heal.abstract |
Most statistical procedures, including parameter estimation and hypothesis testing, are based on a tacit assumption of a statistical sample consisted of independent random variables. This is not consistent with geophysical processes, which usually exhibit a strong temporal dependence, often of long range. Such dependence implies substantial negative bias in the estimation of statistical parameters of dispersion, e.g., variance, as well as parameters of dependence, e.g., autocorrelation. Failure to account for this bias leads to distorted picture of the underlying process and results in erroneous modelling and prediction. Here we demonstrate the impact of neglecting dependence in parameter estimators by using synthetic examples from stochastic processes with sort- and long-range dependence, as well as rainfall datasets that exhibit high temporal dependence. We also propose a methodology to correctly account for the bias. |
en |