Constraining the response of the hydrological cycle, land surface and regional weather to global change (HYDRA)

气专委CMIP-3和CMIP-5模型的相互比较侧重于大气-海洋耦合系统对特定排放量或浓度情景的大尺度温度响应的不确定性，将缩小尺度、水文循环和生物圈的不确定性视为额外的误差来源。这种完全耦合的方法只允许相对粗分辨率的全球模型和有限的合奏大小，大的系统误差和低信噪比的水文变量，持续在初始化的十年期预报的问题。因此，降水观测在制约降水预报方面只发挥了很小的作用，导致极端降雨发生频率等高影响变量的不确定性范围很大，而且可能是非物理性的。因此，CMIP风格的模拟被认为是我们的关键利益相关者之一，保险风险建模的用途有限。这种不确定性在很大程度上可能是不必要的，因为我们知道海表温度和冰盖（SSTIC）在过去几十年中是如何演变的，以及水文循环是如何响应的，因此我们应该直接使用这些信息来约束大气和陆地表面参数。此外，所有研究都表明，在未来几十年内，大规模外部驱动的SSTIC变化的自由度范围有限。因此，一个强大的补充方法CMIP的“排放情景驱动”的范例是“温度情景驱动”的方法下，一系列的大规模的SSTIC的变化被用来驱动更高分辨率的模型，直接或放松在一个简单的模型的海洋混合层。这使得在最近的观测期间可以进行更大的总体观测，减少偏差，从而更系统地探索大气、水文和地表响应的不确定性。与同年的观测结果进行详细比较，包括从卫星获得的大气层顶部通量，应能对大气和地表参数施加比耦合模式自由运行和比较仅限于大尺度气候学和最近趋势时可能施加的约束更严格的约束。模拟降水趋势与观测结果进行定量比较的一个关键挑战是降水特征（如辐合区）位置的系统偏差。我们将使用为神经成像开发的图像扭曲技术来解决这个问题，该技术已在试点规模上得到证明，以纠正气候模型中的特征位置偏差。这些也将提供一个强大的工具来检测外部驱动的功能位置的变化，如哈德利环流的扩张。我们将运行大量的全球大气/陆地表面模型，这些模型由过去60年观测到的SSTIC驱动，以及来自广泛来源的到2040年的预测变化，包括CMIP-3，CMIP-5和UKCP 09以及climateprediction.net扰动物理集合。将利用对过去60年的重复模拟，去除人为影响的估计特征，以确定最近观测到的降水、地表变量和径流变化在多大程度上可归因于人类影响。数千人的集合将使水文和陆地表面变量的分布得以详细绘制，在很大程度上消除了对其基本统计矩的预测中的随机不确定性。这些模拟的一个代表性子集将用于驱动欧洲的嵌套区域模型，并用于驱动径流模型，以评估其效用的洪水和干旱风险建模的输出。建模框架将提供给国际合作伙伴，以解决其他地区的问题。所有模拟将使用climateprediction.net公共资源分布式计算进行，最大限度地减少其成本和环境影响。

The IPCC's CMIP-3 and CMIP-5 model inter-comparisons focus on uncertainty in the large-scale temperature response of the coupled atmosphere-ocean system to a given emissions or concentration scenario, treating uncertainty in downscaling, in the hydrological cycle and in the biosphere as additional sources of error. This fully coupled approach only permits relatively coarse-resolution global models and limited ensemble sizes, with large systematic errors and low signal-to-noise in hydrological variables, problems which persist in initialized decadal forecasts. As a result, precipitation observations play only a minor role in constraining precipitation forecasts, resulting in large and potentially unphysical ranges of uncertainty on high-impact variables such as the frequency of occurrence of extreme rainfall. Hence CMIP-style simulations have been found to be of limited use by what should be one of our key stakeholders, insurance risk modeling. Much of this uncertainty may be unnecessary since we know how sea surface temperatures and ice cover (SSTICs) have evolved over the past few decades and how the hydrological cycle has responded, so we should be using this information directly to constrain atmospheric and land-surface parameters. Moreover, all studies suggest a limited range of degrees of freedom in the large-scale externally-driven SSTIC change over the next few decades. Hence a powerful complementary approach to the CMIP 'emissions scenario driven' paradigm is the 'temperature scenario driven' approach under which a range of large-scale SSTIC changes are used to drive higher-resolution models, either directly or by relaxation in a simple model of the ocean mixed layer. This allows much larger ensembles and reduced bias over the recent observational period, providing a more systematic exploration of uncertainty in the atmospheric, hydrological and land-surface response. Detailed comparison with observations for the same years, including satellite-derived top-of-atmosphere fluxes, should allow much tighter constraints to be placed on atmospheric and land-surface parameters than is possible when coupled models are run free and comparisons are restricted to large-scale climatology and recent trends. A key challenge in quantitative comparison of simulated precipitation trends with observations is systematic biases in the location of precipitation features such as convergence zones. We will address this using image-warping techniques developed for neuro-imaging which have been demonstrated on a pilot scale to correct feature-location biases in climate models. These will also provide a powerful tool to detect externally-driven shifts in feature location, such as an expansion of the Hadley circulation. We will run large ensembles of global atmospheric/land-surface models driven with observed SSTICs over the past 60 years together with projected changes to 2040 derived from a broad range of sources, including CMIP-3, CMIP-5 and the UKCP09 and climateprediction.net perturbed physics ensembles. Repeat simulations of the past 60 years with the estimated signature of anthropogenic influence removed will be used to address how far recent observed changes in precipitation, land-surface variables and run-off can be attributed to human influence. Multi-thousand-member ensembles will allow the distribution of hydrological and land-surface variables to be mapped in detail, largely eliminating stochastic uncertainty from predictions of their underlying statistical moments. A representative subset of these simulations will be used to drive nested regional models over Europe, and the output used to drive run-off models to evaluate their utility for flood and drought risk modeling. The modeling framework will be made available to international partners to address other regions. All simulations will be performed using climateprediction.net public resource distributed computing, minimizing both their cost and environmental impact.