Reconstructing Multi-Decadal Forest Disturbances: A Spatio-Temporal Transformer Approach

2026-06-05

AI Summary

This study addresses the limitations of conventional forest disturbance monitoring approaches, which rely on pixel-level time-series analysis and neglect spatial context, often yielding low-accuracy and spatially inconsistent long-term disturbance maps. For the first time, a spatiotemporal vision Transformer architecture is applied to reconstruct forest disturbances across the conterminous United States from 1984 to 2022, integrating multisource data from Landsat, Sentinel-1, and Sentinel-2. The model jointly captures temporal dynamics and spatial neighborhood relationships under weak supervision, effectively suppressing noise and spatial artifacts. Evaluated on a newly curated labeled dataset (n=300) and an independent wildfire boundary dataset (n=706), the approach achieves an annual disturbance detection accuracy of 98.2%, with F1 scores of 75.8% and 47.3%, respectively—significantly outperforming traditional methods.

Abstract

Accurate monitoring of forest disturbances is essential for understanding carbon dynamics and land management, yet traditional approaches typically rely on pixel-wise analysis of satellite time-series, ignoring spatial context. We present a deep learning framework that maps 38 years (1984-2022) of forest disturbance across the contiguous United States by modeling temporal trajectories and spatial neighborhoods simultaneously. By leveraging a vision transformer architecture, our approach effectively filters noise from weak supervision signals to produce spatially coherent disturbance maps. We perform exhaustive evaluations across multiple satellites (Landsat, Sentinel-1, Sentinel-2) and temporal windows (38 years and the more recent 6 years), validating performance against a novel, manually annotated validation dataset (n=300) and independent fire perimeter dataset (n=706). The results highlight the complexity of the task: while our spatio-temporal model demonstrates high precision (up to 98.2% for +-1 year detection on MTBS and up to 71.3% on the CONUS validation datasets, with F1-scores up to 75.8% and 47.3%, respectively) and effectively reduces spatial artifacts, it exhibits performance trade-offs across different disturbance regimes compared to pixel-wise baselines. Our method offers a promising foundation for consistent forest monitoring.

Problem

• forest disturbance
• spatio-temporal reconstruction
• satellite time-series
• spatial context
• carbon dynamics

Innovation

• spatio-temporal transformer
• forest disturbance mapping
• weak supervision
• vision transformer
• multi-decadal monitoring