A surface cover change detection method based on the Australian Geoscience Data Cube

Abstract

We describe a surface cover change detection method based on the Australian Geoscience Data Cube (AGDC). The AGDC is a common analytical framework for large volumes of regularly gridded geoscientific data initially developed by Geoscience Australia (GA), the National Computational Infrastructure (NCI) and the Commonwealth Scientific and Industrial Research Organisation (CSIRO). AGDC effectively links geoscience data sets from various sources by spatial and temporal stamps associated with the data. Therefore, AGDC enables analysis of generations of consistent remote sensing time series data across Australia. The Australian Reflectance Grid 25m is one of the remote sensing data sets in the AGDC. The data is currently hosted on the high performance computational cloud at the National Computational Infrastructure. Our change detection method takes advantage of temporally rich data in the AGDC, applying time series analysis to identify changes in surface cover. The aim of this study is to develop a modelling framework addressing these issues, and to improve the efficiency and effectiveness of data modelling processes for the AGDC. The framework adopts a modular design, taking advantages of standardisation of data structures provided by the AGDC. The basic unit in the framework is a modelling module, which applies generic statistical functions or machine learning algorithms on a spatial-temporal partition of remote sensing data. Under the framework, a typical workflow of a modelling process consists of a sequence of connected modules. Such modular design offers both flexibility and reusability. To detect change we apply a series of modules, which are independent of each other. The modules include: - a pixel quality mask and time series noise detection mask, which detects and filters out noise in data; - classification modules based on a random forests algorithm, which classifies pixels into specific objects using spectral information;