In this proof of concept study, we investigate the potential of utilizing machine learning techniques for CryoSat-2 (CS2) radar altimeter waveform classification in order to derive melt information. It has been found that the presence of liquid water in the snowpack causes subtle changes in the CS2 altimeter waveform. However, deriving melt information from CS2 waveforms is complex as many snowpack properties such as density, water content, grain size, shape, and layering influence the waveform. Training data is derived by spatiotemporally matching CS2 measurements with MODIS land surface temperature measurements. We propose a 1D time convolution network with a fully connected classifier tail for CS2 waveform classification. In addition, a non-deep learning model is implemented, providing a baseline. One of the main challenges is the high-class imbalance and therefore limited number of melt samples, as surface temperatures on the interior of Greenland rarely reach the freezing point. We propose shift augmenting the waveform and limit the number of model parameters to avoid over-fitting. Results indicate that extracting melt information from CS2 radar altimeter waveforms is feasible.

MODIS 2020 averaged land surface temperature (left) and CS2-net prediction, where 1 represents melt and 0 no-melt (right).