Periodic autoregressive models in flow regulating reservoirs design

Abstract

Historical streamflow series may not present critical drought periods. As a result, volumes of flow regulating reservoirs estimated from historical data can lead to reservoirs design underestimation. Periodic stochastic models can be used to generate multiple different sequences of flows that present statistical properties close to those corresponding to the historical series but including more critical dry periods. The different volume values estimates allow probabilistic analysis and definition of regulating reservoir volumes based on periodic flows synthetic sequences occurrence risks. This study aims at assessing the performance of different periodic autoregressive models in generating sequences of flows and reproducing regulating reservoirs volumes. Periodic multiplicative autoregressive moving average models (PMIX), complete or parsimonious, showed better performances than traditional periodic autoregressive and autoregressive moving average models, for most historical streamflow series considered. The consideration of synthetic series generated by using PMIX multiplicative models can reduced risks of underestimation of reservoirs volumes.