耕地地力评价方法及模型分析  

Evaluation method and model analysis for productivity of cultivated land

作　　者：闫一凡[1,2] 刘建立[1] 张佳宝[1] 

机构地区：[1]中国科学院南京土壤研究所,南京210008 [2]中国科学院大学,北京100049

出　　处：《农业工程学报》2014年第5期204-210,共7页Transactions of the Chinese Society of Agricultural Engineering

基　　金：973计划课题(2011CB100506);国家自然科学基金项目(41171179;41001127)

摘　　要：耕地地力的定量评价和分等定级是测土配方施肥的重要内容,也是实现农田地力定向培育和精准农作的基础。该文从地力评价指标筛选、评价单元划分与赋值、评价指标的权重确定等方面介绍了国内外耕地地力评价的主要流程和重要研究进展,对中国农业部推荐方法(特尔斐法-层次分析法)、神经网络法、支持向量机和决策树法等评价方法的原理及其优劣进行了较系统的述评。进一步地,还对该领域目前存在的指标体系通用性、评价结果可比性、数据缺失等问题及可能的解决方案作了探讨。在未来的耕地地力评价工作中,应将传统的层次分析法与近年兴起的分类与回归树等数据挖掘新技术相结合,建立起更为客观、全面的地力定量评价模型,为中国精准农业生产提供方法学参考。Quantitative evaluation, classification and gradation of cultivated land productivity are important for implementing formula fertilization, guiding the oriented soil fertility cultivation and precision farming. In this paper, the definition and main processes of cultivated land productivity were introduced ranging from indexes selection, evaluation unit division and assignment, index weight determination and gradation. Different evaluation methods of land productivity using machine learning technique confirmed with good results were summarized such as China＆#39;s ministry of agriculture recommended method, Delphi-analytical hierarchy process, soil productivity index, support vector machine, artificial neural network, and decision tree. Their use methods, advantages and disadvantages were analyzed. In general, these machine learning techniques are objective and can easily overcome Delphi’s subjective effect. Farmland soil fertility survey and quality evaluation are popular. However, some potential problems occurred, for example that evaluation index system is lack of universality, results of evaluation cannot be compared for different city, even county if the evaluation methods are different, and work is hard to be done in some remote mountainous areas where the economy and science fall far behind other regions. These problems were discussed and some possible solutions were proposed such as applying classification and regression trees in remote mountainous areas to enhance coefficient of utilization of data set based on mechanism for handling missing values. Finally, the paper analyzed if average annual yields used as target variables of these new machine learning techniques are feasible and reasonable. If the answer was yes, how to integrate these new techniques into traditional evaluation and classification methods of cultivated land productivity may become the possible direction for study. It hoped that this article would provide valuable information on methodology for evaluation, classification, and gradat

关 键 词：土地利用 土壤 数学模型 耕地地力 评价方法 评价指标 权重系数 

分 类 号：S158[农业科学—土壤学]