| Phyllostachys pubescens is widely distributed in collective forest areas in the south of China.As of the eighth forest resource inventory statistics,its area has reached 4.3 million hm2,accounting for 74%of the total area of all bamboo forests.It has an absolute advantage in all types of bamboo resources in China and has a comparative high economic,ecological and social benefits.The chlorophyll and nitrogen contents of vegetation leaves are important indicators of vegetation health and growth status,and are closely related to soil fertility.Estimating the chlorophyll and nitrogen content of phyllostachys pubescens forests has certain instructions for indirect estimation of the growth stress,nitrogen deficiency,and indirect estimation of soil fertility of phyllostachys pubescens forests.In this paper,Dagan Town,which is at Shunchang County in Fujian Province of China,was used as a study area.Linear estimation models,random forest estimation models and a support vector machine estimation models of chlorophyll and nitrogen content were respectively established by exploring sensitive spectral feature parameters based on excavating hyperspectral massive information excavating with using spectral analysis method at leaf and canopy level,and accuracy was comprehensively verified.Then the best estimate models for chlorophyll and nitrogen content were identified at leaf and canopy level,respectively.The main conclusions are as follows:(1)By analyzing the correlations among chlorophyll content of leaf with the original spectrum,first-order differential spectrum and vegetation index of leaves,the original spectra R593,R594 and R599,first-order differential spectra DR654,DR797,DR907 and DR11028 as well as the eight vegetation indices such as ratio vegetation index Ⅰ(RVIⅠ),ratio vegetation index Ⅱ(RⅥⅡ),greenness Normalized Vegetation Index(GNDⅥ),Simple ratio vegetation index(SRI),pigment-specific normalized difference(PNSD)and so on were all used as the spectral characteristic parameters.Leaf chlorophyll estimation models were established and their accuracies was comprehensively validated.The results showed that a one-dimensional linear model constructed with the vegetation index RⅥⅡ was adopted as theY=13.17*RⅥⅡ-28.166 had the best estimation effect,whose fitting coefficient(R2)was 0.5099,root mean square error(RMSE)was 8.2102μg/cm2,the forecast bias value(bias)was 0.4846,and the overall accuracy was 78.06%.(2)The correlations among leaf nitrogen content with leaf initial spectrum,first-order differential spectrum and vegetation indexes were analyzed.The original spectrum R387,first-order differential spectrum DR663,the six vegetation indexes including normalized vegetation index(NDⅥg-b),structure insensitive pigment index(SIPI),photochemical reflection index(PRI),pigment ratio index(PPR)were used as sensitive feature parameters.One-dimensional linear model,stepwise regression model,random forest(RF)algorithm and support vector machine(SVM)were applied respectively to constructs the leaf nitrogen content estimation model,and through a comprehensive comparison of the accuracy of the estimation results,it was found that the combined spectral parameters based on 6 sensitive features,SVM model which the penalty factor C and the nuclear parameter sigma were set to 3 and 0.1 respectively was the optimal.The coefficient of determination of the measured value and the predicted value was 0.7997,the overall accuracy was 93.87%,the root mean square error was 0.0343μg/cm2,and the prediction bias value was 0.0083.(3)Using the PROSAIL model to simulate the canopy reflectance,and compared with the measured reflectivity of the phyllostachys pubescens canopy,it was found that the simulated canopy reflectance was consistent with the measured canopy reflectance curve.There was a little fluctuation from about 400 to 700 nm,and after 700nm,the reflectivity of the model simulation was almost the same as the measured reflectivity of the phyllostachys pubescens canopy,even overlapping.It indicated that the PROSAIL model can well simulate the canopy reflectance of the phyllostachys pubescens.(4)Quantitatively analyzing the effects of various parameters of the PROSAIL model on the simulated canopy reflectance showed that the effect of LAI on canopy reflectivity of the model was most prominent in the band from 900 to 1400 nm.In the visible light range from 400 to 760nm,the sensitivity of chlorophyll(Cab)was second only to LAI and carotenoid(Car)had a relatively narrow range of influence and its sensitivity to leaf area index(LAI)and chlorophyll(Cab)differ greatly.The equivalent water thickness(Cw)mainly influenced the band after 1400 nm.The dry matter content(Cm)mainly influenced the band after 800 nm;in the band range from 400 to 1100 nm,the sensitivity from high to low was LAI>Cab>Cm>Cw>N>Car>ALA.(5)The leaf and canopy reflectance were compared,meanwhile,the correlation between the leaves chlorophyll content with the vegetation index,and between the canopy chlorophyll content with the vegetation index were analyzed.The results showed that the canopy reflectivity was generally smaller than the reflectivity of the leaves.At the leaf level,At the leaf level,chlorophyll and vegetations indexes including RⅥⅠ.RⅥⅡ、NDⅥ800-680、SRI、GNDⅥ、GNDVI801-550、MSR705、GNDⅥ801-550、PSND reached a significant correlation with P<0.01,but at the canopy level,chlorophyll was correlated with Viopt,modified triangular vegetation index(MTⅥ2),optimized soil-adjusted vegetation index(OSⅥ),red-edge vegetation stress index(RVSI),PSND and so on in all eight vegetation indices at P<0.05,which indicated that the chlorophyll content of phyllostachys pubescens had different degree of response at Leaf and canopy level.(6)The estimation models of canopy chlorophyll content and nitrogen content were constructed respectively by using a linear model,stepwise regression model,random forest and support vector machine estimation model,and the accuracies were verified.When estimating the canopy chlorophyll content,it was found that the the random forest estimation model whose mtry and ntry set 3 and 1000 respectively had the best effect.The determination coefficient of the model was 0.4560,the root mean square error was 0.028μg/cm2,the prediction bias value was-0.0208,and the overall accuracy was 69.59%.When estimating the canopy nitrogen content,it was found that the support vector radial basis model with the penalty factor C and the nuclear parameter Sigma set to 2 and 0.1 respectively had the best estimation effect.The estimated coefficient of estimated value of the model with measured value was 0.5257,the root mean square error was 0.1083μg/cm2,the prediction bias value was0.0674,and the overall accuracy was 82.52%. |
PDF Full Text Request