A Statistical Approach to Testing Equatorial Ocean Models with Observed Data

Publisher Website
Google Scholar
Abstract
A model testing procedure based on multivariate statistical analysis has been developed to provide an objective measure of the fit between ocean model simulations and observations, taking into account the uncertainties in the atmospheric forcing and the inaccuracies in the oceanic data. The method is applied to the seasonal variations in an ocean model of the tropical Atlantic. A wind-driven linear multimode model with a simple mixed-layer is tested against surface currents estimated from ship drift. The uncertainties in the observations, and in the model response due to random errors in the wind stress and its interannual variability create a substantial indeterminacy in the available sample, but it is shown that they do not explain the large discrepancies that are found between observed and modeled seasonal surface currents. Nor are the uncertainties in the wind stress bulk formulation sufficient to account for the model-data differences. These can then be attributed unambiguously to the oversi... Abstract A model testing procedure based on multivariate statistical analysis has been developed to provide an objective measure of the fit between ocean model simulations and observations, taking into account the uncertainties in the atmospheric forcing and the inaccuracies in the oceanic data. The method is applied to the seasonal variations in an ocean model of the tropical Atlantic. A wind-driven linear multimode model with a simple mixed-layer is tested against surface currents estimated from ship drift. The uncertainties in the observations, and in the model response due to random errors in the wind stress and its interannual variability create a substantial indeterminacy in the available sample, but it is shown that they do not explain the large discrepancies that are found between observed and modeled seasonal surface currents. Nor are the uncertainties in the wind stress bulk formulation sufficient to account for the model-data differences. These can then be attributed unambiguously to the oversi...

This publication has 0 references indexed in Scilit: