Estimation Of Forest Biomass In Beijing Using Landsat 8 OLI And ALOS-2 PALSAR-2 Data

Forest biomass reflects the material cycle of forest ecosystem and is an important indicator to measure the changes of forest structure and function.Accurate estimation of forest biomass is the research basis of measuring forest carbon reserves,which is of great significance to better understand carbon cycle and improve the efficiency of forest policy and management activities.Optical remote sensing image has rich spectral and texture features,which can obtain forest horizontal structure information.SAR has advantages in obtaining forest vertical information.Therefore,the combination of optical image and SAR to establish biomass estimation model to give full play to the advantages of the two kinds of data has great development potential in improving the fitting effect and accuracy of regional forest biomass estimation model.Based on the Landsat 8 OLI image and ALOS-2 PALSAR-2 data,combined with the survey data of forest sample plots,the modeling variables of the two remote sensing data were extracted.The forest biomass estimation models of coniferous forest,broad-leaved forest and mixed forest with different variable sets(only Landsat 8 OLI image,only ALOS-2 PALSAR-2 data,combination of the two images)were established by using k nearest neighbor method and random forest method.The accuracy of forest biomass estimation by different modeling methods,different variable combinations and different forest types was compared,and the most suitable estimation model for Beijing area was discussed,and the distribution of forest biomass in Beijing was statistically analyzed.The main research contents and results are as follows:(1)There are three sets of variables for modeling,which are only Landsat 8 OLI image,only ALOS-2 PALSAR-2 data,and variables subsets that combined with two kinds of data.The results show that the model combined with two kinds of variables(R~2 is 0.50-0.66,root mean square error(RMSE)is 13.43-16.42Mg/ha)is better than the model with single variable(R~2 is 0.38-0.53,RMSE is 14.54-19.26Mg/ha).In the same case,the estimation results of model combined with two kinds of variables are higher than that of single variable.(2)The forest biomass estimation models of k nearest neighbor method and random forest method were constructed.The results show that the random forest model can effectively reduce the error of biomass estimation under the same conditions.Compared with KNN model,R~2 of random forest model increased from 0.38-0.51 to 0.50-0.66,RMSE decreased from 14.46-19.26 Mg/ha to 13.43-16.12 Mg/ha,which improved the model accuracy.Therefore,the random forest model is more suitable for the estimation of forest biomass in Beijing.(3)The average biomass density of Beijing is 51.93Mg/ha.Forest biomass is mainly distributed in the Northwest Mountainous areas such as Yanqing,Huairou,Miyun,etc.;the central area where Xicheng,Dongcheng,Chaoyang,Haidian and other six urban areas are located is mainly urban area,with fewer forests and low biomass value.According to the altitude division,more than 75%of the forest biomass is distributed in the area below 800m,and the rest is distributed in the area above 800m.By analyzing the influence of different variable sets and different methods on the estimation of forest biomass,the optimal model of forest biomass estimation in Beijing is obtained.This method is suitable for the estimation of forest biomass in Beijing and provides basis for planning of relevant forestry decision-making departments.