| Forest,known as the"lungs of the earth",plays an important role in balancing the global ecosystem and is the basis for human survival and development.Forest stock is an important index for evaluating the quantity and quality of forest resources.Traditional forest stock survey based on the field needs a lot of manpower,material resources and a long duration.It is difficult to meet the development needs of modern forestry.Therefore,how to quickly and accurately obtain large-scale forest stock in different regions has become one of the research hotspots in the field of forestry.The rapid development of remote sensing technology can effectively solve the difficult problems of traditional forest stock survey,and provide convenience for forest resource survey and dynamic monitoring.Taking Xiangcheng County,Ganzi Prefecture,Sichuan Province as the research area,using SPOT6remote sensing image as data source,this paper classifies remote sensing image by Maximum Likelihood Classification,extracts coniferous forest distribution information,constructs coniferous forest volume estimation model with spectral information,vegetation index,texture information and topographic information as characteristic variables,and explores the effect of different window sizes and texture characteristics on coniferous forest storage.The influence of the accuracy of the product estimation model is studied,and the optimal estimation model and the optimal texture window are selected.The conclusions are as follows:(1)After the remote sensing image pretreatment,land classification was carried out by Maximum Likelihood Method,the results showed that there were 9 types of land,including coniferous forest land,broad-leaved forest land,mixed coniferous and broad-leaved forest land,shrub forest land,grassland,cultivated land,water area,construction land and other land.The result of image classification is ideal,the overall classification accuracy is 87.64%,and the Kappa index is 0.8579.(2)The results show that the correlation between spectral characteristics,vegetation index of the SPOT 6 remote sensing image and topographic characteristics and coniferous forest volume is strong,while the correlation between texture information and coniferous forest volume is weak.Among the six spectral characteristic variables,single band(B1,B2,B3,B4)and ratio band(B4/B2)were significantly correlated with forest volume,whose correlation coefficients were-0.563,-0.553,-0.629,-0.376 and 0.506;Normalized Differential Vegetation Index(NDVI),Ratio Vegetation Index(RVI)were significantly correlated with forest volume among the three vegetation index characteristic variables,with the correlation coefficient of 0.54.Among the three topographic factors,elevation has a strongly significant correlation with the volume,and the correlation coefficient is 0.641.Among the eight texture feature variables with different size of texture window,only Mean Value(MEA)has a very significant positive correlation with the volume.(3)The results showed that elevation,single band(B1,B2,B3),ratio band(B4/B2),Normalized Differential Vegetation Index(NDVI),Ratio Vegetation Index(RVI)and Mean Value(MEA)had the greatest impact on coniferous forest volume.(4)Texture features of various window sizes have some influence on the accuracy of coniferous forest volume estimation.Mean value(MEAN)has the strongest correlation with coniferous forest volume in eight different texture features.The selection of the optimal texture window is uncertain.The optimal texture window of the coniferous forest estimation model in Xiangcheng County is 9×9.(5)Based on spectral information,vegetation index,texture characteristics and t opographic factors,coniferous forest volume estimation models were established und er 6 texture windows of 3×3,5×5,7×7,9×9,11×11 and 13×13.The optimal estim ation model was obtained by accuracy test,which was:V=146.618+0.0144361×Elevation-0.141139×B1-0.105841×B2-0.0946412×B3+18.5399×B4/B2+183.024×NDVI+14.2387×RVI-12.0595×MEA9.(9×9 window,Estimation accuracy is 86.74%,R~2=0.87971,RMSE=23.695 m~3/hm~2,RE=13.26%)... |
PDF Full Text Request