基于集成机器学习的大理红花大金元种植适宜性区划  

Planting Suitability Zones of Honghuadajinyuan in Dali Prefecture Based on Integrated Machine Learning

作　　者：翁倩文 陈伟强[1] 王德勋[2] 陈逸林 史宏志[4] 马月红[1] 于滢 苑钰珂 WENG Qianwen;CHEN Weiqiang;WANG Dexun;CHEN Yilin;SHI Hongzhi;MA Yuehong;YU Ying;YUAN Yuke(College of Resoures and Environment,Henan Agricultural University,Zhengzhou 450046,China;Yunnan Tobacco Company Dali Prefecture Company,Dali 671000,Yunnan,China;College of Information and Intelligence Engineering,Zhejiang Wanli University,Ningbo 315000,Zhejiang,China;Tobacco College of Henan Agricultural University,Zhengzhou 450046,China)

机构地区：[1]河南农业大学资源与环境学院,郑州450046 [2]云南省烟草公司大理州公司,云南大理671000 [3]浙江万里学院信息与智能学院,浙江宁波315000 [4]河南农业大学烟草学院,郑州450046

出　　处：《中国烟草科学》2025年第2期113-120,共8页Chinese Tobacco Science

基　　金：中国烟草总公司云南省公司科技项目(30802431)。

摘　　要：烤烟特色品种红花大金元生态适应性较差,为了构建大理烟区红花大金元种植适宜性智能区划方法,选取高程、成熟期降水、成熟期均温、水热适度系数、土壤pH、速效钾含量和水溶性氯含量7个指标为生态环境协变量,以752份调查数据为学习样本,采用集成机器学习算法,对大理州红花大金元种植适宜性区划进行研究。结果表明:(1)红花大金元适宜性与环境协变量呈多维非线性关系,适合采用非线性机器学习模型。(2)最优模型为Bagging和Boosting集成学习算法的CHAID决策树。适宜类判别模型中,水热适度系数、土壤速效钾含量和成熟期降水最为重要;适宜级判别模型中,成熟期降水、成熟期均温、高程最为重要。(3)大理州红花大金元种植“最适宜”、“较适宜”和“不适宜”的评价单元数分别为471、456和531个。剑川县、云龙县、洱源县、大理州、巍山县、弥渡县、南涧县大部分地区,宾川县东西两侧、永平县西南部为适宜种植区。(4)经48个样品评吸数据验证,区划结果符合实际。研究结果可为大理州红花大金元种植区划提供决策依据。The flue-cured tobacco variety“Honghuadajinyuan”exhibits poor ecological adaptability.To establish an intelligent zoning method for its cultivation suitability in Dali tobacco-growing area,seven ecological environmental covariates were selected,including altitude,precipitation during the maturity period,mean temperature during the maturity period,hydrothermal coefficient,soil pH,available potassium content,and water-soluble chlorine content.Using 752 survey datasets as training samples and the ensemble of machine learning algorithms,we investigated the cultivation suitability zoning of“Honghuadajinyuan”in Dali Prefecture.Results demonstrated as the follows.(1)The suitability of“Honghuadajinyuan”shows a multidimensional nonlinear relationship with environmental covariates,validating the use of nonlinear machine learning models.(2)The optimal models were CHAID decision trees integrated with bagging and boosting algorithms.In the suitability class model,the hydrothermal coefficient,soil available potassium,and maturity-period precipitation were the most critical indicators;In the suitability level model,maturity-period precipitation,mean temperature during the maturity period,and altitude were prioritized.(3)Among 1458 evaluation units,471 were classified as“most suitable”,456 as“moderately suitable”and 531 as“unsuitable”.Suitable cultivation areas were concentrated in Jianchuan,Yunlong,Eryuan,Dali Prefecture,Weishan,Midu,and Nanjian counties,as well as the eastern and western parts of Binchuan and the southwestern region of Yongping.(4)Validation using sensory evaluation data from 48 tobacco leaf samples confirmed the alignment of zoning results with actual quality.These findings provide a scientific basis for optimizing“Honghuadajinyuan”cultivation zoning in Dali Prefecture.

关 键 词：机器学习 红花大金元 种植适宜性评价 大理州 

分 类 号：S572[农业科学—烟草工业] S126[农业科学—作物学]