Unraveling vegetation carbon storage using STVisX: a spatiotemporal visual explainable AI framework

Understanding vegetation carbon storage dynamics is essential for assessing ecosystem resilience and achieving carbon neutrality, as vegetation governs spatiotemporal variability in carbon sinks and underpins low-carbon planning. Yet existing studies lack the spatial resolution needed to capture fine-scale urban–ecological interactions. Machine learning improves predictive accuracy but remains opaque, and existing explainable AI (XAI) methods rarely preserve the spatial structure of factor contributions or capture their temporal dynamics, limiting their use in rapidly transforming cities. To address these gaps, we introduce the spatiotemporal visual explainable AI (STVisX) framework, which integrates meta-analysis, remote sensing, and XAI to quantify and explain urban vegetation carbon storage. We apply STVisX to Yan’an City, investigating species-level vegetation carbon storage using Sentinel-2 spectral data for twelve dominant species. By combining meta-analysis, remote sensing, and machine learning, we generate high-resolution estimates of seasonal vegetation carbon storage for 2018–2024. STVisX visualizes how the explanatory variables jointly contribute to predicted vegetation carbon storage across seasons and reveals seasonal shifts in their relative contributions, reflecting seasonal relationships between phenology and topography. These insights provide an empirical basis for integrating carbon storage management into urban planning and ecological restoration policies to support greener and more resilient cities.