A hidden Markov model combined with climate indices for multidecadal streamflow simulation

Hydroclimate time series often exhibit very low year-to-year autocorrelation while showing prolonged wet and dry epochs reminiscent of regime-shifting behavior. Traditional stochastic time series models cannot capture the regime-shifting features thereby misrepresenting the risk of prolonged wet and dry periods, consequently impacting management and planning efforts. Upper Colorado River Basin (UCRB) annual flow series highlights this clearly. To address this, a simulation framework is developed using a hidden Markov (HM) model in combination with large-scale climate indices that drive multidecadal variability. We demonstrate this on the UCRB flows and show that the simulations are able to capture the regime features by reproducing the multidecadal spectral features present in the data where a basic HM model without climate information cannot. Key Points <list id="wrcr21149-list-0001" list-type="bulleted"> <list-item id="wrcr21149-li-0001">Stochastic simulation of flow time series with regime-like behavior <list-item id="wrcr21149-li-0002">The method used a gamma HM and multinomial logistic regression model <list-item id="wrcr21149-li-0003">Model captures observed statistics and nonstationary spectral variability <doi origin="wiley" registered="yes">10.1002/(ISSN)1944-7973</doi