uantifying the sensitivity of shallow cumulus cloud fields to a broad range of environmental perturbations

尽管在过去几十年的工作中取得了重大进展，但气候预估仍然受到很大的不确定性的影响，特别是在平衡气候敏感性（ECS）方面，这是衡量全球平衡温度对二氧化碳增加一倍的响应的一种措施。（Schneider et al., 2017）对于给定的排放情景，预测中这种传播的关键原因之一来自浅层海洋云对气候变化的响应，称为“浅层云反馈”。这些云对地球的辐射收支具有特别重要的意义，因为它们的低、暖云顶和高反照率具有净降温影响。（Bony et al., 2004; Bony and Dufresne, 2005; Spill et al., 2019）最近的研究表明，CMIP5模式之间ECS的一半以上差异可以用热带低云短波反射率的变化来解释。（Brient et al., 2016）此外，早期的工作已经证明，浅层云反馈中最大的不确定性来源是暖边界层云对环境变化的敏感性。（Bony和Dufresne， 2005）因此，了解和量化暖浅云场特性对其环境变化的敏感性是至关重要的，以便更好地约束它们在全球气候模式中的不确定性行为。（Schneider et al., 2017）本项目提出以两种方式扩展先前敏感性研究的分析：首先，我们将对瞬态、温暖的浅云场进行一套高分辨率模拟，明确考虑环境和微物理扰动之间可能的共变范围。（Dagan等人，2017年，2018年；Spill等人，2019年）为此，我们将使用拉丁超立方体采样来选择信息量最大的模拟，以便对外部扰动可能组合的不确定性空间进行充分采样。可能干扰的参数包括温度、沉降、CDNC、自由对流层相对湿度或风速。（McKay, Beckman and Conover, 1979; Lee et al., 2011）其次，使用空间填充、信息量最大的模拟集，我们将能够训练和验证高斯过程（GP）模拟器，该模拟器可用于估计跨高维参数空间的云响应。这种拉丁超立方体采样与GP模拟相结合的技术已经在全球气溶胶-气候模式的不确定性量化中取得了巨大的效果，并且该方法已经成熟，可以用于云反馈研究。（Lee et al., 2011, 2012）一旦模拟器被构建和验证，我们将能够轻松地对大范围的环境扰动进行采样，从而使我们能够识别浅层积土层对其环境变化的模拟响应中最敏感的参数。我们还将能够全面探索云对外部参数组合的响应在多大程度上是线性或非线性的，这在以前的研究中是不可能的。最后，预计仿真器方法也将使人们更好地理解CDNC如何变化以及“降水调控器”机制如何作为云反馈的控制。

Despite significant progress being made over previous decades of work, climate projections are still marred by large uncertainties, particularly in the Equilibrium Climate Sensitivity (ECS), a measure of the global equilibrium temperature response to a doubling of CO2. (Schneider et al., 2017)One of the key causes for this spread in predictions - for a given emissions scenario - emanates from the response of shallow, marine clouds to climate change, termed the `shallow cloud feedback`. These clouds are of particular importance for the earth's radiation budget as they have a net cooling influence due to their low, warm cloud tops and high albedo. (Bony et al., 2004; Bony and Dufresne, 2005; Spill et al., 2019)Recent work has shown that over half the variance of ECS between CMIP5 models can be explained by changes in the shortwave reflectance of tropical low clouds. (Brient et al., 2016) Additionally, early work has demonstrated that the single greatest source of uncertainty within the shallow cloud feedback is the sensitivity of warm, boundary layer clouds to changes in their environment. (Bony and Dufresne, 2005)Hence, it is critical to understand and quantify the sensitivity of warm, shallow cloud field properties to changes in their environment, so as to better constrain their uncertain behaviour in global climate models. (Schneider et al., 2017)this project proposes to extend the analysis of previous sensitivity studies in two ways:Firstly, we will conduct a suite of high-resolution simulations of transient, warm shallow cloud fields which explicitly consider the possible range of covariation between both environmental and microphysical perturbations. (Dagan et al., 2017, 2018; Spill et al., 2019) To do this we will use Latin Hypercube sampling to select the maximally informative simulations to conduct in order to adequately sample the uncertainty space of possible combinations of external perturbations. Possible parameters to perturb could include temperature, subsidence, CDNC, free-tropospheric relative humidity or windspeed. (McKay, Beckman and Conover, 1979; Lee et al., 2011) Secondly, using the space-filling, maximally informative set of simulations conducted, we will be able to train and verify a Gaussian process (GP) emulator which can be used to estimate the cloud response across the high-dimensional parameter space. This technique of combining Latin hypercube sampling with GP emulation has already been used to great effect in the quantification of uncertainty in global aerosol-climate models, and the approach is ripe for use in cloud feedback studies. (Lee et al., 2011, 2012) Once the emulator is constructed and verified, we will be able to easily sample a wide range of environmental perturbations, allowing us to identify the most sensitive parameters in the modelled response of shallow cumuli to changes in their environment. We will also be able to comprehensively explore to what extent the cloud response to a combination of external parameters is linear or non-linear, which has not been possible in previous studies. Finally, it is expected that the emulator approach will also enable a better understanding of how CDNC changes and the "precipitation governor" mechanism act as a control on cloud feedbacks.