摘要
传统富硒(Se)土地划分仅依据土壤Se含量,容易导致遗漏和误判,对水稻籽实Se含量进行准确预测,可以更精准地划分富硒产区。本文利用福建省大规模的土壤和水稻籽实样品数据,通过机器学习中的随机森林(RF)模型,建立了水稻籽实Se含量预测模型,并采用网格搜索方法提高模型预测的准确性,旨在为研究区富硒土地区划提供数据支撑。结果表明,福建省表层土壤Se含量平均值为0.34 mg/kg,且具有很强的空间变异性,其含量处于足硒和富硒状态的比例分别是62.83%、27.05%,不存在硒中毒现象。土壤硒对成土母质具有很强的继承性,煤系地层、硅质岩等高硒岩石发育的土壤中Se含量较高。采样区水稻籽实富硒率为44.09%,对应的根系土富硒率仅为10.54%。土壤pH、铁氧化物含量、有机质含量、磷含量、降雨量和温度是影响水稻籽实Se生物富集系数的重要因素。将随机森林模型预测结果与传统的线性回归模型进行对比,随机森林的均方根误差(RMSE)为0.018,低于线性回归的RMSE值(0.096),相关系数(r)为0.957,高于多元线性回归的r值(0.653),随机森林模型的性能显著优于多元线性回归模型。基于优化的随机森林模型,将福建省富硒产区划分为优先发展区和潜在发展区,该省富硒水稻优先发展区主要位于宁德市西北部、三明市西部和漳州市中南部,潜在发展区主要位于南平市中南部和三明市中部。
The traditional classification of selenium-enriched land is based solely on soil selenium content,which may lead to omissions and misjudgments.Accurate prediction of selenium content in rice grains can provide a more precise basis for delineating selenium-enriched production areas.Large-scale sample data of soil and rice grains from Fujian Province were utilized in this study to establish a selenium content prediction model for rice grains through machine learning.A grid search method was employed to enhance the accuracy of the model’s predictions,aiming to provide data support for the regionalization of selenium-enriched soil in the study area.The results indicated that the average selenium content in the surface soil of Fujian Province was 0.34 mg/kg,showing strong spatial variability.The proportions of soil in the sufficient selenium and selenium-enriched states were 62.83%and 27.05%,respectively,with no selenium poisoning observed.Soil selenium content is largely inherited from parent material,with higher selenium content in soils developed from high-selenium rocks such as coalbearing strata and siliceous rocks.The selenium-enriched rate of rice grains in the sampling area was 44.09%,while the selenium-enriched rate of the corresponding root zone soil was only 10.54%.Soil pH,iron oxide content,organic matter content,phosphorus content,rainfall,and temperature were important factors influencing the selenium bioaccumulation coefficient in rice grains.By comparing the prediction results of the random forest model with those of the traditional multiple linear regression(MLR)model,the root mean square error(RMSE)of the random forest model was 0.018,lower than that of the MLR model(0.096),and the correlation coefficient was 0.957,higher than that of the MLR model(0.653).Thus,the performance of the random forest model was significantly superior to that of the MLR model.Based on the optimized random forest model,the seleniumenriched production areas in Fujian Province were classified into priority development areas and potential development areas.The priority development areas for selenium-enriched rice in this province are mainly located in the northwest of Ningde City,the west of Sanming City,and the central and southern parts of Zhangzhou City,while the potential development areas are mainly in the central and southern parts of Nanping City and the central part of Sanming City.
作者
王莹
刘玖芬
杨忠芳
刘晓煌
李子奇
赵晓峰
王斐然
王超
刘佳
WANG Ying;LIU Jiufen;YANG Zhongfang;LIU Xiaohuang;LI Ziqi;ZHAO Xiaofeng;WANG Feiran;WANG Chao;LIU Jia(Natural Resources Comprehensive Survey Command Center of China Geological Survey,Beijing 100055,China;School of Earth Sciences and Resources,China University of Geosciences,Beijing 100083,China;Key Laboratory of Coupling Process and Effect of Natural Resources Elements,Beijing 100055,China;Technology Innovation Center for Analysis and Detection of the Elemental Speciation and Emerging Contaminants,China Geological Survey,Kunming 650111,China;State Key Laboratory of Biogeology and Environmental Geology,China University of Geosciences,Wuhan 430078,China)
出处
《岩矿测试》
北大核心
2026年第2期484-494,共11页
Rock and Mineral Analysis
基金
中国地质调查局地质调查项目(DD20250209101)
中国地质调查局地质调查项目(DD20190301)
中国科协青年人才托举工程项目(YESS20240578)
国家重点研发计划项目(2024YFD1500901)
自然资源部生态地球化学重点实验室基金项目(ZSDHJJ202303)。
关键词
水稻
硒
空间分布
预测模型
机器学习
rice
selenium
spatial distribution
prediction model
machine learning
