Evaluating the influence of extending hydrologic time series in extreme quantile estimation

Abstract

Reliable estimates of quantiles associated with mid-to-large return periods are required in the everyday practice of Hydrologic Engineering. However, the usually small samples pose numerous challenges for inferring such quantiles. Therefore, augmenting sample sizes via extension techniques could be beneficial for statistical inference. This paper attempts to provide a comprehensive assessment of the performance of a collection of such techniques in estimating rare and extreme quantiles. Regression models, such as the ordinary least squares (OLS) approach and the Generalised Linear Models (GLM), as well as techniques specifically designed for time series extension, such as the Maintenance of Variance (MOVE) family, were evaluated by means of Monte Carlo simulations. Results show that, for both two and three-parameter distributional models and any level of association, the MOVE3 and MOVE4 techniques appear to provide the best balance between bias and precision of extreme quantile estimates.