Che-Ming Yang

The Matai'an landslide is the largest landslide event ever recorded in Taiwan, with an estimated volume of approximately 308 million m ³ . The failure resulted in the formation of a landslide dam, whose subsequent breach triggered a catastrophic flood, resulting in 19 fatalities, 5 missing persons, and 157 injuries. Multitemporal remote sensing analysis revealed obvious geomorphic precursor features prior to the landslide, with tensile cracking initiating and progressively developing following the 2024 Mw7.3 Hualien earthquake. Seismic source inversion analysis further indicated that the Matai'an landslide was composed of two distinct source events with different sliding directions, in which the second event predominantly involved westward motion. The two events occurred within 65 s. First, the slope collapsed, with an absolute maximum force magnitude (Fmax) of 1.995 × 10 ¹² N and a high sliding angle ( θ = 38.3°). Second, low-angle ( θ = 6.8°) sliding forces with an Fmax value of 0.908 × 10 ¹² N were resolved by seismic source inversion. This multisource failure mechanism led to stratified material deposition during landslide dam formation, a factor that critically influences the assessment of potential dam-breach mechanisms. By integrating multitemporal imagery with seismic waveform analysis, this study demonstrates that landslide-related geomorphic precursors can be detected prior to failure, while the landslide can be detected seismologically in near real time. Furthermore, the deployment of riverine seismic stations along downstream channels provides a viable approach for early warning of impending landslide-dam breach flooding.

After the 2024 Mw 7.4 Hualien earthquake in eastern Taiwan, the Mataian River watershed experienced a catastrophic sequence of cascading geohazards. This study reconstructs the long-term evolution and failure kinematics of the 2025 Mataian giant landslide and its subsequent dam-breach events. By integrating multi-temporal LiDAR-derived topography, satellite imagery, microseismic signal analysis, and high-resolution UAV surveys, we offer a comprehensive geomorphic and kinematic reconstruction of this complex event. Satellite images are identified a 1,200 m-long tension crack developing along the crown of a paleo-landslide after the 2024 earthquake. On 21 July 2025, a massive failure occurred with a maximum scarp retreat of 120 m and a failure depth of 380 m. Multi-temporal LiDAR differencing estimates a total landslide volume of ~308 million cubic meters. Microseismic records captured a distinct two-stage runout process: an initial dominant southeastward motion toward the Wang Creek tributary, followed by a secondary southward runout ~80 s later along the Mataian River mainstream. The resulting landslide dam reached a height of ~200 m and a maximum depositional thickness of ~325 m. On 23 September 2025, the dam catastrophically breached, with the impounded lake volume plummeting from 91 to 1.15 million cubic meters and causing 19 fatalities and 5 missing persons downstream. Post-breach UAV observations of the residual dam exposed a stratified internal structure of fractured greenschist, quartz-mica schist, and marble, overlain by boulder-gravel deposits layer. Notably, subsequent failures on 21 October and 13 November were concentrated on the right bank. Due to the run-up process during the major event, where the colluvial front collided with the opposing slope, forming a steep and mechanically weak interface. A comprehensive dynamic model of the landslide-to-breach sequence is established. Our findings provide critical insights into the post-failure stability of residual dams and important information for subsequent numerical modeling, physical breach experiments, and the hazard mitigation strategies in similar region.

