Interpolation of precipitation under topographic influence at different time scales
Interpolation of precipitation under topographic influence at different time scales
复制标题
DOI：
10.1002/wrcr.20307
复制
发表时间：
2013-08-01
影响因子：
5.4
通讯作者：
Pegram, Geoffrey
中科院分区：
文献类型：
--
作者：
Bardossy, Andras;Pegram, Geoffrey
本文描述了在单个时间间隔（从一天到一年）内插降雨数据的新方法，使用高斯copula和非对称v-copula，并将海拔作为外生变量进行各种处理。对于较短时间的聚合，零被视为删减变量。对于每个选择的时间步长，降水量的边际分布使用非参数密度估计器建模，而空间依赖结构使用最大似然方法估计。将该方法与普通克里格法和外漂移克里格法等其他常用的地质统计插值方法进行了比较。偏差和误差结构的几个措施已被用于评估方法的有效性，在一系列的比较分裂抽样研究。本研究选择的数据集包括德国三个地区41年来的日降水时间序列，测量范围为126,144公里(2)的1200多个地点。在本文的诸多发现中，最突出的是：(1)降水与地形的相关性随着时间间隔的延长而增加，地形的定向平滑显著改善了降水与地形的相关性；（ii） copula方法在插值质量（偏差和不确定性估计）方面优于kriging方法；（iii）将零作为删减变量的处理提高了每日和候值的插值质量（该地区的月度和年度数据没有干旱期）；（iv） copula方法在一个区间内的目标点上产生估计的完整条件分布，大大改进了由克里格导出的简单不确定性估计；(v)高斯联结在计算效率方面表现最好，结合插值中有用的误差曲线，并且在所有方法中误差估计是最现实的。
In this paper, new methodologies for interpolating rainfall data in individual time intervals (ranging from a day to a year) using Gaussian copulas and unsymmetrical v-copulas, with a variety of treatments of altitude as an exogenous variable, are described. For shorter time aggregations, zeros were treated as censored variables. For each selected time step, the marginal distributions of precipitation amounts were modeled using nonparametric density estimators, while the spatial dependence structures were estimated using a maximum likelihood methodology. The methodology was compared to other common geostatistical interpolators such as ordinary kriging and external drift kriging. Several measures of bias and error structure have been used to assess the efficacy of the methods in a range of comparative split-sampling studies. The data set chosen for the study comprises daily precipitation time series over 41 years in three regions in Germany measured at more than 1200 locations over a large area (126,144 km(2)). Among the many findings in the paper, the ones that stand out are: (i) correlation between precipitation and topography increases with the length of time interval and is significantly improved by directional smoothing of topography; (ii) the copula methods are superior to kriging methods in terms of quality of interpolation (bias and uncertainty estimation); (iii) the treatment of zeros as censored variables improves interpolation quality for daily and pentad values (monthly and annual data in this region present no dry periods); (iv) the copula methods yield full conditional distributions of estimates at a target point, in an interval, improving substantially on the simple uncertainty estimates derived from kriging; and (v) the Gauss copula in particular performs best overall in terms of computational efficiency, combined with useful error profiles in the interpolations, and of all methods is the most realistic in its error estimates.