Extreme fast volcanic SO2 plume height retrieval from UVN sensors

Abstract

Volcanic eruptions pose a major threat not only to the local population but also to aviation. The knowledge of the exact location and height of the SO2 plume is essential for forecast models and also for aviation control since SO2 causes sulfidation in the aircraft engines which might lead to a total engine failure if exposed over a long time. Furthermore for some volcanic events SO2 was found to be a proxy for the much harder to detect volcanic ash, which melts in the airplane engines eventually also leading to a total engine failure. The near-real-time (NRT) information about the amount and location of volcanic SO2, obtained for example from GOME-2, is currently being used by the Volcanic Ash Advisory Centers (VAACs). What is not available today is the near-real-time information about the altitude of the SO2 plume detected by UVN sensors. The published SO2 plume height retrieval algorithms from UVN data are very time consuming and therefore not adequate for NRT applications. In this work we present a novel algorithm called Inverse Learning Machine (ILM) for the extremely fast and accurate retrieval of SO2 plume height. The basic idea of ILM is to find canonical correlations between spectral radiance and the geophysical parameters of interest using radiative transfer model (RTM) simulations. In the case of the SO2 plume height problem, we use LIDORT-LRRS to simulate radiances in the wavelength range 310 to 335 nm for different SO2 amounts and plume height scenarios and then we compute an inversion operator relating radiances with the plume height. The time consuming part of ILM requiring RTM simulations, computation of canonical correlations and determination of inversion operator is performed off-line. The application of the resulting inversion operator to real measurements is extremely fast because it only involves the computation of a few simple matrix operations. The inversion operator found with ILM is then applied to GOME-2 on MetOp-A and -B measured spectra for a number of volcanic scenarios including Kasatochi (2008), Nabro (2011), and Bardarbunga (2014). The SO2 plume height obtained with ILM agrees well with published results from other algorithms and sensors. Finally we show that the accuracy of the volcanic SO2 columns is significantly improved by using the plume height retrieved with ILM to compute appropriate Air Mass Factors.