Classification Of Cultivated Land Productivity At County Scale Based On Discriminant Analysis Technologies

It is the scientific evaluation of the quality of cultivated land that is an important guarantee to enhance the quality of cultivated land.The cultivated land productivity,as a comprehensive indicator of the evaluation of cultivated land quality,being evaluated scientifically and accurately classified is not only an important part of the Financial Subsidy Project for Soil Testing and Formulated Fertilization,but also an important basic work of scientifically formulating agricultural development planning and food security guarantee policies and earnestly strengthening the construction and management of cultivated land quality.At present,the theory and method system of land classification and quality evaluation has been increasingly mature and greatly improved,including various kinds of technologies which are emerging in an endless stream,such as geostatistics,3S??Remote Sensing,Global Position System and Geographic Information System,and machine learning.Discriminant analysis,a method of statistical analysis based on machine learning,however,is rarely used in the field of land productivity evaluation.In this study,discriminant analysis was applied to the evaluation of potential productivity of cultivated land at county scale and its feasibility and rationality was to be verified.In this study,Huixian City,known as "Granary in the north of Henan Province",is the research area,which has a strong agricultural foundation and is a national high-quality wheat production base.Its strategic position to ensure national food security is particularly significant.The main data sources were the samples obtained by the Subsidy Project for Soil Testing and Formulated Fertilization conducted in Huixian County from 2009 to 2012.The surface soil properties of cultivated land,site environment and other data and some related map data were collected.Those closed spots with definite boundaries formed by the superimposition of the soil map and the land-use map were used as the basic evaluation units of cultivated land productivity.A set of indicators of the evaluation of cultivated land productivity was comprised of three quantitative evaluation factors and seven conceptual factors.The Delphi method for conceptual factors and the membership function method for the quantitative counterparts were used to calculate the membership degree of all those evaluating samples.Besides,using Analytical Hierarchy Process,the weight of evaluating indicators relative to cultivated land was calculated.The 2/3 samples were randomly selected as the training data and the rest as the test data.The weight of each evaluation indicator and the membership degree of those evaluating samples in the training data were the basis for establishing the discriminant analysis models.Based on those established models of Fisher,Bayes and Mahalanobis distance discriminant functions,the samples in the test data who had different productivity grade were classified,and the accuracy reached up to 87.10%,87.97% and 89.28% respectively.Through the significance test of the mean of every indicator of cultivated land samples in each group,it was proved that the classification of those samples in each group had been optimized,and the discriminant functions were meaningful and feasible.In addition,the fuzzy c-means clustering method was used to classify those samples in the test data with an accuracy of 84.81% which was relatively lower relative to discriminant analysis.Based on all these results of cultivated land productivity evaluation by each algorithm,it is concluded that the discriminant analysis method is feasible and effective for the classification of cultivated land productivity at county scale.The discrepancies in the number of evaluating samples definitely affect the accuracy of the discrimination and clustering,and the accuracy of the fuzzy c-means clustering and Fisher discriminant analysis is more impaired.