摘要
土壤有机质(SOM)是土壤肥力重要指标之一。快速,无损且准确地预测SOM含量对于保护和提升土壤肥力有重要作用。可见-近红外(vis-NIR)光谱结合偏最小二乘回归(PLSR)模型在土壤属性估测中广泛使用,目的是探讨通过Bootstrap抽样提高PLSR的预测能力和泛化能力。以江西、浙江和湖南三省水稻土为研究对象,采集了523个耕层(0~20 cm)土壤样本,比较偏最小二乘回归(PLSR)和Bootstrap-PLSR两种回归模型在估测SOM的精度和泛化能力;利用确定系数(R^2),均方根误差(RMSE)和性能指标(RPIQ,标准差与四分位间距离的比值)来评估预测的准确度,利用Bootstrap抽样后预测值的95%置信区间和实测值的分布情况、欠拟合和过拟合PLSR和Bootstrap-PLSR回归因子的差异来分析Bootstrap-PLSR模型的泛化能力和稳定性。研究表明:使用Bootstrap-PLSR预测的SOM含量的预测精度(R^2=0.76,RMSE=5.82,RPIQ=2.51)高于PLSR模型(R^2=0.72,RMSE=6.27,RPIQ=2.33)。Bootstrap抽样能够提高SOM含量中间部分的预测精度并且具有较强的建模稳定性,Bootstrap-PLSR具有较强的泛化能力且可以用来选择特征波段。
Soil organic matter(SOM)is an essential soil fertility indictor of paddy soil in the middle-lower Yangtze plain.Rapid,non-destructive and accurate determination of SOM is vital to preventing soil degradation caused by inappropriate land management practice.Visible-near infrared(vis-NIR)spectroscopy with PLSR can be used to effectively estimate soil properties.In this study,523 soil samples were collected from paddy fields in the provinces of Zhejiang,Jiangxi and Hunan,China.Partial least squares regression(PLSR)and PLSR combined with Bootstrap sampling were used to compare the prediction accuracy of SOM based on vis–NIR full bands.The coefficient of determination(R^2),root mean square error(RMSE),and ratio of performance to inter-quartile distance(RPIQ)were used to assess the prediction accuracy.The generalization ability and stability of the Bootstrap-PLSR model were analyzed by using the distribution of the 95% confidence interval of the predicted value and the measured value,the difference in the coefficients between the PLSR and the Bootstrap-PLSR could produce higher accuracy(R^2=0.76,RMSE=5.82,RPIQ=2.51)compared with that of PLSR(R^2=0.72,RMSE=6.27,RPIQ=2.33). The performance of the PLSR and Bootstrap-PLSR in SOM prediction did not differ significantly in the range of higher and lower values but a slight increase could be found in the middle SOM value.The Bootstrap-PLSR could provide the uncertainty of the models and their predictions.Therefore,PLSR coupled with Bootstrap sampling is recommended for prediction of SOM in the middle-lower Yangtze plain.
作者
杨梅花
徐强
赵小敏
YANG Mei-hua;XU Qiang;ZHAO Xiao-min(Yuzhang Normal University,Nanchang 330103,China;Shangrao Vocational&Technical College,Shangrao 334109,China;School of Environmenteal and Land Resource Management,Jiangxi Agricultural University,Nanchang 330045,China)
出处
《江西农业大学学报》
CAS
CSCD
北大核心
2019年第6期1227-1234,共8页
Acta Agriculturae Universitatis Jiangxiensis
基金
国家自然科学基金项目（41361049）
江西省教育厅科学技术研究项目（GJJ181150）~~
