Development of a selection system of the optimal sampling locations for a soil pollution survey.

提出了一种利用地质统计学方法进行土壤污染采样点优选的新方法。评价函数是作为地质统计差异的估计而创建的。无条件模拟地统计蒙特卡罗方法所产生的估计场和虚拟实场。这种差异是通过整个场地中每个位置的浓度差的平方和或整个场地中污染物质的总和差的平方和来评估的。该方法还可以通过从无条件模拟产生的虚拟实场中选择满足给定信息的实场来处理不确定信息。对遗传算法进行了改进，提出了一种总能达到局部解的“逐步移动至最优点（SMBP）”方法。他们的混合方法对于寻找最优解是有效的。研究了最似然法参数估计的特点及存在的问题。结果表明，最大似然法估计的相关尺度往往小于实际的相关尺度，并且利用估计的参数不一定能正确估计未知点的不确定性。日本政府公布的土壤污染实例表明，利用这些实例估计的相关尺度具有对数正态分布。利用这些数据作为经验先验信息，改进了参数估计。

A new method for optimal strategy of sampling locations for soil pollution was developed using geostatistics. The evaluation function was created as a estimate, of the differences between geostatistically. estimated fields and fictitious real fields created by the unconditional simulation, which was a geostatistical Monte Carlo method. The difference was evaluated by sum of squares of the difference of concentration at each location in a whole field or a square of the difference of the sum of a pollution material in a whole field. The method could also treat with uncertain information by selecting fields which could meet the given information from the fictitious real fields created by the unconditional simulation. Improvement of genetic algorithms and development of a new method named "stepwise movement to the best point (SMBP)" which could always reach a local solution were executed. Their hybrid method was shown to be effective for searching the optimal solution. Characteristics 'and problems in the parameter estimation by most likelihood method was also studied. The results showed that the correlation scale estimated by the most likelihood method was apt of be smaller than the real one and that uncertainty at unknown points could not be necessarily estimated properly using the estimated parameters. The examples of soil pollution released by Japanese government showed that correlation scales estimated using the examples had log-normal distribution. Using these data as the experiential prior information, parameter estimation was thought to be improved.