A NONLINEAR DISAGGREGATION METHOD WITH A REDUCED PARAMETER SET FOR SIMULATION OF HYDROLOGIC SERIES

A multivariate dynamic disaggregation model is developed as a stepwise approach to stochastic disaggregation problems, oriented toward hydrologic applications. The general idea of the approach is the conversion of a sequential stochastic simulation model, such as a seasonal AR(1), into a disaggregation model. Its structure includes two separate parts, a linear step-by-step moments determination procedure, based on the associated sequential model, and an independent nonlinear bivariate generation procedure (partition procedure). The model assures the preservation of the additive property of the actual (not transformed) variables. Its modular structure allows for various model configurations. Two different configurations (PAR(1) and PARX(1)), both associated with the sequential Markov model, are studied. Like the sequential Markov model, both configurations utilize the minimum set of second-order statistics and the marginal means and third moments of the lower-level variables. All these statistics are approximated by the model with the use of explicit relations. Both configurations perform well with regard to the correlation of consecutive lower-level variables each located in consecutive higher-level time steps. The PARX(1) configuration exhibits better behavior with regard to the correlation properties of lower-level variables with lagged higher-level variables.