干旱区生态环境知识资源中心

A new Kalman filter is presented in this work, enhanced with abilities and tools obtained from the area of Information Geometry (IG), for the optimization of the results of numerical weather prediction models. The new developments allow the better estimation of the discrepancies between modeled and corresponding observed data sets by categorizing them in the appropriate geometric framework and utilizing the associated geodesics/minimum-length-curves in order to estimate and minimize the necessary cost functions, instead of adopting least square based approximations. Numerical analysis techniques are developed and used for solving the associated boundary value problems. The proposed new IG Kalman filter is evaluated over idealized and real world test cases with very promising results. The systematic error of the simulations is almost eliminated and the forecast uncertainty is critically reduced.