Aligning Initial Conditions With Long‐Term Model Dynamics to Enhance 4DVar‐Driven Medium‐Range Global Forecasts

Abstract Applying the four‐dimensional variational (4DVar) method in nonlinear systems faces conflicts between a lack of long‐term dynamical information using a short assimilation window and a trap of local minima using a long window. Here, an adjusted analysis approach that incorporates a thought of pseudo‐orbit analysis adjusting into 4DVar was proposed, resulting in the enhanced consistency between the initial condition (IC) and the long‐term dynamics of model, and the reduced risk of falling into local minima. This approach tries to find a truer orbit of weather state by minimizing the distance from the 4DVar‐based pseudo‐orbit to the long‐term model trajectory. Sensitivity experiments based on observing system simulation experiment reveal that the new approach with a proper iterative number can produce more reasonable analysis than 4DVar. Real observation experiments further exhibit that this approach can help 4DVar produce ICs that are more consistent with the long‐term dynamics and benefit the forecast skills.