空间软数据及其插值方法研究进展

由于对地观测技术的迅速发展，空间数据的种类和数量增长迅猛，由空间数据反演得到的各种信息日趋膨胀，这些反演结果中的信息不少以软数据的形式出现。在实际应用中，这些软数据往往与空间插值的目标变量具有一定的相关性，甚至成为控制目标变量空间分布特征的重要因素。然而，由于这些数据通常表示为非数值形式，在计算和处理上存在着一定困难，以致被传统的插值方法所忽视，从而造成信息浪费。近来出现的空间软插值方法是一种利用空间软数据作为辅助信息并以改善插值效果的方法，能够较好的处理并利用软数据所隐含的信息，具有较好的应用发展前景。本文根据空间软数据的特点及其分类，系统综述了空间软插值方法及其应用领域。首先分析了空间数据软硬性质的根本区别，论述了软数据的分类和“硬化”方法，然后介绍空间插值模型中对空间软数据的集成方法和原理，最后对空间软插值方法及其应用研究领域进行了展望。

Review on Soft Spatial Data and its Spatial Interpolation Methods

In recent years, as the observation technologies develop rapidly, both type and number of spatial data is increasing, and information retrieved from spatial data expands increasingly, among which includes a large number of qualitative information, for instance, land-use type data, vegetation type data, topographic feature data, which some experts called soft information or soft data. These so-called soft data often have associations with the predicted target variable, even could become one of most important factors that influence the spatial distribution of target variable obviously in some cases, therefore, they can help improve prediction of target variable theoretically. However, in respect that non-numerical soft data can’t be calculated directly and is neglected by traditional spatial interpolation methods, connotative useful information can not be utilized sufficiently and effectively, which results in a mass of wasted information. Lately, soft spatial interpolation technology was proposed, aimed to integrate soft spatial data as auxiliary or second information to help improve interpolation accuracy. According to the characteristics and categories of soft spatial data, this paper aimed to review on soft spatial interpolation methods and their applications. Firstly, we summarized some “harden” methods, hardening the soft spatial data to hard data. Then, we discussed several different type soft spatial interpolation methods afterward, such as simple kriging, cokriging, indicator kriging, ordinary kriging, stratified kriging, kriging with external drift regression, bayesian maximum entropy, inverse distance weighted. After that, prospects of application of soft data and soft spatial interpolations were proposed in the last part.