Stochastic Nonparametric Techniques : Applications to Hydroclimate Modeling

Abstract

Nonparametric techniques for functional (probability density functions or regression) estimation provide a flexible and adaptive data exploration framework. These techniques involve "local estimation" - in that, functional estimation at a point is based on a few nearest neighbors. This provides the ability to capture nonlinearities and non-Gaussian features exhibited by the data. Utility of these techniques will be demonstrated through a variety of hydroclimate applications. These applications include:

1. Scenario Generation
- stochastic weather generation conditioned on large scale climate features.

2. Time series Forecasting
- Ensemble streamflow forecasting using climate indices
- Streamflow simulation for salinity analysis

3. Extreme value analysis
(e.g. flood frequency estimation)