Publications related to the GRACE Missions (no abstracts)

• Sorted by Date • Sorted by Last Name of First Author •

A semi–supervised LSTM framework for spatiotemporal downscaling of GRACE–derived terrestrial water storage anomalies to improve flood monitoring in the Yarlung Tsangpo River basin

Hu, Ying, Cai, Xiaoming, Xu, Yue–Ping, Gu, Haiting, Xie, Jingkai, and Dai, Sirui, 2026. A semi–supervised LSTM framework for spatiotemporal downscaling of GRACE–derived terrestrial water storage anomalies to improve flood monitoring in the Yarlung Tsangpo River basin. Journal of Hydrology, 671:135275, doi:10.1016/j.jhydrol.2026.135275.

Downloads

• from the NASA Astrophysics Data System  • by the DOI System  •

@ARTICLE{2026JHyd..67135275H,
       author = {{Hu}, Ying and {Cai}, Xiaoming and {Xu}, Yue-Ping and {Gu}, Haiting and {Xie}, Jingkai and {Dai}, Sirui},
        title = "{A semi-supervised LSTM framework for spatiotemporal downscaling of GRACE-derived terrestrial water storage anomalies to improve flood monitoring in the Yarlung Tsangpo River basin}",
      journal = {Journal of Hydrology},
     keywords = {GRACE/GRACE-FO, YarlungTsangpo River Basin, Semi-supervised learning, Spatiotemporal downscaling, Flood monitoring},
         year = 2026,
        month = may,
       volume = {671},
          eid = {135275},
        pages = {135275},
     abstract = "{The Gravity Recovery and Climate Experiment (GRACE) and its follow-on
        mission (GRACE-FO) provide essential observations for flood
        monitoring in medium and large basins. However, their
        applications are constrained by coarse spatiotemporal
        resolution, data gaps, and product inconsistencies. In this
        study, GRACE-derived terrestrial water storage anomalies (GRACE-
        TWSA) products from three major agencies were fused using the
        Bayesian three cornered hat (BTCH) to reduce uncertainty in the
        Yarlung Tsangpo River Basin (YTRB), China. Building on this, we
        developed a novel semi-supervised spatiotemporal downscaling
        framework based on long short-term memory networks (SSDF-LSTM).
        This framework leveraged both labeled and unlabeled data,
        applied physical constraint during pseudo-label refinement, and
        adopted a spatially stratified training strategy that preserved
        complete temporal signals. As a result, the monthly
        0.5{\textdegree} TWSA was downscaled to daily 0.25{\textdegree}
        resolution while filling missing data and maintaining high
        consistency with GRACE observations, achieving a correlation
        coefficients (CC) of 0.97, Nash-Sutcliffe efficiencies (NSE) of
        0.92, and root mean square error (RMSE) of 22.26 mm.
        Furthermore, we proposed a daily detrended normalized flood
        potential index (D-DNFPI) which demonstrated higher accuracy
        (0.85) and precision (0.80) than conventional indices. Overall,
        this study established a robust methodological framework for
        enhancing the spatiotemporal resolution of GRACE-TWSA and
        advancing its applications in flood monitoring.}",
          doi = {10.1016/j.jhydrol.2026.135275},
       adsurl = {https://ui.adsabs.harvard.edu/abs/2026JHyd..67135275H},
      adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}

Generated by bib2html_grace.pl (written by Patrick Riley modified for this page by Volker Klemann) on Fri Apr 10, 2026 11:13:48

---

Fri Apr 10, F. Flechtner