Annual 30-m tree mortality dataset across China's Three-North Shelterbelt Forest Program (2019–2025)

Dataset Overview This dataset provides annual tree mortality maps at 30 m spatial resolution across the Three-North Shelterbelt Forest Program (TNSFP) region in northern China, covering the period from 2019 to 2025. The maps quantify the fractional tree mortality rate within forest pixels, ranging from 0 to 100%, and are further classified into four severity tiers: Minimal (0–5%), Mild (5–15%), Moderate (15–25%), and Severe (>25%). Production Methodology The dataset was generated using a unified cross-scale framework that integrates ultra-high-resolution UAV imagery and multi-temporal Sentinel-2 satellite data: 1. Individual tree crown mortality was delineated from UAV RGB imagery using an Edge-enhanced Swin Transformer U-Net (ESTU) deep learning architecture. 2. A coverage-weighted random forest (CW-RF) model coupled with Logit spatial transformation was developed to upscale crown-level observations to regional 30 m satellite scales, mitigating mixed-pixel effects and data imbalance. 3. All mortality estimates were constrained within a stable forest mask derived from the Dynamic World land cover product to exclude non-forest pixels. Spatial resolution: 30 m Coordinate system: WGS84 Data format: GeoTIFF File Structure: `three_north_mortality_30m_2019.tif` – Annual mortality map for 2019 `three_north_mortality_30m_2020.tif` – Annual mortality map for 2020 `three_north_mortality_30m_2021.tif` – Annual mortality map for 2021 `three_north_mortality_30m_2022.tif` – Annual mortality map for 2022 `three_north_mortality_30m_2023.tif` – Annual mortality map for 2023 `three_north_mortality_30m_2024.tif` – Annual mortality map for 2024 `three_north_mortality_30m_2025.tif` – Annual mortality map for 2025