基于Delaunay三角网和地理加权回归的降水空间异常探测方法研究

针对目前已有的降水资料质量控制方法适应性差,降水数据难以模拟局地降水规律的问题。以湖南省为实验区域,利用1985~2014年夏季降水观测数据,定义局部地理空间内降水量发生显著偏离的气象站点为异常,发展了一种结合Delaunay三角网和地理加权回归的降水空间异常探测方法。该方法通过构建不规则三角网来获取各个区域之间稳定合理的地理邻近关系,根据空间邻近关系获取自适应回归带宽,并纳入地理加权回归模型进行计算,最后根据回归残差项的统计分布性质开展空间异常度量与分析。实验表明该方法能有效检测出降水空间异常,可应用于降水数据二次处理时的质量控制阶段,有助于进一步探索局部地理空间内的降水规律。

PRECIPITATION ANOMALY DETECTION METHOD RESEARCH BASED ON DELAUNAY TRIANGULATION AND GEOGRAPHICALLY WEIGHTED REGRESSION

Local precipitation laws were difficult to simulate due to the poor adaptability of current precipitation data quality control methods.Based on the observed summer precipitation data from 35 meteorological stations covering 1985-2014 in Hunan province,a method for detecting spatial outliers of precipitation was developed by combination of Delaunay triangulation and geography weighted regression.In this method,the stable and reasonable geographical proximity relationships between each region was obtained by constructing irregular triangulation,and the adaptive regression bandwidth was obtained according to spatial proximity relationships.Spatial outliers of precipitation were then detected and analyzed according to the statistical distribution characteristics of regression residual,which was calculated by geographically weighted regression method.The results showed that the method could be used to detect spatial outliers of precipitation accurately and effectively.It indicated that the method were meaningful for the secondary treatment of precipitation information and the further exploration of local precipitation laws.