| Soil is an integral part of the earth’s surface ecosystem.It is not only the largest carbon pool but also the basis of agricultural production.Therefore,it is of great theoretical and practical significance to study the spatial distribution of soil properties.In order to describe the spatial distribution of soil properties as accurately as possible,researchers have proposed many schemes in recent years,among which the geostatistical method represented by Kriging method is the most commonly used.With the increasing demand for high-precision interpolation in applications,a series of improved methods have emerged,among which the Bayesian maximum entropy(BME)estimation method began to appear under this background and has gradually been widely used.At present,the usual application mode of BME is to first find auxiliary data,then construct soft data,and finally use this method to spatial prediction.Due to the superiority of the method framework,it can often achieve better results than traditional methods.However,in this workflow,there are several issues worth discussing: firstly,with the advancement of technology,the means of obtaining data are also diversifying,how to use these data into BME estimate,so as to improve data quality and reduce data acquisition costs,has always been an important topic in BME research;secondly,in the face of multiple sources of auxiliary information,in order to improve the accuracy of spatial estimation,how to screen out the part that is closely related to the actual research target variable is also very important;thirdly,Constructing a relationship model between auxiliary variables and research target variables is the basis for subsequent spatial estimation.Facing the existing various types of modeling methods,how to choose a method with high precision,simple calculation,and reasonable estimation error,to construct high-quality soft data,which is also of great significance for the practical application of BME.Based on the above considerations,this paper explores the above issues through three specific cases: estimating the spatial distribution of soil organic carbon based on spectral data,spatial estimation of soil total potassium based on multi-source auxiliary data,spatial estimation and traceability of soil heavy metals,and achieved the following results:(1)Applying the soil reflectance spectrum data to the research of the spatial distribution of soil organic carbon in Florida,the quantitative relationship between the reflectance spectrum and soil organic carbon content was established by Partial least square regression(PLSR)and Random Forest(RF)methods,and then the soft data formed by the prediction errors of the two methods were explored.This data was applied to the Ordinary Kriging(OK)and BME estimates for comparative analysis.The results show that both RF and PLSR can invert the soil organic carbon content in the study area relatively well from the model validation results,and the RF model is slightly more accurate,which proves that the reflectance spectrum can be applied as a test tool for soil property content in the spatial estimation study.The Infinitesimal jackknife method used by RF has high standard error precision of the estimated value,and the soft data interval formed is relatively reasonable,which can better reflect the real data situation,and is a relatively reliable soft data construction method.The spatial estimation results of the soft data generated using BME combined with RF have higher accuracy than OK that directly using prediction data from the spectral model.Therefore,the method of constructing soft data using BME combined with RF can use soil spectral data more strictly,thereby improving the spatial estimation accuracy of soil organic carbon.(2)Integrating multiple environmental factors into the spatial estimation of soil properties,multiple linear regression,geographically weighted regression and random forest methods were used to construct the relationship between environmental factors and soil attributes.On the basis of these models,soft data was constructed and applied to BME spatial estimation,and the BME estimation results was compared with the regression Kriging(RK)and Geographical Weighted Regression(GWR).The main conclusions of the study are as follows:(1)The total potassium content of soils in Shayang County was highly variable,with a trend of higher in the east and lower in the west throughout the region,and the highest content occurred in the northeast of the study area.The variables that were closely related to the total potassium content of soils in the study area were eight continuous variables,LST,FMI,h,NDVI,NDWI,QFD,WTI and β,and two category variables,tidal soil and rice soil.(2)Using the RF method to construct the relationship between environmental variables and soil attributes and combining it with the BME method can obtain optimal spatial estimation results with higher accuracy than the OK method without using environmental data and the RK and GWR estimation results without strictly using environmental data.The RK and GWR use environmental factors in a way that ignores the uncertainty contained in them,and when the regression model is not accurate enough,it may be less accurate spatially than the OK method.estimation accuracy will be inferior to that of the OK method without the use of auxiliary data.(3)The RK and GWR estimation results are more fragmented,which may not truly reflect the spatial distribution of total potassium content in the research area.The spatial distribution map generated by the OK method is too smooth and loses more local details.The 3 BME estimates,on the other hand,are in between the 2 aforementioned,and the estimation results are both holistic and can show the details of local variations.(3)Based on Land use regression(LUR)theory,a BME-LUR mixed model is proposed for soil heavy metal pollution research.Using a variety of statistical analysis,Positive matrix factorization(PMF)model,RF and BME methods,the accumulation of soil heavy metal content,important influencing factors,spatial distribution,sources and proportions were explored,and a new set of soil heavy metal pollution analysis method framework was explored,while deepening the understanding of heavy metal pollution in the study area.The results show that Cd exceeds the standard of severe pollution,Pb and Zn are light pollution,and Cr and Zn basically do not cause pollution as a whole.The results of BME spatial estimation show that the content of Cd in the study area far exceeds the background value of the study area,and the accumulation degree is extremely high.The pollution is mainly distributed in the area where industrial enterprises are concentrated.The accumulation of Cu and Zn is mainly distributed near residential areas.The spatial distribution of Pb is closely related to the road.However,Cr was lower than the background value overall.Spatial estimation results of five soil heavy metals indicate that the BME-LUR hybrid model is better than the LUR model.Using the PMF method and integrating the results of various analyzes,four factors that affect the accumulation of heavy metals in the soil in the study area are extracted,namely industrial pollution,human activities,natural factors and traffic emissions,and their contribution rates are 38.34%,33.82%,and 28.94% and 16.21% respectively.Industrial pollution and human activities are the main factors affecting the accumulation of heavy metals in the region.Through the research on the above-mentioned issues,this paper successfully introduces soil reflectance spectrum as soft data into BME estimation,aiming to provide new soft data sources and construction methods for related research.For studies using relevant environmental variables as soft data,this paper proposes a set of RFbased screening and modeling methods.And for the study of heavy metal pollution in soil,a mixed model based on BME-LUR was proposed to solve the problems of spatial distribution and source analysis of heavy metals.It is expected that based on these new methods and their practical application cases,the source and construction ideas of soft data will be expanded to a certain extent,and then the application of BME method will be promoted,and new solutions will be provided for the issues related to the spatial distribution of soil properties. |
PDF Full Text Request