Impact of localization and inflation on geomagnetic data assimilation

In geomagnetic data assimilation (DA), information on the Earth's core magnetic field is combined with numerical dynamo models to estimate the dynamic state of the deep interior and produce forecasts of future magnetic field variations. We present a series of numerical experiments exploring the use of localization and inflation schemes in an Ensemble Kalman Filter (EnKF) based geomagnetic DA system. Localization and inflation schemes are (often necessary) modifications to ensemble-based DA systems, which can improve performance, particularly when computational expense limits ensemble size (the number of simultaneous model runs). We find that the studied localization and inflation schemes enable small ensembles to not only match, but exceed the performance of a larger ensemble. Detailed analysis of the results show that, without localization and inflation, even the larger ensembles make adjustments during assimilation which are either too strong or too weak, and lead to poorer estimates of the dynamo state and forecast uncertainties. The use of localization and inflation allows one to better control the impact of assimilation on elements of the dynamo state, thus mitigating some of these issues.