Enhancing SAMS : Incorporating Nonparametric Data Analysis tools and evaluating its performance
Recently, BORs research efforts resulted in a user friendly parametric modeling framework, Stochastic Analysis, Modeling and Simulation (SAMS), for stochastic simulation of streamflows on the Colorado Basin a key component for decision making and long-term planning. The main research question is how best to enhance the capability of SAMS and evaluate its performance.
Need and Benefit
SAMS was developed using a parametric Periodic Auto Regressive (PAR) framework by Salas et al. (2000). This is a user-friendly and powerful package that can simulate synthetic streamflow sequences. The Parametric framework can be restrictive in that, it requires the data to be Gaussian distributed and has limited ability to capture nonlinear dependence structure. Typically, the data is first transformed to a Gaussian distribution before the PAR model is fitted for simulations SAMS has options for this transformation. However, these transformations are not automatic and can be quite cumbersome when applying to streamflows from the entire Colorado basin. Recent developments in nonparametric time series modeling (Lall and Sharma, 1996, Tarboton et al., 1997, Rajagopalan and Lall, 1999, Prairie et al., 2002) alleviate some of the shortcomings of the parametric framework.
To this end, we propose to incorporate nonparametric tools into SAMS to provide useful alternatives to the user while simulating synthetic streamflow sequences and also evaluate the performance of SAMS at several sites on the Colorado basin.
Implementing nonparametric techniques in SAMS will provide additional stochastic methods for basin mangers that are simple alternatives to parametric models for stochastic simulation. Nonparametric techniques can remove the subjective nature of transforming non-Gaussian data a typical requirement for streamflow time series. The specific tasks envisaged are incorporate nonparametric tools for:
1. probability density function estimation (marginal and conditional) (Bowman and Azzalini, 1997)
2. smoothing Local Polynomials (Loader, 1999)
3. data transformation (Tarboton, 1994) which can then used in the PAR modeling framework of SAMS
4. time series model for streamflow simulation (Prairie et al., 2003)
5. streamflow simulation conditioned on large scale climate (Prairie, et al., 2003)
Most of the tools for the above tasks along with visualization have been developed and/or are available as packages/subroutines. The main task will be to implement them in SAMS
Contact the Principal Investigator for information about partners.