Study of the Jacobian of an Extended Kalman Filter for soil analysis in SURFEXv5

Abstract

Abstract. An externalised surface scheme like SURFEX allows computationally cheap offline runs. This is a major advantage for surface assimilation techniques such as the Extended Kalman Filter (EKF), where the offline runs allow a cheaper numerical estimation of the observation operator Jacobian. In the recent past an EKF has been developped within SURFEX for the initialisation of soil water content and soil temperature based on screen-level temperature and relative humidity observations. In this paper we make a comparison of the Jacobian calculated with offline SURFEX runs and with runs coupled to the atmospheric ALARO model. Comparisons are made with respect to spatial structure and average value of the Jacobian, gain values and increments. We determine the optimal perturbation size of the Jacobian for the offline and coupled approaches and compare the linearity of the Jacobian for these cases. Results show that the offline Jacobian approach gives similar results as the coupled approach and it allows for smaller perturbation sizes that better approximate this linearity assumption. We document a new case of non-linearities that can hamper this linearity assumption and cause spurious 2Δt oscillations in small parts of the domain for the coupled as well as the offline runs. While these oscillations do not have a detrimental effect on the model run, they can introduce some noise in the Jacobian in the affected locations. The oscillations influence both the surface fluxes and the screen-level variables. The oscillations occur in the late afternoon in summer when a stable boundary layer starts to form near the surface. We propose a filter to remove the oscillations and show that this filter works accordingly.