Classification Of Picea Crassifolia And Sabina Przewalskii Based On Multi-source Remote Sensing Images

Forest ecosystem is an important part of land ecosystem,which plays an important role in controlling soil erosion,regulating climate change,maintaining ecological balance.Mastering the information of dominant tree species is also of great value to forest resources.Picea crassifolia and Sabina przewalskii are the dominant tree species in Qilian Mountain Nature Reserve,both of which are irreplaceable in water conservation and soil and water conservation.It is of great significance to extract the spatial distribution of the two types of trees.Through the Sentinel-2A,Sentinel-1A and Landsat-8 and ASTER DEM,this paper analyzed and extracted 22 feature variables,including spectral features,vegetation index,topographic features,texture features and microwave features.Based on the random forest algorithm and hierarchical classification technology,the paper classified Picea crassifolia and Sabina przewalskii in Qilian Mountain National Park of Gansu Province.By comparing and analyzing the results of different feature combinations,the best feature combination with the high accuracy was selected.It also analyzed the spatial distribution of Picea crassifolia and Sabina przewalskii and made statistics of the area and proportion of them in different functional divisions.The main conclusions are as follows:(1)Compared with single remote sensing image,adding multi-source image and terrain data helps to improve the classification accuracy.The accuracy of Sentinel-2A image features for forest is only 93.36%,whlie adding DEM and Landsat-8 image can improve the overall accuracy by 2.14%and 0.56%.The overall accuracy of Landsat-8features is 81.13%and the kappa coefficient is 0.6733,while Sentinel-2A image is92.45%and the kappa accuracy is 0.8696.Based on the Sentinel-2A image,adding DEM,Landsat-8 image and Sentinel-1A image in turn,it was found that the classification accuracy of the dominant tree species has increased by different degrees.The best combination with highest accuracy includes Sentinel-2A image features,Landsat-8 image features,DEM features and VV polarization backscatter coefficient of Sentinel-1A,whose overall accuracy is 96.50%and the kappa accuracy is 0.9394.(2)The most important topographical feature are elevation and aspect,they are closely related to the habitat characteristics of dominant species.Picea crassifolia prefers humid environment,which is mainly distributed on the shady slope,semi shady slope and humid valley of mountain,while Sabina przewalskii prefers sunny slope or semi sunny slope.The average altitude of spatial distribution for Sabina przewalskii is higher than Picea crassifolia,therefore,the altitude and slope are benefit for identification.S2mean,which is extracted from Sentinel-2A image with higher spatial resolution and richer band information,has a great contribution.Its score is10.85 and the scores of Enhanced Vegetation Index(EVI)and Difference Vegetation Index(DVI)are 3.59 and 4.11.Even the EVI of lower resolution landsat-8 image is also high.In addition,the score of S2b67 from Vegetation Red Edge 2 and Vegetation Red Edge 3 of Sentinel-2A image reached 6.37,which is great significant for identifying dominant tree species.(3)The backscatter coefficients of Sentinel-1A in VV and VH polarization are different in dominant tree species classification.VV polarization can improve the classification accuracy,while VH polarization reduced the classification accuracy.The radar wave intensity returned by the interaction between the dominant species and microwave in VV polarization mode is stronger,while VH polarization mode is weaker.The difference of backscatter coefficient between the two dominant species of VV polarization mode is 2.667882dB,which is greater than VH polarization mode.Consequently,the VV polarization mode contributes more for identifying dominant species.The differences of tree structure,height,biomass and water content make the radar backscatter echo intensity and scattering degree of Picea crassifolia and Sabina przewalskii different.Picea crassifolia has higher tree height,humid habitat,better soil water and water conservation under the forest,less dense gaps in the canopy.It is difficult for radar wave to penetrate the canopy and cause body scattering on its surface,therefore,the backscatter coefficient of Picea crassifolia is higher than Sabina przewalskii.(4)According to the statistical analysis,the main distribution of Picea crassifolia in the study area is 2600-3200 meters above sea level,and Sabina przewalskii is3000-3500 meters above sea level.Both are mostly distributed on slopes and steep slopes with slopes ranging from 16°to 35°.There are 71%of Picea crassifolia distributed the slope from northwest to north,northeast to north,and northeast to east,where are shady slopes and semi-shady slopes.The centralized distribution of Sabina przewalskii is different from Picea crassifolia,50%of Sabina przewalskii is distributed in the area from slope to northeast to southeast,southeast to south,and36%is distributed from south to southwest,southwest to west,west to northwest,and northwest to north.The area of dominant tree species in different functional areas is quite different.,the most area is traditional utilization area,where Picea crassifolia is834.28km~2 and Sabina przewalskii is 65.96km~2,accounting for 86.10%and 60.47%.The least is the core reserve,where Picea crassifolia and Sabina przewalskii are4.17km~2and 10.91km~2,accounting for only 0.43%and 10.01%.The core protected area of dominant tree species is too small,while the traditional use of more human activities and frequent agricultural land activities includes too many dominant tree species.It is recommended to adjust the functional zoning boundaries to manage and protect the dominant tree species better.