[M-GI29] Data assimilation: A fundamental approach in geosciences

[MGI29-05] A Multi-scale Localization for the Local Ensemble Transform Kalman Filter with Attenuation of Ensemble Perturbation

*Shunji Kotsuki^1,2, Craig Bishop³ (1.Center for Environmental Remote Sensing, Chiba University, 2.RIKEN Center for Computational Science, 3.University of Melbourne)

Keywords:Data Assimilation, Numerical Weather Prediction, Multi scale, Localization

Recent numerical weather prediction systems have significantly improved medium-range forecasts by implementing hybrid background error covariance, for which climatological (static) and ensemble-based (flow-dependent) error covariance are combined. While the hybrid approach has been investigated mainly in variational systems, Kotsuki and Bishop (2022) constructed hybrid background error covariance for the local ensemble transform Kalman filter (LETKF) by adding collections of climatological perturbations to the forecast ensemble, and succeeded in improving forecast errors using an intermediate global atmospheric model known as SPEEDY. Kotsuki and Bishop (2022) also proposed the new localization method that attenuates the ensemble perturbation (Z-localization) instead of inflating observation error variance (R-localization). While the Z-localization can apply different localization scales for flow-dependent perturbation and climatological static perturbations, the optimal localization for climatological perturbations differed slightly from that for flow-dependent perturbations.
This study extends the previous study, and explores a multi-scale localization using the Z-localization. We generate smoothed ensemble perturbations based on ensemble forecasts, and construct the background error covariance using both original and smoothed ensemble perturbations. Using the Z-localization, larger localization scales are applied for the smoothed ensemble perturbation than that for the original ensemble perturbations. This multi-scale localization with the Z-localization successfully results in smaller forecast errors than the standard LETKF. The multi-scale localization was also combined with the hybrid background error covariance method. This combined approach showed beneficial impacts for reducing forecasts errors. This talk will include the most recent progress up to the time of the conference.