top of page
Search

Dr. Jianglei Xu’s recent publication on Remote Sensing of Environment

  • Writer: GEOG HKU
    GEOG HKU
  • Dec 1
  • 2 min read

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

 

Congratulations to Dr. Jianglei Xu and colleagues on publishing a research article on Remote Sensing of Environment (Vol. 333, 115135). The title of the article is “Joint estimation of global daily 1 km surface radiation budget components from MODIS observations (2000-2023) using conservation-constrained deep neural networks” and it is now available at https://doi.org/10.1016/j.rse.2025.115135. The product is freely accessible at www.glass.hku.hk.


Accurate characterization of the surface radiation budget (SRB) is essential for understanding the Earth's climate. However, satellite-based SRB products are typically generated by estimating different components separately using various algorithms, resulting in inconsistent uncertainties and poor energy conservation. To address this, the study developed a conservation-constrained multi-task learning densely connected convolutional neural network (MTL-DenseNet) to jointly estimate global daily SRB components at 1 km resolution. By utilizing MODIS observations spanning 2000-2023 and incorporating physical constraints, the method enhances retrieval accuracy and significantly reduces non-conservation issues. Validation against 224 global sites demonstrates that the new product achieves high accuracy (e.g., daily net radiation RMSE of 24.28 Wm-2) and reduces non-conservation by 26.69% compared to the GLASS-MODIS product.


Schematic diagram of the proposed MTL-DenseNet framework structure consisting of densely connected CNN and multi-task MLP.
Schematic diagram of the proposed MTL-DenseNet framework structure consisting of densely connected CNN and multi-task MLP.

Global distributions of annual mean SRB components from the new MTL-SRB product (Terra + Aqua) and their differences from CERES-SYN, ERA5-Land, GLASS-MODIS, and BESS in 2013.
Global distributions of annual mean SRB components from the new MTL-SRB product (Terra + Aqua) and their differences from CERES-SYN, ERA5-Land, GLASS-MODIS, and BESS in 2013.

Reference: Xu, J., Liang, S., Ma, H., Chen, Y., Li, W., Ma, Y., Zhao, X., Jiang, B., Zhang, X., & Guan, S. (2026). Joint estimation of global daily 1 km surface radiation budget components from MODIS observations (2000–2023) using conservation-constrained deep neural networks. Remote Sensing of Environment, 333, 115135. https://doi.org/10.1016/j.rse.2025.115135


Keywords: Surface radiation budgets, Deep learning, Multi-task learning, SRB non-conservation, Joint estimation

Comments

bottom of page