Hidden Hazards: the conditions that potentially enabled the mudflow disaster at Villa Santa Lucía in Chilean Patagonia

Abstract. The evaluation of potential mass wasting in mountain areas is a very complex process because there is not enough information to quantify the probability and magnitude of these events. Identifying the whole chain of events is not a straightforward task, and the impacts of mass wasting processes depend on the conditions downstream of the origin. Additionally, climate change is playing an essential role in the occurrence and distribution. Mean temperatures are continuously rising to produce long term instabilities, particularly on steep slopes. Extreme precipitations events are more recurrent as well as heat waves that can melt snow and glaciers, increasing the water available to unstabilized slopes. In this paper, we present an example that portraits the complexities in the evaluation of the chain of events. On the 16 of December of 2017, a rockslide occurred in the Yelcho mountain range. In that event, 7 million m3 of rocks and soil fell on the Yelcho glacier depositing 2 million m3 on the glacier terminal, and the rest continued downstream, triggering a mudflow that hit Villa Santa Lucia in the Chilean Patagonia, killing 22 people. The rockslide event or similar was anticipated in the region by the National Geological and Mining Survey (Sernageomin in Spanish). However, the effects of the terrain characteristics along the runout area were more significant than what was anticipated. In this work, we evaluate the conditions that enable the mudflow that hits Villa Santa Lucia. We used the information generated by Sernageomin's professional after the mudflow. We carried out geotechnical tests to characterize the soil. We simulated the mudflow using two hydrodynamics software (r-avaflow and Flo-2D) that can handle the rheology of the water–soil mixture. Our results indicate that the soil is classified as volcanic pumices. This type of soil can be susceptible to the collapse of the structure when subjected to shearing (molding), flowing like a viscous liquid. From the numerical modeling, we concluded that r-avaflow performs better than Flo2D. We can reproduce the mudflow satisfactorily using water content in the mixture ranging from 30 to 40 %. Finally, in order to achieve the water content, we need a source of water smaller than 3 million m3 approximately. From the simulations and soil tests, we determined that in the area scoured by the mudflow, there were around 2 789 500 m3 of water within the soil. Therefore, the conditions of the valley were crucial to enhance the impacts of the landslide. This result is relevant because it highlights the importance of evaluating the complete chain of events to map hazards. We suggest that in future hazard mapping, geotechnical studies in combination with hydrodynamic simulation should be included, in particular, when human lives are at risk.


In this study, we want to explore the mechanisms that enable a landslide of 7x10 6 m 3 to evolve to the catastrophic mudflow that hit Villa Santa Lucía in the Chilean Patagonia, resulting in 22 fatalities. The avalanche, which may 95 have been triggered by hydrometeorological conditions and destabilization of the wall around the receding Yelcho glacier, led to the generation of a hyper-concentrated flow at the head of the Burritos River that traveled around ten kilometers and affected 50% of the urban area of Villa Santa Lucia on December 16, 2017. The first hypothesis indicated that the event was possible because of the presence of a glacier lake; however, there was no indication of such features in the area. Therefore, this study seeks to understand the preconditions in the valley right below the 100 avalanche zone that contributed to this hyper-concentrated flow event and what conditions enabled this event without the presence of a glacier lake, which will have a two-fold application. First, it will allow us to understand the mechanisms of the chain of events leading to the 2017 mudflow in Villa Santa Lucia, and second, and probably most important, to update the criteria for mapping risks associated with mudflows in the Chilean Patagonia.

