Del Frate, Fabio

In recent years many studies concerning the monitoring of volcanic activity have been carried out to develop ever more accurate and refine methods which allow to face the emergencies related to an eruption event. In our work we present different approaches for the volcanic ash cloud detection and retrieval using Sentinel-3 Sea and Land Surface Temperature Radiometer (SLSTR) data. As test case the SLSTR image collected on Raikoke volcano the 22 June 2019 at 00:07 UTC has been considered. A neural network based algorithm able to detect and distinguish volcanic and meteorological clouds, and the underlying surfaces, has been implemented and compared with two consolidated approaches: the RGB (Red-Green-Blue) and the Brightness Temperature Difference procedures. For the ash retrieval parameters (aerosol optical depth, effective radius and ash mass), three different methods have been compared: the reliable and consolidated LUT p (Look Up Table) procedure, the very fast VPR (Volcanic Plume Retrieval) algorithm and a neural network based model.

The implications of the COVID-19 outbreak are subjected to an increasing number of studies. So far, air quality trends related to the lockdown due to the pandemic have been analysed in large cities or entire regions. In this work, the region studied is the metropolitan area of Cagliari, which is the main city on the island of Sardinia (Italy) and can be representative of a coastal city that includes industrial settlements. The purpose of the study is to evaluate the effect of restrictions related to the COVID-19 outbreak on air quality levels and the traffic dynamics in this type of urban area. Nitrogen Dioxide (NO₂) levels before, during and after COVID-19 lockdown have been investigated using data acquired from the Sentinel-5P/TROPOMI satellite combined with on-site measurements. Both TROPOMI detected and ground-based data have revealed higher levels of NO₂ before and after the lockdown, compared to those during the period of COVID-related restrictions, in particular in the urban area of Cagliari. On the other hand, NO2 registered in the oil refinery area did not show significant differences associated with lockdown. The correlation of TROPOMI NO₂ tropospheric column with ground data (surface NO2) on a monthly mean basis showed different values based on the background and the highest Pearson's coefficient was of about 0.78 near to the city centre, where traffic can be considered a significant source of emission. In addition, a comparison of the air pollution level with the dynamics of vehicle traffic was investigated. The study highlighted a remarkable correlation between the reduction of the number of vehicles and the corresponding tropospheric NO₂ values that decreased on a weekly mean basis.

The lesson learned from the recent Icelandic Eyjafjallajokull volcanic eruption is the need to obtain accurate near real time retrievals in order to sample the phenomenon evolution. In particular, because of the harming effects of fine volcanic ash particles on aircrafts, the real time tracking of volcanic clouds is a key issue for aviation safety. The current mostly utilized procedure for the ash retrievals is based on the Brightness Temperature Difference (BTD) algorithm, using the 11 and 12 micron channels measurements and radiative transfer model computation. This latter requires many input parameters and is time consuming, preventing the utilization during the crisis phases. In this work a fast and accurate Neural Network (NN) approach has been developed to detect and retrieve volcanic ash cloud properties using multispectral IR measurements. The exploited data come from the Moderate Resolution Imaging Spectroradiometer (MODIS) acquired over Mt. Etna volcano during the 2001, 2002 and 2006 eruptive events. The procedure consists in two separate steps that uses the three MODIS channels 28, 31 and 32: the detection and the ash retrievals. The detection is reduced to a classification problem. In this context several classes can be individuated, such as free sea surface, meteorological clouds, and ash plume. To maintain the solution of the problem as easy as possible we have simplified the scenario identifying only two classes on the MODIS images: 'ash' and 'no ash' pixels. This approach is coherent with the philosophy of this work in which the time passed to obtain the result is a stringent factor. For the ash mass retrieval, the trained network replicates the model. In fact, in order to have a network able to learn a behavior and to represent it through a functional approximation, it is necessary to provide appropriate information by an ensemble of examples. These latter can be obtained from a model if a direct measure is not available. In this work the results obtained with the BTD procedure have been considered. The results obtained from the entire procedure are encouraging, indeed the confusion matrix for the test set has an accuracy greater than 90%. Moreover the ash mass retrieval shows a good agreement with that achieved by BTD procedure.

In this work neural networks (NNs) have been used for the retrieval of volcanic ash and sulfur dioxide (SO2) parameters based on Moderate Resolution Imaging Spectroradiometer (MODIS) multispectral measurements. Different neural networks were built in order for each parameter to be retrieved, for experimenting with different topologies and evaluating their performances. The neural networks' capabilities to process a large amount of new data in a very fast way have been exploited to propose a novel applicative scheme aimed at providing a complete characterization of eruptive products. As a test case, the May 2010 Eyjafjallajókull eruption has been considered. A set of seven MODIS images have been used for the training and validation phases. In order to estimate the parameters associated to the volcanic eruption, such as ash mass, effective radius, aerosol optical depth and SO2 columnar abundance, the neural networks have been trained using the retrievals from well-known algorithms. These are based on simulated radiances at the top of the atmosphere and are estimated by radiative transfer models. Three neural network topologies with a different number of inputs have been compared: (a) three thermal infrared MODIS channels, (b) all multispectral MODIS channels and (c) the channels selected by a pruning procedure applied to all MODIS channels. Results show that the neural network approach is able to estimate the volcanic eruption parameters very well, showing a root mean square error (RMSE) below the target data standard deviation (SD). 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 the networks with less inputs reveal a better generalization performance when applied to independent data sets. In order to increase the network's generalization capability and to select the most significant MODIS channels, a pruning algorithm has been implemented. The pruning outcomes revealed that channel sensitive to ash parameters correspond to the thermal infrared, visible and mid-infrared spectral ranges. The neural network approach has been proven to be effective when addressing the inversion problem for the estimation of volcanic ash and SO2 cloud parameters, providing fast and reliable retrievals, important requirements during volcanic crises.

