EGU21-159
https://doi.org/10.5194/egusphere-egu21-159
EGU General Assembly 2021
© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

A New Algorithm for the Retrieval of Volcanic Ash Cloud Properties using MSG-SEVIRI and Artificial Neural Networks

Dennis Piontek1, Luca Bugliaro1, Christiane Voigt1,2, Adrian Hornby3, Josef Gasteiger4, Ulrich Schumann1, Franco Marenco5, and Jayanta Kar6
Dennis Piontek et al.
  • 1Deutsches Zentrum für Luft- und Raumfahrt, Institut für Physik der Atmosphäre, Oberpfaffenhofen, Germany
  • 2Johannes Gutenberg-Universität Mainz, Institut für Physik der Atmosphäre, Mainz, Germany
  • 3Cornell University, Department of Earth and Atmospheric Sciences, Ithaca, NY, USA
  • 4Universität Wien, Aerosol Physics and Environmental Physics, Vienna, Austria
  • 5Met Office, Exeter, UK
  • 6Science Systems and Applications Inc., Hampton, VA, USA

Artificial neural networks (ANNs) have been successfully applied to various remote sensing problems. Here we use ANNs to detect and analyze volcanic ash clouds pixelwise in MSG-SEVIRI images. Therefore, radiative transfer calculations based on realistic ash properties and atmospheric profiles covering a wide range of possible atmospheric states are performed, and their results are used for the training of the ANNs.

With respect to the volcanic ash properties the role of the complex refractive index (RI) is highlighted: While it can vary strongly between different eruptions, some models use a limited set of RI measurements. Here we sketch a novel method to calculate the RI of volcanic ashes for wavelengths from 5 to 15 µm from measurements of their individual components (i.e. minerals, glasses, gas bubbles) based on generic petrological ash compositions. A comprehensive data set of RIs for volcanic glasses and bulk volcanic ashes of different chemical compositions is derived and used for the ANNs training data set.

The final ANNs with specific tasks (classification, retrieval of optical depth, cloud top height and particle effective radius) are validated against an unseen simulated test data set. This allows us to systematically investigate strengths and weaknesses of the retrievals with respect to cloud properties (e.g. optical thickness), geographic and meteorological conditions. To prove real-world applicability case studies for volcanic ash clouds produced by Eyjafjallajökull (2010) and Puyehue-Cordón Caulle (2011) are considered, and comparisons with lidar and in situ measurements show overall good agreement. As for the training only homogeneous single layer ash clouds were assumed, a sensitivity study was carried out to investigate the impact of the vertical mass profile, multiple layers and the geometrical extent of the clouds on the retrieval results.

Finally, a comparison with a precursor algorithm running operationally at the German weather service (DWD) since 2015 shows that in the case of the Eyjafjallajökull 2010 eruption the new algorithm detects more as well as higher concentrated volcanic ash clouds.

How to cite: Piontek, D., Bugliaro, L., Voigt, C., Hornby, A., Gasteiger, J., Schumann, U., Marenco, F., and Kar, J.: A New Algorithm for the Retrieval of Volcanic Ash Cloud Properties using MSG-SEVIRI and Artificial Neural Networks, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-159, https://doi.org/10.5194/egusphere-egu21-159, 2020.

Corresponding displays formerly uploaded have been withdrawn.