A critical analysis on the uncertainty computation in ground-penetrating radar-retrieved dry snow parameters

The use of the ground-penetrating-radar (GPR) technique to estimate snow parameters such as thickness, density, and snow water equivalent (SWE) is particularly promising because it allows for surveying a large area in a relatively short amount of time. However, this application requires an accurate evaluation of the physical parameters retrieved from the radar measurements, which requires estimating each quantity involved in the computation along with its associated uncertainty. Conversely, the uncertainties are rarely reported in GPR snow studies, even if they represent essential information for data comparisons with other techniques such as the snow rod or snow pit methods. Snow parameters can be estimated from radar data as follows: The snow thickness can be computed from two-way traveltime if the snow average wave velocity is known; the snow density can be estimated from wave velocity using an appropriate mixing formula, and SWE can be computed once these two parameters have been calculated. Starting from published data, we have estimated the accuracy achievable by computing the overall uncertainty for each GPR-retrieved snow parameter and evaluated the influence of the different sources of uncertainties. The computation was made for three antenna frequencies (250, 500, and 1000 MHz) and various snow depths (0–5 m). We find that for snow thicknesses of less than 3 m, the main contribution to the uncertainties associated with snow parameters is given by the uncertainty on two-way traveltime estimation, especially for low antenna frequencies. However, for thicker snow depths, other factors such as the uncertainty on the antenna separation affect the overall accuracy and cannot be neglected. Our studies highlight the importance of the uncertaintiy assessment and suggest a rigorous way for their computation in the field of quantitative geophysics.