Improved Techniques for Targeting Additional Observations to Improve Forecast Skill.

摘要

LONG-TERM GOAL. This project aims at studying advanced targeting techniques to improve weather forecast skill and address the question of the optimization of composite observing systems with adaptive components. OBJECTIVES. In numerical weather prediction (NWP) the computational representation of the atmosphere produces unavoidable uncertainties in both the equations of motion (model error) and the initial state (initial error). These limitations imply forecast errors and uncertainty. Observation targeting aims at better describing the initial state, by using additional observations. The targeted locations are chosen beforehand such that they fall in regions for which a given forecast is sensitive. See Joly (2002) for a overview of the principles of observation targeting and its philosophy. The so-called sensitive regions where to deploy supplementary observations can be computed with several objective techniques. In this work, the computation of sensitive regions is done with the adjoint technique and uses the singular vectors of the propagator of the tangent-linear forecast model (Palmer et al., 1998). Observation targeting has been implemented in several field experiments since 1997. The objective techniques used singular vectors and other adjoint based methods, or ensemble based approaches. Following FASTEX in 1997, adaptive observations have been deployed each winter over the Northern Pacific during the campaigns NORPEX and WSRP (Szunyogh, 2000). It has been verified that the deployment of additional adaptive observations and their assimilation improve the forecasts, on average. However, strong variability among the cases were detected. To make the adaptive observations more efficient, the newly developed targeting techniques had to be optimized. Hence, it has been shown that the impact of targeting depends on assimilation scheme used (Bergot, 2001).