Adutwum et al. (2026) Multi-modal dual-target spatio-temporal graph learning for regional evapotranspiration estimation

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-07-11
• Authors: Gyamfi Kwame Adutwum, ‪Eun‐Sung Chung
• DOI: 10.1016/j.jhydrol.2026.136014

Research Groups

Civil Engineering Department, Seoul National University of Science and Technology

Short Summary

This study presents a multi-modal dual-target spatio-temporal graph neural network (ST-GNN) framework for regional evapotranspiration estimation over the Korean Peninsula, achieving improved accuracy using climate and satellite data fusion.

Objective

• Investigate the feasibility of multi-modal ST-GNN in estimating regional ET with high accuracy

Study Configuration

• Spatial Scale: Regional scale over the Korean Peninsula
• Temporal Scale: Daily to seasonal time scales

Methodology and Data

• Models used: Multi-modal dual-target spatio-temporal graph neural network (ST-GNN)
• Data sources: MODIS-derived surface descriptors, meteorological forcing data from climate reanalysis datasets, FAO-56 Penman–Monteith reference ET (PM ET0)

Main Results

• The multi-modal ST-GNN achieved MAE/RMSE/𝑅2 of 2.197/3.503/0.883 for MODIS ET and 0.473/0.615/0.999 for PM ET0 on the holdout test set.
• Relative to a graph-free CNN + LSTM dual-head baseline, MAE decreased by 33.3% for MODIS ET and 39.0% for PM ET0.

Contributions

• The study demonstrates the effectiveness of multi-modal ST-GNN in estimating regional ET with high accuracy using climate and satellite data fusion.
• The framework's ability to capture spatio-temporal dependence and fuse climate and satellite representations while quantifying their relative contributions is a significant contribution to the field.

Funding

• This research was supported by the National Research Foundation of Korea (NRF) through the Basic Science Research Program (2020R1A2C1007413).

Citation

@article{Adutwum2026Multimodal,
  author = {Adutwum, Gyamfi Kwame and Chung, ‪Eun‐Sung},
  title = {Multi-modal dual-target spatio-temporal graph learning for regional evapotranspiration estimation},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136014},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136014}
}