海洋浮游植物丰度的空间插值优化

在探索性空间数据分析(Exploratory spatial data analysis)和数据转化的基础上,利用反距离加权(Inverse distance weighting,IDW)、径向基函数(Radial basis functions,RBF)、普通克里格(Ordinary Kriging,OK),3种插值方法,对2003年8月获得的珠江口浮游植物丰度数据进行插值运算,并对插值准确度进行交叉验证。结果显示,珠江口浮游植物丰度数据具有离散性大、存在极大和极小值、呈正偏分布等特点。而对数转化能大大减小数据的离散性和不对称性,有效消除插值结果图中各类插值噪音。交叉验证显示,插值精确度OK最高,RBF次之,IDW最低。观察插值结果等值面图,发现3种方法均能较客观地模拟出浮游植物丰度的总体分布趋势,在对局部趋势的模拟上,OK的表现最好。综合评定,OK为最适合珠江口浮游植物丰度数据的插值方法。半变异模型的选择对OK的插值结果影响不明显。在四种半变异模型中,圆形模型的拟合效果最好。

Optimization of the spatial interpolation for marine phytoplankton abundance

To understand the spatial interpolation optimization laws of marine phytoplankton abundance, we compared three different methods of spatial interpolation: (1) inverse distance weighting (IDW), (2) radial basis functions (RBF) and (3) Ordinary Kriging (OK), using the recorded phytoplankton abundance in the Zhujiang Estuary, China, in August 2003. Firstly, exploratory spatial data analysis was used to gain a deeper understanding of the recorded phytoplankton abundance. Then the phytoplankton abundance was log-transformed. Lastly, we generated interpolation surfaces of phytoplankton biomass using the three interpolation methods. The results indicate that the phytoplankton abundance data is charactered by high dispersion, with a few outliers, and a positively skewed distribution. The log-transformation reduces the variances and skewed distribution, and effectively removes various interpolation noises in the interpolation surfaces. The accuracy of the OK is the highest, followed by the RBF, and then the IDW. The interpolation surfaces reveal that all of the three methods correctly show general trends of the phytoplankton abundance by using a series of optimization techniques. But the contours generated by the IDW always bend around the global outliers with excessively great curvature and sometimes even form closed small loops, which maybe cause some interference in identifying the general trends. The contours generated by the RBF are excessively smooth. However, it represents the general trends clearly, although many local trends are lost. The contours generated by the OK are considered the best, as the method can represent both general and local trends accurately. Thus, the OK method is more efficient than RBF and IDW in terms of accuracy and surface representation. The four semi-variance models of the OK do not affect the interpolation results, with the circular model having the best fit to the data.