The remote sensing of volcanic sulphur dioxide (SO2) is important because it is used as a proxy for volcanic ash, which is dangerous to aviation and is generally more difficult to discriminate. This paper presents an Artificial Neural Network (ANN) algorithm that recognizes volcanic SO2 in the atmosphere using hyperspectral remotely sensed data from the IASI instrument aboard the Metop-A satellite. The importance of this approach lies in exploiting all thermal infrared spectral information of IASI and its application to near real-time volcanic monitoring in a fast manner. In this paper, the ANN algorithm is demonstrated on data of the Eyjafjallajökull volcanic eruption (Iceland) during the months of April and May 2010, and on the Grímsvötn eruption occurring during May 2011. The algorithm consists of a two output neural network classifier trained with a time series consisting of some hyperspectral eruption datasets collected during 14 April to 14 May 2010 and a few from 22 to 26 May 2011. The inputs were all channels (441) in the IASI v3 band and the target outputs (truth) were the corresponding retrievals of SO2 amount obtained with an optimal estimation method. The validation results for the Eyjafjallajökull independent data-sets had an overall accuracy of 100% and no commission errors, therefore demonstrating the feasibility of estimating the presence of SO2 using a neural network approach also a in cloudy sky conditions. Although the validation of the neural network classifier on datasets from the Grímsvötn eruption had no commission errors, the overall accuracies were lower due to the presence of omission errors. Statistical analysis revealed that those false negatives lie near the detection threshold for discriminating pixels affected by SO2. This demonstrated that the accuracy in classification is strictly related to the sensitivity of the model. The lower accuracy obtained in detecting SO2 for Grímsvötn validation dates might also be caused by less statistical knowledge of such an eruption during the training phase.

Accurate automatic volcanic cloud detection by means of satellite data is a challenging task and is of great concern for both the scientific community and aviation stakeholders due to well-known issues generated by strong eruption events in relation to aviation safety and health impacts. In this context, machine learning techniques applied to satellite data acquired from recent spaceborne sensors have shown promising results in the last few years. This work focuses on the application of a neural-network-based model to Sentinel-3 SLSTR (Sea and Land Surface Temperature Radiometer) daytime products in order to detect volcanic ash plumes generated by the 2019 Raikoke eruption. A classification of meteorological clouds and of other surfaces comprising the scene is also carried out. The neural network has been trained with MODIS (Moderate Resolution Imaging Spectroradiometer) daytime imagery collected during the 2010 Eyjafjallajökull eruption. The similar acquisition channels of SLSTR and MODIS sensors and the comparable latitudes of the eruptions permit an extension of the approach to SLSTR, thereby overcoming the lack in Sentinel-3 products collected in previous mid- to high-latitude eruptions. The results show that the neural network model is able to detect volcanic ash with good accuracy if compared to RGB visual inspection and BTD (brightness temperature difference) procedures. Moreover, the comparison between the ash cloud obtained by the neural network (NN) and a plume mask manually generated for the specific SLSTR images considered shows significant agreement, with an F-measure of around 0.7. Thus, the proposed approach allows for an automatic image classification during eruption events, and it is also considerably faster than time-consuming manual algorithms. Furthermore, the whole image classification indicates the overall reliability of the algorithm, particularly for recognition and discrimination between volcanic clouds and other objects.

This work shows the potential use of neural networks in the characterization of eruptive events monitored by satellite, through fast and automatic classification of multispectral images. The algorithm has been developed for the MODIS instrument and can easily be extended to other similar sensors. Six classes have been defined paying particular attention to image regions that represent the different surfaces that could possibly be found under volcanic ash clouds. Complex cloudy scenarios composed by images collected during the Icelandic eruptions of the Eyjafjallajökull (2010) and Grimsvötn (2011) volcanoes have been considered as test cases. A sensitivity analysis on the MODIS TIR and VIS channels has been performed to optimize the algorithm. The neural network has been trained with the first image of the dataset, while the remaining data have been considered as independent validation sets. Finally, the neural network classifier’s results have been compared with maps classified with several interactive procedures performed in a consolidated operational framework. This comparison shows that the automatic methodology proposed achieves a very promising performance, showing an overall accuracy greater than 84%, for the Eyjafjallajökull event, and equal to 74% for the Grimsvötn event.

