Aerosol iron solubility in global marine atmosphere projected by deep learning Extended Data Table 3

Dateset of aerosol iron (Fe) solubility projected by the Deep Learning Neutral Network (DLNN) model based on the contents of total Fe, sulfate, nitrate and oxalate in aerosol particles, the particle size and air relative humidity in the ranges of actual atmosphere worldwide reported in the literature. The concentration range of total Fe in global aerosols was between 0.1 ng/m3 and 86000 ng/m3 (Mahowald et al., 2018; Shalley et al., 2018). In the preparation of the dataset, the range was divided into six intervals with different increments according to the occurrence frequency of total Fe concentration in aerosols, namely, 0.1-1 ng/m3 with an increment of 0.1; 1-10 ng/m3 with an increment of 1; 10-200 ng/m3 with an increment of 10; 200-1000 ng/m3 with an increment of 100; 1000-2000 ng/m3 with an increment of 500, and 2000-86000 ng/m3 with an increment of 2000. The concentration range of sulfate and nitrate in aerosol particles was reported from 0 to 110 μg/m3 and 0 to 70 μg/m3, respectively, and their maximum concentrations were the average levels observed in haze aerosols at the city of Xi'an in China (Shen et al., 2009). Here, we set the sulfate and nitrate concentrations as 0-110 μg/m3 with an increment of 5 and 0-70 μg/m3 with an increment of 5, respectively. The concentration range of oxalate in aerosols varied from 0 to 2 μg/m3, and its maximum value was from the coastal city of Qingdao, China (Zhang et al., 2018). The size ranges of particles were divided 9-stage of ≤0.43, 0.43–0.65, 0.65–1.1, 1.1–2.1, 2.1–3.3, 3.3–4.7, 4.7–7.0, 7.0–11, and >11 μm, and the corresponding particle size was set 0.1, 0.54, 0.875, 1.6, 2.7, 4.0, 5.85, 9.0, 55.5 μm, respectively. The air relative humidity was set from 10% to 100% with an increment of 5%. We considered all combinations of the values of the six factors and projected the Fe solubility in aerosols in each case by our constructed and trained DNLL model. In total, we obtained 59,053,994 sets of data including the six controlling factors and the projected Fe solubility (data size about 2.5G). A higher resolution prediction of Fe solubility in global aerosols can be obtained by further reducing the increment of each factor. References 1.Mahowald, N. M. et al. Aerosol trace metal leaching and impacts on marine microorganisms. Nat. Commun. 9, 2614 (2018). 2.Shelley, R. U., Landing, W. M., Ussher, S. J., Planquette, H. & Sarthou, G. Regional trends in the fractional solubility of Fe and other metals from North Atlantic aerosols (GEOTRACES cruises GA01 and GA03) following a two-stage leach. Biogeosciences 15, 2271-2288 (2018). 3.Shen, Z. et al. Ionic composition of TSP and PM2.5 during dust storms and air pollution episodes at Xi'an, China. Atmos. Environ. 43, 2911-2918 (2009). 4.Zhang, S., Shi, J., Yao, X. & Gao H. Distribution of oxalate in atmospheric aerosols and the related influencing factors in qingdao, during winter and spring. Huan jing ke xue 39, 1512-1519 (2018).本数据集为基于文献报道的全球实际大气范围内气溶胶颗粒物总铁、硫酸盐、硝酸盐、草酸盐含量，颗粒物粒径及空气相对湿度区间，通过深度学习神经网络（Deep Learning Neutral Network，DLNN）模型预测得到的气溶胶铁（Fe）溶解度数据集。 全球气溶胶总铁浓度范围为0.1 ng/m³至86000 ng/m³（Mahowald等，2018；Shalley等，2018）。本数据集构建时，根据气溶胶总铁浓度的出现频率，将该区间划分为6个不同步长的子区间：0.1~1 ng/m³，步长0.1；1~10 ng/m³，步长1；10~200 ng/m³，步长10；200~1000 ng/m³，步长100；1000~2000 ng/m³，步长500；以及2000~86000 ng/m³，步长2000。 气溶胶颗粒物中硫酸盐和硝酸盐的浓度范围分别为0~110 μg/m³和0~70 μg/m³，其最高浓度取自中国西安市霾天气气溶胶的观测平均水平（Shen等，2009）。本数据集分别将硫酸盐、硝酸盐浓度设置为0~110 μg/m³（步长5）和0~70 μg/m³（步长5）。 气溶胶中草酸盐浓度范围为0~2 μg/m³，其最大值取自中国沿海城市青岛的观测数据（Zhang等，2018）。 颗粒物粒径区间划分为9级：≤0.43、0.43~0.65、0.65~1.1、1.1~2.1、2.1~3.3、3.3~4.7、4.7~7.0、7.0~11及>11 μm，对应的中位粒径分别设置为0.1、0.54、0.875、1.6、2.7、4.0、5.85、9.0及55.5 μm。 空气相对湿度设置为10%~100%，步长5%。 随后，本研究对上述6个因子的所有取值组合进行枚举，并通过构建并训练的DLNN模型预测每种组合下的气溶胶铁溶解度。最终共获得59053994组数据，包含6个控制因子及预测得到的铁溶解度（数据体量约2.5G）。 通过进一步缩小各因子的步长，可获得全球气溶胶铁溶解度更高分辨率的预测结果。 参考文献 1. Mahowald, N. M. 等. 气溶胶痕量金属淋溶及其对海洋微生物的影响. 《自然-通讯》, 9, 2614 (2018). 2. Shelley, R. U., Landing, W. M., Ussher, S. J., Planquette, H. & Sarthou, G. 北大西洋气溶胶中铁及其他金属的部分溶解度区域特征（GEOTRACES航次GA01与GA03）：经两步浸提实验分析. 《生物地球科学》, 15, 2271-2288 (2018). 3. Shen, Z. 等. 中国西安市沙尘及空气污染事件期间总悬浮颗粒物与PM2.5的离子组成. 《大气环境》, 43, 2911-2918 (2009). 4. Zhang, S., Shi, J., Yao, X. & Gao H. 中国青岛冬春季大气气溶胶中草酸盐的分布及其影响因素. 《环境科学》, 39, 1512-1519 (2018).