From Change Detection to Change Analytics: Decomposing Multi-Temporal Pixel Evolution Vectors (Short Paper)

Abstract

Change detection is a well-established process of detaining spatial and temporal changes of entities between two or more timesteps. Current advancements in digital map processing offer vast new sources of multitemporal geodata. As the temporal aspect gains complexity, the dismantling of detected changes on a pixel-based scale becomes a costly undertaking. In efforts to establish and preserve the evolution of detected changes in long time series, this paper presents a method that allows the decomposition of pixel evolution vectors into three dimensions of change, described as directed change, change variability, and change magnitude. The three dimensions of change compile to complex change analytics per individual pixels and offer a multi-faceted analysis of landscape changes on an ordinal scale. Finally, the integration of class confidence from learned uncertainty estimates illustrates the avenue to include uncertainty into the here presented change analytics, and the three dimensions of change are visualized in complex change maps.