基于随机森林算法的耕地质量定级指标体系研究  

Research on the index system of cultivated land quality grading based on random forest algorithm

作　　者：林子聪 任向宁[1] 朱阿兴 赵鑫[1] 胡月明 LIN Zicong;REN Xiangning;ZHU Axing;ZHAO Xin;HU Yueming(College of Natural Resources and Environment,South China Agricultural University/Institute of Geographic Information Engineering,South China Agricultural University/Key Laboratory of Construction Land Improvement,Ministry of Land and Resources/Guangdong Provincial Key Laboratory of Land Use and Consolidation,Guangzhou 510642,China;Department of Geography,University of Wisconsin-Madison,Madison 53706,USA;College of Agriculture and Animal Husbandry,Qinghai University,Xining 810016,China;School of Resources and Environment,University of Electronic Science and Technology of China,Chengdu 610054,China)

机构地区：[1]华南农业大学资源环境学院/华南农业大学地理信息工程研究所/国土资源部建设用地再开发重点实验室/广东省土地利用与整治重点实验室,广东广州501642 [2]威斯康星大学−麦迪逊分校地理系,威斯康星州麦迪逊53706 [3]青海大学农牧学院,青海西宁810016 [4]电子科技大学资源与环境学院,四川成都610054

出　　处：《华南农业大学学报》2020年第4期38-48,共11页Journal of South China Agricultural University

基　　金：国家重点研发计划(2016YFC0501801,2018YFD1100103);青海省科技计划(2017-ZJ-730);广州市科技计划(201804020034)。

摘　　要：【目的】分析研究区域内的耕地质量差异,优化耕地利用与布局,为耕地保护提供参考依据。【方法】以青海省共和县、都兰县和乌兰县的耕地为研究对象,根据历史及现有文献收集耕地质量的影响因素,采用随机森林算法和相关性分析筛选定级指标并确认权重,通过加权求和法计算定级指数并划分级别,得到定级结果。与常用的特尔菲法定级成果进行比较分析。【结果】随机森林算法得到的变量重要性(I)范围在0.03~11.94,相关性分析结果显示,大部分影响因素间相关性不显著,有8个为显著相关,综合I值和相关性分析结果将30个影响因素收敛为4个纬度下的14个定级指标,其中影响研究区域耕地质量的主要因素为生态系统脆弱性、生长季平均降水和年总太阳辐射量,权重分别为0.11、0.10和0.09,随机森林算法评价结果与实际情况相符。【结论】与常用的特尔菲法比较,随机森林算法稳定性更好,级别指数变幅区间更小,更有利于构建省级空间尺度的耕地级别可比序列。【Objective】To analyze the difference of cultivated land quality in the study region,optimize the use and layout of cultivated land,and provide a reference for cultivated land protection.【Method】Taking the cultivated land in Gonghe County,Dulan County and Wulan County in Qinghai Province as the research object,the influencing factors of cultivated land quality were collected based on the history and existing literature,and the random forest algorithm and correlation analysis were used to screen the grading indicators and confirm the weight.We calculated the grading index and divided the levels by weighted sum method to get the grading result.We compared the results with the grading results of commonly used Delphi method.【Result】The value of variable importance I obtained by random forest algorithm ranged from 0.03 to 11.94.Correlation analysis showed that the correlation between most influencing factors was not significant,eight of which were significant correlation.The 14 rating indicators under four dimensions were astringed from 30 influencing factors.The main factors influencing the quality of cultivated land in the study area were ecosystem vulnerability,mean precipitation of growing season and annual solar radiation amount,with the weights of 0.11,0.10 and 0.09,respectively.【Conclusion】Compared with Delphi method,the random forest algorithm has better stability and smaller level of index variation interval,which is more conducive to construct a comparable sequence of cultivated land levels at provincial spatial scale.

关 键 词：耕地质量评价 定级指标体系 随机森林算法 特尔菲法 

分 类 号：X53[环境科学与工程—环境工程]