Master Thesis Defense Martin Lyskjær Frølund

Title: Perturbation Methods in Ensemble Weather Forecasts (The defense will be in Danish)


The chaotic nature of the atmosphere makes weather prediction a fundamentally difficult task. For that reason, weather prediction centres perform ensembles of forecasts, all with different initial conditions, to estimate the uncertainty of their prediction. Various methods are used to perturb the initial conditions, the choice of which influences the uncertainty estimate.In this work we investigate seven different perturbation methods including the breeding vector (BV) method, orthogonal breeding vector method (BV-EOF), the singular vector (SV) method, the Lyapunov vector method (LLV), and the random field (RF) method. We compare the methods theoretically and through numerical simulations with two low-dimensional non-linear models; the Lorentz-63 model and the Sabra shell model of turbulence. In evaluating the methods, we focus on the spatial/spectral distribution of the perturbations and the ability of the methods to produce large error growth relative to a reference. With the Lorentz-63 model, we find the largest exponential growth rates of the error with the SV method. In general, we observe good agreement with similar studies for all methods except the BV and RF method. For the Sabra shell model, we find the largest exponential growth rates of the error with the BV-EOF and the LLV method.

Supervisor: Peter Ditlevsen