Dr. Cristina Pérez Guillén

Powder snow avalanches (PSAs) radiate infrasound energy, yet the source mechanism remains unclear, limiting hazard monitoring and mitigation with infrasound-based technologies. Here, we analyze a unique data set from a large PSA to improve the understanding of the source mechanism. Through comparison of cluster activity within the airborne layers of the PSA and the recorded infrasound signal in the frequency domain, we demonstrate that infrasound is mainly generated from particle clusters suspended by turbulent eddies or ejected from the denser basal layer. Further correlating infrasound amplitudes with radar-derived spatial distributions of these clusters, we reveal a distributed source extending hundreds of meters behind the avalanche front. Additionally, we establish a relationship between infrasound and kinetic energy of suspended particles. These findings deepen our understanding of the complex dynamics of infrasound generation, offering valuable insights for avalanche detection and early warning strategies, and fundamental comprehension of PSA dynamics.

The destructive power of avalanches demands fast detection and early warning systems to mitigate avalanche risk in mountain areas. Since avalanches are extended moving sources of seismic energy, array systems based on seismic sensors can be used to monitor avalanches over wide areas, regardless of weather and visibility. Operationally, one of the main challenges of these detection systems is the automatic discrimination of avalanche signals from other sources. Various machine learning techniques have, therefore, been used to automatically differentiate avalanche events within seismic recordings. So far, these models mainly relied on expert knowledge and a time-consuming selection process to extract signal attributes from the time and frequency domains. In this study, we therefore investigate the detection performance of models trained with expert-derived features and fully automatic extracted features using autoencoders. Autoencoders are neural networks that efficiently compress (encode) input data, such as seismic signals, down to its  essential features to reconstruct (decode) the original input. Hence, meaningful features can be automatically learnt during this process. To develop the models, we use a catalogue of labelled events recorded by a seismic array system deployed above Davos (Switzerland) since the winter season 2020-2021. We assess the performance of different classifiers developed with data from individual sensors by aggregating the predictions from all array components through ensemble predictions. Our results show that models trained with autoencoder-extracted features performed as well as those relying on expert features and spectrogram images. When tested on a continuous seismic dataset, the autoencoder model based on the frequency content of the signals achieved the best balance between avalanche detection accuracy and false alarm rate. Looking forward, we aim to integrate the best approach into an operational, near-real-time avalanche detection system.

Snow avalanches are the deadliest natural hazards in Switzerland and cause significant economic losses annually. Therefore, avalanche detection and prediction are crucial in ensuring safety and mobility in the Swiss Alps. Operationally, avalanche forecasts are issued daily during the winter season to inform the public and local authorities about the avalanche hazard. Whereas traditional avalanche forecasting has been a human-expert decision-making process entirely, machine-learning models have become increasingly used in recent years. In this context, the Swiss avalanche warning service has been at the forefront of using these innovative data-driven tools to supplement current forecasting practices. These data-driven approaches have shown the potential to provide objective decision tools with higher spatiotemporal resolution than human-based forecasts. Nevertheless, the predictions generated by these machine learning models are opaque, usually referred to as 'black boxes', making it challenging for avalanche forecasters to interpret how variables are combined to make predictions. To address this, we employed an explanation method to interpret predictions from a machine learning model developed to predict danger levels for dry snow conditions in Switzerland. Moreover, precise avalanche data collected from an automated seismic detection system at the Vallée de la Sionne test site (Switzerland) are used to validate these danger level forecasts during the winter season 2020-2021. A novelty in future model implementations would be integrating these timely avalanche detections into forecasting models.

Powder snow avalanches (PSAs) with high-energy airborne layers pose a significant threat to buildings and facilities intended to be protected by infrastructure like dams. However, understanding the factors driving the development of these hazardous suspension layers remains challenging due to limited  xperimental data and understanding of PSA generation mechanisms. Recent advances in infrasound research have revealed that infrasound is primarily generated from particle clusters suspended in the airborne layer of PSAs by turbulent eddies or ejected from the denser basal layer. In addition, the infrasound signal is correlated with the kinetic energy of all active particles within these layers, offering a promising avenue to address this challenge. In this study, we analyze 27 infrasound signals collected at the Vall ´ee de la Sionne test site in Switzerland, covering various PSA dimensions and degrees of powder cloud development. We systematically quantify amplitudes, cumulative intensities, and energy distribution extracted from the infrasound  easurements. These findings are compared with reference data from a high-resolution GEODAR radar to assess the evolution of the powder cloud in both temporal and spatial dimensions. Subsequently, we correlate this information with boundary conditions such as snow cover, and previous avalanche activity to understand their influence on the development of high-energy airborne layers in PSAs. Conversely, we explore how these factors contribute to the decay of the powder cloud, thereby enhancing our ability to assess and mitigate avalanche hazard.

