基于Fisher判别分析算法的县域耕地地力等级预测——以河南省辉县市为例

目的]借用机器学习算法——判别分析算法来简化耕地地力评价工作,探索区域尺度上机器学习方法在地力评价应用的新途径。方法]基于辉县市测土配方施肥财政补贴项目耕地地力评价工作获取的基础数据,依据我国农业部标准《耕地地力调查与质量评价技术规程》(NY/T 1634—2008)和该市耕地地力评价实践经验,选取研究区表层土壤质地、土壤剖面特征、地表砾石度、速效钾、有效磷、有机质含量、灌溉保证率、排涝能力、地貌类型、坡度等10个土壤和立地条件因素作为耕地地力水平的判别变量,构建Fisher典则判别函数模型,对5 922个评价单元的耕地地力状况进行判断分析和归类分级。结果]经对判别结果进行统计验证和回代验证,显示预测判别正确率高达91.4%。结论]在耕地地力评价与分级标准确定的前提下,判别分析算法在区域尺度上对分析耕地地力状况、预测耕地地力等级方面具有独特优势。

Per dictive Clas sification of Cultivat ed Land at Coun ty Sc ale Using Fisher Discriminant Analysis Algorti hms—A Case Study of Huix-ian City,Henan Province

Objective]To simplify the evaluation of cultivated land fertility by applying the machine learning algorithm, which aims to explore a new approach to the application of machine learning method in the evaluation work of cultivated land fertility at regional scale. Method] Based on Technical Specification for Investigation and Quality Evaluation of Cultivated Land Fertility (NY/T 1634—2008) and the local prac-tices of cultivated land evaluation, the methods applied by this study generally are supposed to use the based data obtained by the financial subsidy project for soil testing and formulated fertilization conducted in Huixian City, Henan Province, to establish canonical discriminate func-tions.10 soil and site condition factors including surface soil texture, soil profile characteristics, surface gravel degree, rapidly available po-tassium in soil, available phosphorous in soil, organic matter content in soil, irrigation guarantee rate, capacity for drainage, geomorphic types, and surface slope are selected as the discriminant variables of cultivated land fertility level.By constructing the model of Fisher discrim-inant functions, Fisher discriminant analysis ( FDA) was employed to determine, analyzed and classified land fertility in 5922 sampled sites of the studied region using that Fisher discriminate functions. Result] The results of the methods demonstrate a prediction accuracy reaching up 91.4% after mathematical statistics verification and back substitution verification which means the original data being returned back to the Fisher discriminant functions. Conclusion] Under the premise of identifying the standard of evaluation and classification of cultivated land, the discriminant analysis algorithm has a unique advantage in analyzing and classifying the fertility situation of cultivated land and predicting the grade of cultivated land.