APPLICATION OF NEURAL NETWORK APPROACH FOR RETRIEVAL OF VOLCANIC AEROSOL OPTICAL THICKNESS FROM MULTISPECTRAL REMOTE SENSING DATA

摘要

In this work a Multi Layer Perceptron Neural Network (MLPNN) approach has been used for volcanic aerosol optical thickness (AOT) retrieval. Associative memory techniques have proven their efficacy in forecasting and monitoring of various hazards. Volcanic eruptions are an important scenario in which the capabilities of these algorithms may be of great interest. MLPNN needs examples to learn. We chose Moderate Resolution Imaging Spectroradiometer (MODIS) L1 measurements and Aerosol L2 products as input and output, respectively. In particular we used The AOT at 0.55 μm for both ocean and land product. With regard to the NN training and validation steps, a set of MODIS images was selected, covering the Eyjafjallajokull volcanic eruption occurred during May2010. Neural Network revealed its effectiveness in detection and retrieval of AOT, even in conditions in which the MODIS AOT product is not estimated, because of the cloud masking algorithm used that involves also volcanic ash.