DAWIS
A unified data assimilation framework based on windowed inverse sampling. It introduces a multi-task stochastic interpolant and supports three operational modes: filtering, fixed-lag smoothing, and block smoothing. The framework enables retrospective correction of historical state estimates as new observations become available.
Timeline (last 90 days)
DAWIS (Data Assimilation with Windowed Inverse Sampling) is a unified framework that covers three data assimilation modes—filtering, fixed-lag smoothing, and block smoothing—within a single construction.
DAWIS introduces a multi-task stochastic interpolator that operates over a window composed of sequential states, assigning an independent flow time to each state within the window.
One assimilation cycle in DAWIS consists of two steps: first inverting the entire state window into a set of per-state pivot vectors, then regenerating the window guided by observations.
In DAWIS, the pivot points control whether each state is kept fixed, corrected, or fully regenerated from scratch during the regeneration process.
The construction of DAWIS can absorb the prediction step into the assimilation cycle, eliminating the need for an independent prediction model to advance the system state.
DAWIS is evaluated on multiple nonlinear systems and achieves improvements over both filtering and smoothing baseline methods under sparse, noisy, and nonlinear observation conditions.
The code for DAWIS has been open-sourced at github.com/Erik-Wikingsson/DAWIS.
The scalability and computational cost of the DAWIS method in real large-scale operational systems remain to be examined by future work.
All Facts (8)
The scalability and computational cost of the DAWIS method in real large-scale operational systems remain to be examined by future work.
50%UnverifiedDAWIS introduces a multi-task stochastic interpolator that operates over a window composed of sequential states, assigning an independent flow time to each state within the window.
50%UnverifiedOne assimilation cycle in DAWIS consists of two steps: first inverting the entire state window into a set of per-state pivot vectors, then regenerating the window guided by observations.
50%UnverifiedIn DAWIS, the pivot points control whether each state is kept fixed, corrected, or fully regenerated from scratch during the regeneration process.
50%UnverifiedThe construction of DAWIS can absorb the prediction step into the assimilation cycle, eliminating the need for an independent prediction model to advance the system state.
50%UnverifiedDAWIS is evaluated on multiple nonlinear systems and achieves improvements over both filtering and smoothing baseline methods under sparse, noisy, and nonlinear observation conditions.
50%UnverifiedDAWIS (Data Assimilation with Windowed Inverse Sampling) is a unified framework that covers three data assimilation modes—filtering, fixed-lag smoothing, and block smoothing—within a single construction.
50%UnverifiedThe code for DAWIS has been open-sourced at github.com/Erik-Wikingsson/DAWIS.
50%