Assessment and reduction of climate model errors using data assimilation

Abstract

During this thesis, we develop three methods to assess the error of a climate model by comparing it to observations. The first consists of averaging over a chosen spatial region, and comparing the resulting time series. The second method uses Empirical Orthogonal Function decomposition to obtain a time-dependent (Principal Component) and space-dependent (EOF) component of the model output and observations. We then fix the spatial component by projecting one dataset onto the EOF of the other. This yields two Principal Components which we can compare. The third method uses Kalman filter data assimilation to obtain an estimate for the variance of the model output. We also implement a rudimentary form of performing data assimilation directly on an EOF.