A Parallel and Distributed Approach to the Analysis of Time Series on Remote Sensing Big Data

Abstract

Every day a large number of Earth observation spaceborne and airborne sensors from many different countries provide a massive amount of remotely sensed data. We have entered an era of Remote Sensing Big Data. Those data are used for different applications, such as to identify the Earth's surface changes that are the single most important variable affecting ecological systems and the greatest threat to biodiversity. Time series analysis of remote sensing images has become indispensable to identify these changes and the \textit{Time-Weighted Dynamic Time Warping (TWDTW)} algorithm stands out as one of the best solution found in the literature in this field. However, its computational complexity makes it unfeasible for Remote Sensing Big Data. Moreover, the huge volume of high spatial-temporal resolution remote sensing data cannot be handled by a single computing node. To overcome this limitation, this work proposes a parallel algorithm, named SP-TWDTW (Spatial Parallel TWDTW), that allows the analysis of large scale time series using Manycore architectures (GPU). In order to process massive time series of remote sensing data in a cluster of computers, an approach for distributing the SP-TWDTW processing is introduced in this paper. The SP-TWDTW considers the temporal axis and the spatial auto-correlation to determine the land use mapping in a given region increasing the method accuracy.