Answers are grounded in this paper and cite their sources.
Demonstrates improved land cover classification accuracy in urban areas, indicating enhanced environmental monitoring capabilities.
Land cover mapping underpins environmental monitoring, ecological assessment, and land-use planning. However, most existing approaches assume that land cover remains unchanged within a year, overlooking informative intra-annual dynamics. A pixel converted from water to impervious surface in April is typically labelled impervious surface for the whole year by majority voting, so the initial water state and the transition itself are lost. To address this, we propose a Transformer-based framework for monitoring intra-annual land cover dynamics, namely Intra-Annual Net for Land Dynamics (IAND). IAND takes remote sensing image time series as input and employs an encoder–decoder autoregressive classifier to capture intra-annual transitions without prior assumptions about the number or types of changes. Rather than assigning only the dominant annual class, it outputs a semantic transition sequence such as Water to Impervious Surface. Across multiple representative urban regions worldwide, IAND improves within-year classification accuracy and recovers short-cycle transition sequences more faithfully than a single annual label. It shows encouraging sensitivity to subtle changes in regions of rapid urban expansion or seasonal agriculture, while avoiding false changes in stable areas, thereby complementing conventional annual products. All code and data are available at https://github.com/CUG-BEODL/SAND.
No takes yet. Share an insight, caveat, or question.
Xu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: