Chi-Wen Chen

Climate-normalized rainfall rarity may facilitate comparisons across climates, but pooled rainfall–landslide associations can obscure regional structure and rainfall timing. We examined 994 reported rainfall-induced landslides in Taiwan (1987–2018) using 2-km hourly precipitation and generalized extreme-value climatologies for 1980–2019. Rainfall rarity ( T _max ) was defined as the maximum of 1–72-h duration-specific return periods, not the recurrence interval of an entire storm. Retrospective whole-storm T _max was available for 912 records, including 708 with estimated volume. Within 18 selected spatial cells, 16 contained 393 volume-bearing records. Contextual decomposition yielded a positive between-cell rarity–volume coefficient ( β _between = 0.945, 95% confidence interval (CI) = 0.642–1.248), but no detectable positive within-cell association ( β _within = − 0.105, 95% CI = − 0.288–0.078). Cell-clustered inference was less decisive. The between-cell coefficient decreased to 0.279 after adjusting for absolute 24-h rainfall and to 0.317 after excluding Typhoon Morakot. When rarity was calculated using only rainfall preceding recorded failure, the positive between-cell association was not recovered ( β _between = − 0.276, 95% CI = − 1.115–0.563; 361 volume-bearing records), although analytical populations differed. These findings identify a Morakot-anchored, inventory-specific association that is sensitive to absolute rainfall and exposure timing.

Topographic amplification is widely invoked to explain strong shaking and slope failure in mountainous terrain, yet direct field observations remain limited. Here we analyze 24 regional earthquakes recorded by a four-station strong-motion array spanning a ridge crest, a valley bottom, and two contrasting slope facets on a single ridge in central Japan. Peak ground acceleration (PGA) and frequency-dependent measures reveal markedly different representations of local seismic response. PGA ratios vary among events and are positively associated with epicentral distance; however, earthquake magnitude also increases with distance, so source-size and propagation-distance effects cannot be separated in this catalog. In contrast, the NS slope exhibits a persistent narrow-band station-to-valley Fourier spectral-ratio peak at 3.5–4.0 Hz that remains stable across events and processing choices. Absolute spectra confirm enhanced energy at the slope in this band, while also showing that a relative spectral depression at the valley reference contributes to the large ratio magnitude. Most importantly, the geometric-mean 5%-damped response-spectral ratio reaches 6.67 at 0.277 s (3.61 Hz), compared with a corresponding full-record geometric-mean horizontal PGA ratio of 1.13. These observations show that PGA can substantially under-represent frequency-specific elastic response on mountain slopes, while the responsible mechanism cannot be uniquely attributed to topography or near-surface structure alone.

Background: Landslides pose significant risks to lives and infrastructure. This study developed landslide susceptibility mapping (LSM) for Yilan County, Taiwan, a region highly affected by seismic activity and intense rainfall. A comprehensive inventory of 12,047 landslides (2004–2017) and 12 causative factors spanning topography, hydrology, geology, and seismicity were compiled. Three multivariate statistical methods, Analytical Hierarchy Process (AHP), Frequency Ratio (FR), and Simple Additive Weighting (SAW), were applied and validated using Area Under the Curve (AUC) and a suite of error metrics. Results: The dataset was partitioned into 70% training and 30% testing samples. Multicollinearity analysis confirmed that all causative factors (Pearson correlation < 0.7, variance inflation factor < 10) were suitable for inclusion. Chi-square ranking identified slope gradient and annual rainfall as the two most influential factors. The Jenks natural breaks method classified susceptibility into four levels: low, moderate, high, and very high. Validation results showed that AHP achieved the highest AUC (training: 0.830, testing: 0.839), followed by FR (training: 0.828, testing: 0.828) and SAW (training: 0.824, testing: 0.814). AHP also yielded the lowest error values (mean squared error = 1.7 × 10 ⁻³ , root mean square = 2.7 × 10 ⁻² ) and the highest correlation coefficient ( r = 0.9974). Although differences among methods were relatively small (maximum AUC difference of 0.025), AHP consistently outperformed the others across all metrics. High and very high susceptibility zones, comprising approximately 37.5–49.5% of the study area depending on the method, were concentrated in southern Yilan, particularly in Tiansongpi Township, where steep slopes exceeding 30° coincide with annual precipitation of 3100–4700 mm. Conclusions: This framework provides a practical and reproducible approach for landslide hazard assessment and risk management in similar geological settings worldwide.

