Model Maps Where Rain and Earthquakes Could Trigger Landslides Across China

Landslides rank among the most widespread and destructive geological hazards in China, and their impacts become especially severe when intense rainfall and strong ground motion strike in combination. A new open-access study in the journal Geoenvironmental Disasters presents a scenario-based, physically informed framework for estimating landslide failure probability across the entire country, integrating two well-established physical models rather than relying on the empirical or machine-learning approaches that have dominated national-scale hazard mapping to date.

The research team, led by Siyuan Ma and Renmao Yuan with Xiaoyi Shao and Chong Xu, combined the TRIGRS model, which simulates transient rainfall infiltration and its effect on pore water pressure, with the Newmark model, which estimates permanent slope displacement under seismic shaking. To translate deterministic model outputs into probabilistic metrics, the researchers applied the Rosenblueth point estimate method to incorporate uncertainty in cohesion and internal friction angle for rainfall-induced failure, and used an empirical failure-probability curve calibrated from the 2008 Wenchuan earthquake inventory to convert Newmark displacements into seismic failure probabilities.

The framework was run on a uniform 100-meter computational grid derived from the Shuttle Radar Topography Mission digital elevation model, with soil depth from a global gridded dataset and soil hydraulic and geomechanical parameters drawn from the Harmonized World Soil Database. The authors emphasize that these parameters are representative national-scale averages for each soil class rather than site-specific engineering values, an important caveat when interpreting the results.

Under 15 rainfall scenarios spanning three intensities of 100, 200, and 500 millimeters per day and five durations from 2 to 20 hours, the TRIGRS simulations identified the eastern margin of the Tibetan Plateau, including central and western Sichuan, northern and central Yunnan, and eastern Tibet, together with the Loess Plateau in southern and central Shaanxi, western Shanxi, and western Gansu, as the regions with the lowest factors of safety. Central Taiwan and scattered zones in the Altai and Greater Khingan Mountains also showed low values. As rainfall intensified and lengthened, areas with factors of safety below 1.5 expanded, with the Loess Plateau showing particularly strong sensitivity.

Converting these stability results into failure probabilities revealed that comparatively high rainfall-induced probabilities concentrate along the northeastern margin of the Tibetan Plateau, the southern Loess Plateau, and central Taiwan under low-intensity, short-duration rainfall. With stronger and longer rainfall, the most pronounced increases in high-probability area occurred in northern Shaanxi, western Shanxi, western Henan, northern Gansu, and eastern Liaoning. Using a failure-probability threshold of 0.5, the predicted failure area grew logarithmically with rainfall duration, from roughly 3,700 square kilometers under 100 millimeters per day for 2 hours, about 0.04 percent of China’s land area, to roughly 21,600 square kilometers, about 0.22 percent, under 500 millimeters per day for 20 hours.

For seismic triggering, the team applied the Newmark model under ten spatially uniform peak ground acceleration levels from 0.1 to 1.0 g. The areas of comparatively high failure probability expanded significantly with increasing PGA, concentrating in western Sichuan, northern Shaanxi, and eastern Tibet, where tectonic activity is intense, and extending to southern Shaanxi, western Yunnan, northern Chongqing, and western Hubei at stronger shaking levels. The Loess Plateau and the Longmenshan region of Sichuan showed the most prominent growth in high-probability zones. Under a PGA of 1.0 g, the predicted failure area, defined by a 5-centimeter Newmark displacement threshold, reached approximately 33,000 square kilometers, roughly 0.3 percent of the national territory.

The study’s most striking result concerns compound triggering. The researchers introduced a sequential scenario in which an extreme rainfall event of 500 millimeters per day lasting 20 hours weakens slopes before earthquake shaking arrives, with the rainfall-conditioned factor of safety serving as the initial state for the Newmark calculation. Under this scenario, the predicted failure area climbed from about 33,900 square kilometers at 0.1 g to approximately 419,000 square kilometers at 1.0 g. At 1.0 g, the earthquake-only and extreme-rainfall-only scenarios yielded roughly 33,222 and 21,600 square kilometers respectively, a combined total of about 54,800 square kilometers, equivalent to only around 13 percent of the sequential-scenario result. The authors attribute this to pronounced within-model scenario amplification, as antecedent rainfall markedly enhances slope sensitivity to seismic loading.

Comparing the two triggers, the team found that rainfall and earthquakes can produce comparable modeled failure areas at similar triggering intensities: a PGA of 0.1 g yielded an unstable area of roughly 3,900 square kilometers, nearly matching the 3,700 square kilometers produced by 100 millimeters per day over 2 hours. Fitting an exponential relationship, the rainfall required for equivalent triggering capacity increases with PGA according to the equation Rainfall = 5.36 × e^(3.93 × PGA). The authors stress this is a descriptive within-model comparison conditional on the selected thresholds, not evidence of physical equivalence between the two triggers. They also note that rainfall intensity matters beyond total accumulation: 200 millimeters per day over 2 hours produced a larger unstable area than the same total rainfall delivered as 100 millimeters per day over 4 hours.

The framework carries substantial limitations that the authors discuss candidly. The infinite-slope formulation underlying both models restricts the analysis to shallow soil landslides, excluding deep-seated failures, rock avalanches, and discontinuity-controlled rockslides. The full-infiltration assumption in TRIGRS may overestimate pore-pressure rise where rainfall flux exceeds saturated hydraulic conductivity. The PGA-based Newmark formulation ignores Arias intensity, shaking duration, frequency content, near-fault pulses, and topographic amplification, and the empirical probability curve, capped at an asymptote of about 0.1005, was calibrated on the Wenchuan earthquake and may not transfer to other geological settings. Surface runoff was also not explicitly modeled.

Despite these constraints, the modeled spatial patterns align broadly with previous national-scale assessments, including empirical threshold models, neural-network susceptibility maps, and probabilistic seismic landslide hazard maps, while adding physically interpretable, scenario-responsive information that statistical approaches cannot provide. The authors position the framework as a screening reference for identifying landslide-sensitive areas under prescribed rainfall and ground-motion scenarios and as a contribution to hazard-informed planning, with future work aimed at refining regional geotechnical parameters and developing dynamic rainfall-earthquake coupling simulations for compound disaster assessment.

Subject of Research: Scenario-based physically informed assessment of rainfall- and earthquake-induced landslide failure probability across China

Article Title: Scenario-based physically informed assessment of rainfall- and earthquake-induced landslide failure probability across China

Article References: Ma, S., Shao, X., Xu, C., & Yuan, R. (2026). Scenario-based physically informed assessment of rainfall- and earthquake-induced landslide failure probability across China. Geoenvironmental Disasters, 13(1), Article 53. https://doi.org/10.1186/s40677-026-00408-1

Image Credits: AI Generated

DOI: 10.1186/s40677-026-00408-1

Keywords: landslides, China, TRIGRS, Newmark model, slope stability, failure probability, rainfall scenarios, peak ground acceleration, compound hazards, Loess Plateau, Tibetan Plateau, hazard assessment