Acquisition Of Key Index For Cultivated Land Quality Evaluation By Vis-NIR Spectroscopy

Land is the basic resource of human survival.Cultivated land is the most important part of the land resources,its quantity and quality are pivotal aspects to food security in China.With the rapid development of industrialization and urbanization in China,the reduction of cultivated land quantity and cultivated land quality,as well as the promotion of land consolidation and well-facilitated farmland construction,all put forward new requirements for quality evaluation of cultivated land.However,in the current cultivated land quality evaluation,soil properties are obtained through field sampling and laboratory analysis methods.These methods require long operating time and high costs,which seriously restricts the efficiency of cultivated land quality evaluation.Therefore,how to obtain cultivated land information quickly and reliably is an urgent problem in cultivated land quality evaluation.Based on the proximal soil sensing which can obtain the surface or section information of the cultivated land in real time,this research achieves rapid access for cultivated land quality evaluation index,and tries to use this method on cultivated land quality evaluation work.The research cannot only provide basic information and scientific basis for the evaluation of cultivated land quality,but also help to promote the innovation of land science and technology.In the present study,294 cultivated land soil samples are collected from 18 counties of Jiangxi province.The spectral of the soil by proximal in situ and stationary vis-NIR sensing and the spectral in laboratory test are gathered by utilizing ASD Field Pro FR spectrometer.We also sample them for measurements using traditional chemical methods for 10 soil properties.The feasibility of predicting 10 soil properties using two kinds of spectra has been studied,and different chemo metric methods are used to improve their prediction accuracies.On this basis,more accurate prediction results of soil organic matter and soil texture are used to recalculate the natural quality index and reevaluate the classification of 163 soil samples.Finally,the paper analyzes the consequences of the cultivated land quality,and verifies the feasibility of this method.The main results are as follows:(1)The results found that with laboratory-based spectral,soil organic matter,soil texture can be well predicted while soil TN,AN can be quantitatively predicted using PLSR algorithm.Compared with PLSR algorithm,using LS-SVM methods improve the prediction accuracy of soil properties and can predict soil SOM,TN,texture with high accuracy.The prediction accuracy of soil properties with in situ field spectra is lower than with laboratory spectra,but using LS-SVM methods can also improve the prediction accuracy of soil properties.When using direction standardization(DS)to remove the effects from soil water from in situ field measured spectra,the prediction accuracy of LS-SVM methods improves more.It can predicted soil organic matter and soil texture with high accuracy(for SOM: RPD = 2.06;for soil texture: RPD = 2.25;2.14;2.00).(2)Based on the in situ field spectra,we replaced the values of soil organic matter and texture with their prediction results of LS-SVM methods using DS,than recalculate the cultivated land physical quality index and judge the cultivated land physical grade of 163 samples.Physical quality index was took as evaluation results of vis-NIR spectral identification and cultivated land physical grade results existed was used to verify it.The result showed that they had a 95.09 percent degree of consistency within an error range between plus fifteen to minus fifteen,the maximum difference between them is less than 400,the correlation coefficient of existed physical quality index and prediction index is 0.7674.Compared with physical quality results using laboratory-based SOM,soil texture,the degree of consistency of vis-NIR spectral identification is 99.39 percent within an error range between plus fifteen to minus fifteen and the correlation coefficient is 0.9546.From the comparison result of physical quality grades using laboratory-based SOM,soil texture and vis-NIR spectral identification values,the grades of 99 soil samples remain the same.(3)According to the validation result,the random sampling results shows that the difference between average predicted SOM and existed SOM is less than 10 g/kg,and the laboratory-based measured content is very close to the mean value of the predicted content.In the meantime,the in situ spectral prediction of soil texture is consistent with the result of database,which is above 90%.In general,the study could provide some references to the cultivated quality gradation and farmland quality evaluation of land consolidation,based on the proximal soil sensing,we can get cultivated land information and evaluate the quality of cultivated land.