Spatial Simulation Of Urban Vegetation Carbon Density Based On Optimizing Local Sample Sizes

Urban vegetation is an important part of the urban ecosystem.It can regulate the urban microclimate by absorbing carbon dioxide in the air.Carbon stock is a hot topic in current research.Vegetation carbon density is an important indicator for studying the carbon sequestration capacity and productivity of vegetation,and is essential for regional and global carbon cycle and climate change.Due to the low efficiency and high cost of the traditional manual survey method,the estimation method based on the combination of optical remote sensing image and sample survey data has emerged as a common means to estimate the regional vegetation carbon density.The traditional linear regression model is not universal.Therefore,the development of an accurate vegetation carbon density estimation method is of great significance for the study of regional vegetation carbon density inversion.In this paper,Shenzhen City,Guangdong Province was selected as the study area.The Landsat8 remote sensing images imaged in 2014 were used as the data sources.To measure aboveground carbon density,239 plots were arranged by stratified random sampling to conduct on-the-spot investigation of the carbon density of vegetation in Shenzhen.The Teillet model,C model and SCS+C model were used to topographically correct the image,and the topographic correction method most suitable for the study area was selected to reduce the influence of topographical factors and shadows on carbon density estimation.Taking the measured carbon density as the dependent variable,a total of 281 spectral factors,texture factors and topographic factors of remote sensing images were extracted as independent variables,and correlation analysis was carried out.The factor elimination and variance inflation factor method were used for factor screening to obtain the final use.Based on the independent variables of modeling,logarithmic regression,geographically weighted logarithmic regression,k-NN method and geographically weighted logarithmic regression model based on local sample size optimization method are established to estimate the carbon density of urban vegetation in Shenzhen and obtain the vegetation of Shenzhen.Spatial distribution maps of vegetation carbon density can also be drawn from these models.The main findings are as follows:(1)Topographic correction reduces the effects of topographical factors and shadows.The topographic correction enhances the brightness of the remote sensing image,the shadow of the shady slope is weakened,and the real object condition is displayed,avoiding the phenomenon of "isomorphism".(2)The optimal topographic correction model for the study area was determined.Through the visual inspection,statistical test and scatter plot test,the three topographic correction methods are compared and analyzed.It is found that the topographic correction effect of the SCS+C correction model is better than the Teillet correction model and the C correction model.(3)The modeling factor for vegetation carbon density inversion based on Landsat8 remote sensing image was determined.Combined with the correlation analysis results,the gradual elimination and variance inflation factor method were used for screening.For Landsat8 image,four independent variables were selected from 281 initial variables for the inversion of vegetation carbon density:SR536,B4,TVI,and SR32.Relatively speaking,spectral factors and their derived factors are slightly more correlated with vegetation carbon density than texture factors and topographic factors.(4)The optimal inversion model of vegetation carbon density in Shenzhen based on Landsat8 images was established.In the carbon plant density inversion model selected by the study,from the perspective of model fitting effect and prediction accuracy,the k-NN method has the worst prediction result,the coefficient of determination is the smallest,and the root mean square error is the largest.This is followed by a logarithm regression model with a significantly overestimated and underestimated forecast.Compared with the first two models,the geographic weighted logarithm regression model has higher prediction accuracy,but there are still too high and too low estimates.The geographically weighted logarithm regression model based on the local sample size optimization method can predict the carbon density of urban vegetation in Shenzhen.The prediction coefficient R2 and root mean square error of the prediction model are 0.54 and 13.28(Mg/ha),respectively.It is the optimal inversion model of vegetation carbon density in the study area.(5)Draw a spatial distribution map of vegetation carbon density in Shenzhen.The paper uses 4 vegetation carbon density prediction models to draw a spatial distribution map of vegetation carbon density in Shenzhen,The advantages and disadvantages of the four methods and the rationality of the spatial distribution are analyzed from the perspective of spatial distribution.The spatial distribution of the prediction results of the four models is generally consistent with the regional vegetation distribution characteristics.Among them,the prediction results of the k-NN method are relatively average,but the estimates are too low and too high.The overall prediction results of the logarithm regression model and the geographically weighted logarithm regression model are both low.The prediction results of the geographically weighted logarithm regression model based on the local sample size optimization method are most consistent with the actual distribution.