基于民勤干旱荒漠区地面实测的人工梭梭同化枝高光谱反射率与光合色素数据，经光谱曲线处理和特征参数提取后，采用随机森林(RF)、反向传递人工神经网络(BP-ANN)、支持向量机(SVM)三种常见的机器学习模型模拟叶绿素a、叶绿素b和类胡萝卜素含量，确定梭梭同化枝光合色素含量的最优估算模型，揭示影响光合色素含量的光谱关键变量。结果表明，RF模型在模拟精度上表现最优，对三种光合色素的R2值介于0。73至0。77，RMSE介于0。023至0。302 mg·g-1。其中，RF对叶绿素b的预测精度最高，R2为0。77，RMSE为0。023 mg·g-1。不同光合色素对干旱胁迫的响应各异，导致影响它们的光谱变量也不同。叶绿素a与叶绿素b的最优模型受到吸收位置(Absorption position，AP)、红边位置(Red-edge position，REP)和红边拐点(Red-edge inflection point，REIP)的影响较多，而类胡萝卜素的最优模型更容易受REIP、Vogelmann红边指数(Vogelmann red-edge index 2，VOG2)与水波段指数(Water band index，WBI)的影响。该研究为基于高光谱遥感与机器学习模型反演梭梭同化枝光合色素含量提供依据，服务于人工梭梭林的旱情监测与抚育管理。
Estimation of the Photosynthetic Pigment Content in Assimilated Branches of Haloxylon ammodendron Based on Hyperspectral Data and Machine Learning Methods
Based on the ground-measured hyperspectral reflectance and photosynthetic pigment data of assimi-lated branches of Haloxylon ammodendron in the Minqin desert area,the hyperspectral data were processed with noise reduction and feature parameter extraction.Three common machine learning models,including random forest(RF),back propagation-artificial neural networks,and support vector machine,were used to simulate chlorophyll a,chlorophyll b,and carotenoid contents.The optimal estimation model for photosyn-thetic pigment content in the assimilated branch of Haloxylon ammodendron was determined,and the key spec-tral variables affecting the photosynthetic pigment content were revealed.The results showed that the RF model had the best simulation accuracy,with R2 values ranging from 0.73 to 0.77 and RMSE ranging from 0.023 to 0.302 mg·g-1 for the three photosynthetic pigments.Among them,RF had the highest prediction accuracy for chlorophyll b with R2 value of 0.77 and 0.023 mg·g-1 RMSE.Different photosynthetic pigments had different responses to drought stress,resulting in different spectral variables affecting them.The optimal models of chlo-rophyll a and chlorophyll b were mostly affected by Absorption position(AP),Red-edge position(REP)and Red-edge inflection point(REIP);while the optimal model of carotenoids was more susceptible to the influ-ence of REIP,Vogelmann red-edge index 2(VOG2)and Water band index(WBI).This study established a foundation for retrieving photosynthetic pigment content in assimilated branches of Haloxylon ammodendron using hyperspectral remote sensing and machine learning models.It also contributed to the drought monitoring and rearing management of artificial Haloxylon ammodendron forests.
万方数据
维普
