Search

Past exploration missions have revealed that the lunar topography is eroded through mass wasting processes such as rockfalls, similar to Earth. Despite the abundance of remote sensing and in–situ data many questions about the occurrence and characteristics of lunar rockfalls remain unanswered: Why do lunar rockfalls occur? What endo- and exogenic processes drive lunar rockfalls? What can these processes tell us about the geologic activity and evolution of the Moon? What can rockfall tracks teach us about the shallow lunar regolith and its interaction with the space environment? To help address these questions, we develop deep learning-driven algorithms that allow us to automate the detection and mapping of rockfalls in Lunar Reconnaissance Orbiter’s Petabyte-scale image archive - a uniquely rich resource that has barely been exploited by previous work. Our convolutional neural networks achieve human-level detection performance while reducing the required processing time by several orders of magnitude. By applying transfer learning principles we further boost the performance of our detectors and make them robust enough to map rockfalls on other bodies across the Solar System as well, such as Mars, Ceres, and comet 67P. Using such multi-domain trained algorithms we conduct the first-ever, global scan of the lunar surface and derive the global distribution of rockfalls on the Moon, identifying more than 130,000 individual rockfalls. The global distribution of rockfalls is highly heterogeneous, where the vast majority as well as the highest spatial concentrations of rockfalls occur in young impact craters and on equatorfacing slopes with large thermal gradients. This indicates that rockfall occurrence is mainly controlled by erratic impacts and continuous solar-driven thermal fatigue. We discover that rockfalls occur even in the oldest, pre-Nectarian lunar terranes, indicating that the identified processes - particularly small-scale meteoritic impacts - drive erosion over billions of years. We do not observe significant clusters of rockfalls in the proximity of visible volcanic and tectonic features which suggests that seismic activity has not been a major, global driver of lunar rockfall in the Moon’s recent geologic past. Using satellite images of rockfall boulders and their tracks, we further develop a method to estimate the regolith surface strength and geomechanical properties of yet unexplored regions, such as the polar permanently shadowed regions and large pyroclastic deposits, directly supporting humanity’s current efforts to continue the exploration of the Moon and beyond. This dissertation demonstrates the enormous potential of deep learning-driven applications for planetary science and reveals new insights about the weathering, erosion and evolution of airless planetary bodies in our solar system.

Slope instabilities adversely affect mountainous regions by posing direct and indirect hazards. Landslides can evolve in catastrophic failures that can directly cause loss of lives, damage to properties and to critical infrastructure. Moreover, indirect effects can cause destruction of arable land and, in case of landslide damming of rivers, subsequent floods that can have an impact a long way downstream of the landslide site. A large-scale overview of unstable slopes in mountainous regions is therefore important in order to understand the spatial distribution, to investigate the activity, to create a basis for landslide hazard assessment and to further the science regarding the predisposing and controlling factors. Bhutan was chosen for this study given its relatively poor regional knowledge of unstable slopes, despite its proneness. The main purpose of our work is to understand the spatial distribution of past large rock slope movements in the region, to quantify the recent rates of displacements and to investigate the structural and geological controls. Given the inaccessibility of the region and the large scale of the study, we used remote sensing techniques through all stages of our investigations. In particular, a combination of optical images and high-resolution digital surface model analysis and synthetic aperture radar differential interferometric (DInSAR) techniques was used. The use of DInSAR allowed also to image, quantify and discern slope displacements that are related to soil slide/creep or rock glaciers creep, moreover, it allowed to illuminate areas with reversible displacements related either to freeze-thaw cycles in the permafrost region, or to groundwater table seasonal fluctuations at lower elevations. We produced some new datasets for the region including: 1) a regional rock slope instabilities inventory based on optical images, 2) a rock glaciers inventory based on optical images, 3) an active slope instabilities inventory based on the analysis of individual interferograms, 4) an inventory of active slope instabilities based on multi-temporal DInSAR analyses which includes gravitational and reversible deformation, 5) a regional structural inventory largely based on remote sensing mapping with the inclusion of some in-situ measurements. We present the analysis of individual interferograms aimed at the identification of actively unstable slopes. We perform a geomorphological classification based on the landslide types and we propose a new method to assess the activity likelihood uniquely based on remote sensing techniques. This methodology is based on an articulated