High-resolution data assimilation for two maritime extreme weather events: a comparison between 3D-Var and EnKF

Abstract

Populated coastal regions in the Mediterranean are known to be severely affected by extreme weather events. Generally, they are initiated over maritime regions, where a lack of in situ observations is present, hampering initial condition estimations and, hence, forecast accuracy. To face this problem, data assimilation (DA) is used to improve the estimation of initial conditions and their respective forecasts. Although comparisons between different DA methods have been performed at global scales, few studies have been conducted at high resolution, focusing on extreme weather events triggered over the sea and enhanced by complex topographic regions. In this study, we investigate the role of assimilating different types of conventional and remote sensing observations using the three-dimensional variational (3D-Var) approach and the ensemble Kalman filter (EnKF), which are the most common DA schemes used globally at national weather centers. To this aim, two different events are chosen because of both the different areas of occurrence and the triggering mechanisms. Both 3D-Var and EnKF are used at convection-permitting scales to improve the predictability of two high-impact coastal extreme weather episodes that were poorly predicted by numerical weather prediction models: (a) the heavy-precipitation event IOP13 and (b) the intense Mediterranean tropical-like cyclone Qendresa. Results show that EnKF and 3D-Var perform similarly for the IOP13 event for most of the verification metrics, although, looking at the receiver operating characteristic (ROC) curve and the area under the ROC curve (AUC) scores, EnKF clearly outperforms 3D-Var. However, the ensemble mean of EnKF is generally worse than that of 3D-Var for Qendresa, although some of the ensemble members of EnKF individually outperform 3D-Var, allowing for information to be gained on the physics of the event and hence the benefits of using an ensemble-based DA scheme.