Rainfall-Triggered Shallow Landslides in a Large-Scale Physical Model

Shallow landslides studies are usually extended over landscape scale, where the investigations about geotechnical and hydrological properties of the soil are limited to some local points and not sufficient to assure an in-depth explanation of failure trigger. The physics of the phenomenon is thus minimized, and resolution in space and time is maximized. Such as approach can be useful to predict landslide occurrences for emergency purposes, but it is not effective to interpret the real triggering landslide mechanism. A local scale analysis become needed to achieve an understanding of the processes leading to the failure. Specifically, a full comprehension requires to provide experimental data from a carefully monitored and controlled landslide field site. The present study focuses on a large-scale device aimed at simulating shallow landslides triggered by heavy intensity rainfall. The physical model consists of an artificial hillslope built with a reinforced concrete box: the maximum height is 3.5 m, with length of 6 m and width of 2 m, so that a 2:3 slope can be built. On each lateral side of the box, 50 openings closed with screw caps allow the insertion on properly chosen positions of the control instrumentation (6 tensiometers and 6 Water Content Reflectometer sensors). The monitoring network, connected to an automatic acquisition system, was completed by two piezometers, and two stream gages able to evaluate both the surface runoff and subsurface contributions to the total outflow. The work developed in this study concerns the design and the performance analysis of the main features characterizing the large-scale hillslope model, up to the performance of two landslide experiments on a 60 cm thick sandy soil layer. A rainfall simulator was designed and built to reproduce an intensive precipitation causing the soil collapse. It was realized with a one-loop network equipped with spray nozzles appropriately chosen to minimize the surface splash erosion. In such a way the effects induced by the simulator concern infiltration dynamics without generating top erosion, which could introduce further factors of more difficult understanding. The nozzle configurations on the network were chosen to reproduce i) the desired range of the rainfall intensity, varying from 50 to 150 mm/h, and ii) the spatial uniformity of the produced rain. A careful analysis of the rain sprayed by a single nozzle was developed on a prototype, in order to recognize the main variables affecting the nozzle functioning and performance. Further investigations were then carried out to test the performance of the final full-scale version of the rainfall simulator, highlighting its flexibility for the regulation and the control of the generated rain intensity. Depending on the desired rainfall range, four different configurations of nozzles, distinguished by the number of active nozzles and their location, were chosen to cover the required intensity interval. A careful analysis about the drop diameters was conducted by recurring to an oil mixture poured in Petri dishes that were exposed to the rain. The drop size distribution thus collected characterizes the induced rainfall and was used for a numerical simulation aimed at estimating the impact energy of the drops falling on the soil. The proposed model calculates the trajectories of the particles injected by the nozzle using a constitutive law of sphere aerodynamics in a 3D space. As a result, the rainfall potential erosion and its spatial distribution were assessed, highlighting the limited surface erosion generated by the proposed rainfall simulator. In a second step, a suitable device was realized to calibrate the WCR (Water Content Reflectometer) sensors. It consists of a 0.6 x 0.5 x 0.6 cubic meters Plexiglas box containing the soil with the top exposed to rainfall and the bottom sustained by a perforated base. The calibration of the WCR sensors pointed to obtain an effective law for an accurate assessment of the water infiltration evolution in the soil during the landslide experiments. Several tests were performed with varying porosity values of the sand sample placed into the Plexiglas box, where three tensiometers and as many as WCR probes were arranged. The final results suggest a calibration relationship linearly depending on the WCR output signal and porosity. Two experiments on the artificial slope were then performed by applying two different porosities of the soil during the placement. The chosen soil consists of a fine sand with high particle size uniformity. The first porosity was obtained by dumping the sand without applying compacting action, such that the sand was in loose conditions. In a second case, the sand was compacted to yield a dense sand. The two experiments were carried out by applying the rainfall at until the sand collapse. The observation of the experiments and the analysis of the recorded data allow to examine the hydrological dynamics leading to the landslide and the triggering factors. With loose sand, the failure occurred suddenly without warning signs; at the failure, the soil appeared like a viscous fluid and the tensiometers recorded an instantaneous peak of the water pressure head. In the case with dense sand, the failure occurred really slowly, and some local detachments of top layer preceded the advance of the whole sand volume. A numerical model solving Richards equation was used to reproduce the hydrological processes leading to failure in the two experiments. A numerical inverse method was adopted to improve the reliability of the numerical solution with respect to the data recorded from the experiments. The comparison reveals a good agreement between the experimental and numerical results for the loose sand experiment. In the case regarding dense sand, the limits of Richards solution does not allow to reach an acceptable agreement with experimental recorded. The causes might be linked with the affection of the air phase in sand pores and the incipient deformation of the soil matrix at micro-scale