Regional evaluation of three day snow depth for avalanche hazard mapping in Switzerland

Abstract

The distribution of the maximum annual three day snow fall depth <i>H₇₂</i>, used for avalanche hazard mapping according to the Swiss procedure (<i>Sp</i>), is investigated for a network of 124 stations in the Alpine part of Switzerland, using a data set dating back to 1931. Stationarity in time is investigated, showing in practice no significant trend for the considered period. Building on previous studies about climatology of Switzerland and using an iterative approach based on statistical tests for regional homogeneity and scaling of <i>H₇₂</i> with altitude, seven homogenous regions are identified. A regional approach based on the index value is then developed to estimate the <i>T</i>-years return period quantiles of <i>H₇₂</i> at each single site <i>i</i>, <i>H_72i(T)</i>. The index value is the single site sample average &mu;_{<i>H_72i</i>}. The dimensionless values of <i>H^*_72i=H_72i / &mu;_{H_72i}</i> are grouped in one sample for each region and their frequency of occurrence is accommodated by a General Extreme Value, GEV, probability distribution, including Gumbel. The proposed distributions, valid in each site of the homogeneous regions, can be used to assess the <i>T</i>-years return period quantiles of <i>H^*_72i</i>. It is shown that the value of <i>H_72i(T)</i> estimated with the regional approach is more accurate than that calculated by single site distribution fitting, particularly for high return periods. A sampling strategy based on accuracy is also suggested to estimate the single site index value, i.e. the sample average &mu;_{<i>H_72i</i>}, critical for the evaluation of the distribution of <i>H_72i</i>. The proposed regional approach is valuable because it gives more accurate snow depth input to dynamics models than the present procedure based on single site analysis, so decreasing uncertainty in hazard mapping procedure.