摘要
利用2006-2008年2 403个国家气象观测站地面雨量计的逐日降水量资料,采用与网格点最近的观测站有、无降水确定该网格点有、无降水和Barnes插值方案确定网格点降水大小的混合插值方案,得到全国空间分辨率为0.1°×0.1°(约10km×10km)的逐日降水量格点数据,在此基础上通过交叉检验方法统计格点数据的误差,从相关系数、平均偏差、平均绝对误差、平均相对误差、均方根误差、复合相对误差、观测均方根、插值均方根、相对方差比率和降水概率密度函数分布等方面研究了在该混合插值方案下得到的我国逐日降水格点数据的误差分布,分析了误差统计量的时空变化特征。交叉检验的统计结果表明:(1)观测站逐日降水量的估计序列(即在不使用该观测站实际观测情况下由周围观测站插值得到网格数据,再由网格数据反插观测站估计值)与实际观测序列比较,两者的相关系数为0.81,平均偏差为-0.02mm.d-1,平均绝对误差为1.3mm.d-1,平均相对误差为58.67%,均方根误差为4.5mm.d-1,复合相对误差为41.17%,相对方差比率为93.12%,即格点化的降水资料具有较高的精度。(2)观测站逐日降水量的估计序列与实际观测序列的概率密度分布比较接近,插值误差基本符合正态分布。(3)逐日降水格点误差的空间分布规律是:平均绝对误差和均方根误差从我国西北向东南逐步增大;而平均相对误差的分布特点是东部特别是东部平原小、西部相对大;相关系数的分布特点是东部高、西部相对低。(4)误差统计量的逐月变化规律是:绝对误差和均方根误差明显表现为随月降水总量的增减而相应增减,一般是7月达最大,12月为全年最小值;而相关系数的大小与降水性质有关,一般大尺度系统性降水多时(11月)相关系数偏高,局地对流性降水偏多时(8月)相关系数偏低。
Based on daily precipitation of 2 403 gauge observations in China for the periods from 2006 to 2008,a gridded dataset with 0.1°×0.1°（about 10 km×10 km） is established using the methods by combining the Barnes interpolation and observation precipitation frequency.Then the precision of gridded data is assessed by examining cross-validation statistics of the correlation coefficient（COR）,the mean bias error（MBE）,the mean absolute error（MAE）,the mean relative error（RMAE）,the root mean square error（RMSE）,compound relative error（CRE）,observation mean square deviation σ（O）,interpolation estimated mean square deviation σ（I）,the relative variance ratio（RVAR） and the probability density function（PDF） distribution,the temporal and spatial characteristics of the statistics are investigated and the annual cycle of interpolation error for the gridded data is discussed.The results show that daily gridded precipitation has a higher precision with the correlation coefficient at 0.81,the mean bias error is-0.02 mm·d-1,the mean absolute error is 1.3 mm·d-1,the mean relative error is 58.67%,the root mean square error is 4.5 mm·d-1 and compound relative error is 41.17%,and the relative variance ratio is 93.12%.Moreover,the PDF distribution of interpolation precipitation is closer to observation as well as the interpolation error is the normal distribution.The mean absolute errors and the root mean square errors for gridded precipitation increase from northwest China to southeast China.The mean relative errors are small in the east of China,especially in the flatland but large in the west of China.In the meanwhile,the correlation coefficients show high in the east part of China and relatively low in the west part of China.In addition,the annual cycle of monthly mean absolute errors and the root mean square errors for gridded precipitation are increasing with the augment of monthly total precipitation,the maximum errors in July while the minimum in December in general.But the precipitation characteristics,such as convection activities,have a heavy impact on the distribution of error statistics.The correlation coefficient,reaching its peak in November and valley in August,associates with the precipitation characteristics that the high correlations are related to large-scale and systematic precipitation while the low correlations are related to local the convective precipitation.
出处
《高原气象》
CSCD
北大核心
2011年第6期1615-1625,共11页
Plateau Meteorology
基金
国家973项目(2010CB950603
2010CB833503)
国家自然科学基金项目(40875029
41005002)共同资助
关键词
降水
空间插值
交叉检验
误差统计量
Precipitation
Spatial interpolation
Cross-validation
Error statistic
