Localization With Satellite Radiances In An Ensemble Kalman Filter For Tropical Cyclones

China is among the countries that are severely influence by tropical cyclones.These disastrous weather systems threaten the lives and properties of residents in coastal regions.In adapting the impact of tropical cyclones,it is crucial to accurately forecast their tracks and intensities.A widely used data assimilation algorithm is called the ensemble Kalman filter,which has proved its validity by improving tropical cyclone forecasts through assimilating radiance observation.Yet given the current computational ability,a localization algorithm is necessary to limit the impact of observations only to state variables that are physically close to the observations.However,as satellites observe the integral column from above,their observation locations and impact scopes are unclear.This study implements an adaptive localization algorithm along with the model space localization scheme in a regional assimilation-forecast system,and refer their effects to the traditional observation space Gaspari-Cohn scheme.First,the adaptive localization is implemented through the Global Group Filter(GGF)scheme,with additional experiments using adaptive localization functions that separate TC and non-TC regions or that vary with time.These experiments better capture the multi-scale and time-varying features of tropical cyclones.The GGF method estimates localization functions for each satellites and channels,by minimizing the sampling errors of correlations between observations and state variables.The adaptive localization is applied in the case of Yutu(2018).When forecasts with different localization schemes are verified against observations,the adaptive schemes have smaller errors compared to the control experiment(without adaptive schemes),especially in regions closed to the tropical cyclone track.The adaptive methods also produce better tropical cyclone forecasts.The rapid intensification process is better simulated,and the structures of tropical cyclone Yutu(2018)is more coherent,including stronger warm cores,wetter conditions near the eyewall,and increased radial wind and tangential wind.In general,the adaptive localization functions,especially those separating TC and non-TC regions and those varying with time,are proved to have great advantages over the traditional methods.The differences between model space localization and observation space localization is also compared in the paper.The model space localization is implemented through a modulation procedure.The raw ensembles multiple the eigenvalues and eigenvectors of the localization matrix to create an extended ensemble.This extended ensemble is then the new background for assimilation.The advantages of model space localization is already proved in the NCEP FV3 Global Forecast System(GFS).However,the behaviors of the model space scheme in the regional model is not in accordance with those in the global model.When forecasts are verified against satellite radiances,model space localization has smaller errors.It also produces smaller errors when verified against the conventional wind observations.However,model space localization is worse than observation space localization when forecasts are verified against conventional temperature and water vapor observations.As for tropical cyclones,the model space localization only slightly improves track forecasts,while it barely improves intensity forecasts.Several reasons can account for this phenomenon,such as error from topography and the number of eigenvalues for localization matrix.State variables with significant topographic characteristics have significant different surface pressure than those without.The localization matrix used for this site differs more from the true matrix.In addition,the number of eigenvalues limits the performance of model space localization as well.The errors are larger in regions where background covariances are large,the localization is less accurate when the model has lower predictability.In conclusion,this paper proves the advantages of adaptive localization in a regional model for tropical cyclone forecasts.The model space localization in a regional model performs not as well as that in the global model.Topography and localization matrix cut-off can account for the inferior forecasts.