decision tree, which takes into account the number of sensors used for the analysis, the number of interferograms in which displacements are identified, the satellite orbit geometry, vegetation cover, geometrical distortions, estimated velocity and temporal coverage of the data. This method allows lowering or maintaining the confidence regarding the activity of each identified slope instability by analysing what would be the detection expectation. It maximises the III level of information retrievable from standard interferometric analyses and is new and applicable to other mountainous regions. We use multi-temporal analyses based on SAR data to show the potential of the technique in identifying specific instabilities of which no previous knowledge existed. Here we show a methodology to combine a model-based atmospheric correction and an empirical correction to remove stratified, long wavelength signals from the cumulative displacement maps, without an atmospheric correction on the interferograms used to generate them. This method takes advantage of the local nature of slope processes leading to surface displacements to separate them from the long-wavelengths signals related to artifacts. This method allows to identify small-scale reversible displacements related with freeze-thaw cycles in the permafrost region and, moreover, it provides insights into the possibility of observing the dynamics of natural annual ground water table variations on valley flanks, which, had not yet been explored with multi-temporal satellite based InsAR analyses. We show the analysis performed with the combination of our structural dataset and landslides dataset aimed at investigating predisposing factors to large rock slope instabilities in the region. The methodologies developed here involve the generation of a compound structural dataset that makes use of different sources of information. The latter include remotely mapped fault traces and planar valley flanks, but also some field measurements of foliation and faults. The methodology used to compile this dataset addresses the biases deriving from uneven sampling of the different data types across the area and allows to make the dataset more balanced for systematic kinematic analyses. We apply kinematic analyses across the structural domains identified in the study region in a probabilistic way, in order to investigate the regional structural control on large rock slope instabilities formation. These methodologies are also new, and applicable to other mountainous regions where remote sensing analyses are the main source of information. We identify a structural and lithological control on rock slope instabilities across large parts of the study region. A higher predisposition to failure seems to be associated with the presence of specific sets of regional faults and foliation structures. The average orientation of one of such fault sets corresponds to the orientation of the Lingshi fault, a major known structure in the northwest of the region. Moreover, there seems to be a lithological control both on past rock slide activity and on more recent activity in association with shales, quartzite and limestone belonging to the Tethyan sediments and found in the northwest of the region. The analysis by means of DInSAR of rock slope activity in the recent past (our observation window is between 2006 and 2011) shows that the majority of active rock slope instabilities identified have rates which correspond to the slow to very slow categories, with maximum displacement rates of up to 160 mm/year, with only very few exceptions. IV These findings, combined with the field observations of a large number of past rock slide deposits, covered by thick soils, point to the fact that active unstable rock slopes are less abundant in the region in recent times than in the past. We postulate that this may be related to the lack in recent times of major earthquakes that could have previously caused widespread landslide activity at a given point in time. We have also observed that rock glaciers in northwestern Bhutan creep downslope with velocities that are on the whole lower than what has been observed in the Alps in recent years. This could be due to either a less pronounced recent warming trend than at higher latitudes or to less water being available to penetrate into the frozen unconsolidated material supersaturated with interstitial ice and to deeper subsurface shear horizons.

The valley flanks around the Great Aletsch Glacier (Valais, Switzerland) contain several rock slope instabilities, concentrated around the current extent of the glacier tongue. These alpine rock slopes have been exposed to cycles of glaciation and retreat, leading to changing slope geometries and stress conditions. Processes leading to the formation of rockslides, rock avalanches or progressive slow movements with time since deglaciation, the impact of cyclic ice-loading and the dominant driving mechanisms have been studied for many years but are partly still unexplained. Additionally, the triggering events and causes of failures of landslides in deglaciated or currently deglaciating valleys are poorly understood. Often, the absence of observable triggering mechanisms presents a significant problem for hazard management and science. Better knowledge of subsurface processes, even if only reconstructed from surface observations and deformations, is essential to improve our understanding of paraglacial rock slope