Chlorophyll is an important participant in the photosynthetic process.Chlorophyll has been used to study and evaluate the forest productivity and forest health and is the basis of study forest ecosystem carbon sequestration ability.Therefore,it is very important to accurately estimate the chlorophyll content.Multiple linear regression models,neural network and support vector machine methods were used to establish the chlorophyll content estimating model based on the measured leaf chlorophyll content and corresponding reflectance spectrum in this study.The advantages and disadvantages of those models was compared and analyzed to provide a theoretical basis for estimating chlorophyll content based on remote sensing technology.The result showed that estimating model established using Erf-BP neural network method was the best one with the accuracy of up to 94.46%,the root mean square error(RMSE) of which was 3.321μg/cm2.The model established using support vector machine method was better with the accuracy of 88.74%,RMSE of which was 5.705μg/cm2.The model established using multiple stepwise regression method was restively poor with the accuracy and RMSE of 92.41% and 13.354μg/cm2 respectively.Erf-BP neural network was concluded to be the best method with high fitting precision and good stability by the comparison.%叶绿素是植物光合作用的主要参与者,是分析和评估森林生产力、森林健康和研究森林生态系统固碳能力的基础,准确的估算森林叶绿素含量极为重要.本研究利用实测叶片光谱和对应的叶绿素含量,分别使用多元统计、神经网络和支持向量机方法建立叶绿素含量估算模型,并比较和分析不同估算模型的优缺点,为开展相关研究提供参考.研究结果表明:三种建模方法中以Erf-BP神经网络模型最优,拟合精度达到了94.46%,RMSE为3.321 μg/cm2;支持向量机方法次之,拟合精度在92.41%,RMSE为5.705 μg/cm2;以多元逐步回归方程精度最低,为88.74%,且RMSE偏大,达到了13.354 μg/cm2.通过拟合精度和均方根误差2个指标可以明显地比较出Erf-BP神经网络具有最好的拟合精度和稳定性.
展开▼
