Application Of Non-Parametric Statistics And Artificial Intelligence Technology In Spatial Variability Of Water-Soil

Spatial variability Theory is the focus of attention in world's frontier science and technology , This paper discussed the spatial variability of soil water and salt, selected two typical example water-saving irrigation area within a large irrigation area-China Hetao irrigation District as example, opened up the new principles and new methods of the water-soil spatial variation. The main results of this study are:In order to solve the majority of small sample size is limited, outlier causing a partial distribution of the statistical characteristics, this paper introduced Robust Kriging of RGS , but it still has a significant smoothing effect. Meanwhile studying Impact Point, results shown that: the impact point identification method and specific identification method is consistent, oulier identification should based on analysis of the Impact Point.Try to introduced the BP neural network technology of ANN into the environmental monitoring, ordered new methods for the spatial variability of soil and water research. After a large number of exploratory studies, established BP neural network model of the soil salt. The results shown that BP method can overcome the smoothing effect of OK, it has no strict sampling distribution and parameters requirements for original data, the issue does not involve oulier, it is simple and practical, but BP neural network has no a significant statistical analysis of test features.Initially introduced the main method-Indicator Kriging (abbreviated IK) of Non-Parametric Geostatistics (NPG) to monitor soil water and salt ,and carried out simulation and valuation of soil water and salt spatial and temporal distribution at test area of cultivated land and wasteland of salt before summer irrigation. discussed . instructions of threshold selection method that there is no theoretical analysis, the instructions of the soil water and salt threshold can be referenced. Emphasized on the IK method indicator kriging model and the parameters of the systems analysis, author found that the average indicator kriging probability, instructions threshold, and indicating variogram is closely related, just considering from the increased probability of prediction accuracy, moisture indicator threshold value should be smaller than the median threshold, that of salt should be higher than the median. IK method has unique estimate effect, but still has a certain degree of smoothing effect, the larger estimate variance of the indicator probability. Meanwhile used firstly the disjunctive kriging (abbreviated DK) method to estimate soil water and salt, contrasted IK, DK and Probability Kriging(abbreviated PK). In view of ANN does not have a significant statistical analysis, while the IK method still has smooth phenomenon, this paper discussed the integration of indicator kriging and the BP neural network technology, in accordance with ideas of (Non-Parametric) statistical and Neural Network, author proposed an new innovative artificial neural Indicator Kriging technology. Used Artificial indicator kriging for the soil water and salt evaluation, it is found that: Artificial Neural Indicator Kriging has advantages of IK and ANN, no statistical assumptions, not involving oulier, but it can get approximation of non-linear function, the outlier impact is limited in sub-region, so it can naturally reduced influence of outlier, smoothing effect is significantly weaken, the nonlinear approximation problem is solved, author recommends it to monitor soil water and salt and extends to other related work.