Natural disasters such as landslides and rockfalls are mostly difficult to study because of the impossibility of making in situ measurements due to their destructive nature and spontaneous occurrence. Seismology is able to record the occurrence of such events from a distance and in real time. In this study, we show that, by using a machine learning approach, the mass and velocity of rockfalls can be estimated from the seismic signal they generate.

We show that Icelandic long-runout landslides with longitudinal ridges represent good analogues of Martian landforms. The large record of long-runout landslides with longitudinal ridges emplaced after the Last Glacial Maximum in Iceland offers a unique opportunity to study the possible relation between the development of these landforms and environmental conditions. This could have implications for reconstructing Martian paleoclimatic and paleoenvironmental conditions.

In the Grands Causses area (Southern France), we study the relationship between the evolution of the river, its incision through time, and the location of the nearby caves. It is commonly accepted that horizontal caves are formed during a period of river stability (no incision) at the elevation of the river. Our original results show that it is wrong in our case study. Therefore, another model of cave formation is proposed that does not rely on direct river control over cave locations.

In this paper, we propose a framework for generating risk maps that provide the probabilities of erosion due to river migration. This framework uses concepts from probability theory to learn the river migration model's parameter values from satellite data while taking into account parameter uncertainty. Our analysis shows that such geomorphic risk estimation is more reliable than models that do not explicitly consider various sources of variability and uncertainty.

Understanding how volcanic edifices develop drainage basins remains unexplored in landscape evolution. Using digital evolution models of volcanoes with varying ages, we quantify the geometries of their edifices and associated drainage basins through time. We find that these metrics correlate with edifice age and are thus useful indicators of a volcano’s history. We then develop a generalized model for how volcano basins develop and compare our results to basin evolution in other settings.

Ice cliffs are melt hot spots that increase melt rates on debris-covered glaciers which otherwise see a reduction in melt rates. In this study, we show how surface runoff streams contribute to the generation, evolution, and survival of ice cliffs by carving into the glacier and transporting rocky debris. On Kennicott Glacier, Alaska, 33 % of ice cliffs are actively influenced by streams, while nearly half are within 10 m of streams.

This study investigates subsurface weathering zones, revealing their structure through shear wave velocity variations. The research focuses on the arid climate of Pan de Azúcar National Park, Chile, using seismic ambient noise recordings to construct pseudo-3D models. The resulting models show the subsurface structure, including granite gradients and mafic dike intrusions. Comparison with other sites emphasizes the intricate relationship between climate, geology, and weathering depth.

Accurately identifying the presence and absence of landforms is important to inferring processes and testing numerical models of landscape evolution. Using synthetic scenarios, I show that the Matthews correlation coefficient (MCC) should be favored over the F₁ score when comparing accuracy across scenes where landform abundances vary. Despite the resilience of MCC to imbalanced data, strong sensitivity to the size and shape of features can still occur when truth and model data are misaligned.

We evaluated the intensity and spatial extent of landscape change in the Hengduan Mountains by identifying areas where river network reorganization is occurring or expected in the future. We combine four metrics that measure topographic imbalances at different spatial and temporal scales. Our study provides a deeper understanding of the dynamic nature of the Hengduan Mountains landscape and associated drivers, such as tectonic uplift, and insights for applying similar methods elsewhere.

In this paper, we use machine learning to automatically outline landforms based on their characteristics. We test several methods to identify the most accurate and then proceed to develop the most accurate to improve its accuracy further. We manage to outline landforms with 65 %–75 % accuracy, at a resolution of 10 m, thanks to high-quality/high-resolution elevation data. We find that it is possible to run this method at a country scale to quickly produce landform inventories for future studies.

The sensitivity to the wave and sea-level forcing sources in predicting a 6-month embayed beach evolution is assessed using two different morphodynamic models. After a successful model calibration using in situ data, other sources are applied. The wave source choice is critical: hindcast data provide wrong results due to an angle bias, whilst the correct dynamics are recovered with the wave conditions from an offshore buoy. The use of different sea-level sources gives no significant differences.