首页 >  2003, Vol. 7, Issue (5) : 364-371

摘要

全文摘要次数: 3404 全文下载次数: 22
引用本文:

DOI:

10.11834/jrs.20030505

收稿日期:

2002-05-08

修改日期:

2002-06-27

PDF Free   HTML   EndNote   BibTeX
线性回归模型估算水稻叶片叶绿素含量的适宜性分析
1.南京师范大学地理科学学院,南京江苏210097;2.浙江大学,浙江杭州310029
摘要:

利用PROSPECT模型模拟水稻叶片叶绿素含量从20.0μg/cm^2变化到40.0μg/cm^2时的叶片光谱特性,利用FCR模型模拟叶面积指数(LAI)为1,2,…,7时,不同地面状态下,4个不同观测方向的水稻冠层反射率。利用LAI为1,3,5,7时的模拟值,采用多元逐步回归分析法,从不同观测方向建立叶片叶绿素含量与冠层反射率(见)及其变化式ln(1/Rλ),R’λ的多元线性回归模型,并用复相关系数和均方根差评价拟合精度,认为ln(1/Rλ)以及从天顶方向的拟合效果最好。利用从天顶方向建立的回归模型,预测叶片叶绿素含量,认为将该回归模型应用于其它方向是不合适的,从天顶方向预测时,预测精度受地面状态的影响,但总的说来,预测精度呈现随LAI的增大而提高的趋势。

The Robustness of Linear Regression Model in Rice Leaf Chlorophyll Concentration Prediction
Abstract:

We present a modeling approach to assess the robustness of remotely derived spectrometric equations predictive of rice leaf chlorophyll concentration to the view direction. Our methodology uses two radiative transfer models that operate at lea(f PROSPECT)and canop(y FCR)levels. It includes three stages:(1)Simulation canopy bi-directional reflectance on varies leaf chlorophyll concentration,varies leaf area index( LAI)and varies understory;(2)establishment of predictive relationships of chlorophyll concentration with stepwise regression;(3)assessment of the robustness of these relationships. First,we simulate rice leaf spectrum by PROSPECT model. The parameters used in the model are:protein concentration(0.0008g / cm2),cellulose concentration(0.0049g / cm2),water equivalent thick(0.02cm)and leaf structure paramete(r 1.7),while chlorophyll concentration varises from 20μg / cm2 to 40μg / cm2,the change step is 0.1μg / cm2; simulate rice canopy bi-directional reflectance by FCR model. The parameters used in the model are:leaf relative linear size(0.40),model inclination(81),model eccentricity(0.9985),sun zenith angle(32o),sun azimuth angle(0o),and sun direct radiation / total radiation(80%),the changed parameters are LAI,view direction and understory spectrumcharacter. LAI varies from 1 to 7. The change step is 1. There are four view directions:nadir view direction(both view zenith angle and azimuth angle equal to 0o),hot spot view direction(view zenith angle equals to 32o,azimuth angle equals to 0o),oblique view direction(view zenith angle equals to 32o,azimuth angle equals to 90o)and specular view direction( view zenith angle equals to 32o azimuth angle equals to 180o). The reflectance of three types of understory are measured in 1999. Second,the established multiple linear regression model by stepwise regression analysis uses simulation values (LAI are 1,3,5 and 7,chlorophyll concentration varies from 20μg / cm2 to 40μg / cm2,three types of understory)on different directions to predict chlorophyll concentrations. The considered factors are canopy reflectance Rλ ,derived ln(1 / Rλ )and R'λ. The regression equations are established by SPSS software. The selected wavelengths are upon F probability and root mean square erro(r RMSE). Only when rejection probability is less than 0.05,the wavelength can be selected. To make the equation simple,we just chose three wavelengths which partial correlation index is the biggest for each regression equation. Then,the robustness is valued by compound coefficient of correlation(r2)and RMSE . For the equations establishment by the three factors (Rλ ,ln(1 / Rλ ),and R'λ ),r2 on nadir view direction are 0.905,0.916 and 0.883,on hot spot view direction are 0.774,0.962 and 0.747,on oblique view direction are 0.563,0.941 and 0.572,on specular view direction are 0.881,0.937 and 0.883,RMSE on nadir direction are 1.97,1.68 and 1.99,on hot spot view direction are 2.75,1.13 and 2.92,on oblique view direction are 3.83,1.41 and 3.80,on specular view direction are 2.00,1.45 and 1.98. Those stress that the robustness are strongest on ln(1 / Rλ )and nadir direction. The equations on nadir direction are: chl = 19.882 + 5747.957R410 - 3770.780R415 - 414.002R705 (1)chl = 74.631 - 224.236ln(1 / R410 )+ 190.960ln(1 / R415 )+ 39.24ln(1 / R715 ) (2) chl = 59.526 - 149370.7R'420 + 131394.74R'555 + 28899.908R'585 (3) in the equations,chl indices chlorophyll concentration. We predict chlorophyll concentrations by nadir direction equations use simulated reflectance on other directions and see bigger RMSE occurred(for example,the biggest RSME is 26.74 on specular view direction). It stresses that the equations built on nadir view direction are unsuitable for other directions’prediction. When we predict chlorophyll concentrations by nadir direction equations use simulated reflectance on nadir direction( LAI equal to 2,4 and 6,chlorophyll concentration varies from 20μg / cm2 to 40μg / cm2,three types of understory),we find that the predictive accuracy is affected by different understoy,but tends to be robust when LAI increases (for example,on understory A,the RSME of equation(1)equals to 1.81,1.23 and 1.25 when LAI equals to 2,4 and 6,but on understory B it’s 0.34,0.15 and 0.09). The prediction made by measured rice canopy reflectance proves this result too.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群