Authors: Hai Lan*, University of Maryland
Topics: Cyberinfrastructure, Geographic Information Science and Systems, Remote Sensing
Keywords: remote sensing, Apache Spark, cloud computing, image processing
Session Type: Paper
Start / End Time: 5:00 PM / 6:40 PM
Room: Roosevelt 5, Marriott, Exhibition Level
Presentation File: No File Uploaded
Research on attaining sustainable ecosystems is one of the key interests in environmental science. The remote sensing imagery dataset, which is one of the most important data sources in environmental science, has been widely used to support such research by measuring and explaining natural phenomena in multiple ecosystems. Currently, advances in observational platforms that are routinely generating massive amounts of remotely sensed datasets are increasingly providing further support to answering scientific questions in those fields with unprecedented details. However, a grand challenge is to process those datasets into valuable information. In fact, most of the studies with pure remote-sensing data are usually limited to a small study area, with only a few scenes of data, or to low-resolution remotely sensed images for large-area experiments. The goal of this study is to propose a solution based on big data processing framework that is able to process multiple sources of large remotely sensed images with various spatial, spectral and temporal scales in large areas, even on the global scale. By exploiting Apache Spark, we successfully implement and test our parallel in-memory image processing tool to rapidly classify large volumes of multi-scale remotely sensed images and detect changes on the time series. Three different scales of Landsat 8 data have been applied and the results show that ~107.4 gigabyte five-scene time series image-sets can be processed in 1018 seconds in the cloud environment. This study also indicates theoretically all remote sensing imagery can be processed on this framework directly, with only minor adjustments.