processes. During the course of this PhD thesis the extraordinary activation of the deep-seated gravitational slope deformation (DSGSD) of Moosfluh was witnessed and failure mechanisms, spatial and temporal displacement patterns and paraglacial conditioning factors were studied in great detail. During the first stage of this PhD thesis, a very accurate displacement monitoring system was developed for the high alpine environment of Aletsch. This monitoring system consists of two automatic total stations coupled with Global Navigation Satellite System (GNSS) sensors, meteo sensors and a webcam. It is designed to work all year in harsh conditions with extreme temperature changes, heavy rainfall and snowstorms. The reflectors and the stations have to be protected against rockfall and snow avalanches. In the first phase of this PhD the focus was set on the design and implementation of this total station monitoring system and the processing of high accuracy 3D displacement data. We identify key monitoring goals, discuss stability of pillar foundations, the influence of protective housing and compare calculated to observed accuracies. The selected total station monitoring is based on a lightweight but stable foundation and a total station instrument protection without optical refraction of the light beam and mechanic protection of reflector prisms. The selected monitoring sites presented in this study do not only focus on instabilities, like the large and rapid rock slope instability of Moosfluh and the slow‐moving instability of Driest, but also on seasonally reversible rock slope deformations along several profiles in stable rock across the Great Aletsch Glacier valley. During the second stage of this PhD thesis, the paraglacial history and structure of the Moosfluh Landslide from 1850 to 2016 was investigated. The geomorphological evolution of the Moosfluh slope was reconstructed with photogrammetric models generated from historical aerial photographs. This multi-temporal landslide analysis showed that the bulk displacement and internal deformation at Moosfluh is accommodated mainly by toppling composed of shear slip along the steeply dipping Alpine foliation and extensional faulting in the crest area. Numerous uphill-facing scarps, scarps, tension cracks, toe bulging, graben-structures and displacement of moraine deposits, evidence post-Egesen landslide displacements and an acceleration of movements since the LIA and especially since 2007. Together with digital image correlation and total station monitoring, an increase of displacement rates from a few millimeter per year until 1990, to several meter per day in September 2016 was retraced. Balanced longitudinal sections through the Moosfluh Landslide are based on observed ground surface displacements and geometrical and mechanical constraints. A simple limit-equilibrium analysis helped explore changes in rock slope stability caused by glacial ice retreat and changing groundwater levels, showing that the simulated factor of safety drops non-linearly from the LIA maximum (1.12) to the year 2007 (1.02), when the ice thickness at the landslide toe melted down to 100 m. During the third stage of my PhD thesis, the unexpected acceleration of the Moosfluh DSGSD, which started in September 2016 and posed hazards to a nearby cable car station and an adjacent dammed lake 4 km downstream of the Moosfluh slope, was studied in detail. With our high accuracy displacement monitoring system, installed in 2013 and 2014, it was possible to record the progressive evolution of the Moosfluh Landslide from toppling to sliding in unprecedented detail. The formerly slowly moving toppling mode DSGSD of about 100 Mio m³ and up to 170 m depth developed into a fast moving landslide through the formation of several shallow (30-50 m) secondary slides. These secondary landslides started their development at the toe of the slope, and evolved upslope with time, as evidenced by time series of surface displacement vectors. Digital image correlation (DIC) of webcam images and helicopter-borne photogrammetry allows for detailed mapping of landslide boundaries and active morphological features. For kinematic analyses of toppling and sliding block a new method called ‘plunge angle analysis’ was developed. This approach allows to quantify the underlying failure mechanisms (toppling, sliding and mixed mode deformations) and assign depths and inclination angles of toppling/sliding planes from measured surface displacement vectors. The analysis of displacement vector time series allows the identification of landslide volumes and rupture plane depths. The results and developed method provide a better understanding not only of the Moosfluh Landslide but of all slopes affected by mixed toppling and sliding failure mechanisms. This study assembled a unique dataset documenting the development of a 'paraglacial' landslide from a slow moving DSGSD into an active and hazardous landslide. Combining multiple methods for reconstruction and current surface deformation monitoring strengthened our understanding of mechanisms and long-term controls occurring in currently deglaciating environments. In addition this thesis provides a new basis for implementing and operating a surface deformation monitoring system in high alpine environment, understanding the role of glacial ice load on rock slope instabilities and assessing present day hazards of mass movements in which toppling and sliding mechanisms coexist.

