The Development of a Dust Mineralogy Map from Satellite Retrievals and Implementation in WRF-Chem †

: Mineral dust particles are key ingredients of the atmosphere. They interact in atmospheric physics and chemistry and have important implications for human health. Therefore, it is important to examine the properties of these aerosols, including their ambient concentrations, size distributions, shape and mineral composition. In this work, we use satellite remote sensing from Sentinel 2A and EMIT missions to derive the mineralogical composition of surface areas, and we describe the development of a new module to represent the atmospheric life cycle of individual dust minerals in WRF-Chem. In the ﬁrst step, the GMINER30 mineralogical database is implemented in WRF-Chem to describe the emission, transport, dry and wet deposition of different mineral types.


Introduction
A broad spectrum of environmental processes, such as radiation, cloud formation and ocean fertilization, and human health are affected by the presence of mineral dust.The transport of dust particles is dictated by the prevailing meteorological conditions, as well as the composition and physiochemical properties of the particles themselves.The latter factors are determined by the soil mineralogy in the source region.To develop a more refined mineralogical categorization that can significantly improve the dust transport estimations from numerical models and prepare for their implications on weather, biogeochemistry and health, we have worked to achieve two goals: (i) derive a finer mineralogical partition of the source regions through the utilization of high-resolution multi-spectral (Sentinel 2) [1] and hyperspectral (EMIT-NASA) EO datasets [2]; (ii) implement the existing GMINER30 mineralogical database [3] in the WRF-CHEM model and perform sensitivity tests.

Mineralogy from Multispectral (Sentinel 2A) and Hyperspectral (EMIT) Satellite Retrievals
The broader area of Lake Chad in Africa was our selected test-bed for the calculation of mineralogical abundances.Satellite estimates were derived for specific dates in Spring and Autumn in order to efficiently exclude areas of dense vegetation (NDVI > 0.3) and identify a number of minerals via spectral indices.The reference spectrum of minerals related to dust was derived from the USGS Spectral Library v7 [4] and analyzed for signature reflectivity characteristics in specific wavelengths, upon which a number of custom band ratios were created.From Sentinel 2 estimates, Alteration, Ferric Oxides and All Iron were calculated (as both Plagioclase and Orthoclase in the Feldspar group are featureless in the specific bands).As the number of bands in the Sentinel 2A estimates are limiting to identifying individual minerals, an approach of calculating mineralogical categories was preferred instead.Ferric Oxides includes minerals such as Hematite, Goethite and Jarosite, whereas All Iron includes both ferrous as well as ferric oxides of iron.The Alteration index defines areas that are rich in clay content.These three categories can be seen in Figure 1.(ii) implement the existing GMINER30 mineralogical database [3] in the WRF-CHEM model and perform sensitivity tests.

Mineralogy from Multispectral (Sentinel 2A) and Hyperspectral (EMIT) Satellite Retrievals
The broader area of Lake Chad in Africa was our selected test-bed for the calculation of mineralogical abundances.Satellite estimates were derived for specific dates in Spring and Autumn in order to efficiently exclude areas of dense vegetation (NDVI > 0.3) and identify a number of minerals via spectral indices.The reference spectrum of minerals related to dust was derived from the USGS Spectral Library v7 [4] and analyzed for signature reflectivity characteristics in specific wavelengths, upon which a number of custom band ratios were created.From Sentinel 2 estimates, Alteration, Ferric Oxides and All Iron were calculated (as both Plagioclase and Orthoclase in the Feldspar group are featureless in the specific bands).As the number of bands in the Sentinel 2A estimates are limiting to identifying individual minerals, an approach of calculating mineralogical categories was preferred instead.Ferric Oxides includes minerals such as Hematite, Goethite and Jarosite, whereas All Iron includes both ferrous as well as ferric oxides of iron.The Alteration index defines areas that are rich in clay content.These three categories can be seen in Figure 1.On the other hand, the 285 narrow spectral bands of EMIT reflectance products allow a significantly more refined partition in the identification of specific minerals, as presented in Figure 2. The Level 2a product that is currently available provides surface reflectance, which is derived by screening clouds and correcting for atmospheric effects.By utilising the L2A estimates and resorting to the aforementioned custom band ratios in Table 1, we identified a number of minerals that relate to the dust particle uptake.In 2023, the Level 2b product is expected to offer mineralogy data derived from fitting reflectance spectra after screening for non-mineralogical components, so we could input these categories into a global Numerical Weather Prediction (NWP) model.

Name Chemical Formula
Ratio in Wavelengths (nm) On the other hand, the 285 narrow spectral bands of EMIT reflectance products allow a significantly more refined partition in the identification of specific minerals, as presented in Figure 2. The Level 2a product that is currently available provides surface reflectance, which is derived by screening clouds and correcting for atmospheric effects.By utilising the L2A estimates and resorting to the aforementioned custom band ratios in Table 1, we identified a number of minerals that relate to the dust particle uptake.In 2023, the Level 2b product is expected to offer mineralogy data derived from fitting reflectance spectra after screening for non-mineralogical components, so we could input these categories into a global Numerical Weather Prediction (NWP) model.

