Data assimilation

Data assimilation combines observations with a model’s predictions to estimate the evolving state of a system. It accounts for uncertainty in both sources of information.

Bayesian inference: A framework for updating probabilities using evidence and prior information. Data assimilation updates a model-based estimate with observations using the same logic of prior estimates and new evidence.
Numerical weather prediction: Forecasting atmospheric conditions by numerically solving equations for the atmosphere. It is a major modeling setting in which data assimilation initializes evolving simulations.
Weather forecasting: The prediction of future atmospheric conditions from observations and physical models. Weather forecasts depend on assimilated observations to estimate the atmosphere at forecast start.
Data fitting: The adjustment of a model or its parameters to reduce discrepancies with observed data. Unlike ordinary fitting, assimilation estimates a changing state while incorporating model dynamics and uncertainty.
Data assimilation in chaotic systems: The estimation of states in systems whose trajectories diverge rapidly from small initial differences. Chaotic growth can quickly amplify analysis errors and challenge forecast reliability.
Kalman filter: An algorithm that estimates a changing state in a linear system with Gaussian errors. It gives a foundational recursive method for combining model forecasts and observations.
State estimation: The inference of a system’s hidden state from incomplete, noisy measurements. Data assimilation is state estimation when a dynamical model supplies information between observations.
Ocean data assimilation: The integration of ocean observations with circulation models to estimate ocean conditions. It applies assimilation to sparse, irregular measurements of currents, temperature, and salinity.
Numerical simulation: The computational imitation of a system’s behavior using a mathematical model. A simulation can evolve from assumed conditions without using observations to correct its state.
Uncertainty quantification: The assessment of uncertainty in model inputs, outputs, and predictions. Reliable assimilation requires credible estimates of observation and forecast uncertainty.

Linked from 11 pages