Identifying the radiation belt source region by data assimilation

[1] We describe how assimilation of radiation belt data with a simple radial diffusion code can be used to identify and adjust for unknown physics in the model. We study the dropout and the following enhancement of relativistic electrons during a moderate storm on 25 October 2002. We introduce a technique that uses an ensemble Kalman filter and the probability distribution of the forecast ensemble to identify if the model is drifting away from the observations and to find inconsistencies between model forecast and observations. We use the method to pinpoint the time periods and locations where most of the disagreement occurs and how much the Kalman filter has to adjust the model state to match the observations. Although the model does not contain explicit source or loss terms, the Kalman filter algorithm can implicitly add very localized sources or losses in order to reduce the discrepancy between model and observations. We use this technique with multisatellite observations to determine when simple radial diffusion is inconsistent with the observed phase space densities indicating where additional source ( acceleration) or loss ( precipitation) processes must be active. We find that the outer boundary estimated by the ensemble Kalman filter is consistent with negative phase space density gradients in the outer electron radiation belt. We also identify specific regions in the radiation belts (L* approximate to 5 - 6 and to a minor extend also L* approximate to 4) where simple radial diffusion fails to adequately capture the variability of the observations, suggesting local acceleration/ loss mechanisms.