We propose a targeted observation strategy, based on the influence matrix diagnostic, that optimally selects where additional observations may be placed to improve ionospheric forecasts. This strategy is applied in data assimilation observing system experiments, where synthetic electron density vertical profiles, which represent those of Constellation Observing System for Meteorology, Ionosphere, and Climate/Formosa satellite 3, are assimilated into the Thermosphere‐Ionosphere‐Electrodynamics General Circulation Model using the local ensemble transform Kalman filter during the 26 September 2011 geomagnetic storm. During each analysis step, the observation vector is augmented with five synthetic vertical profiles optimally placed to target electron density errors, using our targeted observation strategy. Forecast improvement due to assimilation of augmented vertical profiles is measured with the root‐mean‐square error (RMSE) of analyzed electron density, averaged over 600 km regions centered around the augmented vertical profile locations. Assimilating vertical profiles with targeted locations yields about 60%–80% reduction in electron density RMSE, compared to a 15% average reduction when assimilating randomly placed vertical profiles. Assimilating vertical profiles whose locations target the zonal component of neutral winds (Un) yields on average a 25% RMSE reduction in Un estimates, compared to a 2% average improvement obtained with randomly placed vertical profiles. These results demonstrate that our targeted strategy can improve data assimilation efforts during extreme events by detecting regions where additional observations would provide the largest benefit to the forecast.

We assess the performance of an ensemble Kalman filter for data assimilation and forecasting of ion density in a model of the ionosphere given noisy observations of varying sparsity. The domain of the numerical model is a mid-latitude ionosphere between 80 and 440 km. This domain includes the D–E layers and the peak in the F layer in the ionosphere. The model simulates the time evolution of an ion density field and the coupled electrostatic potential as charge-neutral winds from gravity waves propagate up from the stratosphere. Forecasts are generated for an ensemble of initial conditions, and synthetic observations, which are generated at random locations in the model domain, are assimilated into the ensemble at time intervals corresponding to about a half-period of the gravity wave. The data assimilation scheme, called the local ensemble transform Kalman filter (LETKF), incorporates observations within a fixed radius of each grid point to compute a unique linear combination of the forecast ensembles at each grid point. The collection of updated grid points forms the updated initial conditions (analysis ensemble) for the next forecast. Even when the observation density is spatially sparse, accurate analyses of the ion density still can be obtained, but the results depend on the size of the local region used. The LETKF is robust to large levels of Gaussian noise in the observations. Our results suggest that the LETKF merits consideration as a data assimilation scheme for space weather forecasting.

The dynamics of many models of physical systems depend on the choices of key parameters. This paper describes the results of some observing system simulation experiments using a first-principles model of the Earth’s ionosphere, the Thermosphere Ionosphere Electrodynamics Global Circulation Model (TIEGCM), which is driven by parameters that describe solar activity, geomagnetic conditions, and the state of the thermosphere. Of particular interest is the response of the ionosphere (and predictions of space weather generally) during geomagnetic storms. Errors in the overall specification of driving parameters for the TIEGCM (and similar dynamical models) may be especially large during geomagnetic storms, because they represent significant perturbations away from more typical interactions of the earth-sun system. Such errors can induce systematic biases in model predictions of the ionospheric state and pose difficulties for data assimilation methods, which attempt to infer the model state vector from a collection of sparse and/or noisy measurements. Typical data assimilation schemes assume that the model produces an unbiased estimate of the truth. This paper tests one potential approach to handle the case where there is some systematic bias in the model outputs. Our focus is on the TIEGCM when it is driven with solar and magnetospheric inputs that are systematically misspecified. We report results from observing system experiments in which synthetic electron density vertical profiles are generated at locations representative of the operational FormoSat-3/COSMIC satellite observing platforms during a moderate (G2, Kp = 6) geomagnetic storm event on September 26–27, 2011. The synthetic data are assimilated into the TIEGCM using the Local Ensemble Transform Kalman Filter with a state-augmentation approach to estimate a small set of bias-correction factors. Two representative processes for the time evolution of the bias in the TIEGCM are tested: one in which the bias is constant and another in which the bias has an exponential growth and decay phase in response to strong geomagnetic forcing. We show that even simple approximations of the TIEGCM bias can reduce root-mean-square errors in 1-h forecasts of total electron content (a key ionospheric variable) by 20–45%, compared to no bias correction. These results suggest that our approach is computationally efficient and can be further refined to improve short-term predictions (∼1-h) of ionospheric dynamics during geomagnetic storms.