Assimilation of temperature into an isopycnal ocean general circulation model using a parallel ensemble Kalman filter

Temperature data from the Tropical Atmosphere and Ocean (TAO) array are assimilated into the Pacific basin configuration of the Poseidon quasi-isopycnal ocean general circulation model (OGCM) using a multivariate ensemble Kalman filter (EnKF) implemented on a massively parallel computer architecture. An assimilation algorithm whereby each processing element (PE) solves a localized analysis problem is used. The algorithm relies on a locally supported error-covariance model to avoid the introduction of spurious long-range covariances associated with small ensemble sizes and to facilitate its efficient parallel implementation on a computing platform with distributed memory. Each time data are assimilated, multivariate background-error statistics estimated from the phase-space distribution of an ensemble of model states are used to calculate the Kalman gain matrix and the analysis increments. The resulting cross-field covariances are used to compute temperature, salinity and current increments. The layer thicknesses are left unchanged by the analysis. Instead, they are dynamically adjusted by the model between successive analyses. Independent acoustic Doppler current profiler data are used to assess the performance of the temperature data assimilation. The temperature analyses are also compared to analyses obtained with a univariate optimal interpolation (UOI) algorithm and to a control run without temperature assimilation. The results demonstrate that the multivariate EnKF is both practical and effective for assimilating in situ and remotely sensed observations into a high resolution ocean model in a quasi-operational framework.