Ensemble Kalman filters for large geophysical applications

At the same time that the Kalman filter (KF) was being developed, the United States produced its first operationally useful numerical weather prediction in 1958 using a single-layer barotropic model. Over the next 40 years, incredible increases in computing power have allowed numerical weather prediction (NWP) models to become far larger and more comprehensive so that modern models have tens of millions of state variables. Throughout this period, operational NWP models always taxed the capabilities of the largest computer systems available.