Remote Sensing Inversion And Spatial Distribution Pattern Research Of Forest Biomass In Wangqing Area

The development of forestry remote sensing technology has provided a convenient and effective method for the long time series estimation of forest biomass of large area.The higher resolution of Landsat data contributes to the identification of forest types and the extraction of forest ranges,the rich band information reflects the characteristics of forest vegetation and provides a good data source for forest biomass monitoring.In this study,Wangqing area,Jilin Province,was used as the research area.The 1994,2004 and 2014 Landsat images were used to extract the spectral and texture features,DEM geographic data was used to extract the topographic factors.Using support vector machine to classify the land cover and identify the forest type.At the same time,the biomass model of different forest types was retrieved by linear(stepwise regression)method and nonlinear(BP neural network and support vector machine for regression)method,and the forest biomass model with strong generalization ability was selected to predict the forest biomass in the study area.Analyzed the changes of forest biomass from 1994 to 2014 in the study area combined with the distribution map of forest biomass.The main contents and results are as follows:(1)Processing Landsat image data by seamless stitching,radiation calibration,atmospheric correction,geometric correction,cutting and etc.The field biomass was calculated according to the growth rate of biomass of different tree species.(2)Using the Landsat image data and DEM data,the remote sensing feature variables and the geometric characteristic variables were extracted respectively.The support vector machine classification method was used to classify the land cover types,evaluating the classification accuracy and collecting the forest land area.The results showed that the radial basis function SVM training and prediction classification are the best when C = 2 and g = 0.5,and the SVM is used to identify the forest type.The classification accuracy of the 1994,2004 and 2014 was 85.31%,88.98%,91.46%.(3)Based on the spectral parameters of Landsat data and the terrain parameters of the study area,the biomass estimation model of different forest types was established by stepwise regression analysis.The results showed that the prediction accuracy of the biomass model of needle-leaved forest,broad-leaved forest and mixed forest was 0.746,0.517 and 0.655.The accuracy of the three models was low,linear model estimation of forest biomass is not accurate.It's necessary to use non-linear models for estimating forest biomass.(4)Using principal component analysis,select the main parameters.The BP-neural network and the support vector machine for regression(SVR)were used to invert the forest biomass models,and the accuracy and applicability of the mode was evaluated.The results showed that the prediction accuracy of the biomass model of needle-leaved forest,broad-leaved forest and mixed forest obtained by BP-neural network was 0.769,0.842 and 0.867.The prediction accuracy of the biomass model of needle-leaved forest,broad-leaved forest and mixed forest obtained by SVR was 0.902,0.937 and 0.899.On the whole,the SVR model has better fitting and prediction accuracy,and the model predictive performance is stable and the generalization ability is strong.(5)Through the SVR model,estimating the forest biomass of 1994,2004 and 2014,and outputting the forest Biomass Distribution.The forest biomass values of the study area at each time point were analyzed,and the change of forest biomass in 20 years was analyzed according to the forest biomass map.The spatial distribution and variation of forest biomass were analyzed according to the elevation and slope of the study area.The results showed that the trend of forest biomass change was reduced first and then increase in 20 years,which was similar to that of forest land area.The average annual growth rate of forest biomass reached 0.32 t/hm2 from 1994 to 2014.The forest biomass in Wangqing area is mainly distributed in the range of 300?1000m above sea level and 0 °?35 °slope.In 20 years,the forest biomass changes are mainly concentrated in the middle and low altitude and the 6 °?35 °slope areas.