A Bayesian Retrieval of Greenland Ice Sheet Internal Temperature from Ultra-wideband Software-defined Microwave Radiometer (UWBRAD) Measurements 

Thursday, 17 December 2015
Poster Hall (Moscone South)
Yuna Duan1, Michael T Durand2, Kenneth C Jezek3, Caglar Yardim4, Alexandra Bringer3, Mustafa Aksoy3 and Joel Johnson3, (1)Ohio State University Main Campus, School of Earth Sciences, Columbus, OH, United States, (2)Ohio St Univ-Earth Sciences, Columbus, OH, United States, (3)Ohio State University Main Campus, Columbus, OH, United States, (4)Ohio State University Main Campus, Electroscience Laboratory and Department of Electrical Engineering, Columbus, OH, United States
The ultra-wideband software-defined microwave radiometer (UWBRAD) is designed to provide ice sheet internal temperature product via measuring low frequency microwave emission. Twelve channels ranging from 0.5 to 2.0 GHz are covered by the instrument. A Bayesian framework was designed to retrieve the ice sheet internal temperature from UWBRAD brightness temperature (Tb) measurements for the Greenland air-borne demonstration scheduled for summer 2016. Several parameters would affect the ice sheet physical temperature. And the effective surface temperature, geothermal heat flux and the variance of upper layer ice density were treated as unknown random variables within the retrieval framework. Synthetic brightness temperature were calculated by the snow radiation transfer models as a function of ice temperature, ice density, and an estimate of snow grain size in the upper layers. A incoherent model-the Microwave Emission Model of Layered Snowpacks (MEMLS) and a coherent model were used respectively to estimate the influence of coherent effect. The inputs of the radiation transfer model were generated from a 1-D heat-flow equation developed by Robin and a exponential fit of ice density variation from Borehole measurement. The simulated Tb was corrupted with white noise and served as UWBRAD observation in retrieval. A look-up table was developed between the parameters and the corresponding Tb. In the Bayesian retrieval process, each parameter was defined with its possible range and set to be uniformly distributed. The Markov Chain Monte Carlo (MCMC) approach was applied to make the unknown parameters randomly walk in the parameter space. Experiment results were examined for science goals on three levels: estimation of the 10-m firn temperature, the average temperature integrated with depth, and the entire temperature profile. The 10-m temperature was estimated to within 0.77 K, with a bias of 0.6 K, across the 47 locations on the ice sheet; the 10-m “synthetic true” temperatures ranged from 246 – 262 K. The vertically-averaged temperature RMSE was 1.0 K, with a bias of 0.5 K. In future experiments, other parameters, such as the correlation length of snow grains, will be included.