The Study On Spatial Variability Characteristics Of Soil Salinity And Management Zoning In Alar Reclamation Area

The degree of soil salinity is a key parameter for irrigation quotas on farmland,but soil salinity has strong spatial variability in both horizontal and vertical directions.Therefore,irrigation of salinized farmland at regional scale needs to consider the degree of soil salinity and its spatial variability for management zoning,and adopt variable irrigation for different management zones in order to effectively improve the efficiency of agricultural water utilization.In the southern border region,agricultural irrigation water often ignores the difference of salinity degree of different fields and adopts uniform irrigation volume for irrigation,which is not only easy to cause salt damage due to incomplete salt pressure of local farmland or waste water resources by excessive drenching.Therefore,soil salinity differences are used as the key factor of zoning for precise management zoning,so that precise irrigation of farmland can be guided according to the salinity zoning in which the field is located for the purpose of efficient use of agricultural water resources.In this study,the apparent electrical conductivity and measured salinity data obtained by EM38-MK2 were used to study the optimal salinity inversion model based on the apparent electrical conductivity using linear modeling method and machine modeling method,and the soil apparent electrical conductivity data at different depths were used to discern the types of salinity aggregation in different areas within the reclamation area and to clarify the salinity distribution characteristics of soil profiles for different land uses.The optimal inversion model of salinity is also used to invert the surface soil salinity data in the reclamation area,and analyze and study the spatial variability characteristics of soil salinity.Accordingly,the salinization precision management partition of farmland in the reclamation area was divided by the object-oriented segmentation method using the comprehensive use of multi-source information from the star-land.The research results show that:(1)The model with the combination of EC_H and EC_V as independent variables has the best performance,the model with EC_H as independent variable has the second best performance,and the model with EC_V as independent variable has the worst performance,both under the linear and machine learning models.Under the same independent variables,the performance of the machine learning models is better than that of the linear models,and the predictive ability of the linear models is generally poor,with PRD less than 1.40,while the RPD of the machine learning models BPNN,SVM and RF are all greater than 1.6,and with the combination of EC_H and EC_V as the independent variables,the model accuracy of RF as the modeling method is the best,and the R~2 and RPD of its validation set are the largest,respectively 0.84 and 2.53,the model accuracy of BPNN as the modeling method is the second best,and the model accuracy of SVM as the modeling method is the worst.(2)The mean values of apparent electrical conductivity increased with depth and the coefficients of variation decreased with depth,and all coefficients of variation were greater than 69%,all of which belonged to strong variability,and the spatial heterogeneity of soil apparent electrical conductivity was influenced by both random and structural factors.In the Alar reclamation area,the proportion of soil salinity in the bottom aggregation type profile is the largest,about 64%,followed by the uniform type,and the smallest in the surface aggregation type,about 27%;under different land use methods,the soil profile salinity in arable land is mainly bottom aggregation,accounting for 58%of the arable land area,the proportion of soil profile salinity types in garden and forest grassland is roughly the same as that in arable land,while the proportion of surface aggregation type profile in unutilized land is the largest The proportion of soil profile salinity in unused land is the highest,and the bottom aggregation profile is reduced by about 35%and the surface aggregation profile is increased by about 41%compared with the bottom aggregation profile in other land use types,indicating that there are some differences in soil profile salinity under different land use patterns.(3)After the management partitioning of the Alar Reclamation area using multi-source information,the average coefficient of variation of each partition is less than 26%,which is about 67%lower than that of the whole study area,indicating that the object-oriented multiscale partitioning algorithm can effectively delineate homogeneous areas.Comparing the zoning results,we found that the zoning results based on single soil information are more fragmented than other zoning results,and the homogeneity within and heterogeneity between zoning areas are lower,which is also not conducive to the development and effective implementation of water rights allocation schemes.By adding vegetation information or environmental factors to the partitioning,the partitioning results become regular,and the homogeneity within partitions and heterogeneity between partitions are improved.Management partitioning based on multiple sources of information,with more regular partitioning results,not only has the highest homogeneity within and heterogeneity between partitions,but also facilitates the implementation of variable irrigation.In conclusion,management partitioning based on multiple sources of information was the best,and the results of this partitioning facilitated the implementation of variable irrigation while incorporating vegetation information and environmental factors affecting the spatial distribution of salinity,and the highest homogeneity within and heterogeneity between partitions.