Volcanic SO2 plume height retrieval from UV sensors using a full-physics inverse learning machine algorithm

Abstract

Precise knowledge of the location and height of the volcanic sulphur dioxide (SO2) plume is essential for accurate determination of SO2 emitted by volcanic eruptions. Current SO2 plume height retrieval algorithms based on ultraviolet (UV) satellite measurements are very time-consuming and therefore not suitable for near-real-time applications. In this work we present a novel method called the full-physics inverse learning machine (FP-ILM) algorithm for extremely fast and accurate retrieval of the SO2 plume height. FP-ILM creates a mapping between the spectral radiance and the geophysical parameters of interest using supervised learning methods. The FP-ILM combines smart sampling methods, dimensionality reduction techniques, and various linear and non-linear regression analysis schemes based on principal component analysis and neural networks. The computationally expensive operations in FP-ILM are the radiative transfer model computations of a training dataset and the determination of the inversion operator - these operations are performed off-line. The application of the resulting inversion operator to real measurements is extremely fast since it is based on calculations of simple regression functions. Retrieval of the SO2 plume height is demonstrated for the volcanic eruptions of Mt. Kasatochi (in 2008) and Eyjafjallajökull (in 2010), measured by the GOME-2 (Global Ozone Monitoring Instrument - 2) UV instrument on-board MetOp-A.