Assessment of the uncertainty of evaporation estimation models using Bayesian Model Averaging and their impact on reservoir operation

Abstract

Abstract Accurate estimation of evaporation losses is vital for efficient reservoir operation and optimal water allocation for various purposes. Due to multiple sources of error, uncertainty assessment is considered a fundamental step before using the outputs of models. This study aims to reduce the uncertainty in evaporation estimation models and investigate the influence of their uncertainty on the Zayandeh-Rud reservoir operation. The Bayesian Model Averaging (BMA) approach was used to merge predictions of eight evaporation models, and the Monte Carlo sampling method was conducted to derive 90% uncertainty intervals. The behavior of this system was examined using statistical performance indices such as time-based reliability, resilience, and vulnerability. The results of the study indicated that the models had different abilities in estimating evaporation and cannot fully estimate the pan evaporation data without including errors. The application of the BMA technique resulted in a reduced error rate and more accurate prediction of evaporation in all months. Moreover, applying different evaporation estimation methods affected the magnitude of the failure of the system and the speed of recovery from failure to a satisfactory state and caused an increased resilience index and a decreased vulnerability index.