Assimilation of surface temperature in a land-surface model

An extended Kalman filter (EKF) scheme, for the assimilation of near-surface temperature observations in a hydrological model, is developed. The formulation is based on the modification of the diffusion equation of heat flux into the ground. Both model and measurement uncertainties are incorporated. It is found that, in addition to the model error, the accuracy in specification of the initial error covariance has an important bearing on the performance of the assimilation scheme. An inadequate specification can result in decreased rather than enhanced model performance. Study suggests that an equilibrium error covariance, arrived at by allowing the model to run with an arbitrary value for a long enough time such that the resulting perturbations subside, can capture the basic correlation structure between the different layers. An arbitrarily scaled equilibrium error covariance, to capture the large initial uncertainty, provides significantly improved performance.