Quality-screened GNSS benchmarking finds limited distance-graph gains for landslide displacement forecasting

Ground-based Global Navigation Satellite System (GNSS) networks can track landslide motion at hourly resolution, but missing records and uneven station quality complicate multi-station forecasting. We used a long-term Huangtupo landslide monitoring record to test whether a distance-graph prior improves displacement-increment forecasts after retrospective station admission. Shared temporal baselines and graph models were trained with identical inputs, rolling-origin splits, and random seeds. The Shared GRU achieved a mean absolute error of 0.2755 mm, slightly below 0.2774 mm for the Graph-fusion GRU; this 0.0019 mm difference is much smaller than the 0.1 mm record resolution and estimated increment-noise scale. Expanding the quality-screened node set narrowed the branch gap but did not reverse the ranking, and the small early 12 h gap did not persist in later folds. Leave-one-station-out, topology, and coherence controls showed heterogeneous station effects but no stable evidence that admission eligibility or distance proximity ordered graph-specific predictive value. A secondary GNSS/InSAR correspondence screen also found no station clearly separated from stable-ground and permutation controls. For this single slope, retrospectively admitted node set, and distance-graph protocol, the results support quality-screened benchmarking before deploying graph-based landslide forecasting models.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-20
DOI
https://doi.org/10.1038/s41598-026-79027-7
Primary Topic
Landslides and related hazards
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Quality-screened GNSS benchmarking finds limited distance-graph gains for landslide displacement forecasting

Mengshuang Huang, Jin Yuan, Yufei Zhang, Yu Zhu et al.
4 links
Scientific Reports
Landslides and related hazards
article

Quality-screened GNSS benchmarking finds limited distance-graph gains for landslide displacement forecasting

Mengshuang Huang, Jin Yuan, Yufei Zhang, Yu Zhu, Linhu Wu, Bingyun Zhong, Qinwen Tan
article en

Abstract

Ground-based Global Navigation Satellite System (GNSS) networks can track landslide motion at hourly resolution, but missing records and uneven station quality complicate multi-station forecasting. We used a long-term Huangtupo landslide monitoring record to test whether a distance-graph prior improves displacement-increment forecasts after retrospective station admission. Shared temporal baselines and graph models were trained with identical inputs, rolling-origin splits, and random seeds. The Shared GRU achieved a mean absolute error of 0.2755 mm, slightly below 0.2774 mm for the Graph-fusion GRU; this 0.0019 mm difference is much smaller than the 0.1 mm record resolution and estimated increment-noise scale. Expanding the quality-screened node set narrowed the branch gap but did not reverse the ranking, and the small early 12 h gap did not persist in later folds. Leave-one-station-out, topology, and coherence controls showed heterogeneous station effects but no stable evidence that admission eligibility or distance proximity ordered graph-specific predictive value. A secondary GNSS/InSAR correspondence screen also found no station clearly separated from stable-ground and permutation controls. For this single slope, retrospectively admitted node set, and distance-graph protocol, the results support quality-screened benchmarking before deploying graph-based landslide forecasting models.

Scientific Reports
China University of Geosciences (CN), China Yangtze Power Co., Ltd. (China) (CN)
National Natural Science Foundation of China, China University of Geosciences
Sustainable cities and communities
Landslides and related hazards
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon