Northwest European Seasonal Weather Prediction from Complex Systems Modelling

北大西洋上空的大气环流和急流（巨大的气流）强烈影响欧洲西北部的季节性天气状况。最近的极端季节以独特的急流模式为特征，而急流的强度和位置与英国和西北欧的极端天气状况密切相关。季节性天气特征对人们的生计和经济有重大影响，例如，2013/14年冬季英国约有15亿英镑的损失，因此提前几个月做出可靠的季节性预报将对社会有重大好处。季节性天气条件也对农业、粮食安全、能源供应、公共卫生/福祉和恶劣天气规划产生重大影响。直到最近，北大西洋大气的变化被认为主要是由于不可预测的波动。然而，在巨型超级计算机上运行的动态（即基于物理的）季节预报系统最近在预报技能方面取得了一些进展，主要是冬季预报。影响北大西洋大气环流和急流变化的因素很多；可能的影响可大致分为海冰范围和积雪变化、北大西洋海面温度变化、热带影响（如El-Niño南方涛动）、高层大气（平流层）环流变化、太阳能量变化和火山爆发等影响。这些急流变率的驱动因素可以相互反对或加强，它们之间有相互作用的迹象。急流变率的驱动因子呈现季节变化，不同季节的急流变率驱动因子具有明显的特征。虽然一些观测到的驱动因素可以在气候系统的计算机模型中重现，但提高对最近确定的北大西洋急流驱动因素的理解对于在西北欧季节性气候预测中取得进展至关重要。政府资助的研究重点是动态预报系统；然而，这样的预测并不总是准确的。此外，尽管最近努力评估和改进它们的性能，但动态模式预报在夏季表现不佳。在中纬度地区，包括英国和西北欧，统计预测一直被忽视；然而，最近在“机器学习”的保护伞下，先进统计技术的发展已经发生在气候科学界之外，并且相对快速和廉价地实施。因此，将复杂的统计方法应用于季节预报问题有相当大的余地。使用一种名为NARMAX（一种机器学习，其结果是高度可解释的）的已建立的复杂系统建模方法的新应用，该项目旨在显著改善当前的季节性预测，将熟练的季节性预测扩展到冬季以外的季节，确定有助于预测技能的因素，在区域基础上开发西北欧的季节性预测。并评估熟练的概率季节性预测对感兴趣的最终用户（如农业食品行业）的好处。我们的项目计划有效地建立在我们最近在《皇家气象学会季刊》上发表的有希望的试点研究结果的基础上。我们对NARMAX的新应用可能会显著提高预报技能，并有助于为下一代动态季节预报系统的开发提供信息。我们还寻求吸引季节性预测的最终用户，主要关注改进的季节性预测对农业食品工业的影响：反映我们在这一领域的联系，但也因为与其他关键领域相比，它的研究相对较少。

The atmospheric circulation and jet stream (giant current of air) over the North Atlantic strongly influence seasonal weather conditions over Northwest Europe. Recent extreme seasons have been characterised by distinctive jet stream patterns, and jet strength and location is closely linked with extreme weather conditions experienced across the UK and Northwest Europe. Seasonal weather characteristics have major effects on people's livelihoods and the economy, for example about £1.5 billion in the UK in winter 2013/14, so producing reliable seasonal forecasts some months ahead would have significant benefits for society. Seasonal weather conditions also have major impacts on agriculture, food security, energy supply, public health/wellbeing, and severe weather planning. Until recently, North Atlantic atmospheric variability was thought to be largely due to unpredictable fluctuations. However, dynamical (that is, physics-based) seasonal forecasting systems run on giant supercomputers have led to some recent advances in forecasting skill, mainly for winter forecasts. Many factors appear to influence North Atlantic atmospheric circulation and jet-stream changes; possible influences can be broadly grouped into effects from variations in sea-ice extent and snow cover, North Atlantic sea-surface temperature variations, tropical influences such as the El-Niño Southern Oscillation, changes in the higher atmosphere (stratosphere) circulation, changes in energy from the Sun, and volcanic eruptions. These drivers of jet stream variability can oppose or reinforce one another, and there are indications of interactions between them. Drivers of jet-stream variability show seasonal variation, and distinctive drivers of jet-stream variability operate in different seasons. While some observed drivers can be reproduced in computer models of the climate system, improved understanding of more recently identified drivers of the North Atlantic jet stream is crucial for making progress in Northwest Europe seasonal climate predictions.The focus of government-funded research is on dynamical forecast systems; however, such forecasts are not always accurate. Furthermore, despite recent efforts to assess and improve their performance, dynamical model forecasts show little skill in summer. In the mid latitudes, including the UK and Northwest Europe, statistical forecasting has been neglected; however, recent developments in advanced statistical techniques, under the umbrella of 'machine learning', have taken place outside the climate-science community and are relatively quick and cheap to implement. There is thus considerable scope for applying complex statistical methods to the seasonal forecasting problem. Using a novel application of an established complex systems modelling approach called NARMAX (a type of machine learning, the results of which are highly interpretable), this project seeks to significantly improve current seasonal forecasts, extend skillful seasonal forecasting to seasons beyond winter, identify factors that contribute skill to the forecast, develop seasonal forecasts for Northwest Europe on a regional basis, and assess the benefits of skillful probabilistic seasonal forecasts to interested end users such as the agri-food industry. Our project plan effectively builds on promising pilot study results that we have recently published in the Quarterly Journal of the Royal Meteorological Society. Our novel application of NARMAX is likely to significantly improve forecast skill and help to inform development of the next generation of dynamical seasonal forecasting systems. We also seek to engage end users of seasonal forecasts, focusing mainly on the effects of improved seasonal forecasts on the agri-food industry: reflecting our links in this field but also because it has been relatively little studied compared with other key areas.