期刊名称：Tellus A: Dynamic Meteorology and Oceanography
电子版ISSN：1600-0870
出版年度：1998
卷号：50
期号：4
页码：391-410
DOI：10.3402/tellusa.v50i4.14536
摘要：While forecast models and analysis schemes used in numerical weather prediction have becomegenerally very successful, there is an increasing research interest toward improving forecast skillby adding extra observations either into data sparse areas, or into regions where the verifyingforecast is most sensitive to changes in the initial analysis. The latter approach is referred to as‘‘targeting’’ observations. In a pioneering experiment of this type, the US Air Force launcheddropwindsondes over the relatively data sparse Northeast Pacific Ocean during 1–10 February1995. The focus of this study is the forecast sensitivity to initial analysis differences, forced bythese observations by using both the adjoint method (ADJM) and quasi-inverse linear method(QILM), which are both useful for determining the targeting area where the observations aremost needed. We discuss several factors that may affect the results, such as the radius of themask for the targeted region, the basic flow and the choice of initial differences at the verificationtime. There are some differences between the adjoint and quasi-inverse linear sensitivitymethods. With both sensitivity methods it is possible to find areas where changes in initialconditions lead to changes in the forecast. We find that these two methods are somewhatcomplementary: the 48-h quasi-inverse linear sensitivity is reliable in pinpointing the region oforigin of a forecast difference. This is particularly useful for cases in which the ensemble forecastspread indicates a region of large uncertainty, or when a specific region requires careful forecasts.This region can be isolated with a mask and forecast differences traced back reliably. Anotherimportant application for the QILM is to trace back observed 48-h forecast errors. The 48-hadjoint sensitivity, on the other hand, is useful in pointing out areas that have maximum impacton the region of interest, but not necessarily the regions that actually led to observed differences,which are indicated more clearly by QILM. At 72 h, the linear assumption made in bothmethods breaks down, nevertheless the backward integrations are still very useful for pinningdown all the areas that would produce changes in the regions of interest (QILM) and the areasthat will produce maximum sensitivity (ADJM). Both methods can be useful for adaptiveobservation systems.
