Analysis of dynamic data assimilation for atmospheric phenomena. Effect of the model order

Abstract

The problem of weather forecasting based on the use of a priori information, from a first principles model, in conjunction with on-line measurements is addressed. Our derivation of an estimation algorithm evidences that understanding and exploitation of available estimation techniques require ingredients from physical knowledge of the process, tools to represent partial differential equations (as a reduced set of ordinary differential equations), numerical error propagation and conditional estimation in statistics. It is concluded that updating of the model error statistics and the number and location of measurements play an important role on the estimator performance. A suitable treatment of the preceding issues leads to an estimator whose accuracy, size, numerical treatment, computational effort and measurement mesh can be chosen so that performance is guaranteed while keeping simplicity and ease of implementation.