Location
Villa Santa Lucia is located in the valley of the Frio River, 75 kilometers south from Chaitén (closer town), along the Carretera Austral, in the Los Lagos region, Chile. The avalanche event started at the head of the Burritos river basin (43.413°S, 72.367°W) that runs to the west of Villa Santa Lucía (Figure 1). It begins at the eastern slope of the western side of the regional Andes. 110 This area was under tectonic modeling associated with the Liquiñe-Ofqui Fault System (LOFS), forming valleys with a NS-trend, such as the Frio River Valley and Yelcho Lake, which have been eroded by glacial processes in the Pleistocene and, subsequently, filled by volcanic, alluvial and river processes (Sernageomin, 2018c). The climate of the area presents intense thermal variations, high summer temperatures, and freezing temperatures in the winter period; the rainfall reaches 3,000 mm annually, decreasing to the east (CECS, 2017). 115 https://doi.org/10.5194/nhess-2019-419 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. which should be assessed at greater detail. High flood hazard would affect only the flood plain adjacent to Villa Santa Lucia, although some degree of danger in the town itself must be evaluated on a more detailed scale." This shows that the area of Villa Santa Lucía and its surroundings were and are prone to mass wasting phenomena.
Although the town itself was not in danger, it was mentioned that more detailed studies were needed in the area. Sernageomin (2018b) compile a series of reports; several studies are mentioned that help to understand the origin 155 and explain the event as a whole. On December 16, 2017, a 7 million cubic meter volcanic rock slide detached ( Figure 3, point 1) falling on the Yelcho glacier. The glacier tongue sits on an intrusive formation that has a drop of about 80 meters. The glacier has been steadily retreating during the last decade. Sernageomin (2018c) indicated that there might have been a small lake and water available under the glacier at the terminal, but there is no conclusive information. We checked satellite images before the events, and there is no indication of a lake above the intrusive 160 unit. Two million cubic meters of the material stayed above the intrusive, and five million cubic meters of material continue downstream, sliding on the intrusive unit at an angle of more than 70 degrees. At this point the material encounter the Burrito River that has high slopes, so the detritus flow continued along the river at high velocity  Additionally, the soil was almost saturated, which added water to the mix, transforming the avalanche flow into a mudflow. Moreover, dense forest was present, which added a significant amount of biomass to the mix. Then the 170 mudflow reaches an area with low slopes at a distance of 8.6 km from its origin through the Burritos River to the east. In its trajectory, the mudflow crossed Route 7 (Carretera Austral) in a stretch estimated of 2 km and filled an old wetland. In that sector, the flow was channeled in a canyon oriented north-south toward Villa Santa Lucia in a section of 1.5 km (Figure 1 and 3

Fieldwork and Geotechnical sampling
A fieldwork campaign was carried out in January 2019 to map in high resolution the area using an unmanned aerial 220 vehicle (UAV). We used aerial photogrammetry to produce a high-resolution DEM for some parts of the study area.
Aerial photogrammetry allows creating 3D models from 2D images, obtaining geometric characteristics of the objects they represent. We used a UAV Inspire II with a Zenmuse X4 camera to capture aerial images (Figure 9).
We did not have differential GPS at the moment of this survey to take control points to correct the DEM generated. Therefore, we just used this DEM to corroborate that the DEMs with lower resolution and freely available were able 225 to capture specific features such as canyons or small changes in slopes that could affect the hydrodynamic simulation and the path of the flood. We also generated a high resolution mosaic to observe details of the flood. We  The tests performed to determine the geotechnical properties of the primary materials were: direct shear test, unconfined compression, density, and soil classification. The parameters obtained in each laboratory test will serve as constrains for the numerical modeling of the hyper-concentrated flow of Villa Santa Lucía.

Numerical Modeling
The modeling of the hyper-concentrated flow is carried out using two software: r-avaflow and Flo-2D, which we describe below: R.avaflow is developed in two formats for its environment and operation, r.avaflow expert and r.avaflow professional. The latter represents an autonomous version with reduced functionalities. It operates through a graphical user interface (GUI). In this study, we used the professional version (Mergili et al., 2017). This model represents a complete open-source computational framework based on a geographic information system (GIS) that 255 offers a two-phase flow model. Also, consider the drag of material along the flow path. These characteristics facilitate the simulation of complex mass flows, as well as chained processes and interactions.
For the propagation of the flow, we used the Pudasaini model (Pudasaini, 2012), which is a two-phase mass flow model. Solids and fluids materials can be dragged from the bottom and incorporated into the flow. The r.avaflow output consists mostly of raster maps of solid and fluid flow heights, velocities, pressures, kinetic energies, and 260 dragged heights.

Flo2D
Flo-2D is a two-dimensional finite differences model ( errors. Therefore, we used it for numerical simulation. We resampled the resolution of ALOS PALSAR to 30 meters to speed up the simulations.

Models calibration
For the calibration of the models, we used three criteria (1) Flood area, which was mapped after the event using  Table 1 for the values.

