Research On The Applicability Of 4-Scale Model Spatial Scale

As a representative of geometric optics,4-Scale model describes forest canopy reflectance from four scales(tree distribution,canopy morphology,branch and leaf structure),takes into account the distribution of internal canopy pores and overlapping and shadow effects between canopy layers,and has strong universality,which can make the forest canopy reflectance simulation more accurate,it is suitable for any canopy condition with high precision,forest canopy reflectance is very important for forest research,it can not only accurately interpret vegetation types,but also greatly affect forest biophysical and biochemical parameters obtained from inversion,including leaf area index(LAI),biomass,canopy density,photosynthetic active radiation absorption coefficient,chlorophyll and carbon sequestration parameters,these variables greatly affect the simulation of the carbon cycle in forest ecosystems.With the wide application of high spatial resolution remote sensing data,the observation information such as the spatial distribution of trees within the pixel will be inaccurate,and the parameters such as the distribution of trees within the pixel range are exactly required by the model,the inaccurate parameter simulation of the forest canopy reflectance of the 4-Scale model will cause large errors and even lead to the failure of the model,therefore,it is necessary to clarify the applicable spatial scale of 4-Scale model to simulate forest canopy reflectance,namely model calibration,which is conducive to the high-precision inversion of biophysical and biochemical parameters.This paper conducted an in-depth study on the spatial scale of 4-Scale model suitable for high-precision inversion parameters,the study site was Maoershan Experimental Forest Farm in Shangzhi City,Heilongjiang Province,the forest canopy reflectance with 10 m,20m,30 m,40m and 50 m pixel sizes was simulated in each 100m×100m sample plot of coniferous forest and broad-leaved forest.Remote sensing data source is Sentinle-2,local mean method,nearest neighbor method,bilinear interpolation method and cubic convolution method were used to transform the image scale.The simulated values of forest canopy reflectance at different spatial scales and the pixel reflectance extracted from remote sensing images were compared and analyzed,it is concluded that coniferous forest and broad-leaved forest are suitable for using 4-Scale model to simulate the spatial scale of forest canopy reflectance.The results are as follows:(1)The sensitive parameters of the 4-Scale model were dry height,crown height,crown radius,LAI,clumping index(Ω)and spatial distribution parameters of trees,LAI and Ω had scale effect,no scale effect was found in 20 m broad-leaved forest,a direct method was used to convert LAI,the scale conversion of Ω was weighted according to the species composition of the plot.(2)The Spectral Response Function of Sentinel-2 can be simulated and almost fully fitted by using the Gaussian function of eighth order,which is very close to the real spectral response value,it can correct the reflectance of leaves and background,and minimize the error of simulated forest canopy reflectance.(3)The nearest neighbor method,bilinear interpolation method and cubic convolution method are used for up-scale conversion of remote sensing images.The 20 m spatial scale range of broad-leaved forest is homogeneous,then the scale is transformed by local mean method,the mean value,standard deviation and Peak Signal-to-Noise Ratio(PSNR)were used to evaluate the spectral information carried by the converted remote sensing images,the results show that the cubic convolution method has the best effect of preserving spectral information and covers the most abundant information.(4)The 4-scale model underestimated the forest canopy reflectance of pixels as a whole.Coniferous forest and broad-leaved forest were different in spatial scales suitable for highprecision inversion parameters.The 4-scale model showed the best applicability at 40 m for coniferous forest and 30 m for broad-leaved forest.The mean and standard deviation of the difference between the simulated values and the reflectance values of the remote sensing pixels were the smallest in the red and near-infrared bands.Simulation like yuan reflectivity and remote sensing reflectance at two wavelengths as yuan root mean square error(RMSE)and mean absolute error(MAE)of RMSE=0.0122 respectively,MAE=0.0122,RMSE=0.1458 and MAE=0.1432.Broad-leaved forest RMSE=0.0157,MAE=0.0154,RMSE=0.1635,MAE=0.1616.(5)The simulation results of coniferous forest and broad-leaved forest at 10 m scale were not stable,and the size rules of mean and standard deviation were inconsistent,and the difference between RMSE and MAE in the same band was large.The simulation effect at 20 m scale is the worst and the number of outliers is the largest.The RMSE and MAE in red and near-infrared band are larger.