Kaiqi Du, Guilong Xiao, Jianxi Huang, Xia Jing, Xiaoyan Kang, Jianjian Song, Quandi Niu, Haixiang Guan, Xuecao Li, Yelu Zeng
Abstract
Satellite-derived solar-induced chlorophyll fluorescence (SIF) provides critical insights into large-scale ecosystem functions. However, inherent trade-offs between satellite scan range and spatial resolution, coupled with incomplete coverage and irregular temporal sampling, constrain its utility for fine-scale ecological studies. In this study, we present a monthly 500-meter resolution SIF dataset for China (CNSIF, 2003–2022), reconstructed using a deep learning framework integrating high-resolution Landsat/Sentinel-2 surface reflectance and thermal infrared data. CNSIF accurately captures spatial patterns of vegetation photosynthetic activity and reveals a significant annual growth trend (0.054 mW m⁻² sr⁻¹ nm⁻¹ year⁻¹). Validation against tower-based SIF demonstrates its ability to track monthly photosynthetic dynamics across diverse ecosystems, with R² ranging from 0.324 (p < 0.01) to 0.947 (p < 0.001). A strong correlation with tower-based GPP (R² = 0.55, p < 0.001) further highlights its utility for carbon flux estimation. Comparative analyses show CNSIF’s superiority over existing high-resolution SIF products in resolving fragmented landscapes, reducing spatial artifacts, and improving delineation of fine-scale features (e.g., winter wheat fields, urban boundaries) in heterogeneous ecosystems. CNSIF's higher-resolution estimation of photosynthetic activity offers a promising tool for monitoring vegetation dynamics and assessing fragmented agricultural production. It enables the incorporation of ecosystem fragmentation effects into earth observation and carbon cycle systems. CNSIF is publicly available at https://doi.org/10.6084/m9.figshare.27075145.
Keywords
Solar-induced chlorophyll fluorescence; Deep learning; High-resolution; Vegetation dynamics; Heterogeneous ecosystems
1-s2.0-S0168192325004885-main.pdf