Data Length Requirements for Observational Estimates of Land-Atmosphere Coupling Strength

Abstract:

A pressing problem in the climatologic and hydrologic sciences relates to the realism of models: how closely do models capture the real world? This question can only be answered with comparisons to observational datasets, yet these datasets are typically spatially sparse and temporally brief. This paper addresses the question of how long a dataset needs to be to give a reliable assessment of the “true” value of various metrics of land-atmosphere coupling strength. We demonstrate that the amount of data required to obtain robust estimates of metrics assessing relationships between variables is greater than that necessary to constrain means of directly-measured observables such as temperature or precipitation. Moreover, while the addition of unbiased noise does not significantly alter the mean of a directly observable quantity, inclusion of such noise degrades metrics assessing connections between variables, yielding a unidirectional and negative impact on metric strength estimates. This analysis suggests that longer records of surface observations are required to correctly estimate land-atmosphere coupling strength than are required to estimate mean values of the observed quantities.