Fig. 1: Mineralogy of dust aerosols.

Existing soil atlases inadequately represent laboratory dust mineralogy, whereas data from the Earth’s surface Mineral dust source InvesTigation (EMIT) project enhance spatial accuracy for iron oxides. a Laboratory data at selected 19 sites. Black dash lines represent volume fractions of 0.25, 0.50, and 1.0, respectively. b Locations of the 19 sites grouped by seven geographical zones where the laboratory soil samples were taken to generate dust aerosols⁴². c Similar to (a) but for simulated mineral fractions at each of the 19 sites with one of the existing soil atlases (C1999)¹⁰ for comparison with the laboratory data. Mass fractions of the minerals rereported in the literature were converted to volume fractions (a and c) using their densities: illite-2750 kg m⁻³, kaolinite-2600 kg m⁻³, smectite-2350 kg m⁻³, hematite-5260 kg m⁻³, goethite-3860 kg m⁻³, quartz-2660 kg m⁻³, calcite-2710 kg m⁻³, feldspar-2560 kg m⁻³, gypsum-2300 kg m⁻³, chlorite-2950 kg m⁻³, and dolomite-2840 kg m⁻³. Iron oxides consist of both hematite and goethite (a, c), which is not distinguished in C1999. The laboratory mineralogy data were used with the assumption that regards all clays as kaolinite for sites where the partitioning between illite and kaolinite is unavailable⁴². Sensitivity tests on this and some other assumptions, e.g., all as illite or smectite, suggest minor impacts on the comparison here and on the conclusion of this study. d The comparison of iron oxides (mass fractions) for dust aerosols converted from soil atlases that were created based on high-bound³ iron oxides identified by existing global soil atlases (C1999¹⁰ and J2014¹¹: royal blue and light sea green, respectively) and from EMIT (dark orange) retrievals (vertical bars: standard deviation over pixels around the sampling sites; see Methods for the data processing) against those from laboratory (the black solid line is the 1:1 line. Black and gray dash lines represent a factor of 2 and 4 differences, respectively). Metrics used to measure the model performance on reproducing the volume or mass fractions of the in-common minerals include the Pearson correlation coefficient (“*” denotes significance at the 95% confidence level) and root mean square error (RMSE).