Advances in data assimilation methods for weather and environmental forecasts and climate simulations

在大气和海洋科学中，模型和观测结果被结合在现代数据同化系统中，以产生分析，这是我们对任何给定时间大气状态的最佳估计。从历史上看，需要进行分析来为数值天气预报（NWP）模型提供初始条件，因为天气预报中的误差归因于初始条件的误差。人们很快意识到，将模型模拟与观测结果进行比较可以提供信息来查明模型中的弱点，从而有助于改进 NWP 模型。近年来，数据同化方法取得的大部分进展旨在能够使用目前从多个卫星仪器获得的大量数据以及来自地面仪器的数据。由于数据同化方法取得了重大进步，每 24 小时就会使用数百万个数据进行分析。作者进行的研发最终实现了 4D 同化系统，即所谓的“4D-Var”，该系统能够从时间和空间的大气采样观测中提取信息。风信息可以通过对大气成分的观测来推断。气象天气事件发展的前兆得到了更好的解决，从而减少了重大灾害性天气系统的漏报。二十年来，作者致力于开发和实施新的先进变分方法，这些方法于 1997 年首先成为加拿大环境部业务分析和预报套件的一部分，该项目的第一阶段于 2005 年完成，实施了至今仍在使用的 4D-Var。在数值天气预报取得的成功的基础上，相同的数据同化系统被用来重做最近的分析，即所谓的重新分析，现在是气候模型验证的关键组成部分。我自己的观点是，气候模型应该用于产生短期预测，作为后续分析的先验。重复这个循环，通过与观测值的不断比较，可以获得信息并用于更好地理解物理过程及其相互作用，并根据观测值验证这一点。本提案中提出的研究涉及数据同化方法的特定方面，这些方面可能会影响分析的质量。结果将有助于评估这可能对最终结果产生的影响：改进天气预报和分析。这支持了改善地球系统观测和建模的长期观点，以更好地了解其演化，改善高影响天气预报并通过与观测一致的建模来评估气候变化。现在可以使用准运营数据同化和预测系统进行测试，该系统目前在加拿大计算平台上运行，这是作者在过去六年中领导的一项工作。这为大学研究项目提供了广泛的可能性，而这些项目只能在加拿大环境部、法国气象局或作者合作的欧洲中期天气预报中心等运营数值天气预报研究中心进行。将解决的具体问题涉及集成方法正确捕获导致 4D-Var 成功的时间维度的能力。由于预测重大天气事件的发展极其重要，因此将研究大气不稳定前兆的可观测性，以找出如何设计观测策略来检测它们并改进对此类事件的预报。

In atmospheric and oceanic sciences, models and observations are combined in modern data assimilation systems to produce analyses which are our best estimate of the state of the atmosphere at any given time. Historically, the analyses were needed to provide initial conditions to numerical weather prediction (NWP) models as the error made on weather forecasts were attributed to error in the initial conditions. It was soon realized that comparing model simulations to observations offers information to pinpoint weaknesses in the model which helped to improve the NWP model. Much of the advances made in recent years in data assimilation methods aimed at being able to use the vast amount of data now obtained from several satellite-based instruments on top of those from ground-based instrments. Millions of data are used every 24-h to produce analyses because of the significant advances made in data assimilation methods. The author conducted the R&D which led to the implementation of a 4D assimilation system, the so-called "4D-Var", which is capable to extract information from observations sampling the atmosphere in both time and space. Wind information can be inferred from observations of atmospheric constituents. Precursors to the development of meteorological weather events are better resolved which resulted in a reduction of missed forecasts of significant severe weather systems.For twenty years, the author worked on the development and implementation of new advanced variational methods which have been part of the operational analysis and forecast suite of Environment Canada first in 1997, for the first stage of the project which was completed in 2005 with the implementation of 4D-Var which is still used to this day. Building on the success obtained in NWP, the same data assimilation systems were used to redo the analyses of the recent past, the so-called reanalyses which are now a key component of the validation of climate models. My own view is that climate models should be used to produce short-term forecasts used as an a priori to do the subsequent analysis. This cycle is repeated and this is through this constant comparison to observations that information can be obtained and used to better understand physical processes and their interactions and validate this against observations.The research presented in this proposal is concerned with specific aspects of data assimilation methods which could impact the quality of the analyses. The results will help to make an assessment of the impact this may have on the end result: improving weather forecasts and analyses. This supports the long term view of improving the observation and modelling of the Earth system to better understand its evolution, improve high impact weather forecasts and assess climate changes through modelling consistent with observations. This can now be tested with a quasi-operational data assimilation and forecast system which is now running on Compute Canada platforms, an effort led by the author for the last six years. This opens a wide range of possibilities for research projects in universities which were only possible within operational NWP research centres like that of Environment Canada, Météo-France or the European Centre for Medium-range Weather Forecasts with which the author collaborates. The specific questions will be addressed concerns the ability of ensemble approaches to capture correctly the temporal dimension that was responsible for the success of 4D-Var. As forecasting the development of significant weather events is extremely important, the observability of precursors to atmospheric instability will be examined with a view of finding out how to design an observation strategy that could detect them and improve the forecast of such events.