Extraction of regions with similar temporal evolution using Earth Observation Big Data. Application to water turbidity dynamics

Earth Observation domain benefited, during the last decades, from numerous programs and initiatives of various space agencies and other engaged stakeholders. As a result, the amount of available satellite data is exponentially increasing each day. Manipulating and processing Big Data in order to obtain valuable and synthetic information requires new and adapted methodologies. An innovative approach was developed in order to extract classes of pixels with similar evolution for specific parameters of interest, over long-time periods. The algorithm was applied to time series of satellite derived turbidity products from the Danube Delta coastal area, with a total of approximately 5700 individual scenes. An unsupervised classification is performed by means of an original, customized method, able to perform on such extremely large dataset. Classes of Pixels with Similar Evolution in Time are extracted, using a combination of Fuzzy Logic and Longest Common Subsequence approach. The algorithm is very versatile, being able to work with scenes contaminated with clouds, with stripes and areas of no data, therefore in the presence of occultation and noise. The results, seen as synthetic overviews, are then compared to local patterns of turbidity evolution in the area of interest.