Volcanic ash detection and retrievals using SLSTR. Test case: 2019 Raikoke eruption

When an eruption event occurs it is necessary to accurately and rapidly determine the position and evolution during time of the volcanic cloud and its parameters (such as Aerosol Optical Depth-AOD, effective radius-Re and mass-Ma of the ash particles), in order to ensure the aviation security and the prompt management of the emergencies.

Here we present different procedures for volcanic ash cloud detection and retrieval using S3 SLSTR (Sentinel-3 Sea and Land Surface Temperature Radiometer) data collected the 22 June at 00:07 UTC by the Sentinel-3A platform during the Raikoke (Kuril Islands) 2019 eruption.

The volcanic ash detection is realized by applying an innovative machine learning based algorithm, which uses a MultiLayer Perceptron Neural Network (NN) to classify a SLSTR image in eight different surfaces/objects, distinguishing volcanic and weather clouds, and the underlying surfaces. The results obtained with the NN procedure have been compared with two consolidated approaches based on an RGB channels combination in the visible (VIS) spectral range and the Brightness Temperature Difference (BTD) procedure that exploits the thermal infrared (TIR) channels centred at 11 and 12 microns (S8 and S9 SLSTR channels respectively). The ash volcanic cloud is correctly identified by all the models and the results indicate a good agreement between the NN classification approach, the VIS-RGB and BTD procedures.

The ash retrieval parameters (AOD, Re and Ma) are obtained by applying three different algorithms, all exploiting the volcanic cloud “mask” obtained from the NN detection approach. The first method is the Look Up Table (LUT_p) procedure, which uses a Radiative Transfer Model (RTM) to simulate the Top Of Atmosphere (TOA) radiances in the SLSTR thermal infrared channels (S8, S9), by varying the aerosol optical depth and the effective radius. The second algorithm is the Volcanic Plume Retrieval (VPR), based on a linearization of the radiative transfer equation capable to retrieve, from multispectral satellite images, the abovementioned parameters. The third approach is a NN model, which is built on a training set composed by the inputs-outputs pairs TOA radiances vs. ash parameters. The results of the three retrieval methods have been compared, considering as reference the LUT_p procedure, since that it is the most consolidated approach. The comparison shown promising agreement between the different methods, leading to the development of an integrated approach for the monitoring of volcanic ash clouds using SLSTR.

The results presented in this work have been obtained in the sphere of the VISTA (Volcanic monItoring using SenTinel sensors by an integrated Approach) project, funded by ESA and developed within the EO Science for Society framework [https://eo4society.esa.int/projects/vista/].