A neural network approach for the simultaneous retrieval of volcanic ash parameters and SO₂ using MODIS data

Abstract

Abstract. In this work neural networks have been used for the retrieval of volcanic ash and SO2 parameters based on Moderate Resolution Imaging Spectroradiometer (MODIS) multispectral measurements. Different neural networks were built for each parameter to be retrieved, experimenting different topologies and evaluating their performances. As test case the May 2010 Eyjafjallajokull eruption has been considered. A set of six MODIS images have been used for the training and validation phases. In order to estimate of the parameters associated with volcanic eruption such as ash mass, effective radius, aerosol optical depth and sulphur dioxide columnar abundance, the neural networks have been trained by using the retrievals obtained from well known algorithms based on simulated radiances at the top of the atmosphere estimated from radiative transfer models. Three neural network's topologies with a different number of inputs have been compared: (a) only three MODIS TIR channels, (b) all multispectral MODIS channels and (c) only the channels that were selected by a pruning procedure applied to all MODIS channels. Results show that the neural network approach is able to reproduce very well the results obtained from the standard algorithms for all retrieved parameters, showing a root mean square error (RMSE) computed from the validation sets below the target data standard deviation (STD). In particular the network built considering all the MODIS channels gives a better performance in terms of specialization, mainly on images close in time to the training ones, while, as expected, the networks with less inputs reveals a better generalization performance when applied to independent datasets. In order to increase the network generalization capability, a pruning algorithm has been also implemented. Such a procedure permits to operate a features selection, extracting only the most significant MODIS channels from images. The results of pruning revealed that obtained inputs, for all the retrieved parameters, correspond to the TIR channels sensitive to ash, plus some other channels in the visible and mid-infrared spectral ranges. The artificial neural network approach proved to be effective in addressing the inversion problem for the estimation of volcanic ash and SO2 cloud parameters, providing fast and reliable retrievals, which are important requirements during the volcanic crisis.