Cycles of glaciation drive in-situ stress changes in underlying bedrock as glaciers advance, erode, and retreat, generating damage in adjacent rock slopes and influencing paraglacial slope stability. Glacial debuttressing is frequently implicated as a trigger for paraglacial rock slope failures, despite commonly observed large lag-times between deglaciation and the timing of failure and often without clear mechanical reasoning. Rock slope damage generated during glacial cycles is hypothesized to have a strong role in preparing rock slope failures, however, the mechanics of paraglacial rock slope damage remain poorly characterized. Glacial cycles mechanically load and unload proximal rock slopes by the changing weight of ice, and in addition produce strongly varying thermal and hydraulic rock-surface boundary conditions tied to the fluctuating glacier. Bedrock beneath temperate glacier ice maintains near isothermal surface temperatures at ~0 °C. Glacier retreat exposes rock walls to new thermal boundary conditions with strongly varying daily and seasonal cycles, a transition we term paraglacial thermal shock. Temperature changes generate thermal strain, inducing thermo-mechanical stresses capable of generating rock mass damage. In addition, high subglacial water pressures near the ice overburden level prevail at the base of temperate glaciers, and affect groundwater conditions in proximal valley flanks. Groundwater recharge by precipitation and snowmelt raises the water table seasonally, which is superposed on changes in hillslope groundwater tied to varying glacial ice elevations. Changing cleft water pressures control effective stresses and the strength of rock mass discontinuities. Together, these thermo-hydro- mechanical stresses act in concert with glacial loading cycles to generate rock slope damage, preparing slopes for future failure. We study thermo-hydro-mechanical induced stresses and resulting rock slope damage during repeat glacial cycles in the valley of the Great Aletsch Glacier in Switzerland. Following Lateglacial deglaciation, the surrounding valley rock slopes in the Aletsch region experienced several minor glacier cycles during the Holocene. The foliated gneissic rock mass of the Aletsch valley contains several large rock slope instabilities with a concentration around the retreating, present-day glacier tongue. Surface exposure dating of the Driest instability head scarp reveals a Mid-Holocene initialization age (7.4 ± 0.7 ky), matching post-Egesen / pre-Little Ice Age relative ages for the majority of other rock instabilities in the Aletsch Valley. To investigate progressive rock slope damage induced during glacier cycles, we used detailed, conceptual numerical models closely based on our Aletsch Valley study area. Modeled glacier scenarios represent mapped ice fluctuations at Aletsch, while rock mass strength parameters applied in our models are based on local rock mass characterization. Ground surface temperature measurements, monitoring of subglacial water pressures in ice boreholes, regional spring-line mapping, and monitoring of rock slope deformation at Aletsch each contribute to parameterizing and validating our thermal and hydraulic model boundary conditions. Our simulations reveal that purely mechanical loading and unloading of rock slopes by ice during glacial cycles generates relatively limited new damage. This result supports our view that glaciers make a poor buttress for adjacent slopes due to the ductile behavior of ice over long time scales. However, ice fluctuations in our models do increase the criticality of fractures in adjacent slopes (bringing them closer to the failure envelope), which may in turn increase the efficacy of additional fatigue processes. On the other hand, bedrock erosion during glaciation (i.e., rock debuttressing) promotes significant new rock slope damage during first deglaciation. The amount of initial damage, inherited from pre-glacial, ice-free topographic and in-situ stress conditions, strongly controls the susceptibility of the slope to new damage from ice loading. The slope response during glacial cycles is path-dependent and varies in damage kinematics: glacier advance in our models enhances toppling failure while glacial retreat promotes sliding. Changing thermal boundary conditions during glacier retreat and advance in our models affects the temperature regime in the adjacent rock slopes. Thermal strain from long-term temperature changes induces stresses at depths exceeding 100 m, generating significantly more rock slope damage than predicted for purely mechanical loading cycles. Thermal expansion of the rock mass due to warming after glacier retreat causes increased stresses propagating fractures, while cooling during glacier advance results in contraction, reducing joint normal stresses and promoting toppling. First time exposure to seasonal temperature cycles during deglaciation induces a strong but shallow damage front that follows the retreating ice margin. Glacial loading cycles in parallel with thermal stresses (i.e., thermo-mechanical fatigue) are capable of generating significant rock slope damage. We extend our models by accounting for changing groundwater conditions in proximal valley rock slopes tied to high subglacial water pressures. Glacier loading cycles in parallel with long-term mountain water table variations generate substantial fracture propagation. Major damage occurs during initial ice occupation and first glacier retreat, while subsequent readvances result in minor damage. Superposition of annual groundwater cycles (i.e., hydro-mechanical fatigue) strongly increases rock slope damage during glacial loading cycles, destabilizing the toppling-mode valley flank in our models. The kinematics and dimensions of the predicted instability closely resemble observed characteristics of major landslides in the field at Aletsch. Our results extend simplified assumptions of glacial debuttressing, demonstrating in detail how thermo-hydro-mechanical stresses acting in concert with glacier cycles drive progressive rock mass failure preparing future paraglacial slope instabilities.