VADUGS: a neural network for the remote sensing of volcanic ash with MSG/SEVIRI trained with synthetic thermal satellite observations simulated with a radiative transfer model

Abstract. After the eruption of volcanoes around world, monitoring dispersion ash in atmosphere is an important task for satellite remote sensing since represents a threat to air traffic. In this work we present novel method, tailored Eyjafjallajökull but applicable other eruptions as well, that uses thermal observations SEVIRI imager aboard geostationary Meteosat Second Generation detect clouds and determine their mass column concentration top height during day night. This approach requires compilation extensive data set synthetic train artificial neural network. done by means RTSIM tool combines atmospheric, surface properties runs automatically large number radiative transfer calculations entire disk. The resulting algorithm called “VADUGS” (Volcanic Ash Detection Using Geostationary Satellites) has been evaluated against independent simulations. VADUGS detects ash-contaminated pixels with probability detection 0.84 false-alarm rate 0.05. concentrations are provided correlations up 0.5, scatter 0.6 g m−2 smaller than 2.0 small overestimations range 5 %–50 % moderate viewing angles 35–65∘, 300 zenith close 90 or 0∘. heights mainly underestimated, smallest underestimation −9 between 40 50∘. Absolute errors 70 high correlation coefficients 0.7 concentrations. A comparison spaceborne lidar CALIPSO/CALIOP confirms these results: For six overpasses over cloud from Puyehue-Cordón Caulle volcano June 2011, shows similar features corresponding data, coefficient 0.49 overestimation 55 %, although still uncertainty CALIOP. another both retrievals provide plausible results, being able further away volcano, sometimes missing thick vent. run operationally at German Weather Service application also presented.