トップページ 計算科学研究センターとは 人材育成 計算科学eラーニング データ同化 International Symposium/School on Data Assimilation International Symposium on Data Assimilation-Online (ISDA-Online) "Feature Data Assimilation - Theory, Algorithms and Examples"
"Feature Data Assimilation - Theory, Algorithms and Examples"
The International Symposium on Data Assimilation Online (ISDA-Online) 8 Jan,2021
"Data Assimilation Methodology"
International Symposium on Data Assimilation - Online
The classical setup of atmospheric data assimilation employs direct (in-situ) or indirect remote sensing measurements (e.g. satellite radiances) within a variational data assimilation framework to find the best possible state of some dynamical system as initial condition for forecasting. Modern ensemble data assimilation systems exploit the covariance information provided by an ensemble of forcasted states within their analysis step. Hybrid approaches such as EnVAR combine the advantages of both worlds. When classical continuous variables are measured directly or indirectly, the classical approach has been proven to provide high-quality assimilation and forecasting results.
With the growth of temporally and spatially high-resolution measurement data such as 3D-Volume RADAR as well as hyper-spectral satellite radiances and further temporally high-resolved remote sensing techniques, both variational as well as ensemble-based approaches are significantly challenged by the strong non-linearity of atmospheric processes linked to cloud formation and precipitation processes. The complex processes, which take place for example in thunderstorms, do not allow to fully fit the full dynamical behaviour of a process to observed phenomena based onmeasurements.
The goal of feature or object data assimilation is to move from the assimilation of snapshots of some process recorded by classical direct or indirect measurements to the assimilation of properties or features of a whole process. Often, the process leads to the formation of objects such as clouds or thunderstorms, which have a typical behaviour with a life cycle which consists of birth, growth, stability and decay.
Here, we describe a proper mathematical framework for the description and assimilation of features of phenomena and objects within an ensemble data assimilation systems. Starting from a generic Bayesian approach we describe the natural derivation of feature assimilation methods. We then discuss the design and properties of feature or object forward operators and their use within an ensemble Kalman filter (LETKF) or particle filter (LAPF/LMCPF) based assimilation system. Examples will be shown for the popular Lorenz 63 & 96 benchmark systems as well as for the convective scale ICON model, which is in preparation for operational use of the COSMO consortium with about 40 weather services and, in particular, by Deutscher Wetterdienst from Q1/2021.
講師プロフィール
他の講義を探す
(2021年1月8日)