New Statistical Techniques for Probabilistic Weather Forecasting

行业、机构和公众在预测强降水、暴风雪、洪水或其他破坏性天气现象时做出理性决策，需要有关用户对天气预报的信任程度的信息。因此，提高气象学家量化预报不确定性的能力，以满足社会对可靠信息日益增长的期望，是至关重要的。本研究的长期目标是为下一代概率预测系统奠定方法学基础。具体目标是发展和测试(i)一套用于天气变化概率预测的统计技术，以及（ii）一套用于验证概率预测的性能指标。这项基本技术被称为贝叶斯输出处理器（BPO），它将处理数值天气预报（NWP）模型的输出，并将其与气候数据最佳地融合在一起，以量化预测的不确定性。扩展的技术，称为贝叶斯集成处理器（BPE），将处理NWP模型输出的集成。这些技术将利用贝叶斯统计理论、多元分布、估计方法和集合预测的最新进展。每种技术将以三个版本进行开发和测试，用于(i)二元预测（例如，降水发生指标），（ii）多类别预测（例如，降水类型指标），以及（iii）连续预测（例如，降水数量取决于降水发生、温度、能见度、天花板高度、风速）。初步试验将包括制作和验证未来3天的概率定量降水预报（PQPFs）。评价新技术的主要基准将是目前在业务预测中使用的已有30年历史的模式输出统计技术。这些新的、最先进的概率预测技术的预期影响将改善对所有主要天气变化的预测，从而增加社会效益。特别是，由BPO或BPE制作的可靠和信息丰富的PQPFs，适合水文模型的要求，将能够产生概率河段预报、概率洪水预报和具有明确说明的探测概率的洪水警报。

Rational decision making by industries, agencies, and the public in anticipation of heavy precipitation, snow storm, flood or other disruptive weather phenomenon, requires information about the degree of confidence that the user can place in a weather forecast. It is vital, therefore, to advance the meteorologist's capability of quantifying forecast uncertainty to meet the society's rising expectations for reliable information.The long-term goal of this research is to lay down a methodological foundation for the next generation of probabilistic forecasting systems. The specific objective is to develop and test (i) a set of statistical techniques for probabilistic forecasting of weather variates and (ii) a set of performance measures for verification of probabilistic forecasts. The basic technique, called Bayesian Processor of Output (BPO), will process output from a numerical weather prediction (NWP) model and optimally fuse it with climatic data in order to quantify uncertainty about a predictand. The extended technique, called Bayesian Processor of Ensemble (BPE), will process an ensemble of the NWP model output.The techniques will harness recent advances in Bayesian statistical theory, multivariate distributions, estimation methods, and ensemble forecasting. Each technique will be developed and tested in three versions, for (i) binary predictands (e.g., indicator of precipitation occurrence), (ii) multi-category predictands (e.g., indicator of precipitation type), and (iii) continuous predictands (e.g., precipitation amount conditional on precipitation occurrence, temperature, visibility, ceiling height, wind speed). The primary test will involve the production and verification of probabilistic quantitative precipitation forecasts (PQPFs) for up to 3 days ahead. The primary benchmark for evaluation of the new techniques will be the 30-year old Model Output Statistics (MOS) technique used currently in operational forecasting.The expected impacts of these new, state-of-the-art techniques for probabilistic forecasting will be improved forecasts of all major weather variates, and hence increased societal benefits. In particular, reliable and informative PQPFs, produced by the BPO or the BPE and suited to requirements of hydrologic models, will enable the production of probabilistic river stage forecasts, probabilistic flood forecasts, and flood warnings with explicitly stated detection probabilities.