Statistical postprocessing of ensemble forecasts for various weather quantities

将统计后处理模式应用于数值天气预报（NWP）模式输出的集合，以提高预报的可靠性和确定性预报的校正率，已成为一种普遍的做法。各种国家的最先进的后处理模型是可用的，他们的应用享有越来越多的成功。然而，许多标准模型被设计为在单个空间位置或时间点获得单个天气变量的概率预报。在这种单变量方法中，可能的多变量依赖关系没有明确地解释。我们计划在两个不同的方向上扩展和修改现有的集成后处理模型。一方面，我们将开发适用于非高斯分布天气变量的现有最先进模型的版本。另一方面，我们将把高斯和非高斯模型扩展到一个多元的环境，包括不同天气量、空间位置和时间点之间的依赖关系。在天气预测过程中，对这些类型的依赖结构进行解释性建模，对各种经济和社会部门越来越重要，这些部门需要对未来的天气事件（如风速）进行物理连贯的预测，降水或水文事件作为河流排放。多元统计后处理模型的发展是一个非常活跃的研究领域。因此，我们计划在统计软件环境R中实现开发的模型，并将新的实现与现有的实现相协调，以使现有的和新的模型可供一大群潜在的用户和研究人员使用。

It has become commmon practise to apply statistical postprocessing models on ensembles of numerical weather prediction (NWP) model outputs in to improve the forecast realiability and calibration of the deterministic predictions. A variety of state-of-the-art postprocessing models are available and their application enjoys increasing success. However, many of the standard models are designed to obtain a probabilistic forecast for a single weather variable, at a single spatial locations or time points. In this univariate approaches, possible multivariate dependencies are not explicitly accounted for.We plan to extend and modify existing ensemble postprocessing models in two different directions. On the one hand we will develop versions of existing state-of-the-art models that are suitable for non-Gaussian distributed weather variables. On the other hand we will extend the Gaussian and non-Gaussian models to a multivariate setting incoroprating dependencies among different weather quantities, spatial locations and time points.Explicitly modelling these types of dependence structures in the process of weather prediction is of increasing importance for various economic and social sectors, which require physically coherent predictions of future weather events such as wind speed, precipitation or of hydrological events as river discharge. The development of multivariate statistical postprocessing models is a highly active area of research. Thererfore, we plan to implement the developed models within the statistical software environment R and harmonize the new implementations with existing ones to make existing and new models available to a large group of potential users and researchers.