Spatial Distribution Of Forest Carbon Storage In Maoershan Region Based On GWRK Model

Forest carbon storage have an important impact on global climate change.The limitation of previous models is that it does not account for inherent spatial correlation structure of residual and non-stationary of forest carbon storage.This study is based on ETM+remote sensing image data of Maoershan Experimental Forest Farm of Northeast Forestry University and 193 fixed data of J982 Maoershan Forest Type ? survey,soil data and 30m resolution DEM data for forest carbon stocks in Maoershan area.Model fitting,firstly,through the correlation analysis,the variables were initially screened,and the OLS model was established by using stepwise regression to select the entropy of elevation,vegetation index and gray level co-occurrence matrix,and all of them were significantly correlated with forest carbon storage.The GWR model is established in the least squares model,and the parameters of the GWR model are tested for non-stationarity.The spatial correlation analysis is performed on the residual part of the GWR model by the semivariogram.The form of the semivariogram is determined as Gaussian function.Less than 25%,there is spatial autocorrelation.The common Kriging method is used to interpolate the residuals to construct the GWRK model.At the same time,the model prediction accuracy of the least squares model(OLS)and the geographically weighted regression model(GWR)is compared.The simulation algorithm is used.The spatial distribution of the parameters of the GWR model was was visually analyzed.The residuals of the GWR model and the soil data were used for variance analysis to determine the significant differences in the residuals of the GWR models under different soil types.The results show:(1)The average value of forest carbon storage in the study area is 70.31 t·hm-2.By comparing the accuracy of forest carbon storage estimation in the Maoershan area by comparing the three models,the average absolute error(MAE)and root mean square error of GWRK are known.(RMSE)is lower than the OLS model and the GWR model.The mean error(ME)of the GWRK model is lower than the GWR model,which is similar to the OLS model.The prediction accuracy(Acc)of the GWRK model is 83.2%.which is 6%and 10%higher than the OLS model(73.7%)and the GWR model(77.3%),respectively,and the fitting accuracy is significantly improved.(2)Spatial analysis of forest carbon stocks estimated by GWRK model using the slope direction and altitude distribution of the Maoershan area:in the low-altitude areas where humans live,the carbon storage of forests is correspondingly low due to human disturbance;The gradual increase of forest carbon storage value gradually increases;the forest carbon storage values estimated by different slope GWRK models are different,and the average carbon storage of slope is greater than the value of shady slope.(3)The spatial distribution of forest carbon stocks estimated by the three models is compared locally.Compared with the other two methods,the GWRK model has more varied values in each region,and the transition between high and low values is smoother and does not exist.Obvious plaque phenomenon,the corresponding estimation is more consistent with the topographic transformation.(4)The simulation data of the simulation algorithm is compared with the spatial distribution of the GWR model parameters and the error term.The spatial distribution of the coefficients of the GWR model proves that the GWR model effectively overcomes the influence of spatial heterogeneity on the model.The spatial distribution of the difference shows strong spatial autocorrelation,and the block-to-base ratio of the GWR model residual term is less than 25%,indicating that the GWR model does not consider the influence of the spatial correlation structure inherent in the residual on the prediction of the model.