Implementation of GMINER30 Database in WRF-Chem
To represent atmospheric transport as well as the dry and wet deposition mechanisms of the different mineral components of desert dust, we developed a dust mineralogy module in the framework of the WRF-Chem regional model [5], which we updated with the MODIS-NDVI active dust sources definition, as described in [6].In order to achieve this, we implemented the global 30sec GMINER30 high-resolution mineralogical gridded database of dust-productive soils for atmospheric dust modeling [3].This dataset includes a mean global distribution of the soil mineral composition and is appropriate for implementation in global and regional numerical studies.The distribution of the effective mineral content in soil in percentages is given for quartz, illite, kaolinite, smectite, feldspar, calcite, hematite and gypsum.The mineral fraction is weighted in terms of the clay and silt content in the soil.To derive the mass size distribution for each emitted mineral, we followed the process described in [7], where, for the normalized mass size distribution for each emitted mineral, we assumed that aggregates are homogeneous mixtures of minerals with similar fragmentation properties.The modeled surface mineralogical composition is shown in Figure 3, as obtained via the implementation of GMINER30 in WRF-Chem.Important spatial variability is evident for most minerals, such as kaolinite and quartz, throughout the Saharan and Arabian deserts, which is in accordance with earlier studies [3].The developed module is able to handle various datasets with minimal tampering, and therefore, additional mineralogical databases from satellite missions (e.g., Sentinel 2 and EMIT) will be used as inputs in the model as soon as they become available.As an example, the partitioning of total dust to specific elements (in this case, quartz) is shown in Figure 4.As shown in this plot, the variability of quartz particles for a typical desert dust episode depends on both the atmospheric circulation and the surface mineralogy.

Implementation of GMINER30 Database in WRF-Chem
Τo represent atmospheric transport as well as the dry and wet deposition mechanisms of the different mineral components of desert dust, we developed a dust mineralogy module in the framework of the WRF-Chem regional model [5] , which we updated with the MODIS-NDVI active dust sources definition, as described in [6].In order to achieve this, we implemented the global 30sec GMINER30 high-resolution mineralogical gridded database of dust-productive soils for atmospheric dust modeling  most minerals, such as kaolinite and quartz, throughout the Saharan and Arabian deserts, which is in accordance with earlier studies [3].The developed module is able to handle various datasets with minimal tampering, and therefore, additional mineralogical databases from satellite missions (e.g., Sentinel 2 and EMIT) will be used as inputs in the model as soon as they become available.As an example, the partitioning of total dust to specific elements (in this case, quartz) is shown in Figure 4.As shown in this plot, the variability of quartz particles for a typical desert dust episode depends on both the atmospheric circulation and the surface mineralogy.most minerals, such as kaolinite and quartz, throughout the Saharan and Arabian deserts, which is in accordance with earlier studies [3].The developed module is able to handle various datasets with minimal tampering, and therefore, additional mineralogical databases from satellite missions (e.g., Sentinel 2 and EMIT) will be used as inputs in the model as soon as they become available.As an example, the partitioning of total dust to specific elements (in this case, quartz) is shown in Figure 4.As shown in this plot, the variability of quartz particles for a typical desert dust episode depends on both the atmospheric circulation and the surface mineralogy.

Figure 1 .
Figure 1.Alteration, Ferric Oxides and All Iron Oxides, as calculated using Sentinel data.Black color indicates no identification and red color indicates high identification of each mineral.

Figure 1 .
Figure 1.Alteration, Ferric Oxides and All Iron Oxides, as calculated using Sentinel data.Black color indicates no identification and red color indicates high identification of each mineral.

Figure 2 .
Figure 2. Calcite, Feldspar, Hematite, Clays, Smectite, Kaolinite, Ilite, Gypsum and Phosphorus, as identified from the custom band ratios from EMIT.Black color indicates no identification and red color indicates high identification of each mineral.

Figure 2 .
Figure 2. Calcite, Feldspar, Hematite, Clays, Smectite, Kaolinite, Ilite, Gypsum and Phosphorus, as identified from the custom band ratios from EMIT.Black color indicates no identification and red color indicates high identification of each mineral.

Figure 3 .
Figure 3. Percentage distribution of the effective mineral contents in soil for (a) iron, (b) feldspars, (c) kaolinite and (d) quartz.

Figure 4 .
Figure 4. Desert dust concentration at the surface (a) and the corresponding quartz mineral concentration (b).

Figure 3 .
Figure 3. Percentage distribution of the effective mineral contents in soil for (a) iron, (b) feldspars, (c) kaolinite and (d) quartz.

Figure 3 .
Figure 3. Percentage distribution of the effective mineral contents in soil for (a) iron, (b) feldspars, (c) kaolinite and (d) quartz.

Figure 4 .
Figure 4. Desert dust concentration at the surface (a) and the corresponding quartz mineral concentration (b).

Figure 4 .
Figure 4. Desert dust concentration at the surface (a) and the corresponding quartz mineral concentration (b).

Table 1 .
Custom spectral indices for EMIT.