生态学报2026,Vol.46Issue(19):10743-10755,13.DOI:10.20103/j.stxb.202506251607
基于高光谱与机器学习的镉胁迫下牡丹叶片净光合速率估算
Estimation of net photosynthetic rate in peony leaves under cadmium stress based on hyperspectral imaging and machine learning
摘要
Abstract
To investigate the spectral reflectance characteristics and photosynthetic rate features of peony leaves under different concentrations of heavy metal cadmium stress,a pot experiment was conducted with three cadmium chloride solution concentrations(T0:high concentration;T1:medium concentration;T2:control,CK)using the Paeonia suffruticosa'Luoyang Hong'variety as experimental material.Hyperspectral images of potted plants were acquired,and net photosynthetic rates(Pn)of plant leaves were measured.Multiple machine learning models,specifically Random Forest(RF),Support Vector Machine(SVM),and Extreme Gradient Boosting(XGBoost),along with interpretability analysis(SHAP)methods,were employed to construct estimation models for the Pn in peony leaves.The results indicated that:(1)The Competitive Adaptive Reweighted Sampling(CARS)method effectively selected feature subsets from spectral reflectance,first -order derivatives,vegetation indices,and combined features,with sensitive spectral bands mainly concentrated in the 600-900 nm range;(2)First-order derivatives of spectral reflectance effectively enhanced spectral signal variations and suppressed background noise,improving the model performance by 24.3%and 11.5%compared to models constructed with original bands and vegetation indices,respectively;(3)The XGBoost model constructed with combined features of first-order derivatives and vegetation indices achieved the best performance,with a coefficient of determination(R2)of 0.831 and root mean square error(RMSE)of 0.298 μmol m-2 s-1 on the test set,significantly outperforming other feature combinations;(4)SHAP analysis of the optimal combined feature model revealed that the first- order derivative feature Df104 contributed most to the prediction model with a SHAP value proportion of 29.7%,while the combined SHAP values of vegetation indices RRWVI and GNDVI2 accounted for 10%.In conclusion,the Pn estimation model for peony leaves established in this study provides a methodological reference for the rapid,non-destructive detection of plant growth under heavy metal cadmium stress conditions,and has practical application value for assessing plant tolerance and physiological status under heavy metal pollution.关键词
镉胁迫/高光谱/光合速率/机器学习/SHAP分析Key words
cadmium stress/hyperspectral/photosynthetic rate/machine learning/SHAP analysis引用本文复制引用
刘涛,杨奉源,张江涛,张沛东,王政,张寰..基于高光谱与机器学习的镉胁迫下牡丹叶片净光合速率估算[J].生态学报,2026,46(19):10743-10755,13.基金项目
河南省自然科学基金面上项目(252300421849) (252300421849)
河南省高校科技创新人才支持计划(26HASTIT018) (26HASTIT018)