In recent years, the integration of physical snowpack models coupled with machine-learning techniques has become more prevalent in public avalanche forecasting. When combined with spatial interpolation methods, these approaches enable fully data- and model-driven predictions of snowpack stability or avalanche danger at any given location. This prompts the question: Are such highly detailed spatial model predictions sufficiently accurate for operational use? To explore this, we assess the performance of interpolated, model-based predictions of snowpack stability and avalanche danger, comparing them to human-generated public avalanche forecasts during the 2023/2024 winter season in Switzerland. To do so, we compare human forecasts and model predictions for locations in avalanche terrain (considering coordinates, aspect, elevation) where skiers triggered avalanches (244 events) or which were skied but where no avalanche was triggered (non-events, 3177 data points from GPX tracks). While this data reflects human behavior to some extent, we consider the event ratio as a proxy for the probability of avalanche release due to human load. We observed that with increasing model-predicted danger level or decreasing model-predicted snowpack stability, the event ratio increased. Comparing model predictions with human-made forecasts showed that the predictive performance of two operationally used models was similar to the performance of human avalanche forecasts: both predicted a strong increase in the probability of human-triggered avalanches. In summary, our results indicate that models capture regional patterns of snowpack (in)stability or avalanche danger well, and that these model chains should therefore be systematically integrated in the forecasting process.

Wet-snow avalanches are triggered by the infiltration of liquid water which weakens the snowpack. Wet-snow avalanches are among the most destructive avalanches, yet their release mechanism is not sufficiently understood for a process-based prediction model. Therefore, we followed a data-driven approach and developed a random forest model, depending on slope aspect, to predict the local wet-snow avalanche activity at the locations of 124 automated weather stations distributed throughout the Swiss Alps. The input variables were the snow and weather data recorded by the stations over the past 20 years. The target variable was based on manual observations over the same 20-year period. To filter out erroneous reports, we defined the days with wet-snow avalanches in a stringent manner, selecting only the most extreme active or inactive days, which reduced the size of the dataset but increased the reliability of the target variable. The model was trained with weather variables and variables computed from simulated snow stratigraphy in 38 slopes facing the 4 cardinal directions. While model development and validation were done in nowcast mode, we also studied model performance in 24-hour forecast mode by using input variables computed from a numerical weather prediction (NWP) model. Overall, the performance was good in both nowcast and forecast mode (f1-score around 0.8). To assess model performance beyond the stringent definition of wet-snow avalanche days, we compared model predictions to wet-snow avalanche activity over the entire Swiss Alps, based on the raw data over 8 winters. We obtained a Spearman correlation coefficient of 0.71. Hence, our model represents a step toward the application of support tools in operational wet-snow avalanche forecasting.

Avalanche accidents evidence periods of increased avalanche danger that avalanche forecasters want to pinpoint. Accidents records reveal periods with dangerous situations that may be difficult to grasp in forecasts or manage in the terrain. Hence, analyzing accidents has merit for enhancing avalanche forecasting, danger communication and avalanche education, alike. We analyzed avalanche records from forecasting regions in the French Alps, the Swiss Alps and Tyrol in the Austrian Alps. Including fatal accidents only, we obtained a multinational record that is almost complete with respect to location, date, trigger and simple terrain characteristics, such as slope aspect. A recently developed snow avalanche climatology describes the climatological variations in French Alpine regions. The local conditions were completed with snow cover modelling based on reanalyzed meteorological data. Our results confirm that fatalities with natural release have become rare today. Fatal backcountry avalanches are more frequent on north-facing aspects and occur with dry-snow conditions. Comparisons with the snow climates demonstrate that accidents are more common in regions which are prone to form persistent weak layers. Differences between the three Alpine countries were, for instance, related to the forecast danger level. Blending snow cover modeling with avalanche accident reports, facilitates analyses across borders in an objective manner and can provide a more complete set of information concerning the regional conditions, including regional climate.

