Classification Of Dominant Tree Species Of Urban Vegetation Based On Multi-Source Remote Sensing Data ——A Case Study Of Shenzhen

Urban vegetation is helpful to carbon fixation and oxygen release,climate regulation,heat island mitigation,and habitat provision.It is an important ecological asset and is increasingly being valued by urban ecologists.Vegetation mapping is a crucial approach to quantitatively analyze the quantity and distribution of urban vegetation.Its results can be used for urban ecology research such as ecological survey,resource management,ecosystem service function accounting and quantify ecological effects and processes.Moreover,high-level,large-scale and high-precision mapping results are the prerequisite and foundation for the scientific management of urban green space to enhance the ecological benefits of urban vegetation and the well-being of urban residents.At present,there is no established system for urban vegetation mapping.The traditional top-down natural vegetation mapping method cannot solve the problem of high heterogeneity of urban vegetation.However,the mapping system for urban vegetation has difficulties in distinguish natural vegetation and artificial vegetation accurately.The scale is low,and it is difficult to meet the growing application demands of urban ecology.Although domestic and foreign scholars have carried out tentative work in many regions and based on different data sources.However,the existing work still has big deficiencies in the number and accuracy of classification especially multisource data fusion,classification paradigm and classifiers,etc.The classification process and classification methods need to be explored in depth.Urban vegetation mapping is still a key and difficulty mask in urban ecology.This study based on Google Earth Engine platform uses Shenzhen as the research area,through multi-source satellite remote sensing data,combined with detailed forest land survey data,to explore large-scale,high-precision,and multi-class mapping through by bottom-up classification of dominant species of vegetation.This study’s aim is to explore the accuracy of classification processes and methods such as different data sources,classification features,and classifiers.Under the optimal model,the impact of macro classification frameworks is further studied,such as different classification systems,sample selection methods,and mapping systems in order to provide suggestions and references for vegetation mapping research in other cities.The results of the study show that the highest accuracy is 77.96% and a Kappa coefficient is 0.76 based on pixel-based paradigm,random forest classifier and multitemporal sentinel,DEM,CHM,and Zhuhai OHS-MNF data when 23 dominant species are classified.（1）For the classification process and method,under the multi-source data with a spatial resolution of 10 m,there is low misclassification error in the pixel-based paradigm.And the accuracies of pixel-based paradigm are higher than the object-oriented paradigm under the same classification model.Spectral data is the most important,and multitemporal multi-spectral data is more important than single-period hyperspectral data,followed by phenological data and structural data,while texture information is not reflected its advantages.For urban vegetation mapping,the random forest classifier has the highest accuracy and is more suitable for classification mapping.（2）For classification framework,the classification system needs to be reasonably constructed according to the important value and the limit of the amount of information of the multi-source data to better achieve the trade-off between the classification accuracy and the number of classifications.In addition,the representativeness of the sample directly determines the classification accuracy.The sample construction of stratified random sampling is better than balanced stratification.The formation samples that are closer to the distribution of the true tree species are better for the natural communities.The monodominant samples cannot effectively represent the vegetation information about complex natural plant communities.Based on the method of bottomup classification of dominant species,its accuracy and effect are more practical than traditional top-down,and the result of dominant species classification can be clustered according to application scenarios,which is more suitable for the requirements of urban ecology.