Extreme value modelling applied in hydrology

在设计防洪建筑物或评估洪水损失时，往往根据河流流量测量的短样本做出决策。 选择一个统计频率分布来拟合测量值，然后使用拟合的分布来估计未来的洪水。有许多方法可以将分布拟合到数据。一些分布具有有限的上限，该上限是分布参数（形状、尺度等）的函数。一些分布拟合方法可能会产生与观测数据不一致的上限估计值。当一个或多个样本观测值超过估计分布的上界时，会发生这种情况。这使得估计的上限不切实际。我们将调查这种不一致的一些估计方法与观测数据，并检查其工程风险后果。我们希望建立改进的指导方针之间的统计分布和水文风险分析拟合方法的选择。

In designing flood-control structures, or in estimating flood damage, decisions are often made on the basis of a short sample of river flow measurements. A statistical frequency distribution is chosen to fit the measurements, and the fitted distribution is then used to estimate future floods. Many methods exist to fit distributions to data. Some of the distributions have a finite upper bound that is a function of the distribution parameters (shape, scale, etc.). Some distribution fitting methods may produce an estimate of this upper bound that is inconsistent with the observed data. This occurs when one or more sample observations exceed the upper bound of the estimated distribution. This makes the estimated upper bound unrealistic. We will investigate this inconsistency of some estimation methods with the observed data and examine its engineering risk consequences. We hope to establish improved guidelines for choosing between statistical distributions and fitting methods for hydrological risk analysis.