endpoints) among the non-empty cells.
This reduces the number of considered cells to 378. Figure S1. The 3-day air mass back-trajectories arriving MHD (represented by a filled black square) simultaneously to each cloud case at the cloud base altitude above the ground level during the clean marine conditions. The color scale represents the altitude of each endpoint of back-trajectories. The black box is the main area of interest of the present study.

Figure S2. Comparison between MODIS and SYRSOC cloud microphysical properties (Reff and CDNC).
The red line is the linear regression fit whereas the green if the intercept is forced to zero. The vertical and horizontal bars extend to the 1 st and 3 rd quartiles of each cloud case. The 1:1 line (dashed black) is shown for reference. Figure S3. Spatial distributions of correlation coefficients between Reff measured at MHD and CHL over the NEA Ocean at different time-lags from 0 to 25 days. Only significant correlation coefficients at a 95% confidence level are presented. Figure S4. Same as Figure S3 but for CDNC. Figure S5. Same as Figure S3 but for Na.             Table S1. Seasonal statistics of CHL, Reff, CDNC and Na. The CHL statistics are obtained from daily mean data (2009)(2010)(2011)(2012)(2013)(2014)(2015) within the outlined black box region in Figure 1 (Manuscript). CDNC and Reff statistics are obtained from the considered 52 cloud cases. Most cases are representative of the summer season (24 cases), while 9 cases in both winter and autumn and 10 cases in spring. According to the nonparametric Mann-Whitney U-test, the differences between winter and summer are statistically significant at p<0.05 for CHL, Reff and Na and at p<0.1 for all the reported variables.