Estimation Of Forest Stock Volume In Beijing Based On Remote Sensing

Forest stock volume is one of the crucial stand parameters to evaluate forest resources.Accurately investigating the distribution of forest stock volume is very important to reasonably evaluate the development and utilization status and planning deployment of forest resources.The traditional forest stock volume survey method has the shortcomings of long survey cycle,heavy workload and low survey accuracy.With the rapid development of the national economy,it is increasingly urgent to accurately investigate the distribution of forest volume in the short term.Remote sensing technology can be used to investigate the distribution of forest volume in a relatively short period of time when a small number of plots are set.This paper selects the Beijing as the study area.Sentinel-2 images and the national forest stock volume inventory data were used as data source.61 feature factors of Band reflectance,vegetation index,texture factor was selected from satellite images.The correlation analysis of feature factors and forest stock volume were analyzed,and 18feature factors with the highest correlation are selected as independent variables.With the forest stock volume of a single sample plot as the dependent variable,the estimation model of forest stock volume was established by partial least squares method,support vector machine,random forest,artificial neural network.By analyzing the regression accuracy of each model,the appropriate method to build forest stock volume inversion model was Selected.Combined with the topographic characteristics and field survey results,the spatial distribution of forest volume was briefly analyzed.The main results of this study are as follows:(1)Before quantitative remote sensing inversion,the relative radiometric calibration of multi-view images involved in modeling can reduce the error caused by sensors and other factors,and improve the accuracy of the final model inversion.(2)Among the three feature factors extracted,the correlation between vegetation index and forest stock value is the highest,up to 0.44,followed by single-band reflectance,and the correlation between texture factor features and forest stock value is relatively low.The correlation between texture factor and forest volume increases with the increase of window size.(3)Compared with the four methods used for modeling inversion,the artificial neural network method has the highest accuracy,the root mean square error is 1.18 m~3,and the average absolute error is 0.96 m~3.The determination coefficient is 0.61.Finally,the inversion model constructed by neural network method was used to invert the forest volume of Beijing,and the forest volume of Beijing in 2019 was estimated to be 27.31million m~3.Compared with the results of the ninth forest resource inventory in Beijing(2016),the estimation accuracy of the model was 87.9%.Compared with the predicted forest volume of Beijing in 2019,the estimation accuracy of the model was 93.3%,which proved that the model can predict the value and distribution of forest volume in Beijing.(4)Combined with field investigation experience,by analyzing the distribution of forest volume in different slope directions,it is concluded that the distribution of forest volume in the back shady slope direction is generally higher than that in the sunny slope direction,and the forest volume value in the back shady slope direction is higher than that in the sunny slope direction,reaching 21%.The main reason is that the back shady slope is less affected by solar radiation,which is conducive to the enrichment of water and other nutrients,so as to use forest growth and increase forest volume distribution.