A Comparison of Sea Surface Temperature Perturbation Methods for a Convection Permitting Ensemble Prediction System Over the European Arctic

This article investigates two different methods for perturbing sea surface temperature (SST) in a convection permitting ensemble prediction system based on the AROME-Arctic NWP model. The methods are one that results in perturbations that are purely randomly located and one in which the perturbations are targeted towards locations where the SST errors are thought to be largest. The impact of the magnitude of the perturbations is also tested by scaling the randomly located perturbations to have a similar L1 norm to the targeted perturbations. The impact of the SST error estimate is tested by comparing the method of targeting SST perturbations based on different SST uncertainty estimates. The methods are tested for four high impact weather events over the European Arctic – a polar low, two cold air outbreaks and a severe storm and are verified against near surface observations over land, scatterometer wind speeds over the ocean and against the operational analyses of the model under investigation. It is shown that targeted perturbations generally result in better verification scores when compared with randomly located perturbations. Especially over the ocean it appears that targeting the locations of largest uncertainty can lead to an increased spread without impacting the route mean square error. The results suggest that the impact of SST perturbations over land may be more related to the magnitudes of the perturbations regardless of location, while over the ocean the location of the perturbations becomes more important.