Identifying and interpreting regional signals in tree-ring based reconstructions of snow avalanche activity in the Goms valley (Swiss Alps)

One of the purposes of dendrogeomorphic studies is to provide long and continuous reconstructions of mass movements and to detect climate-induced trends in process activity. The development of regional chronologies—in which information from different sites are aggregated—is often needed to identify process–climate relations and to overcome local-scale specificities, sparse data available for individual sites and to extract a signal that is common to a larger region and possibly driven by past climate fluctuations or large-scale environmental changes. Yet, such chronologies are scarce and consensus neither exists on how to compile local data at the regional scale nor on the methods to be used to extract a common signal. In the case of snow avalanches, existing regional tree-ring studies typically included less than ten paths, and they discriminated years of high/low avalanche activity based on a regional index representing the proportion of disturbed trees in any given year. However, such an index does not account for potential non-stationarities in local tree-ring reconstructions such as e.g., time-varying sample size, decreasing dendrogeomorphic potential of trees after the occurrence of an extremely large, devastating avalanche or socio-environmental changes. Here we combine a dendrogeomorphic approach to reconstruct snow avalanche events in 11 paths located in the Goms valley (Swiss Alps) with an innovative statistical modelling approach. For each path, we compute reconstructions using a 4-step procedure to disentangle potential effects of snow avalanches from disturbance pulses in trees caused by climatic or other exogenous factors. We then process the regional dataset (spanning the period 1766–2014) within a Bayesian hierarchical spatio-temporal framework specifically designed to homogenise time series of avalanche events by i) removing trends related to the decreasing number of living trees back in time and ii) inferring robust trends in mean annual/regional avalanche activity in time and space. This contribution has the merit to introduce a methodological approach allowing rigorous extraction of common, average avalanche signals from snow avalanche paths characterized by heterogeneous process activity. Despite its stringency, we show that 11 avalanche paths may not suffice to yield a signal that is independent from the selection of couloirs. As a result, the approach also does not highlight a clear climatic control of snow avalanche activity but rather points to a complex, yet combined impact of afforestation and management strategies on reconstructed avalanches.