A New Method for Comparing and Matching Snow Profiles, Application for Profiles Measured by Penetrometers

Hardness has long been recognized as a good predictor of snow mechanical properties and therefore as an indicator of snowpack stability at the measured point. Portable digital penetrometers enable the amassing of a large number of snow stratigraphic hardness profiles. Numerous probings can be performed to assess the snowpack spatial variability and to compensate for measurement errors. On a decameter scale, continuous internal layers are typically present in the snowpack. The variability in stratigraphic features observed in the measurement set mainly originates from the measured variations in internal layer thickness due to either a real heterogeneity in the snowpack or to errors in depth measurement. For the purpose of real time analysis of snowpack stability, a great amount of data collected by digital penetrometers must be quickly synthesized into a characterization representative of the test site. This paper presents a method with which to match and combine several hardness profiles by automatically adjusting their layer thicknesses. The objectives are to synthesize the information collected by several profiles into one representative profile of the measurement set, disentangle information about hardness and depth variabilities, and quantitatively compare hardness profiles measured by different penetrometers. Themethodwas tested by using co-located hardness profiles measured with three different penetrometers—the snow micropenetrometer (SMP), the Avatech SP1 and the ramsonde—during the winter 2014–2015 at two sites in the French Alps. When applied to the SMP profiles of both sites, the method reveals a low spatial variability of hardness properties, which is usually masked by depth variations. The developed algorithm is further used to evaluate the new portable penetrometer SP1. The hardness measured with this instrument is shown to be in good agreement with the SMP measurements, but errors in the recovered depth are notable, with a standard deviation of 7.8 cm and a maximum absolute error of 20 cm at one site. Combining several SP1 profiles with our algorithm reduces depth errors to a standard deviation of 3.5 cm and a maximum of 10 cm. On ramsonde profiles, the method is less effective as substantial variability in ram hardness arises from differences between operators.