降水空间插值技术的研究进展

Review on spatial interpolation techniques of rainfall

降水的空间分布信息在水资源管理、灾害预测预警和区域可持续发展研究等方面的地位已越来越重要。然而,空间降水插值却一直是个难题,影响降水的因素很多,如经度、纬度、高程、坡度、坡向、离水体的距离等,建立一个通用的降水插值模型几乎是不可能的。空间降水插值方法很多,优缺点和适用性不同。总体上,降水的空间插值方法有3类:整体插值法(趋势面法和多元回归法等)、局部插值法(泰森多边形法、反距离加权法、克里金插值法和样条法)和混合插值法(整体插值法和局部插值法的综合)。近年来,随着应用数学和人工神经网络技术的发展,很多新方法如BP网络、径向基函数网络等逐渐应用到降水插值中。另外,研究区域和时间尺度的不同决定了所选用的插值方法及模型不同;即使是同一种插值方法,应用于不同的研究区域,所取得的结果也不同。因此,应根据具体的研究区域的自然地理特征,选择不同的插值方法建立不同的插值模型。在各种插值方法中,兼具其它插值方法优点的混合插值法有利于提高降水插值的精度,是未来降水插值研究的一个发展方向。

Rainfall spatial distribution is the important information in many fields, such as water resource management, drought and flood disaster prediction, and regional sustainable development, but its interpolation has been a puzzle because there are many affecting elements, such as latitude, longitude, elevation, distance to water bodies, slope, etc., especially in mountainous regions. It is difficult to build a general rainfall interpolation model for different geographical regions. Several kinds of rainfall interpolation methods were introduced in this paper, including global interpolation methods （ trend surface and multiple regression）, local interpolation methods （Thiessen polygons, inverse distance weighting, kriging and splines）, and mixed methods （combined global and local methods）. Their advantages, disadvantages and applicability were discussed. Recently, with the development of applied mathematics and artificial neural networks （ANN）, some new methods were put forward in the rainfall interpolation, especially the ANN technique, such as Back-Propagation neural networks （ BP network） and Radial Basis Function networks （RBFN）. Because of the uncertainty of rainfall, more detailed geographical and topographical characteristics are needed to improve the precision of predicted rainfall. Detailed topographical characteristics could be provided by a large scale DEM （digital elevation model） or DTM （digital terrain model）, which plays an important role in rainfall interpolation. Different interpolation methods are required in different space or time scales, Even the same rainfall interpolation might get different results in different regions. The mixed methods, combining the advantages of global and local interpolation methods, are useful for improving the interpolation precision, which would be one of the important research fields in the future.