TY - JOUR
TI - Outlier-robust 4-D variational data assimilation based on κ-generalized Gaussian statistics and Kaniadakis entropy
AU - Luo Yu-Chen
AU - Cao Xiao-Qun
AU - Zhou Meng-Ge
AU - Guo Ya-Nan
AU - Peng Ke-Cheng
JN - Thermal Science
PY - 2026
VL - 30
IS - 2
SP - 983
EP - 992
PT - Article
AB - In variational data assimilation, traditional methods generally presuppose  that both background errors and observation errors are distributed according  to the Gaussian distribution. While this assumption simplifies mathematical  operations and facilitates optimization, it demonstrates notable limitations  in practical applications, particularly when dealing with non-Gaussian  errors and outliers. This study proposes an innovative approach to  constructing the objective function based on Kaniadakis entropy and the  κ-generalized Gaussian distribution. The aim of this approach is to enhance  the robustness of the 4-D variational data assimilation method. The  incorporation of Kaniadakis entropy into the objective function effectively  characterizes heavy-tailed error distributions, thereby significantly  improving the suppression of outliers and observation errors and optimizing  the stability of the assimilation process. The proposed method was validated  using the Lorenz-63 model. The findings indicate that the objective function  founded upon the κ-generalized Gaussian distribution attains a substantial  diminution in root mean square error when processing observation data that  encom-passes outliers. Furthermore, the assimilation results demonstrate  enhanced stability and accuracy. This study provides a theoretical  framework and methodological support for non-Gaussian error modeling and  robust data assimilation, offering a novel perspective for data assimilation  research in complex environments.
DO - 10.2298/TSCI2602983L
ER -