Landslides are an example of severe natural disasters that occur worldwide and generate many harmful effects that can affect the stability and development of society. A better-quality susceptibility mapping technique for the landslide risk is crucial for mitigating landslides. However, the use of assemblages of multivariate statistical methods is still uncommon in Indonesia, particularly in the Kepahiang Regency of Bengkulu Province. Therefore, the objective of this study was to provide an improved framework for creating landslide susceptibility map (LSM) using multivariate statistical methods, i.e., the analytical hierarchy process (AHP) method, the simple additive weighting (SAW) method and the frequency ratio (FR) method. In this study, we established a landslide inventory considering 15 causative factors using the area under the curve (AUC) validation method and another evaluation technique. The performance of each causative factor was evaluated using multicollinearity and Pearson correlation analysis with regression-based ranking. The LSM results showed that the most susceptible areas were located in the districts of Kabawetan, Kepahiang, and Tebat Karai. The high landslide risk in these areas could be attributed to the slope conditions in mountainous regions, which are characterized by high annual rainfall and seismic activity. The AUC training values of the AHP, SAW, and FR methods were 0.866, 0.838, and 0.812, respectively. Then, on the validation dataset, the AHP method yielded the highest AUC value (0.863), followed by the SAW (0.833) and FR (0.807) methods. Moreover, the AHP method provided a higher accuracy value, which suggests that the AHP method is more suitable than the other methods. Therefore, our research indicated that all algorithm methods generate a positive impact and greatly improve landslide susceptibility evaluation, especially for the preparation of landslide damage assessments in this study area. Finally, the method proposed in this study could improve the feasibility of LSM and provide support for Indonesian government decision-makers in arranging hazard mitigation measures in the Kepahiang Regency, Indonesia.

Affected by climate change owing to global warming, the frequency of extreme rainfall events has gradually increased in recent years. Many studies have analyzed the impacts of climate change in various fields. However, uncertainty about the scenarios they used is still an important issue. This study used two and four multi-scenarios at the base period (1979–2003) and the end of the 21st century (2075–2099) to collect the top-ranking typhoons and analyze the rainfall conditions of these typhoons in two catchments in northern Taiwan. The landslide-area characteristics caused by these typhoons were estimated using empirical relationships, with rainfall conditions established by a previous study. In addition to counting landslide-area characteristics caused by the typhoons of each single scenario, we also used the ensemble method to combine all scenarios to calculate landslide-area characteristic statistics. Comparing the statistical results of each single scenario and the ensembles, we found that the ensemble method minimized the uncertainty and identified the possible most severe case from the simulation. We further separated typhoons into the top 5%, 5%–10%, and 10%–15% to confirm possible changes in landslide-area characteristics under climate change. We noticed that the uncertainty of the base period and the end of the 21st century almost overlapped if only a single scenario was used. In contrast, the ensemble approach successfully distinguished the differences in both the average values of landslide-area characteristics and the 95% confidence intervals. The ensemble results indicated that the landslide magnitude triggered by medium- and high-level typhoons (top 5%–15%) will increase by 24%–29% and 125%–200% under climate change in the Shihmen Reservoir catchment and the Xindian River catchment, respectively, while landslides triggered by extreme-level typhoons (top 5%) will increase by 8% and 77%, respectively. Still, the uncertainty of landslide-area characteristics caused by extreme typhoon events is slightly high, indicating that we need to include more possible scenarios in future work.

Rainfall thresholds for slope failures are essential information for establishing early-warning systems and for disaster risk reduction. Studies on the thresholds for rainfall-induced landslides of different scales have been undertaken in recent decades. This study attempts to establish a warning threshold for large-scale landslides (LSLs), which are defined as landslides with a disturbed area more massive than 0.1 km². The numerous landslides and extensive rainfall records make Taiwan an appropriate area to investigate the rainfall conditions that can result in LSLs. We used landslide information from multiple sources and rainfall data captured by 594 rain gauges to create a database of 83 rainfall events associated with LSLs in Taiwan between 2001 and 2016. The corresponding rainfall duration, cumulative event rainfall, and rainfall intensity for triggering LSLs were determined. This study adopted the tank model to estimate conceptual water depths (S₁, S₂, S₃) in three-layer tanks and calculated the soil water index (SWI) by summing up the water depths in the three tanks. The empirical SWI and duration (SWI–D) threshold for triggering LSLs occurring during 2001–2013 in Taiwan is determined as SWI = 155.20 − 1.56D and D ≥ 24 h. The SWI–D threshold for LSLs is higher than that for small-scale landslides (SSLs), those with a disturbed area smaller than 0.1 km². The LSLs that occurred during 2015–2016 support this finding. It is notable that when the SWI and S₃ reached high values, the potential of LSLs increased significantly. The rainfall conditions for triggering LSLs gradually descend with increases in antecedent SWI. Unlike the rainfall conditions for triggering SSLs, those for triggering LSLs are related to the long duration–high intensity type of rainfall event.