Soil Basic Fertility Assessment And Organic Matter Content Prediction Based On Support Vector Machine

In the past few decades,growing attentions have been focused on soil quality and fertility.Soil nutrient is an important research object in the field of soil science.As one of the most important nutrient source,soil organic matter(SOM)has become a research topic.SOM content is predicted according to the variation tendency of soil organic matter content in long-term experiment,which provides scientific theoretical basis for the study of SOM.Soil nutrients evaluation is an important part of assessing soil fertility using soil total nitrogen(TN),total phosphorus(TP),total potassium(TK),available nitrogen(AN),available phosphorus(AP),available potassium(AK)etc.Estimating the level of soil quality and soil fertility based on varied nutrient contents could provide scientific and reasonable development and management for land using.Many mathematical models were widely used in different areas so far,and model evaluation is an important application,including comprehensive index method,artificial neural network method,fuzzy mathematical method,different distance clustering method ect.However,these methods can not fit the complex nonlinear relationship between factors and soil fertility well.Manually given in bias reduces the coverage of the evaluation model and the reliability of the results.In recent years,support vector machine(SVM)is a new machine learning technique which was developed based on statistical learning theory to provides an effective method to deal with the above-mentioned disadvantages.VC dimension theory and structural risk minimization theory ensures global optimization,maximum generalization ability and strong extension of model.Hence the module can solve many practical predictions and has become one of the influential achievements in the machine learning area.Combined support vector machine with soil ecology was mainly made in my research:1.Using support vector machine classification theory to evaluate the basic fertility grade by analyzing soil chemical properties with different kernel functions.The result revealed that it is feasible to assess soil basis fertility using this recognition theory.Moreover,the type of kernel functions make little decisive effect on the result of soil basis fertility evaluation.Compared with BP neural network,discriminate method and cluster analysis,support vector machine classification method provides more reality results of soil basis fertility evaluation.2.Using support vector machine to give a prediction of the SOM content of a long term experiment in Fuyang city,Anhui Province.The results revealed that support vector machine regression method was more accurate in doing SOM content prediction when compared with BP neural network and Radial Basis Function neural network.In addition,regression simulation between soil organic matter and crop yield showed positively correlation between soil organic matter and crop yield.3.To improve the reliability of experimental design and the integrity of experiment data,a novel model,multiple hybrid support vector machine model,was proposed to give prediction and classification of soil fertility grade.This research applied the combined machine learning method and support vector machine to soil ecology area to give prediction of soil organic matter content and evaluation of soil fertility grade.These results highlight the feasibility and superiority of support vector machine method.