On April 3, 2024, a strong earthquake with amagnitude ML 7.2 struck Shoufeng Township in Hualien County, Taiwan, causing 18 fatalities due to coseismic landslides. Coseismic landslides have long been a key topic in global geohazard research. To understand their characteristics and causes, it is essential to establish an objective and comprehensive inventory of landslides. Moreover, comparing coseismic landslide inventories from different earthquakes could understand how topographic, seismic, and geological characteristicsinfluence the spatial distribution of these landslides. The 2024 Hualien earthquake induced 3,232 coseismiclandslides with total area of 41.74 square kilometers.These landslides mostly occurred in regions where the peak ground acceleration exceeded 250 gal and the peak ground velocity surpassed 30 cm/s. They were primarily concentrated in areas composed of marble and schist, and predominantly faced east or southeast, with slope angles mainly ranging from 40°-60°.This study further compares the coseismic landslide inventories from the 2024 Hualien earthquake (ML7.2), the 2022 Taitung earthquake (ML6.8), and the 1999 Chi-Chi earthquake (ML 7.6). Except for the 2024 Hualien earthquake, the numbers of coseismic landslides in the other events are 306, and 9,272, respectively, with total landslide areas of 0.93, and 127.8 square kilometers. The slope aspects of the other two earthquakes are southeast and south mainly. The slope gradients range from 40°-60° and 40°-50° for the 2022 Taitung earthquake and the 1999 Chi-Chi earthquake. In terms of lithology, the coseismic landslides associated with the 2022 Taitung earthquake and the 1999 Chi-Chi earthquake occurred in areas characterized by schist and epiclastics, andsandstone and conglomerate, respectively. Therefore, these three coseismic landslide inventories show the influence of different geologic background and topographic relief. This study presented the topographic, geologic, seismic diffrences of coseimic landslide inventories with the regions between metamorphic rock and sedimentary rock in Taiwan. Keywords: 2024 Hualien Earthquake, 2022 Taitung Earthquake, 1999 Chi-Chi Earthquake, Coseismiclandslide inventory.

Seismic signals, with their remote and continuous monitoring advantages, have been instrumental in unveiling various landslide characteristics and have been widely applied in the past decades. However, a few studies have extended these results to provide geologists with pre-survey information, thus enhancing the understanding of the landslide process. In this research, we utilize the deep-seated Cilan landslide (CL) as a case study and employ a series of seismic analyses, including spectrogram analysis, single-force inversion, and geohazard location. These techniques enable us to determine the physical processes, sliding direction, mass amount estimation, and location of the deep-seated landslide. Through efficient discrete Fourier transforms for spectrograms, we identified three distinct events, with the first being the most substantial. Further analysis of spectrograms using a semi-log frequency axis generated by discrete Stockwell transform revealed that Event 1 consisted of four sliding failures occurring within 30 s with decreasing sliding mass. Subsequent Events 2 and 3 were minor toppling and rockfalls, respectively. Geohazard location further constrained the source location, indicating that Events 1 and 2 likely originated from the same slope. Subsequently, the sliding direction retrieved from single-force inversion and the volume estimation were determined to be 153.67° and 557 118 m³, respectively, for the CL. Geological survey data with drone analysis corroborated the above seismological findings, with the sliding direction and source volume estimated to be around 148° and 664 926 m³, respectively, closely aligning with the seismic results. Furthermore, the detailed dynamic process observed in the spectrogram of Event 1 suggested a possible failure mechanism of CL involving advancing, retrogressing, enlarging, or widening. By combining the above mechanism with geomorphological features identified during field surveys, such as the imbrication-like feature in the deposits and the gravitational slope deformation, with video from the event, we can infer the failure mechanism of retrogression of Event 1 after shear-off from the toe. Then, the widening activity was caused by the failure process for subsequent events, like Events 2 and 3. This case study underscores the significance of remote and adjacent seismic stations in offering seismological-based landslide characteristics and a time vision of the physical processes of landslides, thereby assisting geologists in landslide observation and deciphering landslide evolution.

Several studies have suggested that directions of earthquake-triggered landslides might be preferentially oriented according to the seismic waves’ characteristics. Here we further address this issue by analyzing three landslide populations attributed to 1998 Ruei-Li Mw 6.5, 1999 Chi-Chi Mw 7.3 and 2022 Taitung Mw 6.9 earthquakes in Taiwan. In particular, we seek possible linkages between the patterns of co-seismic landsliding (predominant orientations) and the epicenter and fault rupture locations, by exploiting the assumption that surface waves with horizontal particle motions (Love waves) and horizontal shear waves, both characterized by transverse vibrations perpendicular to the direction of wave radiation from the source, are the major agents responsible for earthquake induced slope failures.First, we take the aspect of each landslide source zone as representing the landslide directions. These directions are then statistically evaluated with respect to the epicentre and fault rupture positions for the characteristic segments of the landslide population. In the next step, we consider numerous possible pairs of the landslides to obtain intersections of the lines normal to their aspect directions using custom-designed Python code.At each particular landslide population segment, the landslide displacement directions revealed slight preferential orientation with the maxima sub-perpendicular to the fault rupture. Symmetrically distributed and round landslide population of the Ruei-Li earthquake showed even better results than elongated landslide population of the Chi-Chi earthquake. In all three earthquake cases, the intersections maxima coincided with the maximum slip velocities and/or displacements along the fault ruptures, as revealed by GNSS. These promising results indicate that such an approach might be useful for identifying fault ruptures of old or even prehistoric earthquakes. The research was funded by the Grant Agency of the Czech Republic (GC22-24206J) and Taiwanese National Technological and Science Council (MOST/NTSC 111-2923-M-008-006-MY3).

