Adaptive Observations: A Modeling Study and Fastex Experiment

小行星9634239 数值天气预报已经发展到进一步改进越来越受到观测基础的限制的阶段。 数值天气预报模式依赖于对观测资料的同化，这些观测资料大多来自固定地点. 不过，在某些极端天气情况下，例如飓风或严重的冬季风暴，特别部署的平台会获得额外的观测数据，因为他们认为额外的观测数据会改善预报。 这一策略在飓风路径预测方面无疑是成功的。 因此，探讨是否有针对性的观测策略可以更广泛地应用于天气预报，以及如何优化这种策略是合乎逻辑的。 主要调查人员将调查改进预报的可能性，方法是在预报最能从额外信息中受益的地点和时间有选择地观察大气。 这些位置是通过对数据同化时分析中出现错误的可能性的实时估计以及对预报误差对有无观测的敏感性的估计来确定的。 作为对分析误差可能性的先验估计，将使用6小时或12小时预报集合的成员之间的传播的量度。 这种测量是所谓的“繁殖模式”的幅度。“初步建模研究表明，与随机定位相同数量的观测值相比，基于这些模式的振幅进行目标观测可显著降低预测误差。 进一步的改进可能会获得有针对性的观测的基础上，伴随模型导出的测量的敏感性预测误差的数据在下一个同化周期。 因此，本研究的主要目标是找到最佳的观测目标定位策略的基础上的先验分析误差估计和伴随推导的预测误差灵敏度。 这些战略将使用一系列数值模型进行测试，并作为与国际锋面和大西洋风暴轨迹实验（FASTEX）结合进行的实地实验的一部分。 ***

9634239 Emanuel Numerical weather prediction has advanced to the stage where further improvements are increasingly limited by the observational base. Numerical weather prediction models depend on the assimilation of observational data collected mostly form fixed locations. However, in certain extreme weather situations such as hurricanes or severe winter storms, additional observations are obtained by specially deployed platforms under the supposition that additional observations will improve the forecasts. This strategy certainly has been successful for hurricane track forecasting. It is then logical to explore whether the strategy of targeted observations could be applied more generally to weather forecasting and how such a strategy could be optimized. The principal investigators will investigate the possibility of improving forecasts by selectively observing the atmosphere at the locations and times where the forecast would most benefit from additional information. These locations are determined by real-time estimates of the likelihood of error in the analysis at the time of data assimilation and estimates of the sensitivity of forecast error to the presence or absence of observations. As an a-priori estimate of the likelihood of analysis error, a measure of the spread among members of an ensemble of 6- or 12-hour forecasts would be used. Such measure is the amplitude of the so-called "bred modes." Preliminary modeling studies show that targeting observations based on the amplitude of these modes leads to a significant reduction in the forecast error compared with randomly locating the same number of observations. Further improvements may be gained by targeting observations based on a model adjoint-derived measure of the sensitivity of forecast error to data in the next assimilation cycle. Thus, the primary goal of this research is to find the optimal observation targeting strategy based on both a-priori analysis error estimate and adjoint-derived forecast error sensitiv ities. These strategies will be tested using a hierarchy of numerical models and as part of a field experiment in conjunction with the international Fronts and Atlantic Storm Track Experiment (FASTEX). ***