Improving snow water equivalent estimates with ground penetrating radar by measuring on multiple channels

Abstract: Snow water equivalent (SWE) of a snowpack is often measured along well-chosen transects representative of an area of interest, such as a drainage basin, to capture spatial distribution of SWE, which is of great interest for many applications. For example, it is a useful input to the new generation of hydrological models used for snow melt run-off predictions. A time-effective method to perform such measurements is to conduct them along one or several transects using a ground penetrating radar (GPR) operated from e.g. a snowmobile. Traditionally, a single-channel radar system has been used to estimate SWE from the radar wave two-way travel time via a linear formula, which can be calibrated for a particular snowpack with one or several manual measurements of snow density; this method typically relies on the assumption of a dry snowpack. However, if an unknown amount of liquid water is present in the snow, or if the snow density or the liquid water content varies substantially along the transect, SWE estimates are likely to be inaccurate.A different approach is to use a multi-channel GPR system with an array of antennas that makes it possible to simultaneously measure two-way travel time of several radar pulses that form a common mid-point (CMP) gather. Then the snow depth and the radar wave propagation velocity can be determined at each point with the CMP method under the assumption of a single-layer snowpack with parallel snow and ground surfaces. With liquid water content known or assumed to be zero, the snow density can be estimated from the propagation velocity via an empirical formula for mixtures, thus solving the problem of spatial variation in snow density. Finally, SWE is calculated from the snow depth and density. However, the CMP method is known to be sensitive to measurement errors in two-way travel time and to violations of its assumptions; and for a wet snowpack, the need to know the liquid water content at each measurement point to accurately estimate snow density presents a problem if the liquid water content varies along the transect.In this thesis, two methods that improve SWE estimates obtained with GPR are presented, both of which rely on measuring on multiple channels to obtain a CMP gather at each measurement point. The first method mitigates the impact of errors in CMP calculations on density estimates by establishing a depth-to-density function from the CMP data for all measurement points along a transect. This function, specific for each transect, is then used to determine snow density from snow depth. The second method (the PDA method) improves SWE estimates of wet snowpacks by determining liquid water content at each measurement point from path-dependent attenuation of two radar signals in the CMP gather. Both methods have been tested in field experiments and the sensitivity of the PDA method to built-in assumptions and measurement errors has been investigated in simulations.The field experiment conducted to test the first method has demonstrated that by applying a depth-to-density function, the accuracy of SWE estimates for a dry snowpack can be improved substantially. For the transect in the experiment, snow density and SWE estimated directly with the CMP method were overestimated by 34% and 36% on average; and when a depth-to-density function was used, snow density was underestimated by 2% and SWE was overestimated by less than 1%. The error was determined by comparison with manual measurements.In the field experiment conducted to test the PDA method, for a snowpack with the mean liquid water content of about 5 vol.%, the mean error in SWE was 16%, compared to 34% and 31% for two reference methods that both assumed liquid water content to be zero. Separately, the performed simulations suggest that the PDA method is very sensitive to measurement errors when liquid water content is close to zero; in such cases, one of the methods that assume dry snow should be used instead of the PDA method.