Seismic signals, with their remote and continuous monitoring advantages, have been instrumental in unveiling various landslide characteristics and have been widely applied in the past decades. However, a few studies have extended these results to provide geologists with pre-survey information, thus enhancing the understanding of the landslide process. In this research, we utilize the deep-seated Cilan Landslide (CL) as a case study and employ a series of seismic analyses, including spectrogram analysis, single force inversion, and geohazard location. These techniques enable us to determine the physical processes, sliding direction, mass amount estimation, and location of the deep-seated landslide. Through efficient discrete Fourier transform for spectrograms, we identified three distinct events, with the first being the most substantial. Further analysis of spectrograms using a semi-log frequency axis generated by discrete Stockwell transform revealed that Event 1 consisted of four sliding failures occurring within thirty seconds with decreasing sliding mass. Subsequent Events 2 and 3 were minor toppling and rockfalls, respectively. Geohazard location further constrained the source location, indicating that Events 1 and 2 likely originated from the same slope. Subsequently, the sliding direction retrieved from single force inversion and volume estimation was determined to be 153.67º and 557,118 m³, respectively, for the CL. Geological survey data with drone analysis corroborated the above seismological findings, with the sliding direction and source volume estimated to be around 148° and 664,926 m³, respectively, closely aligning with the seismic results. Furthermore, the detailed dynamic process observed in the spectrogram of Event 1 suggested a possible failure mechanism of CL involving advancing, retrogressing, enlarging, or widening. Combining the above mechanism with geomorphological features identified during field surveys, such as the imbrication-like feature in the deposits and the gravitational slope deformation, with event video, infers the failure mechanism of retrogression of the Event 1 after shear-off from the toe. Then, the widening activity was caused by the failure process for subsequent events, as Events 2 and 3. This case study underscores the significance of remote and adjacent seismic stations in offering seismological-based landslide characteristics and a time vision of the physical processes of landslides, thereby assisting geologists in landslide observation and deciphering landslide evolution.

Seismic signals, with their remote and continuous monitoring advantages, have been instrumental in unveiling various landslide characteristics and have been widely applied in the past decades. However, a few studies have extended these results to provide geologists with pre-survey information, thus enhancing the understanding of the landslide process. In this research, we utilize the deep-seated Cilan Landslide (CL) as a case study and employ a series of seismic analyses, including spectrogram analysis, single force inversion, and geohazard location. These techniques enable us to determine the physical processes, sliding direction, mass amount estimation, and location of the deep-seated landslide. Through efficient discrete Fourier transform for spectrograms, we identified three distinct events, with the first being the most substantial. Further analysis of spectrograms using a semi-log frequency axis generated by discrete Stockwell transform revealed that Event 1 consisted of four sliding failures occurring within thirty seconds with decreasing sliding mass. Subsequent Events 2 and 3 were minor toppling and rockfalls, respectively. Geohazard location further constrained the source location, indicating that Events 1 and 2 likely originated from the same slope. Subsequently, the sliding direction retrieved from single force inversion and volume estimation was determined to be 153.67º and 557,118 m³, respectively, for the CL. Geological survey data with drone analysis corroborated the above seismological findings, with the sliding direction and source volume estimated to be around 148° and 664,926 m³, respectively, closely aligning with the seismic results. Furthermore, the detailed dynamic process observed in the spectrogram of Event 1 suggested a possible failure mechanism of CL involving advancing, retrogressing, enlarging, or widening. Combining the above mechanism with geomorphological features identified during field surveys, such as the imbrication-like feature in the deposits and the gravitational slope deformation, with event video, infers the failure mechanism of retrogression of the Event 1 after shear-off from the toe. Then, the widening activity was caused by the failure process for subsequent events, as Events 2 and 3. This case study underscores the significance of remote and adjacent seismic stations in offering seismological-based landslide characteristics and a time vision of the physical processes of landslides, thereby assisting geologists in landslide observation and deciphering landslide evolution.

