空间变参数成矿预测信息综合模型研究

Due to the complexity of the geological processes and the multiple phases of mineralization, the spatial distribution of the factors for mineral prospectivity mapping often show structural, non-stationary and singularity, and the correlations between the factors and their contribution to the mineralization are changing with the change time and space. Thus it is difficult to suit all locations by using the model obtained based on all samples, which is common in traditional methods for mineral prospectivity mapping. The validity of a predicting model is determined by the quality of the training samples, the higher the degree of exploration, the better the sample representativeness, but the exploration differences are not well considered in existing mineral prospectivity mapping models. In order to contribute to the solution of these two problems, this study will introduce two kinds of spatial weights to existing mineral prospectivity mapping model, i.e. neighboring relation weights and exploration weights. A new spatially weighted technique will be developed for integrating the two kinds of spatial weights mentioned above, and spatially variable parameter model will be established based on this technique. In the new model, the optimized local window parameters will be obtained at each location, and all local models will be established based on its own optimized anisotropy parameters, which will reduce the model uncertainty and improve the accuracy and efficiency of the results in the context of complex geology background like covering areas. More instructions for mineral prospecting can be mined by mapping the parameters of the predicting variables, and this in turn can provide important references to explore the spatial patterns of metallogenic model and deduce the specific space-time condition for the formation of the deposits. Software modules will also be developed for the application of the new model in the study area and the promotion of the new model.

由于地质过程的复杂性和成矿作用的多期次性，成矿预测要素的空间分布往往呈结构性、非平稳性和奇异性，各要素之间相关关系及其对成矿的贡献度亦随时空转换而变化，传统基于全局样本得到的预测模型难以适应不同位置；模型的建立有赖于训练样本，样本勘查程度越高则其代表性越好，而现有成矿预测模型未能充分顾及到勘查程度差异对预测模型的影响。为此，本研究将在现有成矿预测模型基础上加入对空间邻近关系及区域勘查程度的考量，并通过适当的量化算法对这两种空间权重进行量化和综合，发展新的空间加权技术和空间变参数成矿预测模型；新模型将在每个局部位置确定最优窗口，基于最优窗口建立成矿预测模型，可有效提高覆盖等复杂地质背景下成矿预测的精度和效率；通过对预测变量参数进行空间制图和可视化，可以挖掘更多找矿信息，并为探索成矿模式的空间变化以及反推矿床形成的时空条件提供重要参考；还将开发相关软件模块，为研究区成矿预测和模型推广提供支持。

随着地勘工作的推进，大量露天矿、浅表矿被发现和开采，找矿难度增大，覆盖区找矿和深部找矿越来越受到重视。我国存在大面积戈壁、沙漠、黄土、红壤等中新生界覆盖区，而已发现大中型矿床数量却很少，找矿潜力大。由于覆盖的影响，地质调查工作难以开展，物化探手段受到限制，所获取的数据受覆盖背景影响难以反映下伏矿体信息。因此，必须从技术层面挖掘潜力。除了勘测技术手段的升级，发展适合覆盖区等复杂情况下成矿预测信息处理的数据综合模型至关重要，也是数学地质研究的使命担当。传统成矿预测信息综合模型属于全局模型，其特点是在整个研究区采用同一套预测参数与模型。成矿是复杂的非线性动态过程，基于地质过程的成矿预测要素空间分布往往呈结构性、非平稳性和奇异性，它们的内在关系及其对成矿的贡献程度亦随时空环境的转换而变化，基于全局样本得到的模型难以适应所有局部；模型的建立有赖于训练样本，已知样本的勘查程度越高则样本代表性越好，而现有预测模型未能很好顾及研究区勘查程度对于预测模型的影响，同样给成矿预测带来偏差和不确定性。针对上述问题，本研究发展空间变参数成矿预测模型。新模型在每个局部确定最优化窗口，基于最优窗口建立成矿预测模型，以最大限度减少模型不确定性，提高预测精度，对充实现有成矿预测理论方法体系和提高成矿预测精度具有重要意义。通过对预测变量参数进行空间制图，还可以挖掘更多找矿信息，为探索成矿模式的空间变化以及反推矿床形成的时空条件提供重要参考。本研究还开发了相关软件模块，为研究成果的应用推广提供支持。此外，空间变参数模型不仅可以用于回归分析以及矿产资源评价领域，对其它空间统计模型和其它领域相关问题的解决也有一定启示。在项目执行过程中，申请人成功将空间变参数模型运用到植被恢复潜力评价等领域，扩展了本研究的应用领域，提升了本研究的意义与价值。