Scientific Results
Figure \ref{229688} below summarizes the performance of random forests across 4 years of training and test sets. For the purpose of this comparison, only pixels that have nonzero snow amounts for all 4 years are included. All features and snow depths are standardized to a mean of 0 and a standard deviation of 1 before being fed into the model. We used cross-validation to evaluate performance within the same year. For evaluation of performance across years, the model was trained on all the data available in the snapshot for a year. A prediction was then generated based on the same feature inputs and unstandardized according to the mean and standard deviation of the new year's snow depth, and a root mean-squared error (RMSE) on this unstandardized prediction was computed.