A Multiscale Alignment Method for Ensemble Data Assimilation with Displacement Errors

High-resolution models nowadays simulate phenomena across many scales and pose challenges to the design of efficient data assimilation (DA) methods that reduce errors at all scales. Smaller-scale features experience rapid nonlinear error growth that gives rise to displacement errors, which cause sub-optimal DA performance. In this talk, I will introduce a multiscale alignment (MSA) method designed to reduce displacement errors in DA. The method is based on the idea of finding solution incrementally from large to small scales in an iterative manner, which is inspired by the computer vision literature. Consider when nonlinearity at small scales gives rise to position errors of coherent features in the model state, the MSA method performs data assimilation in large scale first, taking advantage of the fact that large-scale errors are more linear, and then utilize the analysis increments at larger scales to reduce the position errors at smaller scales. As a proof of concept, the MSA method is tested with a quasi-geostrophic model, and significant improvement to analysis accuracy is found compared to the existing ensemble DA methods. The MSA method is implemented in the Data Assimilation Research Testbed (DART), and further experiments are being conducted for a tropical cyclone case using the Weather Research and Forecasting (WRF) model. I will discuss the preliminary findings from these experiments and implications for the future development of a multiscale data assimilation framework.