Seismic signals, with their remote and continuous monitoring advantages, have been instrumental in unveiling various landslide characteristics and have been widely applied in the past decades. However, a few studies have extended these results to provide geologists with pre-survey information, thus enhancing the understanding of the landslide process. In this research, we utilize the deep-seated Cilan Landslide (CL) as a case study and employ a series of seismic analyses, including spectrogram analysis, single force inversion, and geohazard location. These techniques enable us to determine the physical processes, sliding direction, mass amount estimation, and location of the deep-seated landslide. Through efficient discrete Fourier transform for spectrograms, we identified three distinct events, with the first being the most substantial. Further analysis of spectrograms using a semi-log frequency axis generated by discrete Stockwell transform revealed that Event 1 consisted of four sliding failures occurring within thirty seconds with decreasing sliding mass. Subsequent Events 2 and 3 were minor toppling and rockfalls, respectively. Geohazard location further constrained the source location, indicating that Events 1 and 2 likely originated from the same slope. Subsequently, the sliding direction retrieved from single force inversion and volume estimation was determined to be 153.67º and 557,118 m³, respectively, for the CL. Geological survey data with drone analysis corroborated the above seismological findings, with the sliding direction and source volume estimated to be around 148° and 664,926 m³, respectively, closely aligning with the seismic results. Furthermore, the detailed dynamic process observed in the spectrogram of Event 1 suggested a possible failure mechanism of CL involving advancing, retrogressing, enlarging, or widening. Combining the above mechanism with geomorphological features identified during field surveys, such as the imbrication-like feature in the deposits and the gravitational slope deformation, with event video, infers the failure mechanism of retrogression of the Event 1 after shear-off from the toe. Then, the widening activity was caused by the failure process for subsequent events, as Events 2 and 3. This case study underscores the significance of remote and adjacent seismic stations in offering seismological-based landslide characteristics and a time vision of the physical processes of landslides, thereby assisting geologists in landslide observation and deciphering landslide evolution.

Seismic signals, with their remote and continuous monitoring advantages, have been instrumental in unveiling various landslide characteristics and have been widely applied in the past decades. However, a few studies have extended these results to provide geologists with pre-survey information, thus enhancing the understanding of the landslide process. In this research, we utilize the deep-seated Cilan Landslide (CL) as a case study and employ a series of seismic analyses, including spectrogram analysis, single force inversion, and geohazard location. These techniques enable us to determine the physical processes, sliding direction, mass amount estimation, and location of the deep-seated landslide. Through efficient discrete Fourier transform for spectrograms, we identified three distinct events, with the first being the most substantial. Further analysis of spectrograms using a semi-log frequency axis generated by discrete Stockwell transform revealed that Event 1 consisted of four sliding failures occurring within thirty seconds with decreasing sliding mass. Subsequent Events 2 and 3 were minor toppling and rockfalls, respectively. Geohazard location further constrained the source location, indicating that Events 1 and 2 likely originated from the same slope. Subsequently, the sliding direction retrieved from single force inversion and volume estimation was determined to be 153.67º and 557,118 m³, respectively, for the CL. Geological survey data with drone analysis corroborated the above seismological findings, with the sliding direction and source volume estimated to be around 148° and 664,926 m³, respectively, closely aligning with the seismic results. Furthermore, the detailed dynamic process observed in the spectrogram of Event 1 suggested a possible failure mechanism of CL involving advancing, retrogressing, enlarging, or widening. Combining the above mechanism with geomorphological features identified during field surveys, such as the imbrication-like feature in the deposits and the gravitational slope deformation, with event video, infers the failure mechanism of retrogression of the Event 1 after shear-off from the toe. Then, the widening activity was caused by the failure process for subsequent events, as Events 2 and 3. This case study underscores the significance of remote and adjacent seismic stations in offering seismological-based landslide characteristics and a time vision of the physical processes of landslides, thereby assisting geologists in landslide observation and deciphering landslide evolution.