abstract In the present work, analysis techniques of satellite data in the TIR (Thermal Infrared) are shown, in the framework of volcano monitoring, in particular concerning the estimation of physical quantities related to volcanic ash clouds, ash mass, effective radius, optical thickness at 11 microns (Aerosol Optical Thickness) and the mass of sulfur dioxide, SO2, at 8.7 microns, present in the atmosphere due to volcanic eruptions. MODIS (Moderate Resolution Imaging Spectroradiometer) multispectral data is analysed, using an inversion model based on Multi Layer Perceptron Neural Networks (MLPNN). A network was built for each parameter to be retrieved. Additionally, for volcanic ash, a network for the classification of “ash image pixels” was implemented, which was then used to mask the estimates. Several network topologies were compared in terms of their performance. Concerning the training phase and testing of the networks, two MODIS images were selected covering the eruption of the Icelandic volcano Eyjafjallajokull, which took place from April to May 2010 and was one of the most disastrous natural hazards in recent years. In particular, the image acquired on May 8th 2010, at 13:20 was selected for training. The networks obtained were then applied to an image of May 9th, 2010, 12:25 UTC. The classification NNs were trained with the volcanic ash classification map obtained with the Bright Temperature Difference (BTD) algorithm, assumed to be error free. The neural networks for the quantitative estimation of the parameters associated with volcanic ash, mass, effective radius AOT and SO2, were instead trained with maps obtained using estimation algorithms based on simulated radiances at the top of the atmosphere (TOA), generated in turn applying a radiative transfer model (RTM) to remote sensing data. The networks proved very effective in solving the inversion problem related to the estimation of the parameters of the volcanic cloud, settling the crucial issue related to false alarms in the detection of volcanic ash. Furthermore, once the training phase is complete, NNs provide a faster inversion technique, useful for the applications. From this point of view the technique satisfies the need to respond quickly as a result of disastrous natural hazards, such as volcanic eruptions. In addition, the comparison between network topologies revealed that, for a given truth, a network with few inputs, but containing information on the physics, is better able to model nonlinear functional relations, proving more robust and therefore more able to generalize the phenomenon. Instead, a network ingesting all the sensor bands would probably require pruning to improve its ability to generalize. Future activities include testing the effectiveness of the technique under different lighting conditions (night images) and on other types of multispectral data, such as that provided by high temporal resolution sensors like SEVIRI-MSG, on board the METEOSAT second Generation satellites. The latter would be particularly suitable considering its exceptional quick response characteristics for real-time monitoring of the atmosphere. The use of hyperspectral data, recently used for the estimation of parameters associated with volcanic clouds, is also under consideration for future work.

Volcanic ash clouds detection and retrieval represent a key issue for aviation safety due to the harming effects on aircraft. A lesson learned from the recent Eyjafjallajokull eruption is the need to obtain accurate and reliable retrievals on a real time basis. In this work we have developed a fast and accurate Neural Network (NN) approach to detect and retrieve volcanic ash cloud properties from the Moderate Resolution Imaging Spectroradiometer (MODIS) data in the Thermal InfraRed (TIR) spectral range. Some measurements collected during the 2001, 2002 and 2006 Mt. Etna volcano eruptions have been considered as test cases. The ash detection and retrievals obtained from the Brightness Temperature Difference (BTD) algorithm are used as training for the NN procedure that consists in two separate steps: ash detection and ash mass retrieval. The ash detection is reduced to a classification problem by identifying two classes: “ashy” and “non-ashy” pixels in the MODIS images. Then the ash mass is estimated by means of the NN, replicating the BTD-based model performances. A segmentation procedure has also been tested to remove the false ash pixels detection induced by the presence of high meteorological clouds. The segmentation procedure shows a clear advantage in terms of classification accuracy: the main drawback is the loss of information on ash clouds distal part. The results obtained are very encouraging; indeed the ash detection accuracy is greater than 90 %, while a mean RMSE equal to 0.365 t km−2 has been obtained for the ash mass retrieval. Moreover, the NN quickness in results delivering makes the procedure extremely attractive in all the cases when the rapid response time of the system is a mandatory requirement.

The basic idea of this paper relies on the concurrent exploitation of the capabilities of neural networks (NNs) and SAR interferometry (InSAR) for the characterization of a seismic source and the estimation of its geometric parameters. When a moderate-to-strong earthquake occurs, we can apply the InSAR technique to compute a differential interferogram. The earthquake is generated by an active seismogenic fault having its own specific geometry. The corresponding differential interferogram contains, in principle, information concerning the geometry of the seismic source that the earthquake comes from. To perform the inversion operation, a novel approach based on NNs is considered. This requires the generation of a statistically significant number of synthetic interferograms necessary for the network training phase. Each of them corresponds to a different combination of fault geometric parameters. After the training, the network is ready to perform, in real time, the inversion on new differential interferograms. This paper illustrates such a methodology and its validation on a set of experimental data.