2016 Ocean Sciences Meeting

February 21-26

This tutorial aims to introduce to the oceanographic modeling community newly developed approaches to uncertainty quantification (UQ).  Ocean and atmospheric models, like-all models, have numerous sources of uncertainties that must be accounted for when analyzing the model's output. The exploration of the large uncertain parameter space via ensemble simulations is the central computational issue, and is the pre-requisite to estimating the mean of the model output and its uncertainty. Here we present the most common two approaches to UQ: one based on polynomial chaos methods and the other on Gaussian Process regression.  Both techniques are ensemble based and share a common paradigm which can briefly be described as follows: Use a small ensemble to construct an accurate and computationally efficient proxy for the model, and perform the statistical calculations on the proxy.  The efficiency of the proxy allows us to simulate large ensembles, of O(10^6), to compute accurately the output statistics.  This large ensemble would not be possible had the original (and expensive) ocean general circulation model been used.  The tutorial will cover the following topics: Construction of non-intrusive Polynomial Chaos proxies; Sensitivity Analysis using Polynomial chaos; Proxy Construction using Gaussian process regression; and Some sample applications and comparisons.

• OD - Ocean Observing and Data Management
• PO - Physical Oceanography/Ocean Circulation