Xu et al. (2025) Joint estimation of global daily 1 km surface radiation budget components from MODIS observations (2000−2023) using conservation-constrained deep neural networks

Identification

• Journal: Remote Sensing of Environment
• Year: 2025
• Date: 2025-11-16
• Authors: Jianglei Xu, Shunlin Liang, Han Ma, Yongzhe Chen, Wenyuan Li, Yichuan Ma, Xiang Zhao, Bo Jiang, Xiaotong Zhang, Shikang Guan
• DOI: 10.1016/j.rse.2025.115135

Research Groups

• Jockey Club Laboratory of Quantitative Remote Sensing, Department of Geography, University of Hong Kong, Hong Kong Special Administrative Region of China
• The State Key Laboratory of Remote Sensing Science, Jointly Sponsored by Beijing Normal University and Institute of Remote Sensing Applications of Chinese Academy of Sciences, Beijing, China
• College of Global Change and Earth System Sciences, Beijing Normal University, Beijing, China

Short Summary

This study developed conservation-constrained deep neural network models to jointly estimate global daily surface radiation budget (SRB) components at 1 km resolution from MODIS observations (2000–2023). The method significantly improves the accuracy and conservation of SRB component retrievals compared to existing products, facilitating a better understanding of their coordinated variation.

Objective

• To jointly estimate global daily surface radiation budget (SRB) components at 1 km spatial resolution from MODIS observations (2000–2023) using conservation-constrained deep neural networks, addressing issues of varying uncertainties and poor conservation in separately estimated SRB products.

Study Configuration

• Spatial Scale: Global, 1 km spatial resolution
• Temporal Scale: Daily, spanning 2000–2023

Methodology and Data

• Models used: SRB conservation constraint multi-task learning densely connected convolutional neural network models
• Data sources: MODIS observations (reflectance from bands 1–5, 7, 19; thermal radiance from bands 28–29, 31–34), ancillary information (elevation, solar-viewing geometry), GLASS-MODIS surface longwave radiation

Main Results

• Validation against 224 sites over three years showed the following daily RMSEs:
  • Downward shortwave radiation: 29.38 W⋅m⁻²
  • Upward shortwave radiation: 20.73 W⋅m⁻²
  • Net shortwave radiation: 23.14 W⋅m⁻²
  • Downward longwave radiation: 19.98 W⋅m⁻²
  • Upward longwave radiation: 15.69 W⋅m⁻²
  • Net longwave radiation: 14.70 W⋅m⁻²
  • Net radiation: 24.28 W⋅m⁻²
• The method improves underestimations of downward and net shortwave radiation in MCD18A1, GLASS-MODIS, and BESS products.
• Accuracy of downward and net longwave radiation retrievals is better than those from GLASS-MODIS and CERES-SYN.
• The estimates reduce non-conservation by 26.69 % compared to GLASS-MODIS.
• The SRB estimates exhibit great spatio-temporal consistency with other SRB products, except for regional reflected solar radiation and net radiation.

Contributions

• Developed a novel SRB conservation constraint multi-task learning densely connected convolutional neural network model for joint estimation of SRB components.
• Enhanced retrieval accuracy by effectively utilizing cross-domain features from all SRB components.
• Addressed and significantly reduced non-conservation issues in SRB retrievals through the explicit constraint of SRB conservation during model training.
• Generated a new global daily 1 km SRB product (2000–2023) with high accuracy and conservation, freely accessible for climate, ecology, and hydrology applications.

Funding

• Not specified in the provided text.

Citation

@article{Xu2025Joint,
  author = {Xu, Jianglei and Liang, Shunlin and Ma, Han and Chen, Yongzhe and Li, Wenyuan and Ma, Yichuan and Zhao, Xiang and Jiang, Bo and Zhang, Xiaotong and Guan, Shikang},
  title = {Joint estimation of global daily 1 km surface radiation budget components from MODIS observations (2000−2023) using conservation-constrained deep neural networks},
  journal = {Remote Sensing of Environment},
  year = {2025},
  doi = {10.1016/j.rse.2025.115135},
  url = {https://doi.org/10.1016/j.rse.2025.115135}
}