Satellite radar altimeters like CryoSat-2 estimate sea ice thickness by measuring the return-time of transmitted radar pulses, reflected from the sea ice and ocean surface, to measure the radar freeboard. Converting freeboard to thickness requires an assumption regarding the fractional depth of the snowpack from which the radar waves backscatter (alpha). We derive sea ice thickness from CryoSat-2 radar freeboard data with incremental values for alpha, for the 2010-2021 winter periods. By comparing these to sea ice thickness estimates derived from upward-looking sonar moorings, we find that alpha values between 35-70% result in the best representation of interannual variability observed over first-year ice, reduced to <40% over multi-year ice. The underestimating bias in retrievals caused by optimising this metric can be removed by reducing the waveform retracking threshold to 20-50%. Our results pave the way for a new generation of ‘partial penetration’ sea ice thickness products from radar altimeters.