One purpose of landslide research is to establish early warning thresholds for rainfall-induced landslides. Insufficient observations of past events have inhibited the analysis of critical rainfall conditions triggering landslides. This difficulty may be resolved by extracting the timing of landslide occurrences through analysis of seismic signals. In this study, seismic records of the Broadband Array in Taiwan for Seismology were examined to identify ground motion triggered by large landslides that occurred in the years 2005 to 2014. A total of 62 landslide-induced seismic signals were identified. The seismic signals were analyzed to determine the timing of landslide occurrences, and the rainfall conditions at those times – including rainfall intensity (I), duration (D), and effective rainfall (Rt) – were assessed. Three common rainfall threshold models (I–D, I–Rt, and Rt–D) were compared, and the crucial factors of a forecast warning model were found to be duration and effective rainfall. In addition, rainfall information related to the 62 landslides was analyzed to establish a critical height of water model, (I − 1.5) ⋅ D = 430.2. The critical height of water model was applied to data from Typhoon Soudelor of 2015, and the model issued a large landslide warning for southern Taiwan.

Rainfall-triggered landslides are one of the most deadly natural hazards in many regions. Seismic recordings have been used to examine source mechanisms and to develop monitoring systems of landslides. We present a semiautomatic algorithm for detecting and locating landslide events using both broadband and short-period recordings and have successfully applied our system to landslides in Taiwan. Compared to local earthquake recordings, the recordings of landslides usually show longer durations and lack distinctivePandSwave arrivals; therefore, the back projection method is adopted for the landslide detection and location. To identify the potential landslide events, the seismic recordings are band-passed from 1 to 3 Hz and the spectrum pattern in the time-frequency domain is used to distinguish landslides from other types of seismic sources based upon carefully selected empirical criteria. Satellite images before and after the detected and located landslide events are used for final confirmation. Our landslide detection and spatial-temporal location system could potentially benefit the establishment of rainfall-triggered landslide forecast models and provide more reliable constraints for physics-based landslide modeling. The accumulating seismic recordings of landslide events could be used as a training dataset for machine learning techniques, which will allow us to fully automate our system in the near future.

Landslide-generated seismic waves (landslide-quakes), exhibiting distinctive waveforms and frequency characteristics, can be recorded by nearby seismometers. Implementing an automatic classifier for landslide-quakes could help provide objective and accurate initiation times of landslides with efficiency. This study collected and analyzed the time information of 214 landslide seismic records due to 33 documented landslide events, from the Broadband Array in Taiwan for Seismology (BATS). In addition, equal numbers of earthquake and noise signals were also incorporated. The 642 seismic signals and time information were carefully examined using the random forest algorithm to create an automatic landslide-quake classifier. By validating the signal attributes of the landslide, earthquake, and noise events, specifically in the time and frequency domains, it was shown that the proposed classifier can reach an accuracy (the proportion of all correctly classified events to the total number of events) of 91.3%. To further evaluate the applicability of the automatic classifier, landslide-quakes generated during the devastating Typhoon Morakot (2009) and Typhoon Soudelor (2015) were also verified, showing that the sensitivity of the classifier is higher than 98%.

Fluctuations in the sediment yielding rate within mountainous areas are found after extreme precipitations. These changes are associated with rainfall-driven landslides and can be evaluated through observations of sediment transported in river water. In this study, we assess the differential impact level and duration of an extreme rainfall event caused by the 2009 Typhoon Morakot in Taiwan. Using a time series of suspended sediment data and landslide inventories within three affected catchments, our results found that the proportion of the typhoon-generated sediment discharges reached 58% of the 2009 annual sediment discharges, and the annual sediment discharges could be 15-fold higher than the average annual sediment discharges prior to the typhoon. The impact of the typhoon on suspended sediment discharge lasted for around four months. It is further revealed that a significant increase in the number of landslides was attributed to the impact of the extreme typhoon event, and the critical rainfall condition triggering landslides, based on the relationship between rainfall and landslide number and the rainfall intensity-duration thresholds, declined for 4–5 years.

Seismological data plays a crucial role in timely slope failure hazard assessments. However, identifying rockfall waveforms from seismic data poses challenges due to their high variability across different events and stations. To address this, we propose RockNet, a deep-learning-based multitask model capable of detecting both rockfall and earthquake events at both the single-station and local seismic network levels. RockNet consists of two sub-models: the single-station model, which computes waveform masks for earthquake and rockfall signals and performs earthquake P and S phase picking simultaneously on single-station seismograms, and the association model, which determines the occurrences of local seismic events by aggregating hidden feature maps from the trained single-station model across all stations. Since rockfall data is relatively scarce and may not be sufficient to train a deep learning model effectively, we augment the dataset with abundant non-rockfall data and add additional tasks to promote shared interpretability and robustness. RockNet is trained and tested on a local dataset collected from the Luhu tribe in Miaoli, Taiwan, achieving macro F1-scores of 0.983 and 0.990 for the single-station model and the association model, respectively. Furthermore, we evaluate RockNet on an independent dataset collected from the Super-Sauze unstable slope region in France, and it demonstrates good generalization performance in discriminating earthquake, rockfall, and noise with a macro F1-score of 0.927. This study highlights the potential of deep learning in leveraging diverse types of inputs for seismic signal detection even with limited training data.