Improving nonlinear observation assimilation through iterative filters: idealized experiments with radar-like operators

• Autores: Gacitua Gutiérrez, Jorge; Ruiz, Juan; Pulido, Manuel; Dillon, María Eugenia; García Skabar, Yanina; Maldonado, Paula; Otsuka, Shigenori; Amemiya, Arata; Pajarola, Renato; Miyoshi, Takemasa
• Año de publicación: 2025
• Idioma: inglés
• Tipo de recurso: documento de conferencia
• Materia: Meteorología; ensemble Kalman filter; nonlinearity; radar data assimilation
• Institución: Universidad Nacional de La Plata

Assimilating nonlinear observations—such as radar reflectivity—into ensemble-based systems remains a challenge. Standard ensemble Kalman filters, including LETKF, are derived under assumptions of linearity and Gaussianity. However, when the observation operator is highly nonlinear, it often leads to filter divergence or poor performance (Lawson and Hansen, 2004). Recent advances have explored the use of iterative or tempered techniques to enhance robustness in such regimes (Carrassi et al., 2018). Likelihood tempering, in particular, has shown promise by progressively assimilating observational information in multiple steps, thereby reducing the impact of nonlinearity in any single update (Kurosawa and Poterjoy, 2021). In this study, we apply and evaluate this technique in a simplified yet dynamically relevant context using the Lorenz-96 model with a radar-like observation operator.
Facultad de Ciencias Astronómicas y Geofísicas