Study On Biomass Estimation Of Wetland Vegetation In Longbaotan Area Based On Multi-angle And Hyperspectral CHRIS Data

Vegetation biomass is a significant indicator of the structure function and health condition of wetland ecosystem,which directly mirrors the growth of vegetation communities.In this paper,Longbaotan national wetland reserve in Qinghai Province was taken as the research region.Taking advantage of hyperspectral dimension and multi-angle stereoscopic structure information of CHRIS/PROBA data,we fused different angle NDVInarrow with 0° image using PCA,Gram-Schmidt transform and NNDiffuse fusion methods.The best fusion method was ascertained by subjective and objective assessment.Then vegetation types in Longbaotan wetland were classified by SVM method.In addition,original spectral reflectance,typical narrow band vegetation indices,red edge indices and VInew vegetation index were extracted from +36°,0° and-36° images.Analysing the correlation between these remote sensing factors and fresh and dry weight biomass respectively,exploring the sensitivity of biomass to angle,we established the best model to estimate the biomass of alpine wetland vegetation using regression analysis method.The aboveground biomass was estimated in Longbaotan area.Finally,we built the belowground biomass model of swamp meadow and alpine meadow individually,and then gained the belowground biomass and total biomass of Longbaotan area.The main conclusions of this study are as follows:(1)Three fusion methods were evaluated quantitatively.Results showed that NNDiffuse fusion had the maximum average gradient,information entropy and correlation coefficient.It indicates that NNDiffuse method contains more information,more abundant details and higher legibility.By comparing the results of different angle information fused,it was found that the mean value,standard deviation,average gradient,information entropy,correlation coefficient and OIF value of-36 degree image were slightly higher than +36 degree and ±36 degrees.It indicates that-36° fusion image is the best one.When multi angle information was fused,the more angle image did not include more information.Too many angle information added led to redundance,which would depress the quality of fusion image.Then the support vector machine classifier was carried out on 0 degree image and fusion image.Classification results showed that the overall accuracy of fusion image was 94.64%,2.94% higher than single angle image,and Kappa coefficient increased by 0.0381.The omission error of bare land and herbaceous swamp and the surplus error of lake wetland and alpine grass were significantly improved.(2)By analyzing the correlation between the spectral reflectance of different angle images and aboveground biomass data,we found that the correlation coefficient of +36° image was significantly higher than that of +36° and-36° images,which indicates that +36° image is more sensitive to biomass.The tendency of correlation coefficient of three angle spectral variables was approximately consistent.When the wavelength is less than 725 nm,the spectral reflectance of each angle is negatively related with biomass.When the wavelength is greater than 725 nm,the spectral reflectance and biomass have negative correlation.The correlation coefficient is zero at around 725 nm.(3)Considering the spectral characteristics of vegetation in Longbaotan wetland and the correlation between biomass and band reflectance of hyperspectral data,we established a new vegetation index VInew using band 25,band 33 and band 45.The correlation coefficients between vegetation narrow band indices,red edge indices,VInew index and aboveground biomass were compared with each other.It was found that Datt1,VOG1,REP and VInew index had famous correlation with aboveground biomass.Afterwards the models for retrieving dry weight and fresh weight of aboveground biomass were built by the method of regression analysis.The exponential model based on-36° VInew had the best fitting for dry weight,whose determination coefficient was 0.613.And the exponential model of taking +36° VInew as the independent variable was the fittest one for fresh weight,and the R~2 was 0.693.(4)The models for reflecting the relationship between aboveground biomass and belowground biomass in swamp meadow and alpine meadow were established individually.As a result of sufficient moisture,the root of swamp meadow grows luxuriantly,which causes high belowground biomass.The average value is 9,696.63 g/m~2.But the biomass of alpine meadow is low,and the average is 1,317.27 g/m~2.The distribution of biomass in Longbaotan area increases from west to East.The biomass of swamp meadow is high around the lake,while alpine meadow located on the sides of mountains and roads has the lower values.The dry weight of aboveground biomass in Longbaotan area is 204.19 g/m~2,and the fresh weight is 658.79g/m~2.The average of total biomass is 4,038.73 g/m~2,and the biomass of marsh meadow is 10030.96g/m~2,and alpine grassland is 1,465.59 g/m~2.