One of the main challenges in avalanche forecasting is the complexity of the snowpack and its interactions with the environment. Traditional methods rely on expert knowledge to analyze snow and weather data, field observations and weather predictions, which is time-consuming and partly subjective. In contrast, data-driven models can analyze large amounts of data faster and may identify patterns that are difficult for humans to detect. Such models are based on statistical or machine learning algorithms that learn from past data to make predictions about new situations. Data-driven models are increasingly used in avalanche forecasting, as they can provide more objective and timely predictions, assisting forecasters in decision-making. Here, we present three recently developed models used in operational avalanche forecasting in Switzerland. The data-driven models use machine-learning algorithms with meteorological and simulated snow stratigraphy data as input to predict (1) the avalanche danger level, (2) snowpack instability and natural avalanche probability, and (3) wet-snow avalanche probability. The models were trained on historical data and typically have an accuracy of about 75\%. During the last three winter seasons, we tested these models in operational avalanche forecasting for the Swiss Alps at SLF. Models 1 and 2 were consulted daily, while model 3 only in potential wet-snow avalanche situations. Preliminary results suggest that the models performed equally well in nowcast mode, when driven with measured data, as in forecast mode, when driven with data from numerical weather prediction models. Overall, the positive feedback we received from the forecasters shows that data-driven models can successfully be integrated into operational forecasting systems.

Even today, the assessment of avalanche danger is by and large a subjective yet data-based decision-making process. Human experts analyse heterogeneous data volumes, diverse in scale, and conclude on the avalanche scenario based on their experience. Nowadays, modern machine learning methods and the rise in computing power in combination with physical snow cover modelling open up new possibilities for developing decision support tools for operational avalanche forecasting. Therefore, we developed a fully data-driven approach to assess the regional avalanche danger level, the key component in public avalanche forecasts, for dry-snow conditions in the Swiss Alps. Using a large data set of more than 20 years of meteorological data measured by a network of automated weather stations, which are located at the elevation of potential avalanche starting zones, and snow cover simulations driven with these input weather data, we trained two random forest (RF) classifiers. The first classifier (RF 1) was trained relying on the forecast danger levels published in the official Swiss avalanche bulletin. To reduce the uncertainty resulting from using the forecast danger level as target variable, we trained a second classifier (RF 2) that relies on a quality-controlled subset of danger level labels. We optimized the RF classifiers by selecting the best set of input features combining meteorological variables and features extracted from the simulated profiles. The accuracy of the models, i.e. the percentage of correct danger level predictions, ranged between 74 % and 76 % for RF 1 and between 72 % and 78 % for RF 2. We assessed the accuracy of forecasts with nowcast assessments of avalanche danger by well-trained observers. The performance of both models was similar to the agreement rate between forecast and nowcast assessments of the current experience-based Swiss avalanche forecasts (which is estimated to be 76 %). The models performed consistently well throughout the Swiss Alps, thus in different climatic regions, albeit with some regional differences. Our results suggest that the models may well have potential to become a valuable supplementary decision support tool for avalanche forecasters when assessing avalanche hazard.

Forecasting avalanche danger at a regional scale is a largely data-driven yet also experience-based decision-making process by human experts. In the case of public avalanche forecasts, this assessment process terminates in an expert judgment concerning summarizing avalanche conditions by using one of five danger levels. This strong simplification of the continuous, multi-dimensional nature of avalanche hazard allows for efficient communication but inevitably leads to a loss of information when summarizing the severity of avalanche hazard. Intending to overcome the discrepancy between determining the final target output in higher resolution while maintaining the well-established standard of assessing and communicating avalanche hazard using the avalanche danger scale, avalanche forecasters at the national avalanche warning service in Switzerland used an approach that combines absolute and relative judgments. First, forecasters make an absolute judgment using the five-level danger scale. In a second step, a relative judgment is made by specifying a sub-level describing the avalanche conditions relative to the chosen danger level. This approach takes into account the human ability to reliably estimate only a certain number of classes. Here, we analyze these (yet unpublished) sub-levels, comparing them with data representing the three contributing factors of avalanche hazard: snowpack stability, the frequency distribution of snowpack stability, and avalanche size. We analyze both data used in operational avalanche forecasting and data independent of the forecast, going back 5 years. Using a sequential analysis, we first establish which data are suitable and in which part of the danger scale they belong by comparing their distributions at consecutive danger levels. In a second step, integrating these findings, we compare the frequency of locations with poor snowpack stability and the number and size of avalanches with the forecast sub-level. Overall, we find good agreement: a higher sub-level is generally related to more locations with poor snowpack stability and more avalanches of larger size. These results suggest that on average avalanche forecasters can make avalanche danger assessments with higher resolution than the five-level danger scale. Our findings are specific to the current forecast set-up in Switzerland. However, we believe that avalanche warning services making a hazard assessment using a similar temporal and spatial scale as currently used in Switzerland should also be able to refine their assessments if (1) relevant data are sufficiently available in time and space and (2) a similar approach combining absolute and relative judgment is used. The sub-levels show a rank-order correlation with data related to the three contributing factors of avalanche hazard. Hence, they increase the predictive value of the forecast, opening the discussion on how this information could be provided to forecast users.

