| Urban Open Space is a dynamic ecosystem in cities,which undertakes functions such as carbon fixation and oxygen release,climate regulation,and water conservation.And Urban Open Space also represents the process of urban modernization.The previous studies use two-dimensional indicators such as green area and green coverage to evaluate the urban green space.However,those indicators are difficult to show the ecological function value and environmental benefit of urban green space in three-dimensional level.Otherwise,most researchers get the information of each green space based on traditional quadrat survey,which rely on manual measurement.There are problems such as high time costs,large measurement errors,etc.Therefore,accurately and quickly obtaining green space information in three-dimensional has become an important topic in urban green space resource management.At the same time,the emergence of new technologies such as UAV lidar and UAV multispectral has also provided new opportunities for the study of urban green space.This study based on UAV(Unmanned Aerial Vehicle)Li DAR(Light Detection and Ranging)data,UAV multispectral data,Sentinel-2 data,combined with field survey data,to evaluate the Green Volume in park scale and city scale.In the park scale,the random forest algorithm was used to establish the three-dimensional green volume(TGB)model of trees and shrubs at the single tree level and the leaf area green volume(LAI)model of trees,shrubs and grasses.And the green volume of seven park in Yinchuan City was estimated accurately by those models.In the city scale,TGB and LAI model was established based on the park scale green volume spatial distribution and Sentinel-2 data,and the accuracy of the four algorithms of multiple linear regression,support vector machine,random forest and MLP in the modeling of urban green volume is compared.Finally,the map of green volume in Yinchuan is generated.The main conclusions are as follows:(1)In the single tree parameter extraction,the correct rates of single tree segmentation of trees and shrubs are 89.9%and 92.6%,respectively,and the recall rates are 98.4%and 97.3%by using Li DAR data.Otherwise,based on Li DAR data extraction,the average absolute errors of single tree height of trees and shrubs were2.296 m and 0.544 m,and the average absolute errors of single tree crown area were15.851 m~2 and 1.980 m~2.The results of single tree parameter extraction from Li DAR data were reliable.(2)In park scale,the R~2 of TGB and LAI models based on Random Forest algorithm are both higher than 0.85,and the accuracy of the trees and herbs green volume models are higher than that of shrubs.However,all of the models are underestimate or overestimate.And from the importance of variables,the average height of tree is the best important of the trees green volume models,the OSAVI and LCI are relatively important for shrubs green volume models,the most important parameter in the herbal LAI model is NDRE,and the Red Edge band plays a decisive role in estimating the herbaceous green volume.(3)In the city scale,the Random Forest model have the high accuracy in evaluating the green volume in Yinchuan city,and the MLP model is the second.The R~2 of TGB and LAI model is 0.73 and 0.64,respectively.Otherwise,the green band(560nm)is important in TGB model and the red band(665nm)is important in LAI model.(4)In 2017,the total TGB is 3.42×10~8m~3 and the mean LAI is 0.57 in Yinchuan City;In 2021,the total TGB is 4.95×10~8m~3 and the mean LAI is 0.94 in Yinchuan City.From 2017 to 2021,the amount of green volume in Yinchuan City has increased significantly.The changes of green volume in urban green space is mainly due to the influence of human interference,such as landscaping rectification or urban expansion.This research method can not only simulate the green volume in fine-scale plots simply,quickly and efficiently,which instead of the traditional field survey,but also provides an effective way to estimate urban green volume in large scale accurately and quickly. |
PDF Full Text Request