Geotechnical results
It was not possible to carry out the extraction of unaltered geotechnical samples from the upper part, mainly due to the 5 kilometers of steep terrain, including a section that needed to be climbed to transport the material by foot.
However, it was possible to collect soil and rocks, which, although not in an unaltered condition, help to 295 characterize the soil that originated the hyper-concentrated flow. We extracted soil samples in the middle-low part of For the soil moisture, a wet portion of the soil was extracted and allowed to dry in an oven at 60°C for two days.
With this, it was possible to calculate mass soil water content (by weight) in 1.96 gwater/gsolid 300 Then, a fraction of the soil sample was covered with paraffin to seal all pores and thus determine its density through the submerged weight difference. The wet soil density is 1.24 grams per cm 3, and the dry soil's apparent density is 0.419 grams per cm 3 -the specific weight Gs=2.65 gr. Therefore, the void ratio "e" is 5.324. Using these values, it turned out that the soil porosity is 84.18%, and the water saturation of the soil sample 97.4%; therefore, considering that the total porosity is 84.18%, 81.95% of the total unaltered soil volume corresponded to water. 305 Then a direct shear test was performed in three probes. Each probe was loaded with 2, 4, and 8 kg, respectively, allowing them to consolidate for 24 hours. We recorded the deformations versus time in the first sample to determine the speed for the direct shear test using the square root scale method (Table 2). With these results, we estimated a cohesion value of 22.5 kPa and an internal friction angle of 23.8°. 310

Soil Classification
In order to carry out the classification of the soil, we used two tests. First, we determined soil´s granulometry using the American Society of Testing Materials ASTM D2487 (ASTM, 2017a), and second, we determined soil consistency limits (liquid and plastic) using the American Society of Testing Materials ASTM D4318-17e1 (ASTM, 2017b). For the granulometry, Table 3 shows what percentages pass through the different sieves' sizes. We omitted 315 the larger sieves since 100% of the material passed throughout them. Finally, the liquid and plastic limits of the soil are 50% and 27%, respectively. According to the USCS classification, the soil corresponds to a CH, an inorganic clay of high plasticity, while for the classification of the 320 AASHTO, it is considered to be clay.

Numerical modeling
According to the antecedents of the event, the avalanche fell on the glacier terminus. A fraction of the material stayed deposit between the glacier and the intrusive formation. The difference (about five million cubic meters of sediment) continued downstream, initiating the scouring process and mudflow that lead to the Santa Lucia event. 325 Therefore, the models' domain starts at the base of the intrusive formation with an initial volume of 5.000.000 cubic meters. For the calibration, we matched the observations and the empirical calculations by Sernagomin of the velocities and depositions.

R-Avaflow
First of all, in order to simplify the calibration, we divided the process into two. First, we set the sediment 330 concentration by volume of the flood in 50% and change the drag coefficient, basal friction angle, environmental resistant coefficient, and fluid friction coefficient. Table 4 shows the first set of parameters used. The best set of parameters is 5.75, 2, 0.022 and. 0.0005 for the drag coefficient, base friction angle, environmental 335 resistant coefficient, and fluid friction coefficient, respectively, and the corresponding map result is in Figure 12. https://doi.org/10.5194/nhess-2019-419 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.

Figure 12:
R-avaflow modeling results for a concentration by volume of 50%. Figure 12 shows good agreement with what was reported by (Sernageomin, 2018c). However, the mixture is still largely fluidized, and it does not follow the edges of the inundation. For this reason, we performed simulations with 340 the same input parameters but a varying percentage of initial water. We changed the percentage of water from 20% to 70%. The error for the heights, speeds, and pressures calculated in each model are in Figure 13. From Figure 13, we concluded that a mudflow with a volume of water of 30% could reproduce best the VSL event ( Figure 14).

Flo2d
The main results from Flo-2D for the hyperconcentrated flow in Villa Santa Lucía are presented below. The table below shows the parameters used for modeling.
For the laminar resistance k, we used the default number 3000. The influence of the value of does not affect simulations significantly compared to other parameters related to flow resistance (Hsu et al., 2010). The specific 355 gravity of sediments Gs is equal to 2.65 g cm -3 . For the sediment concentration by volume, we used values between 40% to 50%, which correspond to hyper-concentrated flows. The error for the heights, speeds, and pressures calculated in each model are in Figure 15. Moreover, Figure 16 shows the extension of the best model using Flo2D.

Geotechnical sample
The soil was classified as clay with high plasticity (CH), with a friction angle of 23.8 ° and a cohesion value of 22.5 365 kPa, which are characteristic of a consolidated clay. However, the soil extracted had various types of soils and minerals in its composition, which made this a rare soil that is not described or classified by USCS.
For the granulometry, 72% of the soil passed through sieve No. 200. Mainly sand and volcanic soils did not pass sieve No 200. Similar results were founded by Gonzalez-Pulgar (2012) in volcanic soils that have an internal soil friction angle of 25° and a cohesion value of 2.9 kPa. Moreover, our results were consistent with their high water 370