A stochastic hydrology framework for the management of multiple reservoir systems

Long-term planning and management of large hydrosystems, such as multiple reservoir systems, under hydrological uncertainty continues to be a very difficult task. Stochastic processes and stochastic simulation are the most reliable methodologies for the study of hydrosystems under a wide range of hydroclimatic inputs and for the risk assessment of different management policies. Climate change scenarios and, more specifically, drought scenarios can be incorporated into stochastic models by either modifying the historical statistical characteristics or better, assuming large timescale random fluctuations. Such fluctuations can be equivalently modelled as long-term persistence by means of a specified autocorrelation structure. Using these ideas, a comprehensive stochastic methodology is developed and implemented in an integrated software package named Castalia. The methodology is based on a two-level multivariate simulation-forecast scheme. In the higher level it enables preservation of important features on an annual timescale, such as hydrologic persistence. In the lower level it enables reproduction of features on a monthly or sub-monthly timescale, such as periodicity. The above methodology was applied for the study of the water supply system of Athens, which contains four reservoirs. Several scenarios were examined, which allowed a detailed investigation of uncertainty and risk associated with the system.