Towards a new method for estimating englacial attenuation

Sea level rise projections for the second half of this century exhibit considerable uncertainties, which complicates the implementation of climate change adaptation strategies. These uncertainties stem, in part, from the reliance of ice-flow models on insufficiently constrained parameters such as the englacial temperature and the state of the ice-bed interface. In principle, both parameters can be inferred from radar measurements as the attenuation of the radar signal in the ice is a proxy for the englacial temperature and the strength of the basal reflection depends on the conditions at the basal interface. Here, we focus on developing a new method for inferring attenuation rates from radar measurements for two reasons: (1) existing methods typically provide only depth-averaged attenuation rates and exhibit a strong method dependence of inferred attenuation rates from the same radar dataset, and (2) a better estimate of attenuation rates could additionally improve the interpretation of the basal reflection strength since it relies on attenuation correction. Most contemporary methods infer depth-averaged attenuation rates from the variation of reflection strength of either internal reflectors or the bed reflector with depth. These methods rely on strong assumptions such as comparable reflectivity of internal reflectors or spatially constant reflectivity along the bed reflector. To overcome the dependence on these assumptions, we suggest a different approach that learns the relationship between radar measurements and attenuation rates directly from the data. Due to the lack of radar measurements with known attenuation rates, we simulate realistic radar data with known attenuation rates. We apply Neural Posterior Estimation, a Bayesian machine learning framework, to then infer attenuation rates from radar measurements. Ideally, this approach would not only yield depth-averaged attenuation rates, but also attenuation rate profiles. Here, we present the first results of our work.