Volcanic cloud detection and retrieval by micro-millimetre-waves and thermal-infrared satellite observations

The characterization of the eruption source parameters (EPS) of explosive eruptions is of vital importance to prevent damages, mitigate environmental impact and reduce aviation risks.  We consider highly explosive eruptions with a Volcanic Explosive Index (VEI) greater than 3. During these eruptions, a great number of volcanic particles are ejected into the atmosphere where they can remain suspended for several weeks. Satellite passive sensors can be adopted to monitor volcanoes due to their high spatial and temporal resolution. 

In this work we combine the Microwave (MW) and Millimetre-wave (MMW) observations with Thermal-InfraRed (TIR) radiometric data from Low Earth Orbit (LEO) satellites to have a complete characterization of the volcanic clouds. MW-MMW passive sensors are adopted to detect larger volcanic particles (i.e. size bigger than 20 µm) by working at lower frequencies. TIR observations are employed to study smaller particles due to the sensor settings which work at smaller wavelengths. We describe new physical-statistical methods together with machine learning techniques aiming at detecting and retrieving volcanic clouds masses of 2015 Calbuco, 2014 Kelud as well as other eruptions having high explosive activities worldwide. Concerning the detection, we compare the well-known split-window methods with a machine learning algorithm named Random Forest (RF). This work highlights how the machine learning model is suitable to automatically identify tephra contaminated pixels by combining different spectral information (i.e. MW-MMW and TIR) coming from different satellite platforms. Indeed, we used data coming from: Advanced Technology Microwave Sounder (ATMS) and Visible Infrared Imaging Radiometer Suite (VIIRS) sensors on board the Suomi-NPP LEO satellite; Microwave Humidity Sounder (MHS) and Advanced Very High Resolution Radiometer (AVHRR) sensors on board the Metop series.  In terms of retrieval, the new developed Radiative Transfer Model Algorithm (RTM_A) is designed to estimate the total columnar content (TCC) and in turn the mass, for both MW-MMW and TIR observations. The synthetic BTs (simulated by RTM_A) are linked with the observed BTs to retrieve the volcanic clouds features. In this respect, two minimization techniques, the Maximum Likelihood Estimation (MLE) and the Neural Network (NN) architecture, are also compared and discussed. Results show a good comparison of the mass obtained using the MLE and NN methods for all the analysed bands but also with previous studies on the deposit as well as other validated satellite retrieval methods. 

In conclusion, this work shows how the machine learning model can be an effective tool for volcanic cloud detection and how the synergic use of the TIR and MW-MMW observations can give more accurate estimates of the near source volcanic cloud.