Research On The Sampling Strategies For Soil Properties And A Method For Spatial Local Outlier Detection

The optimal sampling strategy is the equalization of the sampling strength,the analysis cost and the precision experiment.That is,the minimum experimental cost is used to maximize the experimental precision.Sampling is not only an important means of soil features analysis,but also widely used in geographical research,environmental assessment,resource inspection and other fields,is the basis and premise for experiments.The scientificity of the sampling plan not only has great influence on the progress of the research work,but also determines the accuracy and scientificity of the research results.So it is important and difficult for each researcher to pay attention to how to scientifically optimize the sampling strategy scientifically.With the expansion of the sampling area,the development of sensor equipment and sampling methods,more and more items are needed for soil characteristics research,and the required precision is also higher.It's should be tough challenges to pre-process the large sample data set and high latitude data set.The detection of outliers is essential to pre-processing of data.In investigations that use sampling,it is essential that the effects of outliers on spatial variability and appropriate sample size be addressed.For datasets with spatial distribution properties,outliers will inevitably lead to a certain degree of bias in the research results,and even have an impact on the research results.Therefore,it is necessary to detect and screen outliers in the dataset to facilitate the preprocessing of the dataset and further analyze the mechanism of the outliers.Therefore,it is necessary to explore a method for spatial local outlier suitable for soil properties.Most of the current sampling designs at home and abroad are optimized by mathematical models.These designs are better to provide references for layout of multiple sampling and later monitor stations but not suitable for preliminary sampling.In order to overcome these shortcomings,a method of soil sampling design based on priority was proposed in this study,which is able to empower each synergistic factor,to transfer information completely and to accommodate a wide range of qualitative and quantitative synergistic factors in aided design.Aiming at the method,a representative verification based on clustering method was presented in the paper.To avoid the difficulty of measurement and operation between qualitative and quantitative synergistic factors,the Euclidean distance formula and central point formula between data sets were improved.And the improved close degree and compactness-separation effect were also used as the verification indexes.Then the priority sampling point set was compared with the hierarchical sampling point set and the random sampling point set.The independent verification based on Kriging interpolation was used as a contrast proof.The results showed that the information carried by the priority sampling point set is closer to the total sample point set than the hierarchical sampling point set and the random sampling point set.The sampling design method in this study can meet the needs of region sampling,improve the sampling efficiency and quality,and provide references for the designs of the other soil sampling schemes.In order to understand the extent to which the presence of outliers affects the spatial variability of the subject and the number of reasonable samples,This study compared the performance of global outlier detection,local outlier factor(LOF)model,local distance-based outlier factor(LDOF)model,and spatial local outlier factor(SLOF)model on investigations of Soil Hydrodynamic Parameters and analyzed the effects of outliers in terms of statistical characteristics of data,spatial variability,and appropriate sample size.Results: Removal of outliers resulted in a decrease of the appropriate sample size to some extent.Datasets with outliers removed had much better interpolation outcomes than the original and the SLOF model had the best performance among all tested methods.Independent tests of models suggest that fitting errors of values and fitting deviations of tendencies both affected the interpolation outcomes.For the data obtained by spatial sampling,especially the data set with uneven sampling density distribution,the traditional outlier sampling detection method can not meet the detection requirements.Existing spatial-outlier detection algorithms separate spatial and non-spatial attributes,and the number of neighbors k cannot fully reflect the value of a spatial attribute.In addition,these algorithms strongly depend on the user's preset initial value,and there is a very strong edge effect.Therefore,a method for spatial local-outlier detection based on slope(SLODS)is proposed,which makes full use of the information provided by spatial and non-spatial attributes to more sensitively capture local outliers.The algorithm utilizes the inherent range of the dataset to replace the traditional k,and cyclic k value increase-and-decrease calculation is introduced to reduce the dependency on users.Compared with the SLOF algorithm,the SLODS algorithm has higher accuracy,can better adapt to the characteristics of spatially distributed datasets,and is better at detecting outliers.The optimal sampling strategy of soil properties and the spatial outlier detection algorithm have completed the basic work from sampling strategy to data preprocessing under large-scale conditions,and provided sampling strategies and deviations for soil characteristics.The method of group sample detection provides a theoretical basis for the influence of outliers on spatial datasets,and provides certain conditions for further soil characteristics research in the future.