The ground motion generated by a snow avalanche is a complex natural phenomenon produced by the interaction of the avalanche with the snow cover and the terrain. Nowadays, avalanches are recorded using seismic sensors that provide data that can be used to obtain information about the characteristics of the source (type of flow regime), the basic dynamic processes (erosion and deposition) and the approximate size of the avalanche. In order to better understand the connection between seismic signals and avalanche dynamics, we analysed data from thirty avalanches at the Vallee de la Sionne test site in Switzerland. Seismic sensors were configured linearly both in and outside the avalanche path. For each seismic signal, the duration, peak ground velocity of the envelope, and both the intensity and the frequency content were compared with the avalanche flow regimes and the thicknesses of the snow cover measured using frequency-modulated continuous wave radar, as well as with the avalanche run-out distances obtained from photographs. We showed that the frequency content of the seismic signal can be used to infer the avalanche flow regime and to distinguish between dense and energetic turbulent flows. Furthermore, the seismic signal duration of both dense and energetic flows can be correlated to avalanche run-out distance. If the snow cover at the seismic sensor location is less than approximately 2 m and energy absorption does not significantly weaken the intensity of the signal, the avalanche run-out distance can also be deduced from the peak ground velocity and its intensity. These results suggest that the flow type characteristics and the run-out distances of avalanches can be feasibly inferred using only seismic data. Such analysis of seismic data can be usefully employed in avalanche monitoring and management.

Snow avalanches are moving sources of infrasonic and seismic energy. They can be triggered by many different mechanisms that include the shaking produced by earthquakes. The forces induced by an earthquake can cause an increase in the load down the slope and can also decrease the shear strength and both effects can cause the release of an avalanche. This phenomenon represents an important hazard associated with earthquakes in snow-covered mountain areas with high seismicity.
On 6 December 2010 a snow avalanche was released at the experimental site of Vallée de la Sionne (VDLS) in Switzerland seconds after a local earthquake of magnitude M_L 3.1 with the hypocenter in France, approximately 43 km from the avalanche starting zone. The seismic and infrasound signals generated by the earthquake and the snow avalanche were recorded by an array of sensors installed at VDLS. This paper analyses these data and shows that the avalanche was possibly triggered by the earthquake. This analysis also allows us to determine the characteristics of the avalanche (type and path). The infrasound data shows that the time of the avalanche release coincided with the arrival of the seismic waves of the earthquake. We calculate the values of the ground vibration parameters (PGD, PGV, PGA, PSA, I_a and TD) measured at the release area of the avalanche and compare them with those of two other earthquakes that did not trigger an avalanche. To evaluate the influence of the snowpack stratigraphy with the effectiveness of the earthquakes to trigger an avalanche, we simulate the snow cover using the one-dimensional snow cover model SNOWPACK. The weather and snow cover conditions of the days on which these events occurred are compared and used to evaluate the snowpack stability and the consequent likelihood of avalanche activity. The snowpack stability is the primary factor that determines whether an avalanche may be triggered by minor earthquakes. We conclude that when the snowpack is only marginally stable then the displacement caused by even a small earthquake could be enough to trigger an avalanche. Furthermore, the analysis of the other two, even stronger, earthquakes shows that in stable conditions no avalanche was triggered

The full-scale avalanche dynamics test site Vallée de la Sionne (VdlS) is providing scientists and engineers with essential data to understand and model avalanche motion. The site has been in operation since the winter 1997/98 and represents the cornerstone of snow avalanche dynamics research at the SLF. Numerous international teams including, University of Durham (England), IRSTEA (France), University of Barcelona (Spain), University College London (England), University of Sheffield (England), and BFW, Department of Natural Hazards (Austria) have their instrumentation at the site. The site is equipped with a unique collection of field-hardened sensors to measure avalanche velocity, impact pressure, flow depth and density. At VdlS pre- and post-event field campaigns with manual measurements are coupled with advanced remote sensing techniques to obtain a detailed picture of snow avalanche flow. Pulse doppler and frequency modulated continuous wave radars measure velocity and track important flow features over the entire slope. Seismic and acoustic sensors monitor the avalanche activity and provide triggering for the acquisition system. Since the winter season 2012/13 data of five nearby automatic weather stations are used to perform snow cover simulations with the numerical model Alpine 3D. This allows reproducing snow cover parameters such as snow depth and temperature in almost real-time over the entire slope. We use the data to improve our understanding of the physics of avalanches and to investigate important practical questions such as the role of snow cover entrainment in determining run-out, and how avalanches and obstacles interact.