基于随机森林的中大尺度农田产能提升潜力评价——以广东省为例  

Potential Evaluation of Increasing Farmland Productivity on Medium and Large Spatial Scales Based on Random Forests:A Case Study of Guangdong Province

作　　者：陈旭飞 任向宁[2] 张池[2] 咸春龙[3] 冯雪珂 马涛 刘健美 CHEN Xu-fei;REN Xiang-ning;ZHANG Chi;XIAN Chun-Iong;FENG Xue-ke;MA Tao;LIU Jian-mei(Guangdong Land Consolidation and Rehabilitation Center,Guangdong Guangzhou 510630,China;College of Natural Resources and Environment,South China Agricultural University,Guangdong Guangzhou 510642,China;College of Economics and Management,South China Agricultural University,Guangdong Guangzhou 510642,China)

机构地区：[1]广东省土地开发整治中心,广东广州510630 [2]华南农业大学资源环境学院,广东广州510642 [3]华南农业大学经济管理学院,广东广州510642

出　　处：《西南农业学报》2019年第9期2133-2140,共8页Southwest China Journal of Agricultural Sciences

基　　金：广东省垦造水田关键技术与应用研究(GDGTKJ2018003)

摘　　要：【目的】中大空间尺度上农田产能提升潜力的客观准确评价,对于预测区域粮食安全保障能力至关重要,但现有方法过于追求大而全,评价指标多重共线性显著,相互干扰噪声强,指标体系离散性明显。【方法】以广东省2011-2015年高标准农田建设前后农田可实现产能潜力的变化值为因变量,以文献法整理的17个影响因素为自变量,采用随机森林模型测度的影响因素重要性,构建农田产能提升潜力评价指标体系,并以广东省为例研究区进行测算验证。【结果】①根据随机森林模型测度结果,不同因素对农田产能提升潜力重要性(I值)在0.21~12.71,其中水源与农田匹配系数与土壤性质的重要性大于基础设施配套状况、交通状况。②根据随机森林模型测度结果,地形坡度、水源与农田匹配系数、有机质含量、有效土层厚度、剖面构型等5个指标的重要度大于5.00,为农田产能提升潜力评价的核心指标。③经测算,广东省农田产能提升潜力在1.04×lO^4~6.06×lO^4 hm^2,农田产能具有较大的提升潜力。【结论】中大空间尺度上农田产能提升潜力与地形水文、土壤条件、农田基础设施状况、交通状况等有关,但不同因素的重要性存在较大差异,通过随机森林模型构建的评价指标体系降维收敛效果显著,其测算结果具有较强的科学性和合理性。【Objective】The objective and accurate evaluation of farmland productivity enhancement potential on medium and large spatial scales is crucial to the prediction of regional food security capacity.However,the existing methods are too big and comprehensive,with significant multicollinearity of evaluation indicators,strong mutual interference noise and obvious dispersion of index system.【Method】In Guangdong province,in 2011-2015,before and after the construction of high standard farmland farmland which can realize the change of the capacity potential value of the dependent variable,by the method of literature of 17 factors as independent variables,the influence factors of using random forest model to measure the importance of construction of farmland productivity promotion potential evaluation index system,taking Guangdong province as an example verifies the measure in the study area.【Result】(i)According to the measurement results of the random forest model,the importance(I value)of different factors to farmland productivity improvement potential was between 0.21 and 12.71,in which the importance of matching coefficient between water source and farmland and soil properties was greater than that of supporting infrastructure and traffic conditions.(ii)According to the measurement results of the random forest model,the significance of five indicators,including terrain slope,matching coefficient of water source and farmland,organic matter content,effective soil layer thickness and profile configuration,was greater than 5.00,which was the core index for the evaluation of farmland productivity improvement potential.(iii)According to the calculation,the farmland productivity improvement potential of guangdong province was between 1.04×1O^4-6.06×1O^4 hm^2,and the farmland produc-tivity had great potential for improvement.【Conclusion】Large spatial scales of farmland productivity potential and topography and hydrology,soil conditions,farmland infrastructure,traffic conditions and so on,but there are large dif

关 键 词：农田产能 提升潜力 随机森林 广东省 

分 类 号：S17[农业科学—农业基础科学]