Time Series Change Vector Analysis for Semisupervised Abrupt Land Cover Change Detection

Change detection (CD) in satellite image time series (SITS) is more complex than in bitemporal images due to the higher dimensionality of the data. Utilizing the full dimensionality of the time series remains challenging, particularly with dense SITS. An approach that can minimize dimensions without compromising informational depth is essential. In this article, we present an innovative framework for change vector analysis (CVA) in time series analysis and initial demonstrations of its effectiveness in capturing the spectral–temporal characteristics of changes. Unlike current methods, the proposed approach incorporates a wide range of spectral–temporal information and constructs separate reference matrices for each change type, facilitating an in-depth analysis of change components for CD. Based on the time series change vector (TSCV), the proposed framework extends CVA into the time series perspective, offering novel interpretations for magnitude and direction across temporal and spectral dimensions. The framework’s effectiveness is validated using Sentinel-2 data, demonstrating significant improvements in tackling multiple CD challenges in dense SITS scenarios.