Machine Learning Algorithms Assessment for Snow LWC Retrieval from SAR Data

In this study, the retrieval of snow Liquid Water Content (LWC, %) from C- and X- band SAR data was based on Artificial Neural Network (ANN) and Random Forest (RF). Two approaches were explored for generating a sufficient amount of data to train and test the ANN and RF algorithms: the first strategy was defined as “model-driven”. The second one was defined as “data-driven”. The validation results showed that RF performs better than ANN in terms of correlation coefficient R, regardless of the selected approach ("model driven" R
ANN
= 0.60, R
RF
= 0.68; “data-driven” R
ANN
= 0.50, R
RF
= 0.88 at X-band). Moreover, the RF implementation trained with the “data-driven” approach outperformed the “model-driven” approach in terms of correlation coefficient R (R
RF
= 0.88 at X-band).