基于CARS / RFG 优选植被指数的矮林芳樟叶水势反演模型

Inversion Model of Leaf Water Potential of Cinnamomum camphora Dwarf Forests Based on Vegetation Index Optimized by CARS/RFG

摘要:

叶水势（Leaf water potential, LWP）是反映植物水分状态的关键指标，快速准确监测矮林芳樟LWP对调控林地水分管理策略具有重要意义。本文以矮林芳樟为对象，利用无人机搭载多光谱相机采集冠层光谱数据，通过皮尔逊相关性分析（PCC）、竞争性自适应重加权采样算法（CARS）和随机蛙跳算法（RFG）筛选出有效反映矮林芳樟水分特征且相互间具有较低冗余度的植被指数组合，结合反向传播神经网络（BPNN）、径向基函数神经网络（RBFNN）、随机森林（RF）和极限梯度提升（XGBoost）构建矮林芳樟LWP多光谱反演模型，基于决定系数（R2）和均方根误差（RMSE）筛选最优模型。结果表明：在快速生长阶段（6—7月）和稳定生长阶段（8—9月），RFG-XGBoost均为矮林芳樟LWP反演的最优模型，其中2阶段训练集R2分别为0.9945、0.9930，RMSE分别为0.0811、0.0526 MPa;测试集R2分别为0.8066、0.8225，RMSE分别为0.3484、0.1191 MPa。表明基于无人机多光谱影像的RFG-XGBoost模型在反演矮林芳樟LWP方面精度较高，在快速获取关键信息方面具有较大应用潜力。研究结果可为芳樟林间水分管理提供重要科学依据和技术支持。

Abstract:

Leaf water potential (LWP) is a key indicator reflecting plant water status, and rapid and accurate monitoring of LWP in Cinnamomum camphora dwarf forests is crucial for regulating forestland water management strategies. Canopy spectral data of Cinnamomum camphora dwarf forests were collected by using a drone-mounted multispectral camera. Through Pearson correlation analysis (PCC), competitive adaptive reweighted sampling (CARS) algorithm, and random frog (RFG) algorithm, combinations of vegetation indices that effectively reflect the water characteristics of Cinnamomum camphora dwarf forests with low mutual redundancy were screened out. Then, multispectral inversion models for LWP of Cinnamomum camphora dwarf forests were constructed by combining back propagation neural network (BPNN), radial basis function neural network (RBFNN), random forest (RF), and eXtreme gradient boosting (XGBoost). Finally, the optimal model was selected based on the evaluation criteria of coefficient of determination (R2) and root mean square error (RMSE). The results showed that the RFG-XGBoost model was the optimal one for LWP inversion. For the training set in these two stages, the R2 values were 0.9945 and 0.9930, with RMSE values of 0.0811 MPa and 0.0526 MPa, respectively;for the test set, the R2 values were 0.8066 and 0.8225, with RMSE values of 0.3484 MPa and 0.1191 MPa, respectively. It was concluded that the RFG-XGBoost model based on UAV multispectral images showed high accuracy in inverting LWP of Cinnamomum camphora dwarf forests, demonstrating its application potential in quickly obtaining key information. The research results can provide important scientific basis and technical support for water and fertilizer management in Cinnamomum camphora forests.