We propose a robust and low complexity scheme to estimate and track carrier frequency from signals traveling under low SNR conditions in highly non-stationary channels. These scenarios arise in planetary exploration missions subject to high dynamics, such as the Mars exploration rover missions. The method comprises a bank of adaptive linear predictors supervised by a convex combiner that dynamically aggregates the individual predictors. The adaptive combination is able to outperform the best individual estimator in the set, leading to a universal scheme for frequency estimation and tracking.