Attention-based Convolutional Autoencoders for 3D-Variational Data Assimilation
This work addresses the high computational cost of 3D Variational Data Assimilation for practitioners in environmental modeling and other fields requiring efficient data assimilation.
This paper introduces a 'Bi-Reduced Space' approach for 3D Variational Data Assimilation using Convolutional Autoencoders. The method significantly reduces the computational complexity while maintaining data assimilation accuracy, achieving an O(10^3) reduction in background covariance matrix representation size and improved accuracy compared to existing reduced space methods in a real-world pollution model.
We propose a new 'Bi-Reduced Space' approach to solving 3D Variational Data Assimilation using Convolutional Autoencoders. We prove that our approach has the same solution as previous methods but has significantly lower computational complexity; in other words, we reduce the computational cost without affecting the data assimilation accuracy. We tested the new method with data from a real-world application: a pollution model of a site in Elephant and Castle, London and found that we could reduce the size of the background covariance matrix representation by O(10^3) and, at the same time, increase our data assimilation accuracy with respect to existing reduced space methods.