Research And Application Of High-resolution Digital Mapping Of Soil Key Attributes At National Scale

Soil is the most basic natural resource necessary for human beings to produce food.It is equivalent to water resources and is the basis for ensuring food security.As the most basic means of production and labor for human beings,it is the most important resource on our planet and the foundation of human civilization.China is a traditional agricultural country with a long history of agriculture,but it is a resource-constrained country with scarce cultivated land resources.Protecting cultivated land is one of the basic national policies.Rational use of each inch of cultivated land has important theoretical significance and practical application value for food security and environmental protection.Soil properties are important factors in determining soil quality.Detailed national scale soil information can contribute to food security.Exploring spatial variability of soil properties is of great significance to regional ecology and arable land quality evaluation and arable land resource utilization.Therefore,we urgently need better understand the spatial distribution of certain soil key attributes at the national scale.In the past research,the spatial distribution of soil properties has room for improvement in accuracy and accuracy.In the field of soil science,with the increasing power of tools such as geographic information systems(GIS),global positioning systems(GPS),remote and near-end sensors,and soil mapping techniques,digital soil mapping based on soil landscape models has become an efficient expression of soil.The method of spatial distribution.Based on soil landscape science,based on the second national soil census sample data and multi-environmental variable data,this paper uses a variety of machine learning algorithms to predict the spatial distribution of key properties of high-precision surface soils with high resolution of 90 m in China.On this basis,the membership degree of different indicators is calculated.The comprehensive index evaluation method is used to predict the distribution of cultivated land quality grades in China,and the distribution of cultivated land quality grades in different agricultural areas is discussed.The main research contents and research results of the thesis are divided into the following aspects:(1)According to the complex characteristics of topographic and climatic conditions in China,using data mining technology combined with multi-source remote sensing data,the high-precision spatial distribution characteristics of large-scale soil organic matter in China were studied and the uncertainty was calculated.We find that the prediction results can well reveal the spatial differentiation of organic matter;and the model can well explore the relationship with environmental landscape factors.The correlation coefficient of the model independent verification reached 0.62,and the relative error reached 0.72 g/kg,which was at the same level as the accuracy of organic matter research at other national scales.The results of this chapter found that the organic matter content in paddy fields and dry land was 21.38 g/kg and 18.98 g/kg,respectively.Compared with the SoilGrids data,the overall calculation results of this study are close to the actual data provided by soil fertility in China.The results of organic matter prediction and mapping in this study provide a reference for the monitoring and evaluation of soil surface organic matter content in China in the future.(2)Using stepwise regression combined with artificial neural network technology,screening environmental variables and constructing pedotransfer functions(PTF)of soil bulk density in China to fill the lack of soil bulk density data in soil database.Through screening,a series of soil properties and environmental variables(SOM,Silt,TN,TP,Temperature,MrVBF,Aspect,and Radiation)are used in the PTF function.When the hidden node is 7,the model works best,and the R~2 value is 0.47.By comparison,the study found that our artificial neural network PTF predicts that China's soil bulk density is more accurate than other published PTFs.When different PTFs are applied to soils with different geographical environments,their performance is very different.Using the standardized Euclidean distance to determine the effective domain of the PTF,we predict that 93%of the data is cumulative 97.5%cutoff value.The PTF model of artificial neural network was used to fill the soil bulk density loss,and then the gradient lifting tree model was used to predict the high-resolution spatial distribution map of soil bulk density in China.It was found that the distribution of soil bulk density in China showed an opposite trend in the distribution of organic matter.Our findings will be used in the calculation of soil organic carbon stocks in China,which will provide a more accurate baseline.(3)Based on the digital soil mapping theory and the parallel operation of the tile structure,a hybrid model combining the XGBoost model and the random forest two machine learning techniques is used to predict the spatial distribution of high-precision soil pH in China and to complete the cadmium and mercury in the soil through the threshold.The research shows that when the two models are combined,the accuracy of the model is improved,the accuracy R~2 is 0.72,and the root mean square error is 0.71,which is better than the single model.The parallel computing method using the tile structure greatly improves the computational efficiency,and is of great significance for large-scale and high-precision soil digital mapping research.The results of this study were compared with the global soil pH maps generated by SoilGrids and HWSD,and our results presented more detailed spatial variability at small scales.The spatial distribution study of soil pH completed in this study provides important basic data for predicting the temporal and spatial trends of soil acidity in the future,and provides a benchmark for assessing land use and climate change.(4)Using remote sensing and digital mapping as technical support,select 12quantitative factors as the evaluation index of cultivated land quality,construct the evaluation system of cultivated land quality in China,and use the membership function combined with the cumulative method as the comprehensive index evaluation model of cultivated land quality,and complete the cultivated land quality grand map in China.It is found that the spatial distribution of cultivated land quality in China is not balanced.The quality of cultivated land in the middle and lower reaches of the Yangtze River is better than that in other regions,and the quality of cultivated land in different agricultural areas is obviously different.The percentage of cultivated land of different grades obtained by the study is relatively close to the data published by the Ministry of Agriculture in 2014,but it is different.The cultivated land evaluation system based on remote sensing technology and digital mapping completed in this study is an important means of rapid and efficient evaluation.It scientifically quantifies the spatial distribution of cultivated land quality in China,and provides a basis for rational use and management of cultivated land and related land work.