First Rains: Fast-tracking multiscale prediction of rainfall onset across tropical and subtropical regional climates

雨什么时候开始下？地球表面的广大地区在雨季开始之前经历长达数月的干旱期。这些降雨的开始已经定义了数千年来农业日历的开始，然而，气候变化的快速速度正在颠覆当地关于第一场降雨到来的数百年知识。随着地球变暖到目前的二氧化碳水平，发病前的极端高温正在扩大，延迟发病的风险正在增加;这些风险已经承诺，无论未来的二氧化碳排放量如何。这种延误对水、粮食、卫生和能源系统造成严重影响。First Rains提出了一项研究计划，以快速跟踪预报的进展，并在变化无常的第一场降雨面前取得突破，以解锁强大的气候适应能力。降雨开始是（亚）热带气候的一个戏剧性特征，标志着从干燥的土壤和天空到充满雨水的大气的快速状态转换。大雷雨的到来预示着季节的急剧转换。这种到来的时机对农业经济至关重要，但十多年来，它很少成为预测研究方案的唯一重点。这种缺乏重点的情况部分反映了数值模型，到目前为止，只有估计热带雷暴。然而，最近全球季风理论进展的结果表明，爆发的延迟增加。对降雨延迟的预测在南部非洲最为明显，这是该地区季风研究最少的地区。至关重要的是，很少有研究让当地的预测专家参与到全球理论的区域化中来。在预测科学和动力学理论的差距继续阻止提供迫切需要的决策相关的气候适应工作的开始指标。然而，直接模拟雷暴的尖端新大气模型现在可用，最先进的观测提供了迄今为止对地球系统最全面的估计，机器学习（ML）工具正在提供强大的新方法来探索这些数据。这些都是缩小发病研究差距所需的工具，并在发病预测方面提供急需的进步。First Rains将从两个方面实现这一目标。将与建模中心密切合作，对新的对流尺度大气模型进行严格试验，以确定新发现的提前数天至数周预测发病的能力。识别出的模型弱点将反馈给模型开发人员。仔细分析对流规模的区域动态和可预测性，将确保风险最大的国家获得最大利益。第二条研究线将侧重于改进第一场降雨的时空统计特征，这对于业务决策比单一的确定开始日期更重要。统计ML算法的创新使用将有助于观察和模型中的这种发作特征。ML方法的应用还将提供强有力的方法来确定这些数据中最重要的发病可预测性来源。这些对最先进的地球观测和对流尺度模型的分析将有助于确定从几天到几个月的预测提前期内的预测技能，并指出进一步提高这一技能的目标。将区域性的动力学理论推进到地方尺度将统一对流尺度模型和观测分析方法，在基础预测研究方面取得的突破将在与最易受变化无常的第一场雨影响的国家的专家的密切合作下取得成功。FLF +3年将支持将预测进展纳入整个食品-水-健康关系的现有国内气候适应和传播网络。First Rains将解决一个基本的预测科学问题，并满足一个长期而紧迫的社会需求：生成气候信息，以便有效地适应更温暖的世界。

When will the rains start? Vast regions of Earth's surface experience months-long dry periods before the start of the rainy season. Onset of these rains has defined the start of agricultural calendars for millennia, however, the rapid rate of climate change is upending cen-turies of local knowledge about the arrival of the first rains. Pre-onset heat extremes are amplifying and the risk of delayed onset is increasing as the planet warms to current CO2 levels; these are risks already committed to irrespective of future CO2 emission. Dire impacts on water, food, health and energy systems accompany such delays. First Rains sets out a research programme to fast-track advances in onset prediction and make the breakthroughs integral to unlocking robust climate adaptation in the face of fickle first rains.Rainfall onset is a dramatic feature of (sub)tropical climates signalling a rapid regime switch from desiccated soils and skies to rain-filled atmospheres. This sharp switch between seasons is heralded by arrival of large thunderstorms. Timing of this arrival is critical for agricultural economies and yet it has rarely been a sole focus of prediction research programmes for over a decade. This lack in focus partly reflects numerical models that, until now, only estimated tropical thunderstorms. And yet, results from recent global monsoon theory advances point to increased delays of onset. Projections of delayed rainfall are most stark in southern Africa, the least studied of the regional monsoons. Critically, little research has engaged local forecast experts here in efforts to regionalise global theory. Gaps in both prediction science and dynamical theory continue to prevent provision of urgently needed decision-relevant onset metrics to climate adaptation efforts. However, cutting-edge new atmospheric models that directly simulate thunderstorms are now available, state-of-the-art observations provide the most comprehensive estimates the Earth System to date, and machine learning (ML) tools are providing powerful new ways to explore these data. These are the tools needed to close onset research gaps and deliver the urgently needed advance in onset prediction.First Rains will pursue this goal from two fronts. New convective-scale atmospheric models will be rigorously trialled, in close collaboration with modelling centres, to determine new-found capabilities in predicting onset days to weeks in advance. Identified model weaknesses will be fed back to model developers. Careful diagnosis of convective-scale regional dynamics and predictability will ensure maximum benefit to the most at-risk countries. The second line of research will focus on improving characterisation of the spatio-temporal statistics of the first rains, which are more important for operational decisions than a single defined onset date. Innovative use of statistical ML algorithms will aid this onset characterisation in observations and models. Application of ML methods will also provide powerful ways to determine the most important sources of onset predictability in these data. These analyses of state-of-the-art Earth observations and convective-scale models will help determine prediction skill across forecast lead-times from days to months and point to targets for improving this skill further. Advancing the dynamical theory of regional to local-scale onset will unify the convective-scale modelling and observational analysis approaches.The resulting breakthrough in fundamental prediction research will succeed in close collaboration with experts from countries most exposed to fickle first rains. The FLF +3 years will support uptake of the prediction advances into existing in-country climate adaptation and dissemination networks across the food-water-health nexus. First Rains will solve a fundamental prediction science problem and meet a long-standing and urgent societal need: generating climate information to enable effective adaptation to a warmer world.