欢迎访问《生态学杂志》官方网站,今天是 分享到:

生态学杂志

• 方法与技术 • 上一篇    下一篇

基于优化SVR高光谱指数的独尾草叶绿素含量估算

谭林1,2,3,何秉宇1,2,3,刘卫国1,2,3*,庞冬1,2,3#br#   

  1. (1新疆大学资源与环境科学学院, 乌鲁木齐 830046; 2绿洲生态重点实验室, 乌鲁木齐 830046; 3智慧城市与环境建模自治区高校重点实验室, 乌鲁木齐 830046)
  • 出版日期:2017-02-10 发布日期:2017-02-10

Estimation of chlorophyll content of Eremurus chinensis based on optimization support vector regression machine.

TAN lin1,2,3, HE Bing-yu1,2,3, LIU Wei-guo1,2,3*, PANG dong1,2,3#br#   

  1. (1College of Resources and Environment Science, Xinjiang University, Urumqi 830046, China; 2Key Laboratory of Oasis Ecology, Urumqi 830046, China; 3Key Laboratory of Intelligent City and Environmental Modeling of Autonomous Region, Urumqi 830046, China).
     
  • Online:2017-02-10 Published:2017-02-10

摘要: 以位于新疆准噶尔盆地腹地的古尔班通古特沙漠为研究区,测定独尾草幼苗期、开花期的叶片光谱反射率和叶绿素含量,分析24种光谱指数与叶绿素含量之间的相关关系,选用相关性较高的光谱指数建立优化支持向量回归机(SVR)估算模型。结果表明:(1)开花期的叶绿素含量高于幼苗期,主要与植被的光合作用有关,开花期的光谱反射率低于幼苗期,两期的光谱反射率符合普遍植物光谱反射率。(2)在幼苗期,GNDVI(green normalized difference vegetation index)与叶绿素含量相关性最高(R2=0.664);在开花期,GM-2(Gitelson and Merzlyak)与叶绿素含量相关性最高(R2=0.711)。按相关性排序时,在两期中,决定系数排名前7的光谱指数都相同。(3)将7个敏感光谱指数作为输入因子,通过3种优化算法选择最优参数(c,g),建立优化SVR估算模型:幼苗期和开花期,模型精度都较高,PSO-SVR>GA-SVR>GS-SVR,其中PSO-SVR决定系数最高,均方根误差最小。在幼苗期,PSO-SVR决定系数为0.812,均方根误差为0.728,在开花期,PSO-SVR决定系数为0.841,均方根误差为0.247。说明基于PSO-SVR算法优化后的SVR模型精度高误差小,能较好地对叶绿素含量进行估算,且独尾草叶绿素含量开花期的估算比幼苗期的效果要好。本研究为荒漠植被生态特征的监测估算、时空分布和生化参数反演提供了科学依据和技术支持。

Abstract: Spectral reflectance and chlorophyll content of the leaves were determined in the seedling stage and flowering stage of Eremurus chinensis in Gurbantunggut desert located in the center of the Junggar basin. Relationship was developed between 24 kinds of spectral index and chlorophyll, and estimation model was established by support vector regression (SVR) using the optimal spectral index which had high interrelationship. The results showed that (1) the chlorophyll content in flowering stage was higher than that in seedling stage, which was mainly related to the photosynthesis of vegetation, and the spectral reflectance of the flowering stage was lower than that of seedling stage. The spectral reflectance of two phases was consistent with the spectral reflectance of common plants. (2) The GNDVI and chlorophyll content showed the highest correlation with coefficient of determination (R2) values of 0.664, among all the spectral indices. In the flowering stage, the correlation coefficient between GM-2 and chlorophyll content showed the highest correlation with R2 values of 0.711. According to the rankings of correlation coefficients (R2), the first 7 of the spectral index were the same in the two periods. (3) Seven sensitive spectral indices were used as input factors, and the optimal SVR estimation model was established by selecting the optimal parameters (c,g) from 3 kinds of optimization algorithms. The model accuracy was high during both the seedling stage and flowering stage. The model accuracy was ranked as, PSO-SVR>GA-SVR>GS-SVR, with the PSO-SVR coefficient being the highest and the root mean square error (RMSE) the least. In the seedling stage, the PSOSVR and coefficient was 0.812 with the RMSE being 0.728. In the flowering stage, the PSOSVR coefficient was 0.841 with the RMSE being 0.247. The results showed that the SVR model based on PSO-SVR algorithm had high precision with low error, which can estimate the chlorophyll content accurately. Moreover, the estimation of chlorophyll content in the flowering period was better than that in the seedling stage. The result would provide scientific basis and technical support for monitoring and estimating ecological characteristics of desert vegetation, spatial and temporal distribution of desert vegetation, and biochemical parameter inversion.