A metastatistical approach to rainfall extremes
A metastatistical approach to rainfall extremes
复制标题
DOI：
10.1016/j.advwatres.2015.03.001
复制
发表时间：
2015-05-01
影响因子：
4.7
通讯作者：
Ignaccolo, Massimiliano
中科院分区：
文献类型：
--
作者：
Marani, Marco;Ignaccolo, Massimiliano
传统的极端事件统计理论假设一个渐近制度，其中每年的事件数量足够大，以适用于限制广义极值分布。这已被证明不适用于许多实际情况。我们在这里介绍一个元统计极值（MEV）的方法，这是定义在描述“普通”的日降雨量和强度的统计参数的分布。该方法不需要一个渐近假设，并自然占的影响，大部分的分布上的年最大日降雨量的普通事件。建立在现有的观测表明，每日降雨量的分布是威布尔右尾等效，MEV的方法，然后专门产生一个紧凑，易于应用的配方。我们将此配方的Monte Carlo实验的基础上Weibull统计来自3-世纪长的降雨时间序列观察帕多瓦（意大利）。我们发现MEV估计和“观察到的”极端事件发生的频率在合成的时间序列生成的一个很好的协议。GEV和Gumbel估计，相反，表现出系统误差。不同的降雨事件发生率的测试表明，当事件/年的数量增加时，GEV和Gumbel估计偏差略有改善。然而，一个恒定的偏差，GEV和Gumbel估计被认为是（合成）气候的事件的数量和强度的分布是随机变化的。GEV和Gumbel分布的估计均方根误差也大于MEV方法。因此，GEV和Gumbel分位数估计值比MEV估计值更可能远离实际值。最后，新的MEV方法的应用程序的子集的长帕多瓦时间序列识别显着的变化，在百年的时间尺度降雨极端。(C)2015爱思唯尔有限公司版权所有。
The traditional statistical theory of extreme events assumes an asymptotic regime in which the number of events per year is large enough for a limiting Generalized Extreme Value distribution to apply. This has been shown not to be applicable to many practical cases. We introduce here a Metastatistical Extreme Value (MEV) approach which is defined in terms of the distribution of the statistical parameters describing "ordinary'' daily rainfall occurrence and intensity. The method does not require an asymptotic assumption, and naturally accounts for the influence of the bulk of the distribution of ordinary events on the distribution of annual maximum daily rainfall. Building on existing observations showing the distribution of daily rainfall to be Weibull right-tail equivalent, the MEV approach is then specialized to yield a compact and easily applicable formulation. We apply this formulation to Monte Carlo experiments based on Weibull statistics derived from the 3-century long rainfall time series observed in Padova (Italy). We find an excellent agreement between MEV estimates and the 'observed' frequency of occurrence of extreme events in the synthetic time series generated. GEV and Gumbel estimates, on the contrary, exhibit systematic errors. Tests with different rates of occurrence of rainfall events show slight improvements of the GEV and Gumbel estimation bias when the number of events/year is increased. However, a constant bias in GEV and Gumbel estimates is seen for (synthetic) climates where the number of events and the distribution of intensities is varied stochastically. The estimation root mean square error is also larger for the GEV and Gumbel distributions than for the MEV approach. Hence, GEV and Gumbel quantile estimates are more likely to be further away from the actual value than MEV estimates. Finally, the application of the new MEV approach to subsets of the long Padova time series identifies marked variabilities in rainfall extremes at the centennial time scale. (C) 2015 Elsevier Ltd. All rights reserved.